1 REDUCING GREENHOUSE GAS EMISSIONS FROM

2 DEFORESTATION AND DEGRADATION IN

3 DEVELOPING COUNTRIES: A SOURCEBOOK OF
4 METHODS AND PROCEDURES FOR MONITORING,
5 MEASURING AND REPORTING

6 Background and Rationale for the Sourcebook

7 This sourcebook provides a consensus perspective from the global community of earth
8 observation and carbon experts on methodological issues relating to quantifying the green
9 house gas (GHG) impacts of implementing activities to reduce emissions from deforestation
10 and degradation in developing countries (REDD). The UNFCCC negotiations and related
11 country submissions on REDD in 2005-2007 have advocated that methodologies and tools
12 become available for estimating emissions from deforestation with an acceptable level of
13 certainty. Based on the current status of negotiations and UNFCCC approved methodologies,
14 this sourcebook aims to provide additional explanation, clarification, and methodologies to
15 support REDD early actions and readiness mechanisms for building national REDD
16 monitoring systems. It emphasizes the role of satellite remote sensing as an important tool
17 for monitoring changes in forest cover, and provides clarification on applying the IPCC
18 Guidelines for reporting changes in forest carbon stocks at the national level.
19 The sourcebook is the outcome of an ad-hoc REDD working group of “Global Observation of
20 Forest and Land Cover Dynamics” (GOFC-GOLD, www.fao.org/gtos/gofc-gold/), a technical
21 panel of the Global Terrestrial Observing System (GTOS). The working group has been
22 active since the initiation of the UNFCCC REDD process in 2005, has organized REDD expert
23 workshops, and has contributed to related UNFCCC/SBSTA side events and GTOS
24 submissions. GOFC-GOLD provides an independent expert platform for international
25 cooperation and communication to formulate scientific consensus and provide technical
26 input to the discussions and for implementation activities. A number of international experts
27 in remote sensing and carbon measurement and accounting have contributed to the
28 development of this sourcebook.
29 With political discussions and negotiations ongoing, the current document provides the
30 starting point for defining an appropriate monitoring framework considering current
31 technical capabilities to measure gross carbon emission from changes in forest cover by
32 deforestation and degradation on the national level. This sourcebook is a living document
33 and further methods and technical details can be specified and added with evolving political
34 negotiations and decisions. Respective communities are invited to provide comments and
35 feedback to evolve a more detailed and refined technical-guidelines document in the future.
36 Authors

37 Sandra Brown, Winrock International, USA

38 Frederic Achard, European Commission, Joint Research Centre, Institute for Environment
39 and Sustainability, Italy.
40 Barbara Braatz, USA
41 Ivan Csiszar, University of Maryland, USA
42 Sandro Federici, Agenzia per la Protezione dell'Ambiente e per i servizi Tecnici, Italy
43 Ruth De Fries, University of Maryland, USA
44 Giacomo Grassi, European Commission, Joint Research Centre, Institute for Environment
45 and Sustainability, Italy.
46 Nancy Harris, Winrock International, USA
47 Martin Herold, Friedrich Schiller University Jena, Germany
48 Danilo Mollicone, Max-Planck-Institute Jena, Germany
49 Devendra Pandey, Forest Survey of India, India
50 Tim Pearson, Winrock International, USA
51 David Shoch, The Nature Conservancy, USA
52 Carlos Souza Jr., IMAZON, Brazil

53 Publisher

54 GOFC-GOLD Project Office, hosted by Natural Resources Canada, Alberta, Canada

55 Acknowledgments

56 Financial support was provided by The Nature Conservancy to Winrock International to

57 prepare the material on the forest carbon stocks and the methodologies to estimate the
58 carbon emissions as well as to compile and edit the whole report. The European Space
59 Agency, Natural Resources Canada, the National Aeronautics and Space Administration, and
60 the Canadian Space Agency are acknowledged for their support of the GOFC-GOLD REDD
61 working group.
62
 63 Table of Contents

64 Background and Rationale for the Sourcebook ........................................................... 1

65 Authors ............................................................................................................... 2
66 Publisher.............................................................................................................. 2
67 Acknowledgments ................................................................................................. 2
68 Table of Contents .................................................................................................. 3
69 1 PURPOSE AND SCOPE OF THE SOURCEBOOK ............................................................ 5
70 2 ISSUES AND CHALLENGES ..................................................................................... 6
71 2.1 LULUCF in the UNFCCC and Kyoto Protocol .......................................................... 6
72 2.2 Definition of Forests, Deforestation and Degradation ............................................. 7
73 2.3 General Method for Estimating CO2 Emissions ..................................................... 9
74 2.4 Reference Emissions Levels ............................................................................. 11
75 2.5 Roadmap for the Sourcebook........................................................................... 12
76 3 Guidance on Monitoring of Gross Changes in Forest Area .......................................... 13
77 3.1 Scope of chapter............................................................................................ 13
78 3.2 Monitoring of Gross Deforestation .................................................................... 13
79 3.2.1 General recommendation for establishing a historical reference scenario .......... 13
80 3.2.2 Key features ............................................................................................ 13
81 3.2.3 Recommended steps................................................................................. 14
82 3.2.4 Selection and Implementation of a Monitoring Approach ................................ 14
83 3.2.5 National Case Studies ............................................................................... 23
84 3.3 Monitoring of Forest Degradation ..................................................................... 27
85 3.3.1 Direct approach to monitor selective logging ................................................ 28
86 3.3.2 Indirect approach to monitor forest degradation ........................................... 37
87 3.3.3 Systems for mapping active forest fire, burned area and associated emissions .. 41
88 4 ESTIMATION OF CARBON STOCKS ......................................................................... 43
89 4.1 Overview of carbon stocks, and issues related to C stocks ................................... 43
90 4.1.1 Issues related to carbon stocks .................................................................. 43
91 4.1.1.1 The importance of “good” carbon stock estimates....................................... 43
92 4.1.1.2 Fate of carbon pools as a result of deforestation and degradation ................. 44
93 4.1.1.3 The definition of uncertainty for carbon assessments .................................. 45
94 4.1.1.4 The need for stratification and how it relates to remote sensing data ............ 46
95 4.1.2 Overview of Chapter ................................................................................. 46
96 4.2 Which Tier Should be Used? ............................................................................ 47
97 4.2.1 Explanation of IPCC Tiers .......................................................................... 47
98 4.2.2 Data needs for each Tier ........................................................................... 49
99 4.2.3 Selection of Tier ....................................................................................... 50
100 4.3 Stratification by Carbon Stocks ........................................................................ 50
101 4.3.1 Why stratify? ........................................................................................... 50
102 4.3.2 Approaches to stratification ....................................................................... 51
103 4.4 Estimation of Carbon Stocks of Forests Undergoing Change ................................. 55
104 4.4.1 Decisions on which carbon pools to include .................................................. 55
105 4.4.1.1 Key categories ...................................................................................... 55
106 4.4.1.2 Defining carbon measurement pools:........................................................ 57
107 4.4.2 General approaches to estimation of carbon stocks ....................................... 58
108 4.4.2.1 STEP 1: Identify strata where assessment of carbon stocks is necessary- ...... 58
109 4.4.2.2 STEP 2: Assess existing data ................................................................... 59
110 4.4.2.3 STEP 3: Collect missing data ................................................................... 62
111 4.4.3 Guidance on carbon in soils ....................................................................... 64
112 4.4.3.1 Explanation of IPCC Tiers for soil carbon estimates ..................................... 64
113 4.4.3.2 When and how to generate a good Tier 2 analysis for soil carbon.................. 65
114 4.5 Uncertainty ................................................................................................... 70
115 5 Methods for estimating CO2 Emissions from Deforestation and Forest Degradation ....... 72
116 5.1 Scope of this Chapter ..................................................................................... 72
117 5.2 Linkage to 2006 IPCC Guidelines...................................................................... 73
118 5.3 Organization of this Chapter ............................................................................ 74
119 5.4 Fundamental Carbon Estimating Issues ............................................................. 74
120 5.5 Estimation of Emissions from Deforestation ....................................................... 76
121 5.5.1 Disturbance Matrix Documentation.............................................................. 76
122 5.5.2 Changes in Carbon Stocks of Biomass ......................................................... 76
123 5.5.3 Changes in Soil Carbon Stocks ................................................................... 78
124 5.6 Estimation of Emissions from Forest Degradation ............................................... 79
125 5.6.1 Disturbance Matrix Documentation.............................................................. 79
126 5.6.2 Changes in Carbon Stocks ......................................................................... 80
127 5.6.3 Changes in Soil Carbon Stocks ................................................................... 80
128 5.7 Estimation of uncertainties .............................................................................. 80
129 6 Guidance on Reporting ......................................................................................... 83
130 6.1 Issues and challenges in reporting.................................................................... 83
131 6.1.1 The importance of good reporting ............................................................... 83
132 6.1.2 Overview of the Chapter............................................................................ 83
133 6.2 Overview of reporting principles and procedures................................................. 83
134 6.2.1 Current reporting requirements under the UNFCCC ....................................... 83
135 6.2.2 Inventory and reporting principles .............................................................. 84
136 6.2.3 Structure of a GHG inventory ..................................................................... 85
137 6.3 What are the major challenges for developing countries?..................................... 88
138 6.4 The conservativeness approach........................................................................ 89
139 1 PURPOSE AND SCOPE OF THE SOURCEBOOK

140 This sourcebook is designed to be a guide to develop a reference emission and design a
141 system for monitoring and estimating carbon dioxide emissions from deforestation and
142 forest degradation at the national scale based on the requirements for the land use and
143 forest sectors of the UNFCCC.
144 The sourcebook introduces users to: i) the key issues and challenges related to monitoring
145 and estimating carbon emissions from deforestation and degradation ii) the key methods
146 provided in the 2006 IPCC Guidelines for National Greenhouse Gas Inventories for
147 Agriculture, Forestry and Other Land Uses (GL-AFOLU) and the 2003 IPCC Good Practice
148 Guidance for Land Use, Land Use Change and Forestry (GPG-LULUCF) and iii) how these
149 IPCC methods provide the steps needed to estimate emissions from deforestation and
150 degradation.
151 The sourcebook provides transparent methods and procedures that are designed to produce
152 estimates of changes in forest area and carbon stocks from deforestation and degradation,
153 with low uncertainty, in a format that is user-friendly. It is intended to complement the
154 GPG-LULUCF and AFOLU by providing additional explanation, clarification and enhanced
155 methodologies for obtaining and analyzing key data.
156 The sourcebook is not designed as a primer on how to analyze remote sensing data nor how
157 to collect field measurements of forest carbon stocks as it is expected that the users of this
158 sourcebook would have some expertise in either of these areas.
159 The sourcebook was developed considering the following guiding principles:
160 ‰ Relevance: Any monitoring system should provide an appropriate match between
161 known REDD policy requirements and current technical capabilities. Further methods
162 and technical details can be specified and added with evolving political negotiations
163 and decisions.
164 ‰ Comprehensiveness: The system should allow global applicability with
165 implementation at the national level, and with approaches that that have potential
166 for sub-national activities.
167 ‰ Consistency: Efforts have to consider previous related UNFCCC efforts and
168 definitions.
169 ‰ Efficiency: Proposed methods should allow cost-effective and timely implementation,
170 and support early actions.
171 ‰ Robustness: Monitoring should provide appropriate results based on sound scientific
172 underpinnings and international technical consensus among expert groups.
173 ‰ Transparency: The system must open and readily available for third party reviewers
174 and the methodology applied must be replicable.

5
175 2 ISSUES AND CHALLENGES

176 The permanent conversion of forested to non-forested areas in developing countries has had
177 a significant impact on the accumulation of greenhouse gases in the atmosphere1, as has
178 forest degradation caused by high impact logging, over-exploitation for fuelwood, intense
179 grazing that reduces regeneration, wildfires, and forest fragmentation. If the emissions of
180 methane (CH4), nitrous oxide (N2O), and other chemically reactive gases that result from
181 subsequent uses of the land are considered in addition to carbon dioxide (CO2) emissions,
182 annual emissions from land-use change during the 1990s and 2000s accounted for about
183 20-25% of the total anthropogenic emissions of greenhouse gases2.
184 Although tropical deforestation and forest degradation are significant contributors to total
185 anthropogenic greenhouse gas emissions, activities to reduce such emissions are not
186 accepted for generating creditable emissions reductions under the Kyoto Protocol. While the
187 environmental rationale for including such activities is compelling, fundamental issues, such
188 as the development of methodologies that are transparent and reliable and that produce
189 emission estimates that are real, scientifically defensible, and verifiable are needed—this is
190 the focus of this sourcebook.

191 2.1 LULUCF in the UNFCCC and Kyoto Protocol

192 Decisions regarding the framework for REDD remain to be made but it is likely to be based
193 on existing UNFCCC and Kyoto Protocol frameworks for reporting and accounting emissions
194 and removals by Annex I (i.e. industrialized) countries (Table 2.1). Within these
195 frameworks, the Land Use, Land Use Change and Forestry (LULUCF) sector is the only
196 sector where the reporting requirements for the UNFCCC and the Kyoto Protocol are not the
197 same, having different coverage, and therefore reporting guidelines. For the national
198 inventories, estimating and reporting guidelines can be drawn from the Marrakech Accords,
199 1996 IPCC (revised) Guidelines and their 2003 Good Practice Guidance for LULUCF (GPG-
200 LULUCF; Chapter 3). Chapter 4 of the GPG-LULUCF elaborates on methods specific to the
201 Kyoto Protocol inventories. The IPCC has also adopted a more recent set of estimation
202 guidelines that integrate Agriculture and LULUCF to form the Agriculture, Land Use and
203 Forestry (GL-AFOLU) component of the 2006 IPCC Guidelines.

1
Achard, F., H. Eva, H. J. Stibig, P. Mayaux, J. Gallego, T. Richards, and J. P. Malingreau (2002):
Determination of deforestation rates of the world's humid tropical forests. Science 297:999-1002.;
Houghton (2003); Fearnside and Laurance (2004)
2
IPCC (2000), Houghton (2005)

6
204 Table 2.1: Existing frameworks for the Land Use, Land Use Change and Forestry (LULUCF)
205 sector.

Land Use, Land Use Change and Forestry

UNFCCC (2003 GPG and
Kyoto Kyoto-Flexibility
2006 GL-AFOLU)
Six land use classes and Article 3.3 CDM
conversion between them: Afforestation, Afforestation
Forest lands Reforestation, Reforestation
Cropland Deforestation
Grassland Article 3.4
Settlements Cropland management
Wetlands Grazing land management
Other Land Forest management
Revegetation
Deforestation= forest Controlled by the Rules and
converted to another land Modalities (including
category Definitions) of the Marrakesh
Accords

206 2.2 Definition of Forests, Deforestation and Degradation

207 For the new REDD mechanism, many terms, definitions and other elements are not yet clear
208 and no definitions have been agreed on. Although the terms ‘deforestation’ and
209 ‘degradation’ are commonly used they can vary by country. For international negotiations,
210 specific definitions that all Parties can agree to will be needed. These definitions will be
211 agreed through negotiations among the Parties.
212 Decisions for REDD will build on experience and modalities from the UNFCCC national
213 greenhouse gas inventories and Kyoto Protocol inventories for which there are definitions for
214 some terms that are potentially a starting point for considering refined and additional
215 definitions. During the UNFCCC processes there has been agreement on some definitions,
216 and the Marrakesh Accords (MA) prescribes the definitions for the Kyoto Protocol. These
217 definitions could be applicable to REDD but that would be agreed through negotiation in the
218 UNFCCC processes.
219 The definitions as used in UNFCCC and Kyoto protocol and that can be considered for use in
220 REDD are described below.
221 Deforestation - Most definitions characterize deforestation as the long-term or permanent
222 conversion of land from forest use to other non-forest uses. Under Decision 11/CP.7, the
223 UNFCCC defined deforestation as: “..the direct, human-induced conversion of forested land
224 to non-forested land.”
225 Effectively this definition means a reduction in crown cover from above the threshold for
226 forest definition to below this threshold. For example, if a country defines a forest as having
227 a crown cover greater than 30%, then deforestation would not be recorded until the crown
228 cover was reduced below this limit. Yet other countries may define a forest as one with a
229 crown cover of 20% or even 10% and thus deforestation would not be recorded until the
230 crown cover was reduced below these limits.
231 Deforestation causes a change in land cover and in land use. Common changes include:
232 conversion of forests to annual cropland, conversion to perennial plants (oil palm, shrubs),
233 conversion to slash-and-burn (shifting cultivation) lands, and conversion to urban lands or
234 other human infrastructure.

7
235 Degradation – Where there are emissions from forests caused by a decrease in canopy
236 cover that does not qualify as deforestation, it is termed as degradation. Therefore,
237 estimations of degraded areas will be affected by the definition of a “degraded forest”,
238 which is not standardized.
239 The IPCC special report on ‘Definitions and Methodological Options to Inventory Emissions
240 from Direct Human-Induced Degradation of Forests and Devegetation of Other Vegetation
241 Types’ (2003) presents five different potential definitions for degradation along with their
242 pros and cons. The report suggested the following characterization for degradation:
243 “A direct, human-induced, long-term loss (persisting for X years or more) or at least Y% of
244 forest carbon stocks [and forest values] since time T and not qualifying as deforestation”.
245 What thresholds for carbon loss and minimum area affected as well as long term need to be
246 specified to operationalize this definition. In terms of changes in carbon stocks, degradation
247 therefore would represent a measurable, sustained, human-induced decrease in canopy
248 cover, with measured cover remaining above the threshold for definition of forest.
249 Given the lack of a clear definition for degradation makes it difficult to design a monitoring
250 system. However, some general observations and concepts exist and are presented here to
251 inform the debate. Degradation may present a much broader land cover change than
252 deforestation. In reality, monitoring of degradation will be limited by the technical capacity
253 to sense and record the change in canopy cover because small changes will likely not be
254 apparent unless they produce a systematic pattern in the imagery.
255 Many activities cause degradation of carbon stocks in forests but not all of them can be
256 monitored well with high certainty, and not all of them need to be monitored using remote
257 sensing data, though being able to use such data would give more confidence to reported
258 emissions from degradation. To develop a monitoring system for degradation, it is first
259 necessary that the causes of degradation be identified and the likely impact on the carbon
260 stocks be assessed.
261 ‰ Area of forests undergoing selective logging (both legal and illegal) with the presence
262 of gaps, roads, and log decks are likely to be observable in remote sensing imagery,
263 especially the network of roads and log decks. The gaps in the canopy caused by
264 harvesting of trees have been detected in imagery such as Landsat using more
265 sophisticated analytical techniques of frequently collected imagery, and the task is
266 somewhat easier to detect when the logging activity is more intense (i.e. higher
267 number of trees logged; see Section 3.3)). A combination of legal logging followed
268 by illegal activities in the same concession is likely to cause more degradation and
269 more change in canopy characteristics, and an increased chance that this could be
270 monitored with Landsat type imagery and interpretation. The reduction in carbon
271 stocks from selective logging can also be estimated with reasonable certainty without
272 the use satellite imagery using of methods given in the IPCC AFOLU. Legal and illegal
273 selective logging is a common form of change in carbon stocks of forests remaining
274 as forests in many developing countries.
275 ‰ Degradation of carbon stocks by forest fires could be more difficult to monitor with
276 existing satellite imagery and little to no data exist on the changes in carbon stocks.
277 Depending on the severity and extent of fires, the impact on the carbon stocks could
278 vary widely. In practically all cases for tropical forests, the cause of fire will be
279 human induced as there are little to no dry electric storms in tropical humid forest
280 areas.
281 ‰ Degradation by over exploitation for fuel wood or other local uses of wood often
282 followed by animal grazing that prevents regeneration, a situation more common in
283 drier forest areas, is likely not to be detectable from satellite image interpretation
284 unless the rate of degradation was intense causing larger changes in the canopy.

8
285 Forest land – Under the UNFCCC, this category includes all land with woody vegetation
286 consistent with thresholds used to define Forest Land in the national greenhouse gas
287 inventory. It also includes systems with a vegetation structure that does not, but in situ
288 could potentially reach, the threshold values used by a country to define the Forest Land
289 category.
290 The estimation of deforestation is affected by the definitions of ‘forest’ versus ‘non-forest’
291 area that vary widely in terms of tree size, area, and canopy density. Forest definitions are
292 myriad, however, common to most definitions are threshold parameters including minimum
293 area, minimum height and minimum level of crown cover. In its forest resource assessment
294 of 2005, the FAO3 uses a minimum cover of 10%, height of 5m and area of 0.5ha. However,
295 the FAO approach of a single worldwide value excludes variability in ecological conditions
296 and differing perceptions of forests.
297 For the purpose of the Kyoto Protocol4, it was determined through the Marrakech Accords
298 that Parties should select a single value of crown area, tree height and area to define forests
299 within their national boundaries. Selection must be from within the following ranges, with
300 the understanding that young stands that have not yet reached the necessary cover or
301 height are included as forest:
302 ‰ Minimum forest area: 0.05 to 1 ha
303 ‰ Potential to reach a minimum height at maturity in situ of 2-5 m
304 ‰ Tree crown cover (or equivalent stocking level): 10 to 30 %
305 Under this definition a forest can contain anything from 10% to 100% tree cover; it is only
306 when cover falls below the minimum crown cover as designated by a given country that
307 land is classified as non-forest. However, if this is only a temporary change, such as for
308 timber harvest with regeneration expected, the land remains in the forest classification.
309 The definition of forests offers some flexibility for countries when designing a monitoring
310 plan because analysis of remote sensing data can adapt to different minimum tree crown
311 cover thresholds. However, consistency in forest classifications for all REDD activities is
312 critical for integrating different types of information including remote sensing analysis. Using
313 different definitions impact the technical earth observation requirements and could influence
314 cost, availability of data, and abilities to integrate and compare data through time. The
315 specific definition chosen will have implications on where the boundaries between
316 deforestation and degradation occur.

317 2.3 General Method for Estimating CO2 Emissions

318 To facilitate the use of the IPCC GL-AFOLU and GPG reports side by side with the
319 sourcebook, definitions used in the sourcebook remain consistent with the IPCC Guidelines.
320 In this section we summarize key guidance and definitions from the IPCC Guidelines that
321 frame the more detailed procedures that follow.
322 The term “Categories” as used in IPCC reports refers to specific sources of
323 emissions/removals of greenhouse gases. For the purposes of this sourcebook, the following
324 categories are considered under the AFOLU sector:

3
FAO – Food and Agriculture Organization (2006): Global Forest Resources Assessment 2005. Main
Report, www.fao.org/forestry/fra2005
4
UNFCCC (2001): Seventh conference of parties: The Marrakech accords. (Bonn, Germany: UNFCCC
Secretariat) available at http://www.unfccc.int

9
325 ‰ Forest Land Converted to Crop Land, Forest Land Converted to Grass Land, Forest
326 Land Converted to Settlements, Forest Land Converted to Wetlands, and Forest Land
327 Converted to Other Land are commonly equated to deforestation.
328 ‰ A decrease in carbon stocks of Forest Land Remaining Forest Land is commonly
329 equated to forest degradation.
330 The IPCC Guidelines refer to two basic inputs with which to calculate greenhouse gas
331 inventories: activity data and emissions factors. “Activity data” refer to the extent of an
332 emission/removal category, and in the case of deforestation and forest degradation refers to
333 the areal extent of those categories, presented in hectares. Henceforth for the purposes of
334 this sourcebook, activity data are referred to as area change data. “Emission factors” refer
335 to emissions/removals of greenhouse gases per unit activity, e.g. tons carbon dioxide
336 emitted per hectare of deforestation. Emissions/removals resulting from land-use
337 conversion are manifested in changes in ecosystem carbon stocks, and for consistency with
338 the IPCC Guidelines, we use units of carbon, specifically metric tons of carbon per hectare (t
339 C ha-1), to express emission factors for deforestation and forest degradation.
340 The AFOLU guidelines define a methodology for assessing the activity data or the change in
341 area of different land categories. The guidelines describe three different approaches for the
342 area change component (Table 2.2): Approach 1 identifies the total net area change for
343 each land category, but does not provide information on the nature and area of conversions
344 between land uses; Approach 2 involves tracking of land conversions between categories.
345 Both approaches 1 and 2 provide “net” area changes. Approach 3 extends Approach 2 by
346 using spatially explicit land conversion information; thus allowing for an estimation of both
347 “gross” and “net” changes. Because the global interest is on reducing emissions from
348 deforestation, Approach 3 that gives gross deforestation is the only practical approach that
349 can be used for REDD implementation. Furthermore, the estimated area change data will
350 need to be highly accurate and precise so that the overall estimate of the emissions
351 reductions has high certainty.
352 Table 2.2: A summary of which Approach can be used for the activity data and which Tiers
353 for the emission factors for estimating gross emissions of CO2 from deforestation and
354 degradation is shown in the shaded boxes.

Approach for activity data: Area Tiers for emission factors:

change Change in C stocks
1. Non-spatial country statistics (e.g.
FAO) – generally gives net change in 1. IPCC defaults
forest area

2. Based on maps, surveys, and other 2. Country specific data for key
national statistical data factors
3.Spatially specific data from 3.National inventory of key C stocks,
interpretation of remote sensing data repeated measurements of key
stocks through time or modeling
355

356 The emission factors are derived from assessments of the changes in carbon stocks in the
357 various carbon pools of a forest. Carbon stock information can be obtained at different Tier
358 levels (Table 2.2): Tier 1 uses IPCC default values (i.e. biomass in different forest biomes,
359 carbon fraction etc.); Tier 2 requires some country-specific carbon data (i.e. from field
360 inventories, permanent plots), and Tier 3 national inventory-type data of carbon stocks in
361 different pools and assessment of any change in pools through repeated measurements or
362 modeling. Moving from Tier 1 to Tier 3 increases the accuracy and precision of the
363 estimates, but also increases the complexity and the costs of monitoring.

10
364 Chapter 3 of this sourcebook provides guidance on how to obtain the activity data,
365 or gross change in forest cover, with low uncertainty. Chapter 4 focuses on
366 obtaining data for emission factors and providing guidance on how to produce
367 estimates of carbon stocks of forests with low uncertainty suitable for national
368 assessments.

369 According to the IPCC, carbon stocks of the key or significant categories and pools should
370 be estimated with the higher tiers (see also chapter 4.2.3). As the reported estimates of
371 reduced emissions will likely be the basis of an accounting procedure (as in the Kyoto
372 Protocol), with the eventual assignment of economic incentives, Tier 3 should be the level to
373 aim for. In the context of REDD, however, the methodological choice will inevitably results
374 from a balance between the requirements of accuracy/precision and the cost of monitoring.
375 It is likely that this balance will be guided by the principle of conservativeness, i.e. a tier
376 lower than required could be used – or a carbon pool could be ignored - if it can be
377 demonstrated that the overall estimate of reduced emissions are underestimated (see also
378 chapter 6.4). Thus estimates of the forest carbon stocks are needed that are conservative
379 (very low probability to be overestimated) or of low uncertainty.

380 2.4 Reference Emissions Levels

381 The estimate of reductions in emissions from deforestation and degradation requires
382 assessing reference emissions levels against which future emissions can be compared.
383 These reference levels should represent the historical emissions from deforestation and
384 forest degradation in “forested land” at national level.
385 Credible reference levels of emissions can be established for a REDD system using existing
386 scientific and technical tools, and this is the focus of this sourcebook.
387 Technically, from remote sensing imagery it is possible to monitor change with confidence
388 from 1990s onwards and estimates of forest C stocks can be obtained. Feasibility and
389 accuracies will strongly depend from national circumstances (in particular in relation to data
390 availability) i.e. potential limitations are more related to resources and data availability than
391 to methodologies.
392 A related issue is the concept of a benchmark forest area map. Any national program to
393 reduce emissions from deforestation and degradation will need to have an initial forest area
394 map to represent the point from which each future forest area assessment will be made and
395 actual changes will be monitored so as to report only gross deforestation going forward.
396 This initial forest area map is referred to here as a benchmark map. This implies that an
397 agreement will be needed by Parties on deciding on a benchmark year against which all
398 future deforestation and degradation will be measured. The use of a benchmark map will
399 clearly show where gross deforestation is occurring, and clearly show where non-forest land
400 is reverting to forests if at some stage in the future this information becomes relevant.
401 The use of a benchmark map also makes monitoring deforestation (and some degradation)
402 a simpler task. The interpretation of the remote sensing imagery needs to identify only the
403 areas (or pixels) that changed compared to the benchmark map. The benchmark map would
404 then be updated at the start of each new analysis event so that one is just monitoring the
405 loss of forest area from the original benchmark map. The forest area benchmark map would
406 show where forests exist and how they are stratified either for carbon or for other national
407 needs.

11
408 2.5 Roadmap for the Sourcebook

409 The sourcebook is organized as follows:

Estimation of area change

Chapter 3

Estimation of carbon stocks

Chapter 4 Chapter 5
Estimation of CO2
emissions

Chapter 6
Guidance on
reporting of CO2
emissions

410

12
411 3 GUIDANCE ON MONITORING OF GROSS CHANGES IN
412 FOREST AREA

413 3.1 Scope of chapter

414 This chapter presents the state of the art for data and approaches to be used for
415 monitoring forest area changes at the national scale in tropical countries using
416 remote sensing imagery. It includes approaches and data for monitoring both
417 deforestation and forest degradation and for establishing historical reference
418 scenarios.

419 The chapter presents the minimum requirements to develop first order national
420 deforestation databases, using typical and internationally accepted methods. There are
421 more advanced and costly approaches that may lead to more accurate results and would
422 meet the reporting requirements, but they are not presented here.

423 3.2 Monitoring of Gross Deforestation

424 3.2.1 General recommendation for establishing a historical reference scenario

425 As minimum requirement, it is recommended to use Landsat-type remote sensing data (30
426 m resolution) for years 1990, 2000 and 2005 for monitoring forest cover change with 1 to 5
427 ha Minimum Mapping Unit (MMU). It might be necessary to use data from a year prior or
428 after 1990, 2000, and 2005 due to availability and cloud contamination. These data will
429 allow assessing gross deforestation (i.e. to derive area deforested for the period considered)
430 and, if desired, producing a map of national forest area (to derive deforestation rates) using
431 a common forest definition. A hybrid approach combining automated digital segmentation
432 and/or classification techniques with visual interpretation and/or validation of the resulting
433 classes/polygons should be preferred as simple, robust and cost effective method.
434 There maybe different spatial units for the detection of forest and of forest change. Remote
435 sensing data analyses become more difficult and more expensive with smaller Minimum
436 Mapping Units (MMU) i.e. more detailed MMU’s increase mapping efforts and usually
437 decrease change mapping accuracy. There are several MMU examples from current national
438 and regional remote sensing monitoring systems Brazil PRODES (6,25 ha initially, now 1 ha
439 for digital processing), India national forest monitoring (1 ha), EU-wide CORINE land
440 cover/land use change monitoring (5 ha), ‘GMES Service Element’ Forest Monitoring (0.5
441 ha), and Conservation International national case studies (2 ha).

442 3.2.2 Key features

443 The only free global mid-resolution (30m) remote sensing imagery are from NASA (Landsat
444 satellites) for around years 1990, 2000, and 2005 (the mid-decadal dataset 2005/2006 is
445 under preparation) with some quality issues in some parts of the tropics (clouds,
446 seasonality, etc).
447 The period 2000-2005 is more representative of recent historical changes and potentially
448 more suitable due to the availability of complimentary data during a recent time frame.

13
449 Specifications on minimum requirements for image interpretation are:
450 ‰ Geo-location accuracy < 1 pixel, i.e. < 30m,
451 ‰ Minimum mapping unit should be between 1 and 5 ha,
452 ‰ A consistency assessment should be carried out.

453 3.2.3 Recommended steps

454 The following steps are needed for a national assessment that is scientifically credible and
455 can be technically accomplished by in-country experts:
456 1. Selection of the approach:
457 a. Assessment of national circumstances with in particular existing
458 definitions and data sources
459 b. Definition of change assessment approach by deciding on:
460 i. Satellite imagery
461 ii. Sampling versus wall to wall coverage
462 iii. Fully visual versus semi-automated interpretation
463 iv. Accuracy or consistency assessment
464 c. Plan and budget monitoring exercise including:
465 i. Hard and Software resources
466 ii. Requested Training
467 2. Implementation of the monitoring system:
468 a. Selection of the forest definition
469 b. Designation of initial forest area for acquiring satellite data (benchmark
470 map)
471 c. Selection and acquisition of the satellite data
472 d. Analysis of the satellite data (preprocessing and interpretation)
473 e. Assessment of the accuracy

474 3.2.4 Selection and Implementation of a Monitoring Approach

475 Step 1: Selection of the forest definition

476 Currently Annex I Parties use the UNFCCC framework definition of forest and deforestation
477 adopted for implementation of Article 3.3 and 3.4 (see section 2.2) and, without other
478 agreed definition, this definition is considered here as the working definition. Sub-categories
479 of forests (e.g. forest types) can be defined within the framework definition of forest.
480 Remote sensing imagery allows land cover information only to be obtained. Local expert or
481 field information is needed to derive land use estimates.

482 Step 2: Designation of initial forest area for acquiring satellite data
483 Many types of land cover exist within national boundaries. REDD monitoring needs to cover
484 all forest area and the same area needs to be monitored for each reporting period. It is not
485 necessary or practical in many cases to monitor the entire national extent that includes non-
486 forest land cover types. Therefore, a forest mask needs to be designated initially to identify
487 the area to be monitored for each reporting period (referred to in Section 2.2 as the
488 benchmark map).
489 Ideally, an initial wall-to-wall assessment of the entire national extent would be carried out
490 to identify forested area according to UNFCCC forest definitions at the beginning of the
491 reference period (e.g. to be decided by the Parties to the UNFCCC). This approach may not

14
492 be practical for large countries. Existing forest maps at appropriate spatial resolution and for
493 a relatively recent time could be used to identify the initial forest extent.
494

495 Important principles in identifying the initial forest extent are:

496
‰ The area should include all forest within the national reference boundaries
497
‰ A consistent forest extent should be used for monitoring for future reporting
498

499 Step 3: Selection of satellite imagery and coverage

500 Fundamental requirements of national monitoring systems are that they measure changes
501 throughout all forested area, use consistent methodologies at repeated intervals to obtain
502 accurate results, and verify results with ground-based or very high resolution observations.
503 The only practical approach for such monitoring systems is through interpretation of
504 remotely sensed data supported by ground-based observations. Remote sensing includes
505 data acquired by sensors on board aircraft and space-based platforms. Multiple methods are
506 appropriate and reliable for forest cover monitoring at national scales.
507 Many data from optical sensors at a variety of resolutions and costs are available for
508 monitoring deforestation (Table 3.1).
509 Table 3.1: Utility of optical sensors at multiple resolutions for deforestation monitoring
Examples of Minimum
Sensor &
current mapping unit Cost Utility for monitoring
resolution
sensors (change)
SPOT-VGT ~ 100 ha Consistent pan-tropical
(1998- ) annual monitoring to
Coarse
Terra-MODIS ~ 10-20 ha identify large clearings and
(250-1000 Low or free
(2000- ) locate “hotspots” for
m)
Envisat-MERIS further analysis with mid
(2004 - ) resolution
Landsat TM or <$0.001/km²
ETM+, for historical
Primary tool to map
Medium SPOT HRV data
0.5 - 5 ha deforestation and estimate
(10-60 m) IRS AWiFs or $0.02/km²
area change
LISS III to $0.5/km2 for
CBERS HRCCD recent data
IKONOS High to very Validation of results from
Fine
QuickBird < 0.1 ha high coarser resolution analysis,
(<5 m)
Aerial photos $2 -30 /km² and training of algorithms

510 Availability of medium resolution data

511 The USA National Aeronautics and Space Administration (NASA) launched a satellite with a
512 mid-resolution sensor that was able to collect land information at a landscape scale. ERTS-1
513 was launched on July 23, 1972. This satellite, renamed ‘Landsat’, was the first in a series
514 (seven to date) of Earth-observing satellites that have permitted continuous coverage since
515 1972. Subsequent satellites have been launched every 2-3 years. Still in operation Landsat
516 5 and 7 cover the same ground track repeatedly every 16 days.

15
517 Almost complete global coverages from these Landsat satellites are available at low or no
518 cost for early 1990s and early 2000s from NASA5, the USGS6, or from the University of
519 Maryland's Global Land Cover Facility7. These data serve a key role in establishing historical
520 deforestation rates, though in some parts of the humid tropics (e.g. Central Africa)
521 persistent cloudiness is a major limitation to using these data. Until year 2003, Landsat,
522 given its low cost and unrestricted license use, has been the workhorse source for mid-
523 resolution (10-50 m) data analysis.
524 On April 2003, the Landsat 7 ETM+ scan line corrector failed resulting in data gaps outside
525 of the central portion of acquired images, seriously compromising data quality for land cover
526 monitoring. Given this failure, users would need to explore how the ensuing data gap might
527 be filled at a reasonable cost with alternative sources of data in order to meet the needs for
528 operational decision-making.
529 Alternative sources of data include Landsat-5, ASTER, SPOT, IRS, CBERS or DMC data
530 (Table 3.2). NASA, in collaboration with USGS, initiated an effort to acquire and compose
531 appropriate imagery to generate a mid-decadal (around years 2005/2006) data set from
532 such alternative sources. The combined Archived Coverage in EROS Archive of the Landsat 5
533 TM and Landsat-7 ETM+ reprocessed-fill product for the years 2005/2006 covers more than
534 90% of the land area of the Earth. These data will be processed to a new orthorectifed
535 standard using data from NASA’s Shuttle Radar Topography Mission.
536 During the selection of the scenes to use in any assessment, seasonality of climate has to
537 be considered: in situations where seasonal forest types (i.e. a distinct dry season where
538 trees may drop their leaves) exist more than one scene should be used. Inter-annual
539 variability has to be considered based on climatic variability.
540 Optical mid-resolution data have been the primary tool for deforestation monitoring. Other,
541 newer, types of sensors, e.g. Radar (ERS1/2 SAR, JERS-1, ENVISAT-ASAR and ALOS
542 PALSAR) and Lidar, are potentially useful and appropriate. Radar, in particular, alleviates
543 the substantial limitations of optical data in persistently cloudy parts of the tropics. Data
544 from Lidar and Radar have been demonstrated to be useful in project studies, but so far,
545 they are not widely used operationally for tropical deforestation monitoring over large areas.
546 Over the next five years or so, the utility of radar may be enhanced depending on data
547 acquisition, access and scientific developments.
548 In summary, Landsat-type data around years 1990, 2000 and 2005 will most suitable to
549 assess historical rates and patterns of deforestation.

5
https://zulu.ssc.nasa.gov/mrsid
6
http://edc.usgs.gov/products/satellite/landsat_ortho.html
7
http://glcfapp.umiacs.umd.edu/

16
550 Table 3.2: Present availability of optical mid-resolution (10-60 m) sensors
Satellite & Resolution Cost
Nation Feature
sensor & coverage (archive8)
Images every 16 days
to any satellite receiving
Landsat-5 30 m 600 US$/scene
USA station. Operating
TM 180×180 km² 0.02 US$/km2
beyond expected
lifetime.
On April 2003 the
failure of the scan line
corrector resulted in
Landsat-7 30 m 600 US$/scene data gaps outside of the
USA
ETM+ 60×180 km² 0.06 US$/km2 central portion of
images, seriously
compromising data
quality
Data is acquired on
15 m 60 US$/scene request and is not
USA/ Japan Terra ASTER
60×60 km² 0.02 US$/km² routinely collected for
all areas
Experimental craft
IRS-P2 LISS- shows promise,
India 23.5 & 56 m
III & AWIFS although images are
hard to acquire
Experimental; Brazil
CBERS-2 uses on-demand images
China/ Brazil 20 m Free in Brazil
HRCCD to bolster their
coverage.
Algeria/ China/
32 m 3000 €/scene Commercial; Brazil uses
Nigeria/ DMC
160×660 km² 0.03 €/km² alongside Landsat data
Turkey/ UK
Commercial Indonesia &
SPOT-5 5-20 m 2000 €/scene
France Thailand used alongside
HRVIR 60×60 km² 0.5 €/km²
Landsat data
551

552 Utility of coarse resolution data

553 Coarse resolution (250 m – 1km) data are available from 1998 (SPOT-VGT) or 2000
554 (MODIS). Although the spatial resolution is coarser than Landsat-type sensors, the temporal
555 resolution is daily, providing the best possibility for cloud-free observations. The higher
556 temporal resolution increases the likelihood of cloud-free images and can augment data
557 sources where persistent cloud cover is problematic. Coarse resolution data also has cost
558 advantages, offers complete spatial coverage, and reduces the amount of data that needs to
559 be processed.
560 Coarse resolution data cannot be used directly to estimate area of forest change. However,
561 these data are useful for identifying locations of rapid change for further analysis with
562 higher resolution data or as an alert system for controlling deforestation (see section on
563 Brazilian national case study below). For example, MODIS data are used as a stratification

8
Some acquisitions can be programmed (e.g., DMC, SPOT). The cost of programmed data is generally
at least twice the cost of archived data.

17
564 tool in combination with medium spatial resolution Landsat data to estimate forest area
565 cleared. The targeted sampling of change reduces the overall resources typically required in
566 assessing change over large nations. In cases where clearings are large and/or change is
567 rapid, visual interpretation can be used to identify where change in forest cover has
568 occurred. Automated methods such as mixture modeling and regression trees (Box 3.1) can
569 also identify changes in tree cover at the sub-pixel level. Validation of analyses with medium
570 and high resolution data in selected locations can be used to assess accuracy. The use of
571 coarse resolution data to identify deforestation hotspots is particularly useful to design a
572 sampling strategy (see following section).

573 Box 3.1: Mixture models and regression trees

574 Mixture models estimate the proportion of different land cover components within a
575 pixel. For example, each pixel is described as percentage vegetation, shade, and bare
576 soil components. Components sum to 100%. Image processing software packages
577 often provide mixture models using user-specified values for each end-member
578 (spectral values for pixels that contain 100% of each component). Regression trees
579 are another method to estimate proportions within each component based on training
580 data to calibrate the algorithm. Training data with proportions of each component can
581 be derived from higher resolution data. (see Box 3.5 for more details)

582 Utility of fine resolution data

583 Fine resolution (< 5m) data, such as those collected from commercial sensors (e.g.,
584 IKONOS, QuickBird) and aircraft, can be prohibitively expensive to cover large areas.
585 However, these data can be used to calibrate algorithms for analyzing medium and high
586 resolution data and to verify the results — that are they can be used as a tool for “ground-
587 truthing” the interpretation of satellite imagery or for assessing the accuracy.

588 Step 4: Decisions for sampling versus wall to wall coverage

589 Wall-to-wall (an analysis that covers the full spatial extent of the forested areas) and
590 sampling approaches within the forest mask are both suitable methods for analyzing forest
591 area change.
592 The main criteria for the selection of wall-to-wall or sampling are:
593 Wall-to-wall is a common approach if appropriate for national circumstances
594 ‰ If resources are not sufficient to complete wall-to wall coverage, sampling is more
595 efficient, in particular for large countries
596 ‰ Recommended sampling approaches are systematic sampling and stratified sampling
597 (see box 3.2).
598 ‰ A sampling approach in one reporting period could be extended to wall-to-wall
599 coverage in the subsequent period.

18
600 Box 3.2: Systematic and stratified sampling
601 Systematic sampling obtains samples on a regular interval, e.g. one every 10 km.
602 Sampling efficiency can be improved through spatial stratification (‘stratified
603 sampling’) using known proxy variables (e.g. deforestation hot spots). Proxy variables
604 can be derived from coarse resolution satellite data or by combining other geo-
605 referenced or map information such as distance to roads or settlements, previous
606 deforestation, or factors such as fires.
607 Example of systematic sampling Example of stratified sampling

608

609 A stratified sampling approach for forest cover change estimation is currently being
610 implemented within the NASA Land Cover and Land Use Change program. This method
611 relies on wall to wall MODIS change indicator maps (at 500 m resolution) to stratify
612 biomes into regions of varying change likelihood. A stratified sample of Landsat-7
613 ETM+ image pairs is analyzed to quantify biome-wide area of forest clearing. Change
614 estimates can be derived at country level by adapting the sample to the country
615 territory.

616

617 A few very large countries, e.g. Brazil and India, have already demonstrated that
618 operational wall to wall systems can be established based on mid-resolution satellite
619 imagery (see section 3.2.5 for details). Brazil has measured deforestation rates in Brazilian
620 Amazonia since the 1980s. These methods could be easily adapted to cope with smaller
621 country sizes. Although a wall-to-wall coverage is ideal, it may not be practical due to large
622 areas and constraints on resources for accurate analysis.

623 Step 5: Process and analyze the satellite data

624 Step 5.1: Preprocessing
625 Satellite imagery usually goes through three main pre-processing steps: geometric
626 corrections are needed to ensure that images in a time series overlay properly, cloud
627 removal is usually the second step in image pre-processing and radiometric corrections are
628 recommended to make change interpretation easier (by ensuring that images have the
629 same spectral values for the same objects).
630 ‰ Geometric corrections
631 o Low geolocation error of change datasets is to be ensured: relative that is
632 between 2 images – average geolocation error should be < 1 pixel

19
633 o Existing Landsat Geocover data usually provide sufficient geometric accuracy
634 and can be used as a baseline; for limited areas Landsat Geocover has
635 geolocation problems
636 o Using additional data like non-Geocover Landsat, SPOT, etc. there is need to
637 put effort in manual georectification using image to image registration or
638 ground control points (REF)
639 ‰ Cloud and cloud shadow detection and removal
640 o Visual interpretation is the preferred method for areas without complete
641 cloud-free satellite coverage,
642 o Clouds and cloud shadows to be removed for automated approaches
643 ‰ Radiometric corrections
644 o Effort for radiometric corrections depend on the change assessment approach
645 o For simple scene by scene analysis (e.g. visual interpretation), the
646 radiometric effects of topography and atmosphere should be considered in the
647 interpretation process but do not need to digitally normalized
648 o Sophisticated digital and automated approaches may require radiometric
649 correction to calibrate spectral values to the same reference objects in
650 multitemporal datasets. This is usually done by identifying a water body or
651 dark object and calibrating the other images to the first.
652 o Reduction of haze maybe a useful complementary option for digital
653 approaches
654 Step 5.2: Analysis methods
655 Many methods exist to interpret images (Table 3.3). The selection of the method depends
656 on available resources and whether image processing software is available. Whichever
657 method is selected, the results should be repeatable by different analysts.
658 Visual scene to scene interpretation of forest cover change can be simple and robust,
659 although it is a time-consuming method. A combination of automated methods
660 (segmentation or classification) and visual interpretation can reduce the work load.
661 Automated methods are generally preferable where possible because the interpretation is
662 repeatable and efficient. Even in a fully automated process, visual inspection of the result by
663 an analyst familiar with the region should be carried out to ensure appropriate
664 interpretation.
665 A preliminary visual screening of the image pairs can serve to identify the sample sites
666 where change has occurred between the two dates. This data stratification allows removing
667 the image pairs without change from the processing chain (for the detection and
668 measurement of change).
669 Changes (for each image pair) can then be measured by comparing the two multi-date final
670 forest maps. The timing of image pairs has to be adjusted to the reference period, e.g. if
671 selected images are dated 1999 and 2006, it would have to be adjusted to 2000-2005.

672 Visual delineation of land cover entities:

673 This approach is viable, particularly if image analysis tools and experiences are limited. The
674 visual delineation of land cover entities on printouts (used in former times) is not
675 recommended. On screen delineation should be preferred as producing directly digital
676 results. When land cover entities are delineated visually they should also be labeled visually.

20
677 Table 3.3: Main analysis methods for moderate resolution (~ 30 m) imagery
Practical
Method for Method for minimum Advantages /
Principles for use
delineation class labeling mapping limitations
unit
- multiple date preferable - closest to classical
Dot to single date forestry inventories
Visual
interpretation < 0.1 ha interpretation - very accurate although
interpretation
(dots sample) - On screen preferable to interpreter dependent
printouts interpretation - no map of changes
- multiple date analysis
Visual preferable - easy to implement
Visual
delineation 5 – 10 ha - On screen digitizing - time consuming
interpretation
(full image) preferable to delineation - interpreter dependent
on printouts
- selection of common
Supervised
spectral training set from
labeling (with
Pixel based <1 ha multiple dates / images - difficult to implement
training and
classification preferable - training phase needed
correction
- filtering needed to avoid
phases)
noise
- interdependent (multiple - difficult to implement
Unsupervised
<1 ha date) labeling preferable - noisy effect without
clustering +
- filtering needed to avoid filtering
Visual labeling
noise
- multiple date
Supervised
segmentation preferable
labeling (with - more reproducible than
Object based - selection of common
training and 1 - 5 ha visual delineation
segmentation spectral training set from
correction - training phase needed
multiple dates / images
phases)
preferable
- multiple date
Unsupervised segmentation preferable - more reproducible than
clustering + 1 - 5 ha - interdependent (multiple visual delineation
Visual labeling date) labeling of single
date images preferable

678 Multi-date image segmentation:

679 Segmentation for delineating image objects reduces the processing time of image analysis.
680 The delineation provided by this approach is not only more rapid and automatic but also
681 finer than what could be achieved using a manual approach. It is repeatable and therefore
682 more objective than a visual delineation by an analyst. Using multi-date segmentations
683 rather than a pair of individual segmentations is justified by the final objective which is to
684 determine change.
685 If a segmentation approach is used, the image processing can be ideally decomposed into
686 three steps:
687 1. Multi-date image segmentation is applied on image pairs: groups of adjacent
688 pixels that show similar land cover change trajectories between the 2 dates are
689 delineated into objects.
690 2. Objects from every extract (i.e. every date) are classified separately by
691 supervised clustering procedures, leading to two automated forest maps (at date
692 1 and date 2)
693 3. Visual interpretation is conducted interdependently on the image pairs to
694 verify/adjust the label the classes and edit possible classification errors.

21
 Image segmentation is the process of partitioning an image into groups of pixels that
are spectrally similar and spatially adjacent. Boundaries of pixel groups delineate ground
objects in much the same way a human analyst would do based on its shape, tone and
texture. However, delineation is more accurate and objective since it is carried out at the
pixel level based on quantitative values

695 Digital classification techniques:

696 Digital classification applies in the case of automatic delineation.
697 After segmentation, it is recommended to apply two supervised object classifications
698 separately on the two multi-date images instead of applying a single unsupervised object
699 classification on the image pair because two separate land cover classifications are much
700 easier to produce in an unsupervised step than a direct classification of change trajectories.
701 The unsupervised object classification should ideally use a common predefined standard
702 training data set of spectral signatures for each type of ecosystem to create initial
703 automated forest maps (at any date and any location within this ecosystem).

704 General recommendations for image object interpretation methods:

705 Given the heterogeneity of the forest spectral signatures and the occasionally poor
706 radiometric conditions, the image analysis by a skilled interpreter is indispensable to map
707 land cover and land cover change with high accuracy.
708 ‰ Interpretation should focus on change with interdependent assessment of 2 multi-
709 temporal images together.
710 ‰ Existing maps may be useful for stratification or helping in the interpretation
711 ‰ Scene by scene (i.e. site by site) interpretation is more accurate than interpretation
712 of scene or image mosaics
713 ‰ Spectral, spatial and temporal (seasonality) characteristics of the forests have to be
714 considered during the interpretation. In the case of seasonal forests, scenes from the
715 same time of year should be used. Preferably, multiple scenes from different seasons
716 would be used to ensure that changes in forest cover from inter-annual variability in
717 climate are not confused with deforestation.

718 Step 6: Accuracy assessment

719 An independent accuracy assessment is an essential component to link area estimates to
720 crediting system. Reporting accuracy and verification of results are essential components of
721 a monitoring system. Accuracy could be quantified following recommendations of chapter 5
722 of IPCC Good Practice Guidance 2003.
723 Accuracies of 80 to 95% are achievable for monitoring with mid-resolution imagery to
724 discriminate between forest and non-forest. Accuracies can be assessed through in-situ
725 observations or analysis of very high resolution aircraft or satellite data. In both cases, a
726 statistically valid sampling procedure can be used to determine accuracy. Sampling should
727 stratify according to forest type and size of clearings. In the case of in situ observations,
728 practical considerations of accessibility need to be taken into account. While it is difficult to
729 verify change from one time to another on the ground unless the same location is visited at
730 two different time periods, a time series of high (to very high) resolution data can be used
731 to assess accuracy of identifying deforestation. If two time periods are not available,
732 accuracy can be assessed through validation of forest cover at the second date. Both
733 omission (actual deforestation that was not detected) and commission (false detection of
734 deforestation) should be reported.

22
735 Accuracy assessment should be carried out in combination with the national-level analysis.
736 Verification of the monitoring and accuracy assessment by a third party may also be
737 necessary for a crediting system. Because different methods are applicable in different
738 countries, verification of the monitoring by a third party would include review of the
739 appropriateness of the method for the particular forest conditions and deforestation
740 patterns, consistency in the application of the method, adherence to data management
741 standards, and methods for assessing accuracy of the result.
742 For Historical reference periods:
743 ‰ Accuracy assessments are very challenging because reference data of higher quality
744 are usually missing.
745 ‰ If no thorough accuracy assessment is possible, it is recommended to apply the best
746 suitable mapping method in a transparent manner for verification purposes.
747 ‰ A minimum requirement should be to apply a consistency assessment, i.e. the
748 reinterpretation of a sample of the original data in an independent manner (by
749 external experts).
750 ‰ The reference period 1990-2000 is more challenging than 2000-2005 as there are
751 more reference data for more recent periods.
752 For future period:
753 ‰ For future periods, a full accuracy assessment should be planned from the start and
754 included in the cost and time budgets.
755 ‰ It should be based on higher resolution or in-situ data.
756 ‰ More precise guidelines for area change accuracy assessment from scientific
757 community will evolve over time.

758 3.2.5 National Case Studies

759 A. Brazil – annual wall to wall approach

760 The Brazilian National Space Agency (INPE) produces annual estimates of deforestation in
761 the legal Amazon from a comprehensive annual national monitoring program called
762 PRODES.
763 The Brazilian Amazon covers an area of approximately 5 million km2, large enough to cover
764 all of Western Europe. Around 4 million km2 of the Brazilian Amazon is covered by forests.
765 The Government of Brazil decided to generate periodic estimates of the extent and rate of
766 gross deforestation in the Amazon, “a task which could never be conducted without the use
767 of space technology”.
768 The first complete assessment by INPE was undertaken in 1978. Annual assessments have
769 been conducted by INPE since 1988. For each assessment 229 Landsat satellite images are
770 acquired around August and analyzed. Results of the analysis of the satellite imagery are
771 published every year. Spatially-explicit results of the analysis are also publicly available (see
772 http://www.obt.inpe.br/prodes/prodes_1988_2006.htm).

23
773 Box 3.3: Example of result of the PRODES project:
774 Landsat satellite mosaic of year 2006 with deforestation during period 2000-2006
775 Brazilian Amazon window Zoom on Mato Grosso (around Jurunea)

776

777 Forested areas appear in green, non-forest areas appear in violet, old deforestation
778 (1997- 2000) in yellow and recent deforestation (from 2001) in orange-red.

779 PRODES also provides the spatial distribution of critical areas (in terms of deforestation) in
780 the Amazon. For the period August 1999 to August 2000, more than 80% of the
781 deforestation was concentrated in 49 of the 229 satellite images analyzed.
782 A new methodological approach based on digital processing is now in operational phase. A
783 geo-referenced, multi-temporal database is produced including a mosaic of deforested areas
784 by States of Brazilian federation. All results for the period 1997 to 2006 are accessible and
785 can be downloaded from the INPE web site at: http://www.dpi.inpe.br/prodesdigital.
786 Since May 2005, the Brazilian government also has in operation the DETER (Detecção de
787 Desmatamento em Tempo Real) system to serve as an alert in almost real-time (every 15
788 days) for deforestation events larger than 25 ha. The system uses MODIS data (spatial
789 resolution 250m) and WFI data on board CBERS-2 (spatial resolution 260m) and a
790 combination of linear mixture modeling and visual analysis. Results are publicly available
791 through a web-site: http://www.obt.inpe.br/deter/.

792 B. India – Biennial wall to wall approach

793 The application of satellite remote sensing technology to assess the forest cover of the
794 entire country in India began in early 1980s. The National Remote Sensing Agency (NRSA)
795 prepared the first forest map of the country in 1984 at 1:1 million scale by visual
796 interpretation of Landsat data acquired at two periods: 1972-75 and 1980-82. The Forest
797 Survey of India (FSI) has since been assessing the forest cover of the country on a two year
798 cycle. Over the years, there have been improvements both in the remote sensing data and
799 the interpretation techniques. The 10th biennial cycle has just been completed from digital
800 interpretation of data from year 2005 at 23.5 m resolution with a minimum mapping unit of
801 1 ha. The details of the data, scale of interpretation, methodology followed in wall to wall
802 forest cover mapping over a period of 2 decades done in India is presented in Table 3.4.

24
803 Table 3.4. State of the Forest Assessments of India

Forest
Data
Assessment Satellite Sensor Resolution Scale Analysis Cover
Period
Million ha
I 1981-83 LANDSAT-MSS 80 m 1:1 million visual 64.08

II 1985-87 LANDSAT-TM 30 m 1:250,000 visual 63.88

III 1987-89 LANDSAT-TM 30 m 1:250,000 Visual 63.94

IV 1989-91 LANDSAT-TM 30 m 1:250,000 Visual 63.94

V 1991-93 IRS-1B LISSII 36.25 m 1:250,000 Visual 63.89

VI 1993-95 IRS-1B LISSII 36.25 m 1:250,000 Visual 63.34

digital/
VII 1996-98 IRS-1C/1D LISS III 23.5 m 1:250,000 63.73
visual
VIII 2000 IRS-1C/1D LISS III 23.5 m 1:50,000 digital 65.38
IX 2002 IRS-1D LISS III 23.5 m 1:50,000 digital 67.78

X 2004 IRS P6- LISS III 23.5 m 1:50,000 digital 67.70

804

805 The entire assessment from the procurement of satellite data to the reporting, including
806 image rectification, interpretation, ground truthing and validation of the changes by the
807 State/Province Forest Department, takes almost two years.
808 The last assessment (X cycle) used satellite data from the Indian satellite IRS P6 (Sensor
809 LISS III at 23.5 m resolution) mostly from the period November-December (2004) which is
810 the most suitable period for Indian deciduous forests to be discriminated by satellite data.
811 Satellite imagery with less than 10% cloud cover is selected. For a few cases (e.g. north-
812 east region and Andaman & Nicobar Islands where availability of cloud free data during Nov-
813 Dec is difficult) data from January-February were used.
814 Satellite data are digitally processed, including radiometric and contrast corrections and
815 geometric rectification (using geo-referenced topographic sheets at 1:50,000 scale from
816 Survey of India). The interpretation involves a hybrid approach combining unsupervised
817 classification in raster format and on screen visual interpretation of classes. The Normalized
818 Difference Vegetation Index (NDVI) is used for excluding non-vegetated areas. The areas of
819 less than 1 ha are filtered (removed).
820 India classifies its lands into the following cover classes:
All lands with tree cover of canopy density of 70% and
Very Dense Forest
above
Moderately Dense All lands with tree cover of canopy density between 40 %
Forest and 70 % above
All lands with tree cover of canopy density between 10 –
Open Forest
40 %.
All forest lands with poor tree growth mainly of small or
Scrub
stunted trees having canopy density less than 10 percent.
Non-forest Any area not included in the above classes.
821

25
822 The initial interpretation is then followed by extensive ground verification which takes more
823 than six months. All the necessary corrections are subsequently incorporated. Reference
824 data collected by the interpreter during the field campaigns are used in the classification of
825 the forest cover patches into canopy density classes. District wise and States/Union
826 Territories forest cover maps are produced.
827 Accuracy assessment is an independent exercise. Randomly selected sample points are
828 verified on the ground (field inventory data) or with satellite data at 5.8 m resolution and
829 compared with interpretation results. In the X assessment, 4,291 points were randomly
830 distributed over the entire country. The overall accuracy level of the assessment has been
831 found to be 92 %

832 C. Congo basin – example of a sampling approach

833 Analyses of changes in forest cover at national scales have been carried out by the research
834 community. These studies have advanced methodologies for deforestation monitoring and
835 provided assessments of deforestation outside the realm of national governments. As one
836 example, a test of the systematic sampling approach has been carried out in Central Africa
837 to derive area estimates of forest cover change between 1990 and 2000. The proposed
838 systematic sampling approach using mid-resolution imagery (Landsat) was operationally
839 applied to the entire Congo River basin to accurately estimate deforestation at regional level
840 and, for large-size countries, at national level. The survey was composed of 10 × 10 km2
841 sampling sites systematically distributed every 0.5° over the whole forest domain of Central
842 Africa, corresponding to a sampling rate of 3.3 %. For each of the 571 sites, subsets were
843 extracted from both Landsat TM and ETM+ imagery acquired in 1990 and 2000 respectively.
844 The satellite imagery was analyzed with object-based (multi-date segmentation)
845 unsupervised classification techniques.
846 Around 60% of the 390 cloud-free images do not show any forest cover change. For the
847 other 165 sites, the results are represented by a change matrix for every sample site
848 describing four regrouped land cover change processes, e.g. deforestation, reforestation,
849 forest degradation and forest recovery (the samples in which change in forest cover is
850 observed are classified into 10 land cover classes, i.e. “dense forest”, “degraded forest”,
851 “long fallow & secondary forest”, “forest/agriculture mosaic”, “agriculture & short fallow”,
852 “bare soil & urban area”, “non forest vegetation”, “forest-savannah mosaic”, “water bodies”
853 and “no data”). “Degraded forest” were defined spectrally from the imagery (lighter tones in
854 image color composites as compared to dense forests – see next picture).
855 For a region like Central Africa (with 180 Million ha), using 390 samples, corresponding to a
856 sampling rate of 3.3 %, this exercise estimates the annual deforestation rate at
857 0.21 ± 0.05 % for the period 1990-2000. For the Democratic Republic of Congo which is
858 covered by a large-enough number of samples (267), the estimated annual deforestation
859 rate was 0.25 ± 0.06%. Degradation rates were also estimated (annual rate: 0.15 ±
860 0.03 % for the entire basin).
861 The accuracy of the image interpretation was evaluated from the 25 quality control sample
862 sites. For the forest/non-forest discrimination the accuracy is estimated at 93 % (n = 100)
863 and at 72 % for the 10 land cover classes mapping (n = 120). The overall accuracy of the 2
864 regrouped change classes, deforestation and reforestation, is estimated at 91 %. The
865 exercise illustrates also that the statistical precision depends on the sampling intensity.

26
866 Box 3.4: Example of results of interpretation for a 10 km x 10 km sample in
867 Congo Basin
868 Landsat image (TM sensor) of year 1990 Landsat image (ETM sensor) of year 2000

869

870 Image interpretation of year 1990 Image interpretation of year 2000

871

872 Legend: green = Dense forest, light green = degraded forest, yellow =
873 forest/agriculture mosaic, orange = agriculture & fallow.

874 3.3 Monitoring of Forest Degradation

875 Many activities cause degradation of carbon stocks in forests but not all of them can be
876 monitored well with high certainty using remote sensing data. As discussed above in Section
877 2.2, the gaps in the canopy caused by selective harvesting of trees (both legal and illegal)
878 can be detected in imagery such as Landsat using sophisticated analytical techniques of
879 frequently collected imagery, and the task is somewhat easier when the logging activity is
880 more intense (i.e. higher number of trees logged). A combination of legal logging followed
881 by illegal activities in the same concession is likely to cause more degradation and more
882 change in canopy characteristics, and thus an increased chance that this could be monitored
883 with Landsat type imagery and interpretation. The area of forests undergoing selective
884 logging can also be interpreted in remote sensing imagery based on the observations of
885 networks of roads and log decks that are often clearly recognizable in the imagery.
886 Degradation of carbon stocks by forest fires could be more difficult to monitor with existing
887 satellite imagery
888 Degradation by over exploitation for fuel wood or other local uses of wood often followed by
889 animal grazing that prevents regeneration, a situation more common in drier forest areas, is
890 likely not to be detectable from satellite image interpretation unless the rate of degradation

27
891 was intense causing larger changes in the canopy and thus monitoring methods are not
892 presented here.
893 In this section, two approaches are presented that could be used to monitor selective
894 logging: the direct approach that detects gaps and the indirect approach that detects road
895 networks and log decks. (The timber harvesting practice that fells all the trees, commonly
896 referred to as clear cutting, is not considered to be degradation here—it could be considered
897 as deforestation or forest management practice, depending upon the resulting land use.)

898

899 3.3.1 Direct approach to monitor selective logging

900 Mapping forest degradation with remote sensing data is more challenging than mapping
901 deforestation because the degraded forest is a complex mix of different land cover types
902 (vegetation, dead trees, soil, shade) and the spectral signature of the degradation changes
903 quickly (i.e., < 2 years). High spatial resolution sensors such as Landsat and SPOT have
904 been mostly used so far to address this issue. However, very high resolution satellite
905 imagery, such as Ikonos or Quickbird, and aerial digital image acquired with videography
906 have been used as well. Here, the methods available to detect and map forest degradation
907 caused by selective logging and forest fires – the most predominant types of degradation in
908 tropical regions – using optical sensors only are presented.
909 Methods for mapping forest degradation range from simple image interpretation to highly
910 sophisticated automated algorithms. Because the focus is on estimating forest carbon losses
911 associated with degradation, forest canopy gaps and small clearings are the feature of
912 interest to be enhanced and extracted from the satellite imagery. In the case of logging, the
913 damage is associated with areas of tree fall gaps, clearings associated with roads and log
914 landings (i.e., areas cleared to store harvested timber temporarily), and skid trails. The
915 forest canopy gaps and clearings are intermixed with patches of undamaged forests (Figure
916 3.1).

28
917

918 Figure 3.1: Very high resolution Ikonos image showing common features in selectively
919 logged forests in the Eastern Brazilian Amazon (image size: 11 km x 11 km)
920 There are two possible methodological approaches to map logged areas: 1) identifying and
921 mapping forest canopy damage (gaps and clearings); or 2) mapping the combined, i.e.,
922 integrated, area of forest canopy damage, intact forest and regeneration patches.
923 Estimating the proportion of forest carbon loss in the latter mapping approach is more
924 challenging requiring field sampling measurements of forest canopy damage and
925 extrapolation to the whole integrated area to estimate the damage proportion (see section
926 4.X).
927 Mapping forest degradation associated with fires is simpler than that associated with logging
928 because the degraded environment is usually contiguous and more homogeneous than
929 logged areas.
930 The following chart illustrates the steps needed to map forest degradation:

931

29
932 Step 1: Define the spatial resolution
933 Defining the appropriate spatial resolution to map forest degradation due to selective
934 logging depends on the type of harvesting operation (managed or unplanned). Managed and
935 non-mechanized logging practiced in a few areas of e.g., the Brazilian Amazon, cannot be
936 detected using spatial resolution in the order of 30-60 m (Figure 3.2) because these type of
937 logging create small forest gaps and little damage to the canopy. Very high resolution
938 imagery, as acquired with orbital and aerial digital videography, is required to directly map
939 forest canopy damage of these types. Unplanned logging generally creates more impact
940 allowing the detection of forest canopy damage at spatial resolution between 30-60 m.

A B C D

941

942 Figure 3.2. Unplanned logged forest in Sinop, Mato Grosso, Brazilian Amazon in: (A)
943 Ikonos panchromatic image (1 meter pixel); (B) Ikonos multi-spectral and panchromatic
944 fusion (4 meter pixel); (C) Landsat TM5 multi-spectral (R5, G4, B3; 30 meter pixel); and
945 (D) Nornalized Difference Fraction Index (NDFI) image (sub-pixel within 30 m). These
946 images were acquired in August 2001.

947 Step 2: Enhance the image

948 Detecting forest degradation with satellite image usually requires improving the spectral
949 contrast of the degradation signature relative to the background. In tropical forest regions
950 atmospheric correction and haze removal are recommended techniques to be applied to
951 high resolution images. Histogram stretching improves image color contrast and is a
952 recommended technique. However, at high spatial resolution histogram stretching is not
953 enough to enhance the image to detect forest degradation due to logging. Figure 3.2C
954 shows an example of a color composite of reflectance bands (R5,G4,B3) of Landsat image
955 after a linear stretching with little or no evidence of logging. At fine/moderate spatial
956 resolution, such as of the resolution of Landsat and Spot 4 images, a spectral mixed signal
957 of green vegetation (GV), soil, non-photosynthetic vegetation (NPV) and shade is expected
958 within the pixels. That is why the most robust techniques to map selective logging impacts
959 are based on fraction images derived from spectral mixture analysis (SMA). Fractions are
960 sub-pixel estimates of the pure materials (endmembers) expected within pixel sizes such as
961 those of Landsat (i.e., 30 m): GV, soil, NPV and shade endmembers (see SMA Box 1).
962 Figure 3.2D shows the same area and image as Figure 3.2C with logging signature
963 enhanced with the Normalized Difference Fraction Index (NDFI; see Box 3.5). The SMA and
964 NDFI have been successfully applied to Landsat and SPOT images in the Brazilian Amazon to
965 enhance the detection of logging and burned forests (Figure 3.3).
966 Because the degradation signatures of logging and forest fires change quickly in high
967 resolution imagery (i.e., < one year), annual mapping is required. Figure 3.3 illustrates this
968 problem showing logging and forest fires scars changing every year over the period of 1998
969 to 2003. This has important implications for monitoring carbon stocks in degraded forests
970 because old degraded forests (i.e., with less carbon stocks) can be misclassified as intact
971 forests. Therefore, annual detection and mapping the canopy damage associated with
972 logging and forest fires is mandatory to monitoring forest degradation with high resolution
973 multispectral imagery such as SPOT and Landsat.

30
 a 1998 b

Old
Logged
Logged

Logged

c d

Logged and Burned Logged and Burned

e f

Old Logged and Old Logged and

Burned Burned

974

975 Figure 3.3: Forest degradation annual change due to selective logging and logging and
976 burning in Sinop region, Mato Grosso State, Brazil.

31
 977 Step 3: Select the mapping feature and methods
978 Forest canopy damage (gaps and clearings) areas are easier to identify in very high spatial
979 resolution images (Figure 3.2A-B). Image visual interpretation or automated image
980 segmentation can be used to map forest canopy damage areas at this resolution. However,
981 there is a tradeoff between these two methodological approaches when applied to the very
982 high spatial resolution images. Visual identification and delineation of canopy damage and
983 small clearings are more accurate but time consuming, whereas automated segmentation is
984 faster but generates false positive errors that usually require visual auditing and manual
985 correction of these errors. High spatial resolution imagery is the most common type of
986 images used to map logging (unplanned) over large areas. Visual interpretation at this
987 resolution does not allow the interpreter to identify individual gaps and because of this
988 limitation the integrated area – including forest canopy damage, and patches of intact forest
989 and regeneration – is the chosen mapping feature with this approach. Most of the
990 automated techniques – applied at high spatial resolution – map the integrated area as well
991 with only the ones based on image segmentation and change detection able to map directly
992 forest canopy damage. In the case of burned forests, both visual interpretation and
993 automated algorithms can be used and very high and high spatial resolution imagery have
994 been used.

995 Data Needs

996 There are several optical sensors that can be used to map forest degradation caused by
997 selective logging and forest fires (Table 3.5). Users might consider the following factors
998 when defining data needs:
999 ‰ Degradation intensity—is the logging intensity low or high?
1000 ‰ Extent of the area for analysis—large or small areal extent?
1001 ‰ Technique that will be used—visual or automated?
1002 Very high spatial resolution sensors will be required for mapping low intensity degradation.
1003 Small areas can be mapped at this resolution as well if cost is not a limiting factor. If
1004 degradation intensity is low and area is large, indirect methods are preferred because cost
1005 for acquisition of very high resolution imagery may be prohibitive (see section on Indirect
1006 Methods to Map Forest Degradation). For very large areas, high spatial resolution sensors
1007 produce satisfactory estimates of the area affected by degradation.
1008 Finally, the spectral resolution and quality of the radiometric signal must be taken into
1009 account for monitoring forest degradation at high spatial resolution. The estimation of the
1010 abundance of the materials (i.e., end-members) found with the forested pixels, through
1011 SMA, requires at least four spectral bands placed in spectral regions that contrast the end-
1012 members spectral signatures (see Box 3.5).

32
1013 Table 3.5: Remote sensing methods tested and validated to map forest degradation caused
1014 by selective logging and burning in the Brazilian Amazon.
Mapping Spatial
Sensor Objective Advantages Disadvantages
Approach Extent
Does not
Map integrated Labor intensive for large
require
Local and logging area areas and may be user
Visual Landsat sophisticated
Brazilian and canopy biased to define the
Interpretation TM5 image
Amazon damage of boundaries of the
processing
burned forest degraded forest.
techniques
Relatively Harvesting buffers
Detection of
simple to varies across the
Logging Landsat
Map integrated implement and landscape and does not
Landings + TM5 and Local
logging area satisfactorily reproduce the actual
Harvesting ETM+
estimate the shape of the logged
Buffer
area area
Simple and
Map forest intuitive binary
It has not been tested
canopy classification
in very large areas and
damage rules, defined
Decision Tree SPOT 4 Local classification rules may
associated with automatically
vary across the
logging and based on
landscape
burning statistical
methods
Requires two pairs of
Map forest
radiometrically
canopy
Landsat Enhances forest calibrated images and
Change damage
TM5 and Local canopy does not separate
Detection associated with
ETM+ damaged areas. natural and
logging and
anthropogenic forest
burning
changes
It has not been tested
Relatively in very large areas and
Image Landsat Map integrated
Local simple to segmentation
Segmentation TM5 logged area
implement rules may vary across
the landscape
Map forest
Landsat Relatively
Brazilian canopy
Textural Filters TM5 and simple to
Amazon damage
ETM+ implement
associated
Requires very high
Map total computation power, and
Three states logging area Fully automated pairs of images to
Landsat of the (canopy and detect forest change
CLAS TM5 and Brazilian damage, standardized to associated with logging.
ETM+ Amazon (PA, clearings and very large Requires additional
MT and AC) undamaged areas. image types for
forest) atmospheric correction
(MODIS)
Map forest
canopy It has not been tested
Landsat Enhances forest
damage in very large areas and
NDFI+CCA TM5 and Local canopy
associated with does not separate
ETM+ damaged areas.
logging and logging from burning
burning

1015

33
1016 Box 3.5: Spectral Mixture Analysis (SMA)
1017 Detection and mapping forest degradation with remotely sensed data is more
1018 challenging than mapping forest conversion because the degraded forest is a complex
1019 environment with a mixture of different land cover types (i.e., vegetation, dead trees,
1020 bark, soil, shade), causing a mixed pixel problem (see Figure 1x). In degraded forest
1021 environments, the reflectance of each pixel can be decomposed into fractions of green
1022 vegetation (GV), non-photosynthetic vegetation (NPV; e.g., dead tree and bark), soil
1023 and shade through Spectral Mixture Analysis (SMA). The output of SMA models are
1024 fraction images of each pure material found within the degraded forest pixel, known as
1025 endmember. Fractions are more intuitive to interpret than the reflectance of mixed
1026 pixels (most common signature at high spatial resolution). For example, soil fraction
1027 enhances log landings and logging roads; NPV fraction enhances forest damage and
1028 the GV fraction is sensitive to canopy gaps.
1029 The SMA model assumes that the image spectra are formed by a linear combination of
1030 n pure spectra [or endmembers], such that:
n

1031 (1) Rb = ∑ Fi ⋅ Ri ,b + ε b
i =1

1032 for
n

1033 (2) ∑F =1
i =1

1034 where Rb is the reflectance in band b, Ri,b is the reflectance for endmember i, in band
1035 b, Fi the fraction of endmember i, and b is the residual error for each band. The SMA
1036 model error is estimated for each image pixel by computing the RMS error, given by:
1/ 2
⎡ −1 n ⎤
1037 (3) RMS = ⎢n ∑ ε b ⎥
⎣ b =1 ⎦
1038 The identification of the nature and number of pure spectra (i.e., endmembers), in the
1039 image scene is the most important step for a successful application of SMA models. In
1040 Landsat TM/ETM+ images the four types of endmembers are expected in degraded
1041 forest environments (GV, NPV, Soil and Shade) can be easily identified in the extreme
1042 of image bands scatterplots.
1043 The pixels located at the extremes of the data cloud of the Landsat spectral space are
1044 candidate endmembers to run SMA. The final endmembers are selected based on the
1045 spectral shape and image context (e.g., soil spectra are mostly associated with
1046 unpaved roads and NPV with pasture having senesced vegetation) (figure below).
1047 The SMA model results were evaluated as follows: (1) fraction images are evaluated
1048 and interpreted in terms of field context and spatial distribution; (2) the histograms of
1049 the fraction images are inspected to evaluate with the models produced physically
1050 meaningful results (i.e., fractions ranging from zero to 100%). In time-series
1051 applications, as required to monitor forest degradation, fraction values must be
1052 consistent over time for invariant targets (i.e., that intact forest not subject to
1053 phonological changes must have similar values over time). Several image processing
1054 software have spectral plotting and SMA functionalities.

34
1055 Box 3.5: Continuation

1056

1057 Image scatter-plots of Landsat bands in reflectance space and the spectral curves of
1058 GV, Shade, NPV and Soil.

1059 Limitations for forest degradation

1060 There are limiting factors to all methods described above that might be taken into
1061 consideration when mapping forest degradation. First, it requires frequent mapping, at least
1062 annually, because the spatial signatures of the degraded forests change after one year.
1063 Additionally, it is important to keep track of repeated degradation events that affect more
1064 drastically the forest structure and composition resulting in greater changes in carbon
1065 stocks. Second, the human-caused forest degradation signal can be confused with natural
1066 forest changes such as windthrows and phenological changes. Third, all the methods
1067 described above are based on optical sensors which are limited by frequent cloud conditions
1068 in tropical regions. Finally, higher level of expertise is required to use the most robust
1069 automated techniques requiring specialized software and investments in capacity building.

1070 Accuracy assessment

1071 Experience to date on assessing the accuracy of interpretation of selectively logged and
1072 burned areas has shown that it is possible to obtain an accuracy ranging from 86 to 95%
1073 (Table 3.5). Most studies used conventional accuracy assessment based on error matrix.
1074 These studies have used field data and/and or aerial videography imagery as reference data
1075 for the accuracy assessment. Another way to assess the accuracy is to report uncertainty by
1076 combining different sources of errors (e.g., reflectance retrieval, cloud cover, annualization,
1077 manual auditing) to generate the logging map. An example of mapping logging, over a very
1078 large area in the Brazilian Amazon, resulted in an uncertainty of 86% for mapping logging
1079 using a semi-automated approach. But field inspection, in the same study, showed false-
1080 positive and false-negative rates of 5 %.

35
1081 Progress in application of monitoring systems,
1082 Brazil is well-known by its deforestation monitoring systems Prodes
1083 (http://www.obt.inpe.br/prodes/). Currently, a new monitoring system is being developed
1084 to monitor forest degradation, particularly selective logging, named Detex. The demand for
1085 Detex emerged after recent studies confirmed that logging damages annually an area as
1086 large as the area affected by deforestation in this region (i.e., 10,000-20,000 km2/year).
1087 The Detex system will support the management and monitoring of large forest concession
1088 areas in the Brazilian Amazon. All the techniques discussed in this section were developed
1089 and validated in the Brazilian Amazon. Recent efforts to export these methodologies to
1090 other areas are under way. For example, SMA (Box 3.5) and NDFI (Box 3.6) have being
1091 tested in Bolivia with Landsat and Aster imagery. The preliminary results showed that forest
1092 canopy damage of low intensity logging, the most common type of logging in the region,
1093 could not be detected with Landsat. This corroborates with the findings in the Brazilian
1094 Amazon. New sensor data with higher spatial resolution are currently being tested in Bolivia,
1095 including Spot 5 (10 m) and Aster (15 m) to evaluate the best sensor for their operational
1096 system. Given their higher spatial resolution, Aster and Spot imagery are showing promise
1097 for detecting and mapping low intensity logging in Bolivia.

1098 Box 3.6: Calculating NDFI

1099 The detection of logging impacts at moderate spatial resolution is best accomplished
1100 at the subpixel scale, with spectral mixture analysis (SMA). Fraction images obtained
1101 with SMA can enhance the detection of logging infrastructure and canopy damage. For
1102 example, soil fraction can enhance the detection of logging decks and logging roads;
1103 NPV fraction enhances damaged and dead vegetation and green vegetation the canopy
1104 openings. A new spectral index obtained from fractions derived from SMA, the
1105 Normalized Difference Fraction Index (NDFI), enhances even more the degradation
1106 signal caused by selective logging. The NDFI is computed by:

GVShade − ( NPV + Soil )

1107 (1) NDFI =
GVShade + NPV + Soil
1108 where GVshade is the shade-normalized GV fraction given by:
GV
1109 (2) GVShade =
100 − Shade
1110 The NDFI values range from -1 to 1. For intact forest NDFI values are expected to be
1111 high (i.e., about 1) due to the combination of high GVshade (i.e., high GV and canopy
1112 Shade) and low NPV and Soil values. As forest becomes degraded, the NPV and Soil
1113 fractions are expected to increase, lowering the NDFI values relative to intact forest.

1114 Special software requirements and costs

1115 All the techniques described in this section are available in most remote sensing,
1116 commercial and public domain software (refer to the Table that describes image processing
1117 software). The software must have the capability to generate GIS vector layers in case
1118 image interpretation is chosen, and being able to perform SMA for image enhancement.
1119 Image segmentation is the most sophisticated routine required, being available in a few
1120 commercial and public domain software packages. Additionally, it is desired that the
1121 software allows adding new functions to be added to implement new specialized routines,
1122 and have script capability to batch mode processing of large volume of image data.

36
1123 3.3.2 Indirect approach to monitor forest degradation

1124 Often a direct remote sensing approach to assess forest degradation can not be adopted for
1125 various limiting factors (see previous section) which are even more restrictive if forest
1126 degradation has to be measured for a historical period and thus observed only with remote
1127 sensing data that are already available in the archives.
1128 Moreover the forest definition contained in the UNFCCC framework of provisions (UNFCCC,
1129 2001) does not discriminate between forests with different carbon stocks, and often forest
1130 land subcategories defined by Countries are based on concepts related to different forest
1131 types (e.g. specie compositions) or ecosystems than can be delineated through remote
1132 sensing data or through geo-spatial criteria (e.g. altitude). Consequently, any accounting
1133 system based on forest definitions that are not containing parameters related to carbon
1134 content, will require an extensive and high intensive carbon stock measuring effort (e.g.
1135 national forest inventory) in order to report on emissions from forest degradation.
1136 In this context, i.e. the need for activity data (area changes) on degraded forest under the
1137 UNFCCC reporting requirement and the lack of remote sensing data for an exhaustive
1138 monitoring system, a new methodology has been elaborated with the aim of providing an
1139 operational tool that could be applied worldwide. This methodology consists mainly in the
1140 adaptation of the concepts and criteria already developed to assess the world’s intact forest
1141 landscape in the framework of the IPCC Guidance and Guidelines to report GHG emission
1142 from forest land. In this new context the intact forest concept it is no longer related to the
1143 conservation of biodiversity, but has been used as a proxy to identify forest land without
1144 anthropogenic disturbance so as to assess the carbon content present in the forest land:
1145 ‰ intact forests: fully-stocked (any forest with tree cover between 10% and 100% but
1146 must be undisturbed, i.e. there has been no timber extraction)
1147 ‰ non-intact forests: not fully-stocked (tree cover must still be higher than 10% to
1148 qualify as a forest under the existing UNFCCC rules, but in our definition we assume
1149 that in the forest has undergone some level of timber exploitation or canopy
1150 degradation).
1151 This distinction should be applied in any forest land use subcategories (forest stratification)
1152 that a country is aiming to report under UNFCCC. So for example, if a Country is reporting
1153 emissions from its forest land using two forest land subcategories, e.g. lowland forest and
1154 mountain forest, it should further stratify its territory using the intact approach and in this
1155 way it will report on four forest land sub-categories: intact lowland forest; non-intact
1156 lowland forest, intact mountain forest and non-intact mountain forest. Thus a Country will
1157 also have to collect the corresponding carbon pools data in order to characterize each forest
1158 land subcategories.
1159 The intact forest areas are defined according to parameters based on spatial criteria that
1160 could be applied objectively and systematically over all the Country territory. Each Country
1161 according to its specific national circumstance (e.g. forest practices) may develop its intact
1162 forest definition. Here we suggest an intact forest area definition based on the following six
1163 criteria:
1164 ‰ Situated within the forest land according to current UNFCCC definitions and with a 1
1165 km buffer zone inside the forest area;
1166 ‰ Larger than 1,000 hectares and with a smallest width of 1 kilometers;
1167 ‰ Containing a contiguous mosaic of natural ecosystems;
1168 ‰ Not fragmented by infrastructure (road, navigable river, pipeline, etc.);
1169 ‰ Without signs of significant human transformation;

37
1170 ‰ Without burnt lands and young tree sites adjacent to infrastructure objects.
1171 The suggested criteria are almost replicating the criteria that have been already used to
1172 detect intact forest areas all around the world (www.intactforests.org); the differences only
1173 relate to quantitative parameters like minimum extension that moves from 50,000 to 1,000
1174 ha, or the minimum width from 10 to 1 km. In The suggested definition an intact forest is
1175 less affected by forest land fragmentation and thus will require more detailed analysis
1176 compared with the original definition, but should still be applicable all around the world. It
1177 must be noted that if these criteria were to be adopted in the near future, the intact forest
1178 area could not suddenly increase as once degraded (non-intact) a forest land remain
1179 degraded for long time even after the end of human activities.
1180 The adoption of the ‘intact’ concept is also driven by technical and practical reasons. In
1181 compliance with current UNFCCC practice it is the Parties’ responsibilities to identify forests
1182 according to the established 10% - 100% cover range rule. When assessing the condition of
1183 such forest areas using satellite remote sensing methodologies, the “negative approach” can
1184 be used to discriminate between intact and non-intact forests: disturbance such as the
1185 development of roads can be easily detected, whilst the absence of such visual evidence of
1186 disturbance can be taken as evidence that what is left is intact. Disturbance is easier to
1187 unequivocally identify from satellite imagery than the forest ecosystem characteristics which
1188 would need to be determined if we followed the “positive approach” i.e. identifying intact
1189 forest and then determining that the rest in non-intact. Following this approach forest
1190 conversions between intact forests, non-intact forests and other land uses can be easily
1191 measured worldwide through Earth observation satellite imagery; in contrast, any other
1192 forest definition (e.g. pristine, virgin, primary/secondary, etc...) is not always measurable.

1193 Method for delineation of intact forest landscapes

1194 A two-step procedure could be used to exclude non-intact areas and delineate the
1195 remaining intact forest:
1196 1. Exclusion of areas around human settlements and infrastructure and residual
1197 fragments of landscape smaller than 5,000 ha, based on topographic maps, GIS
1198 database, thematic maps, etc. This first step could be done through a spatial analysis
1199 tool in a GIS software (this step could be fully automatic in case of good digital
1200 database on road networks). The result is a candidate set of landscape fragments whit
1201 potential intact forest lands.
1202 2. Further exclusion of non-intact areas and delineation of intact forest lands is done
1203 by fine shaping of boundaries, based on visual interpretation methods of high-
1204 resolution satellite images (Landsat class data with 15-30 m pixel spatial resolution).
1205 Alternatively high-resolution satellite data could be used to develop a more detailed
1206 dataset on human infrastructures, that than could be used to delineate intact forest
1207 boundaries with a spatial analysis tool of a GIS software.

1208 The distinction between intact and non-intact allows us to account for carbon losses from
1209 forest degradation, reporting this as a conversion of intact to non-intact forest. The
1210 degradation process is thus accounted for as one of the three potential changes illustrated
1211 in Figure 1, i.e. from (i) intact forests to other land use, (ii) non-intact forests to other land
1212 use and (iii) intact forests to non-intact forests. In particular carbon emission from forest
1213 degradation for each forest type consist of two factors the difference in carbon content
1214 between intact and non-intact forests and the area loss of intact forest area during the
1215 accounting period. This accounting strategy is fully compatible with the set of rules develop
1216 in the IPCC LULUCF Guidance and AFOLU Guidelines for the sections “Forest land remaining
1217 Forest land”.
1218

38
 intact forests other land use

non-intact forest
1219

1220 Figure 3.4: Forest conversions types considered in the accounting system. The forest
1221 degradation is included in the conversion from intact to non-intact forest, and thus
1222 accounted as carbon stock change in that proportion of forest land remaining as forest land.

39
1223 Figure 3.5 Forest degradation
1224 assessment in Papua New Guinea
1225 The Landast satellite images a and b
1226 are representing the same portion of
1227 PNG territories in the Gulf Province
1228 and they have been acquired
1229 respectively in 26.12.1988 and
1230 07.10.2002. In this part of territory
1231 it is present only the lowland forest
1232 type.
1233 In the image a it is possible to
a)
1234 recognize logging roads only on the
1235 east side of the river, while in the
1236 image b it is possible to recognize a
1237 very well developed logging road
1238 system also on the west side of the
1239 river. The forest canopy (brown-
1240 orange-red colours) does not seem
1241 to have evident changes in spectral
1242 properties (all these images are
1243 reflecting the same Landsat band
1244 combination 4,5,3).
1245 The images a1 and b1 are
a1)
1246 respectively the same images a and
1247 b with some patterned polygons
1248 which are representing the extension
1249 of the intact forest in the respective
1250 dates. In this case a on-screen visual
1251 interpretation method have been
1252 used to delineate intact forest
1253 boundaries.
1254 PNG in order to assess carbon
1255 emission from forest degradation for
1256 this part of its territory, it could
1257 report that in 14 years, 51% of the
1258 existing intact forest land have been b)
1259 converted in non-intact forest land.
1260 Thus the total carbon emission
1261 should be equivalent to the intact
1262 forest loss multiply by the carbon
1263 content difference between intact
1264 and non-intact forest land.
1265 In this particular case, deforestation
1266 (road network) is accounting for less
1267 than 1%.
1268
b1)

40
1269 3.3.3 Systems for mapping active forest fire, burned area and associated
1270 emissions

1271 Forest fires occur annually in all vegetation zones and increasing trends in wildland fire
1272 activity have been reported in many global regions during the most recent 1-2 decades. Due
1273 to the large spatial and temporal variability in fire activity, satellite data provide the most
1274 useful means to monitor fire (Table 3.6). There are several observation objectives relating
1275 the mapping of the extent and activity of current ongoing fires, the area and intensity of
1276 burns, and to predict future fire occurrence and take fire management actions.
1277 Table 3.6: Examples of operational and experimental satellite based observation systems
1278 of active fire, burnt areas and associated emissions
Satellite-based fire
Information and data access
monitoring
Global burnt areas 2000-2007:
www.tem.jrc.it/Disturbance_by_fire/products/burnt_areas/
L3JRC by EC Joint Research
GlobalBurntAreas2000-2007.htm
Center
MODIS fire products:
modis-fire.umd.edu/products.asp
by University of Maryland /
NASA
Globcarbon products:
http://www.fao.org/gtos/tcopjs4.html
By ESA
World Fire Atlas dup.esrin.esa.int/ionia/wfa/index.asp
By ESA
Global Fire Emissions Database
(GFED2) - multi-year burned ess1.ess.uci.edu/%7Ejranders/data/GFED2/
area and emissions ess1.ess.uci.edu/~jranders/data/GFED2/readme.pdf
By NASA
Fire Information for Resource
Management System (FIRMS)
maps.geog.umd.edu/firms/
By University of Maryland /
NASA
Global Fire Monitoring
Center (GFMC)
www.fire.uni-freiburg.de/inventory/burnt%20area.html
By University of Freiburg /
EUMETSAT
Experimental Wildfire Automated
Biomass Burning Algorithm
(GOES WF-ABBA) http://cimss.ssec.wisc.edu/goes/burn/wfabba.html
By University of Wisconsin-
Madison / NOAA
1279

1280 There are several polar and geostationary satellite systems with full operational status and
1281 some experimental systems providing systematic observations that have been used for the
1282 creation of long-term fire mapping data. Major long-term global records of active fires have
1283 been generated by ESA (ATSR World Fire Atlas) and NASA (TRMM and MODIS).
1284 Geostationary fire monitoring has been undertaken using GOES (WF-ABBA) and MSG
1285 SEVIRI (EUMETSAT Active Fire Monitoring and Global Fire Monitoring Center). The only long
1286 term burned area dataset available at the moment is also partly based on active fire
1287 detections (GFED2), but true multi-year burned area products are about to be released
1288 (MODIS, L3JRC, GLOBCARBON). Validation with in situ measurements is limited to only
1289 certain regions and is lacking especially in developing countries. In other regions, calibration
1290 with high resolution satellite data provides the best means for validation. Direct estimating
1291 of carbon emissions from these active fire detections or burned area has improved recently,
1292 with the use of biogeochemical models, but yet fails to capture fine-scale fire processes due

41
1293 to coarse resolutions. With new burned area products this situation will likely be improved in
1294 the next few years.
1295 Fire management occurs at many scales, from the local community to national and
1296 international levels. Fuels (or vegetation) data are basically static for fire management
1297 timescales, but fire and weather data are highly variable over short (hourly) time periods. A
1298 MODIS Rapid Response global near-real time mapping system is in development to notify
1299 protected areas managers of fires in the area of interest. The Fire Information for Resource
1300 Management System (FIRMS) uses data transmitted the MODIS instrument on board
1301 NASA’s Terra and Aqua satellites. These data are processed to produce images and text files
1302 pertaining to active fire locations. NASA and the University of Maryland have already
1303 established a prototype fire early warning system in South Africa which distributes active
1304 fire information to a range of users in the developing countries.

1305 List of key references for Section 3

1306 Achard, F., DeFries, R., Eva H., Hansen M., Mayaux P, Stibig H.-J. (2007): Pan-tropical
1307 monitoring of deforestation. Environmental Research Letters 2 in press
1308 Aksenov, D., Dobrynin, D., Dubinin M., Egorov, A., Isaev A., Karpachevskiy M., Laestadius
1309 L., Potapov P., Purekhovskiy A., Turubanova, S., Yaroshenko, A. (2002): Atlas of
1310 Russia’s intact forest landscapes. Global Forest Watch Russia, Moscow, p 184.
1311 Asner, G. P., Knapp, D. E., Broadbent, E., Oliviera, P., Keller, M., Silva, J. (2005): Selective
1312 logging in the Brazilian Amazon. Science 310: 480–482.
1313 DeFries R., Achard F., Brown, S., Herold, M., Murdiyarso, D., Schlamadinger, B., de Souza
1314 C. (2007): Earth Observations for Estimating Greenhouse Gas Emissions from
1315 Deforestation in Developing Countries. Environmental Science and Policy 10: 385–394.
1316 Duveiller, G., Defourny, P., Desclée, B., Mayaux, P. (2007): Deforestation in Central Africa:
1317 estimates at regional, national and landscape levels by advanced processing of
1318 systematically-distributed Landsat extracts. Remote Sensing of Environment In press
1319 FAO (2006): Global Forest Resources Assessment 2005: Main Report, Food and Agriculture
1320 Organization (FAO). Available at http://www.fao.org/forestry/fra2005
1321 FSI (2004): State of Forest Report 2003. Forest Survey of India (Dehra Dun)
1322 Greenpeace (2006): Roadmap to Recovery: The World's Last Intact Forest Landscapes..
1323 Available at: www.intactforests.org
1324 INPE (2005): Monitoring of the Brazilian Amazonian: Projeto PRODES. National Space
1325 Agency of Brazil. Available at http:// www.obt.inpe.br/prodes/index.html.
1326 IPCC (2003): Good Practice Guidance for Land Use, Land-Use Change and Forestry
1327 (LULUCF). Available at http://www.ipcc-nggip.iges.or.jp
1328 IPCC (2006): Guidelines for National Greenhouse Gas Inventories – Volume 4: Agriculture,
1329 Land Use and Forestry (AFOLU). Available at http://www.ipcc-nggip.iges.or.jp/
1330 Mayaux, P., Holmgren, P., Achard, F., Eva, H., Stibig, H.-J., Branthomme, A. (2005):
1331 Tropical forest cover change in the 1990s and options for future monitoring. Philos.
1332 Trans. Roy. Soc. B 360: 373–384
1333 Mollicone, D., Achard, F., Federici, S., Eva, H., Grassi, G., Belward, A., Raes, F., Seufert, G.,
1334 Stibig, H.-J., Matteucci, G., Schulze, E.-D. (2007): An incentive mechanism for
1335 reducing emissions from conversion of intact and non-intact forests. Climatic Change
1336 83:477–493
1337 Souza, C., Roberts, D. (2005): Mapping forest degradation in the Amazon region with
1338 Ikonos images. Int. J. Remote Sensing 26: 425–429.

42
1339 4 ESTIMATION OF CARBON STOCKS

1340 4.1 Overview of carbon stocks, and issues related to C stocks

1341 Monitoring the location and areal extent of deforestation and degradation represents only
1342 one of two components involved in assessing emissions from deforestation and degradation.
1343 The other component is the emission factors—that is, the changes in carbon stocks of the
1344 forests being deforested and degraded—that are combined with the activity data for
1345 deforestation and degradation.

1346 4.1.1 Issues related to carbon stocks

1347 4.1.1.1 The importance of “good” carbon stock estimates

1348 In the context of REDD, “good” estimates of carbon stocks means that they have low
1349 uncertainty and do no overestimate the true value. A natural preference exists to invest in
1350 refined estimates of areas degraded and deforested, then to combine this accurate picture
1351 with generalized carbon numbers obtained from default look up tables and literature (e.g.
1352 Tier 1 data, see Table 2.2). This is, however, an unsatisfactory strategy because the
1353 accuracy of the area estimate will be lost when paired with unsatisfactory carbon data,
1354 resulting in poor, uncertain estimates of emissions from deforestation and degradation (see
1355 Box 4.1). In reality, the carbon data should be viewed as equally important as the area
1356 data, with data of similar quality paired to produce consistent emissions estimates.

1357 Box 4.1: The Importance of Certainty in Carbon Measurements

1358 To be able to determine if real reductions against the reference case have taken place
1359 at future monitoring periods, it is important that the uncertainty bounds around the
1360 reference case estimate be small. Confidence is generated from the use of good
1361 methods that result in accurate and precise estimates of emission reductions. High
1362 certainty is required both in the estimates of area and in the estimates of the
1363 emissions arising from the given area of deforestation or degradation.
1364 Much of the focus of REDD is on deriving high quality remotely sensed estimates of
1365 area deforested and degraded. The following example shows the importance of an
1366 equal focus on both the area and the carbon stocks (emissions per unit area).

1367

1368 Using the IPCC Tier 1 Simple Propagation of Errors method, despite a constant low
1369 uncertainty of 5% for the area component, the uncertainty of the total final estimate
1370 of emissions is governed by the higher uncertainty in the carbon stock data. Therefore
1371 if uncertainty is not equally low for the two sources of the ultimate deforestation and
1372 degradation emissions, then the investment in the unbalanced half is money poorly
1373 spent.

43
1374 4.1.1.2 Fate of carbon pools as a result of deforestation and degradation
1375 A forest is composed of pools of carbon stored in the living trees above and belowground, in
1376 dead matter including standing dead trees, down woody debris and litter, in non-tree
1377 understory vegetation and in the soil organic matter. When trees are cut down there are
1378 three destinations for the stored carbon – dead wood, wood products or the atmosphere.
1379 ‰ In all cases, following deforestation and degradation, the stock in living trees
1380 decreases.
1381 ‰ Where degradation has occurred this is often followed by a recovery unless continued
1382 anthropogenic pressure or altered ecologic conditions precludes tree regrowth.
1383 ‰ The decreased tree carbon stock can either result in increased dead wood, increased
1384 wood products or immediate emissions.
1385 ‰ Dead wood stocks may be allowed to decompose over time or may, after a given
1386 period, be burned leading to further emissions.
1387 ‰ Wood products over time decompose, burned, or are retired to land fill.
1388 ‰ Where deforestation occurs, trees can be replaced by non-tree vegetation such as
1389 grasses or crops. In this case, the new land-use has consistently lower plant biomass
1390 and often lower soil carbon, particularly when converted to annual crops.
1391 ‰ Where a fallow cycle results, then periods of crops are interspersed with periods of
1392 forest regrowth that may or may not reach the threshold for definition as forest.
1393 Figure 4.1 below illustrates potential fates of existing forest carbon stocks after
1394 deforestation.
Carbon Stock

Trees Dead Wood Soil Carb on Non-tree Wood

Vegetation Products
Before Deforestation
After Deforestation
1395

1396

1397

1398

1399

1400

1401

44
 Deforestation event

Non-Tree Vegetation
Harvested Products
Dead Wood
Soil Carbon
Trees
Carbon Stock

Time
1402

1403 Figure 4.1: Fate of existing forest carbon stocks after deforestation.

1404 4.1.1.3 The definition of uncertainty for carbon assessments

1405 To estimate the carbon stock on the land one has to sample rather than attempt to measure
1406 everything. Sampling is the process by which a subset is studied to allow generalizations to
1407 be made about the whole population or area of interest. The values attained from
1408 measuring a sample are an estimation of the equivalent value for the entire area or
1409 population. Statistics provide us with some idea of how close the estimation is to reality and
1410 therefore how certain or uncertain the estimates are.
1411 There are three critical statistical concepts: bias, accuracy and precision.
1412 Bias is a systematic distortion often caused by flaws in the measurements or sampling
1413 methods.
1414 Accuracy is how close to the actual value your sample measurements are. Accuracy details
1415 the agreement between the true value and repeated measured observations or estimations
1416 of a quantity.
1417 Precision is how well a value is defined. In sampling, precision illustrates the level of
1418 agreement among repeated measurements of the same quantity. This is represented by
1419 how closely grouped the results from the various sampling points or plots are.
1420 A popular analogy is a bull’s eye on a target. In this analogy, how tightly the darts are
1421 grouped is the precision, how close they are to the center is the accuracy. Below in Figure
1422 4-2 (A), the points are close to the center and are therefore accurate but they are widely
1423 spaced and therefore are imprecise. In (B), the points are closely grouped and therefore are
1424 precise and could be biased but are far from the center and so are inaccurate. Finally, in
1425 (C), the points are close to the center and tightly grouped and are both accurate and
1426 precise.

45
1427 (A) Accurate but not precise (B) Precise but not accurate (C) Accurate and precise

1428

1429 Figure 4.2: Illustration of the concepts of accuracy and precision at they apply to estimates
1430 of forest carbon stocks.
1431 When sampling for carbon, measurements that are both accurate (i.e. close to the reality
1432 for the entire population) and precise (closely grouped so the results are highly confident or
1433 have low uncertainty) are needed.
1434 Sampling a subset of the land for carbon estimation involves taking measurements in a
1435 number of locations or ‘plots’ that are distributed randomly or systematically over the area
1436 to avoid any bias in sampling. The average value when all the plots are combined
1437 represents the wider population. A 95 % confidence interval, for example, tells us that 95
1438 times out of a 100 the true carbon density lies within the interval. If the interval is small
1439 then the result is precise –it has low uncertainty.

1440 4.1.1.4 The need for stratification and how it relates to remote sensing data
1441 Carbon stocks vary by forest type, for example tropical pine forests will have a different
1442 stock to tropical broadleaf forest which will again have a different stock to a woodland or a
1443 mangrove forest. Even within broadleaf tropical forests, stocks will vary greatly with
1444 elevation, rainfall and soil type. Then even within a given forest type in a given location the
1445 degree of human disturbance will lead to further differences in stocks. The resolution of
1446 most readily and inexpensively available remote sensing imagery is not good enough to
1447 differentiate between different forest types or even between disturbed and undisturbed
1448 forest, and thus cannot differentiate different forest carbon stocks. Therefore stratifying
1449 forests can lead to more accurate and cost effective emission estimates associated with a
1450 given area of deforestation or degradation (see more on this topic below in section 4.3).

1451 4.1.2 Overview of Chapter

1452 In Section 4.2 guidance is provided on: Which Tier Should be Used? The IPCC GL AFOLU
1453 allow for three Tiers with increasing complexity and costs of monitoring forest carbon
1454 stocks.
1455 In Section 4.3 the focus is on: Stratification by Carbon Stock. As discussed in 4.1.1
1456 stratification is an essential step to allow an accurate, cost effective and creditable linkage
1457 between the remote sensing imagery estimates of areas deforested and estimates of carbon
1458 stocks and therefore emissions. In this section guidance is provided on potential methods
1459 for the stratification of a country’s forests.
1460 In Section 4.4 guidance is given on the actual Estimation of Carbon Stocks of Forests
1461 Undergoing Change. Steps are given on how to devise and implement an inventory.
1462 In Section 4.5 guidance is presented on assessing the Uncertainty resulting from the forest
1463 carbon stock estimations.
1464 Finally in Section 4.6 Case Studies are presented on the entire process of implementing a
1465 carbon stock assessment for REDD. Separate Case Studies are presented for deforestation
1466 and degradation.

46
1467 4.2 Which Tier Should be Used?

1468 4.2.1 Explanation of IPCC Tiers

1469 The IPCC GPG and AFOLU Guidelines present three general approaches for estimating
1470 emissions/removals of greenhouse gases, known as “Tiers” ranging from 1 to 3 representing
1471 increasing levels of data requirements and analytical complexity. Despite differences in
1472 approach among the three tiers, all tiers have in common their adherence to IPCC good
1473 practice concepts of transparency, completeness, consistency, comparability, and accuracy.
1474 Tier 1 requires no new data collection to generate estimates of forest biomass. Default
1475 values for forest biomass and forest biomass mean annual increment (MAI) are obtained
1476 from the IPCC Emission Factor Data Base (EFDB), corresponding to broad continental forest
1477 types (e.g. African tropical rainforest). Tier 1 estimates thus provide limited resolution of
1478 how forest biomass varies sub-nationally and have a large error range (~ +/- 50% or more)
1479 for growing stock in non-industrialized countries (Box 4.2). The former is important because
1480 deforestation and degradation tend to be localized and hence may affect subsets of forest
1481 that differ consistently from a larger scale average (Figure 4.3). Tier 1 also uses simplified
1482 assumptions to calculate emissions. For deforestation, Tier 1 uses the simplified assumption
1483 of instantaneous emissions from woody vegetation, litter and dead wood. To estimate
1484 emissions from degradation (i.e. Forest remaining as Forest), Tier 1 applies the gain-loss
1485 method using a default MAI combined with losses reported from wood removals and
1486 disturbances, with transfers of biomass to dead organic matter estimated using default
1487 equations.

1488 Box 4.2– Error in Carbon Stocks from Tier 1 Reporting

1489 To illustrate the error in applying Tier 1 carbon stocks for the carbon element of REDD
1490 reporting, a comparison is made here between the Tier 1 result and the carbon stock
1491 estimated from on-the-ground IPCC Good Practice-conforming plot measurements
1492 from six sites around the world. As can be seen in the table below, the IPCC Tier 1
1493 predicted stocks range from 33 % higher than a mean derived from plot
1494 measurements to 44 % lower.

1495

47
1496 Figure 4.3 below illustrates a hypothetical forest area, with a subset of the overall forest, or
1497 strata, denoted in light green. Despite the fact that the forest overall (including the light
1498 green strata) has a mean biomass stock of 150 t C/ha, the light green strata alone has a
1499 significantly different mean biomass carbon stock. Because deforestation often takes place
1500 along “fronts” (e.g. agricultural frontiers) that may represent different subsets from a broad
1501 forest type (like the light green strata at the periphery here) higher resolution of forest
1502 biomass carbon stocks is required to accurately assign stocks to where loss of forest cover
1503 takes place. Applying the overall forest stock to the light green strata alone would be
1504 inaccurate, and that source of uncertainty could only be discerned by subsequent ground-
1505 truthing, as compared with precision which is more easily assessed.
1506 Figure 4.3 also demonstrates the inadequacies of extrapolating localized data across a broad
1507 forest area, and hence the need to augment limited existing datasets (e.g. forest
1508 inventories and research studies conducted locally) with supplemental data collection.
1509

200

160
biomass C t per ha

120

80

40

200

160
biomass C t per ha

120

80

40

1510

1511 Figure 4.3: A hypothetical forest area, with a subset of the overall forest, or strata,
1512 denoted in light green.
1513 At the other extreme, Tier 3 is the most rigorous approach associated with the highest level
1514 of effort. Tier 3 uses actual inventories with repeated measures of permanent plots to
1515 directly measure changes in forest biomass and/or uses well parameterized models in
1516 combination with plot data. Tier 3 often focuses on measurements of trees only, and uses
1517 region/forest specific default data and modeling for the other pools. The Tier 3 approach
1518 requires long-term commitments of resources and personnel, generally involving the
1519 establishment of a permanent organization to house the program (e.g. Australian
1520 Greenhouse Gas Office, USDA Forest Service Forest Inventory and Analysis program). The
1521 Tier 3 approach can thus be very expensive in the developing country context, particularly
1522 where only a single objective (estimating emissions of greenhouse gases) supports the
1523 implementation costs. Unlike Tier 1, Tier 3 does not assume immediate emissions from

48
1524 deforestation, instead modeling transfers and releases among pools that more accurately
1525 reflect how emissions are realized over time. To estimate emissions from degradation, in
1526 contrast to Tier 1, Tier 3 uses the stock difference approach where change in forest biomass
1527 stocks is directly estimated from repeated measures or models.
1528 Tier 2 is akin to Tier 1 in that it employs static forest biomass information, but it also
1529 improves on that approach by using country-specific data (i.e. collected within the national
1530 boundary), and by resolving forest biomass at finer scales through the delineation of more
1531 detailed strata. Also, like Tier 3, Tier 2 can modify the Tier 1 assumption that carbon stocks
1532 in woody vegetation, litter and deadwood are immediately emitted following deforestation
1533 (i.e. that stocks after conversion are zero), and instead develop disturbance matrices that
1534 model retention, transfers (e.g. from woody biomass to dead wood/litter) and releases (e.g.
1535 through decomposition and burning) among pools. For degradation, in the absence of
1536 repeated measures from a representative inventory, Tier 2 uses the gain-loss method using
1537 locally-derived data on mean annual increment. Done well, a Tier 2 approach can yield
1538 significant improvements over Tier 1 in precision achieved, and though not as precise as
1539 repeated measures using permanent plots that can focus directly on stock change and
1540 increment, Tier 2 does not require the sustained institutional backing.

1541 4.2.2 Data needs for each Tier

1542 The availability of data is another important consideration in the selection of an appropriate
1543 Tier. Tier 1 has essentially no data collection needs beyond consulting the IPCC tables and
1544 EFDB, while Tier 3 requires substantial mobilization of resources where no national forest
1545 inventory is in place (i.e. most developing countries). Data needs for each Tier are
1546 summarized in Table 4.1.
1547 Table 4.1: Data needs for meeting the requirements of the three IPCC Tiers

Data needs/examples of appropriate

Tier
biomass data
Default MAI* (for degradation) and/or forest
biomass stock (for deforestation) values for
broad continental forest types—includes six
Tier 1 (basic) classes for each continental area to
encompass differences in elevation and
general climatic zone; default values given
for all vegetation-based pools

MAI* and/or forest biomass values from

existing forest inventories and/or ecological
Tier 2 studies.
(intermediate)
Default values provided for all non-tree pools
Newly-collected forest biomass data.

Repeated measurements of trees from

permanent plots and/or calibrated process
Tier 3 (most
models. Can default data for other pools
demanding)
stratified by in-country regions and forest
type, or estimates from process models.

1548 * MAI = Mean annual increment of tree growth

49
1549 4.2.3 Selection of Tier

1550 Tiers should be selected on the basis of goals (e.g. precise measure of emissions reductions
1551 in the context of a performance-based incentives framework), the significance of the target
1552 source/sink, available data, and analytical capability.
1553 The IPCC recommends that it is good practice to use higher Tiers for the
1554 measurement of significant sources/sinks. To more clearly specify levels of data
1555 collection and analytical rigor among sources of emissions/removals, the IPCC Guidelines
1556 provide guidance on the identification of “Key Categories”. Key categories are sources of
1557 emissions/removals that contribute substantially to the overall national inventory and/or
1558 national inventory trends, and/or are key sources of uncertainty in quantifying overall
1559 inventory amounts or trends. Key categories can be further broken down to identify
1560 significant sub-categories or pools (e.g. above-ground biomass, below-ground biomass,
1561 litter, and dead wood) that constitute > 25-30 % emissions/removals for the category.
1562 Due to the balance of costs and the requirement for accuracy/precision in the carbon
1563 component of emission inventories, a Tier 2 methodology for carbon stock monitoring will
1564 likely be the most widely used in both the reference period and for future monitoring of
1565 emissions from deforestation and degradation. Although it is suggested that a Tier 3
1566 methodology be the level to aim for key categories and pools, in practice Tier 3 may be
1567 overly expensive to be widely used, at least in the near to mid term.
1568 On the other hand, Tier 1 will not deliver the accurate and precise measures demanded for
1569 key categories/pools by any mechanism in which economic incentives are foreseen.
1570 However, the principle of conservatism will likely represent a fundamental parameter to
1571 evaluate REDD estimates. In that case, a tier lower than required could be used – or a
1572 carbon pool could be ignored - if it can be soundly demonstrated that the overall estimate of
1573 reduced emissions are underestimated (further explanation is given in chapter 6.4).
1574 Different tiers can be applied to different pools where they have a lower importance. For
1575 example, where preliminary observations demonstrate that emissions from the litter or dead
1576 wood or soil carbon pool constitute less than 25% of emissions from deforestation, the Tier
1577 1 approach using default transfers and decomposition rates is justified for application to that
1578 pool.

1579 4.3 Stratification by Carbon Stocks

1580 Stratification refers to the division of any heterogeneous landscape into distinct sub-sections
1581 (or strata) based on some common grouping factor. In this case, the grouping factor is the
1582 stock of carbon in the vegetation. If multiple forest types are present across a country,
1583 stratification is the first step in a well-designed sampling scheme for estimating carbon
1584 emissions associated with deforestation and degradation over both large and small areas.
1585 Stratification is the critical step that will allow the association of a given area of
1586 deforestation and degradation with an appropriate vegetation carbon stock for the
1587 calculation of emissions.

1588 4.3.1 Why stratify?

1589 Different carbon stocks exist in different forest types and ecoregions depending on physical
1590 factors (e.g., precipitation regime, temperature, soil type, topography), biological factors
1591 (tree species composition, stand age, stand density) and anthropogenic factors (disturbance
1592 history, logging intensity). For example, secondary forests have lower carbon stocks than
1593 mature forests and logged forests have lower carbon stocks than unlogged forests.
1594 Associating a given area of deforestation with a specific carbon stock that is relevant to the

50
1595 location that is deforested or degraded will result in more accurate and precise estimates of
1596 carbon emissions. This is the case for all levels of deforestation assessment from a very
1597 coarse Tier 1 assessment to a highly detailed Tier 3 assessment.
1598 Because ground sampling is usually required to determine appropriate carbon estimates for
1599 the specific areas that were deforested or degraded, stratifying an area by its carbon stocks
1600 can increase accuracy and precision and reduce costs. National carbon accounting
1601 needs to emphasize a system in which stratification and refinement are based on carbon
1602 content (or expected reductions in carbon content) of specific forest types, not necessarily
1603 of forest vegetation. For example, the carbon stocks of a “tropical rain forest” (one
1604 vegetation class) may be vastly different with respect to carbon stocks depending on its
1605 geographic location and degree of disturbance.

1606 4.3.2 Approaches to stratification

1607 There are two different approaches for stratifying forests for national carbon accounting,
1608 both of which require some spatial information on forest cover within a country. In Approach
1609 A, all of a country’s forests are stratified ‘up-front’ and carbon measurements are made to
1610 produce a country-wide map of forest carbon stocks. At future monitoring events, only the
1611 activity data need to be monitored and combined with the pre-estimated difference in
1612 carbon stock values. In Approach B, a full land cover map of the whole country does not
1613 need to be created. Rather, carbon measurements are made at each monitoring event only
1614 in those areas that have undergone change. Which approach to use depends on a country’s
1615 access to relevant and up-to-date data as well as its financial and technological resources
1616 (see Box 4.4 for decision tree). Details of each approach are outlined below.

1617 Approach A: ‘Up-front’ stratification using existing or updated land cover maps
1618 The first step in stratifying by carbon stocks is to determine whether a national land cover
1619 or land use map already exists. This can be done by consulting with government agencies,
1620 forestry experts, universities, etc. who may have created these maps for other purposes.
1621 Before using the existing land cover or land use map for stratification, its quality and
1622 relevance should be assessed. For example:
1623 ‰ When was the map created? Land cover change is often rapid and therefore a land
1624 cover map that was created more than five years ago is most likely out-of-date and
1625 no longer relevant. If this is the case, a new land cover map should be created. To
1626 participate in REDD activities it is likely a country will need to have at least a land
1627 cover map for a relatively recent time (benchmark map—see Chapter 2.4).
1628 ‰ Is the existing map at an appropriate resolution for your country’s size and land
1629 cover distribution? Land cover maps derived from coarse-resolution satellite imagery
1630 may not be detailed enough for very small countries and/or for countries with a
1631 highly patchy distribution of forest area. For most countries, land cover maps derived
1632 from medium-resolution imagery (e.g., 30-m resolution Landsat imagery) are
1633 adequate (cf. Section 3).
1634 ‰ Is the map ground validated for accuracy? An accuracy assessment should be carried
1635 out before using any land cover map in additional analyses. Guidance on assessing
1636 the accuracy of remote sensing data is given in Chapter 3.
1637 Land cover and land use maps are sometimes produced for different purposes and therefore
1638 the classification may not be fully useable in its current form. For example, a land use map
1639 may classify all forest types as one broad ‘forest’ category, which would not be valuable for
1640 stratification unless more detailed information was available to supplement this map.
1641 Indicator maps are valuable for adding detail to broadly defined forest categories (see Box

51
1642 4.3 for examples), but should be used judiciously to avoid overcomplicating the issue. In
1643 most cases, overlaying one or two indicator maps (elevation and distance to transportation
1644 networks, for example) with a forest/non-forest land cover map should be adequate for
1645 delineating forest strata by carbon stocks.
1646 Once strata are delineated on a ground-validated land cover map and forest types have
1647 been identified, carbon stocks are estimated for each stratum using appropriate measuring
1648 and monitoring methods. A national map of carbon stocks can then be created (cf Section
1649 4.4).

1650 Box 4.3: Examples of maps on which a land use stratification can be built
1651 E co lo g i ca l z o ne m a p s
1652 One option for countries with virtually no data on carbon stocks is to stratify the
1653 country initially by ecological zone or ecoregion using global datasets. Examples of
1654 these maps include:
1655 1. Holdridge life zones (http://geodata.grid.unep.ch/)
1656 2. WWF ecoregions (http://www.worldwildlife.org/science/data/terreco.cfm)
1657 3. FAO ecological zones (http://www.fao.org/geonetwork/srv/en/main.home, type
1658 ‘ecological zones’ in search box)

1659
1660 Indicato r maps
1661 After ecological zone maps are overlain with maps of forest cover to delineate where
1662 forests within different ecological zones are located, there are several indicators that
1663 could be used for further stratification. These indicators can be either biophysically- or
1664 anthropogenically-based:
1665 Biophysical indicator maps Anthropogenic indicator maps:
1666 Elevation Distance to deforested land or forest edge
1667 Topography (slope and aspect) Distance to towns and villages
1668 Soils Proximity to transportation networks (roads, rivers)
1669 Forest Age (if known) Rural population density
1670 Areas of protected forest

1671

52
1672 In Approach A, all of the carbon measurements would be made once, up-front, i.e., at the
1673 beginning of monitoring program, and no additional carbon measurements would be
1674 necessary for the remainder of the monitoring period - only the activity data would need to
1675 be monitored. This does assume that the carbon stocks in the original forests being
1676 monitored would not change much over say a couple of decades—such a situation is likely
1677 to exist where most of the forests are relatively intact, have been subject to low intensity
1678 selective logging in the past, no major infrastructure exists in the areas, and/or are at a late
1679 secondary stage (> 40-50 years).
1680 As ecological zone maps are a global product, they tend to be very broad and hence certain
1681 features of the landscape that affect carbon stocks within a country are not accounted for.
1682 For example, a country with mountainous terrain would benefit from using elevation data
1683 (such as a digital elevation model) to stratify ecological zones into different elevational sub-
1684 strata because forest biomass is known to decrease with elevation. Another example would
1685 be to stratify the ecological zone map by soil type as forests on loamy soils tend to have
1686 higher growth potential than those on very sandy or clayey soils. If forest degradation is
1687 common in your country, stratifying ecological zones by distance to towns and villages or to
1688 transportation networks may be useful. (For an example of how to stratify a country with
1689 limited data, see Box 4.5.)

1690 Approach B: Continuous stratification based on a continuous carbon inventory

1691 Where wall-to-wall land cover mapping is not possible for stratifying forest area within a
1692 country by carbon stocks, regularly-timed “inventories” can be made by sampling only the
1693 areas subject to deforestation and degradation. Using this approach, a full land cover map
1694 for the whole country is not necessary because carbon assessment occurs only where land
1695 cover change has already occurred (forest to non-forest, or intact to degraded forest in
1696 some cases). Carbon measurements can then be made in neighboring pixels that have the
1697 same reflectance/textural characteristics as the pixels that had undergone change in the
1698 previous interval, serving as proxies for the sites deforested or degraded, and carbon
1699 emissions can be calculated.

1700 BOX 4.4: Decision tree for stratification approach

yes yes Is this map ground- yes

Do you have an existing Use
Was this map made truthed to
land cover map for the Approach
<5 years ago? acceptable levels of
whole country? A
accuracy?

no no yes no
yes

Are resources
Are resources
available to
available to update
ground-truth this
this map?
map?

Are resources no
available to create a yes
new land cover
map?
no Use no
Approach
B
1701

1702

53
1703 Box 4.5: Forest stratification in countries with limited data availability
1704 An example stratification scheme is shown here for the Democratic Republic of Congo.
1705 Step 1. Overlay a map of forest cover with an ecological zone map (A).
1706 Step 2. Select indicator maps. For this example, elevation (B) and distance to roads
1707 (C) were chosen as indicators.
1708 Step 3. Combine all factors to create a map of forest strata (D).

(A) (B)

1709

(C) (D)

Stratified Forest
Ecological zone/Elevation catagory/Accessibility category ( thousands ha)
Tropical dry/< 1,000 m/<10 km (155 ha)
Tropical dry/< 1,000 m/> 10 km (15 ha)
Tropical moist deciduous/< 1,000 m/<10 km (1,355 ha)
Tropical moist deciduous/< 1,000 m/> 10 km (1,823 ha)
Tropical moist deciduous/> 1,000 m/<10 km (2,446 ha)
Tropical moist deciduous/> 1,000 m/> 10 km (3,864 ha)
Tropical mountain system/< 1,000 m/<10 km (404 ha)
Tropical mountain system/< 1,000 m/> 10 km (466 ha)
Tropical mountain system/> 1,000 m/<10 km (1,885 ha)
Tropical mountain system/> 1,000 m/> 10 km (3,003 ha)
Tropical rainforest/< 1,000 m/<10 km (46,628 ha)
Tropical rainforest/< 1,000 m/> 10 km (77,332 ha)
Tropical rainforest/> 1,000 m/<10 km (845 ha)
Tropical rainforest/> 1,000 m/> 10 km (1,647 ha)
1710

1711

54
1712 This approach is likely the least expensive option as long as neighboring pixels to be
1713 measured are relatively easy to access by field teams. However, this approach is not
1714 recommended when vast areas of contiguous forest are converted to non-forest, because
1715 the forest stocks may have been too spatially variable to estimate a single proxy carbon
1716 value for the entire forest area that was converted. If this is the case, a conservative
1717 approach would be to use the lowest carbon stock estimate for the forest area that was
1718 converted to calculate emissions in the reference case and the highest carbon stock
1719 estimate in the monitoring phase.

1720 4.4 Estimation of Carbon Stocks of Forests Undergoing Change

1721 4.4.1 Decisions on which carbon pools to include

1722 The decision on which carbon pools to monitor as part of a REDD accounting scheme will
1723 likely be governed by the following factors:
1724 ‰ Available financial resources
1725 ‰ Availability of existing data
1726 ‰ Ease and cost of measurement
1727 ‰ The magnitude of potential change in the pool
1728 ‰ The principle of conservativeness
1729 Above all is the principle of conservativeness. This principle ensures that reports of
1730 decreases in emissions are not overstated. Clearly for this purpose both time zero and
1731 subsequent estimations must include exactly the same pools. Conservativeness also
1732 allows for pools to be omitted except for the dominant tree carbon pool and a precedent
1733 exists for Parties to select which pools to monitor within the Kyoto Protocol and Marrakesh
1734 Accords. For example, if dead wood or wood products are omitted then the assumption
1735 must be that all the carbon sequestered in the tree is immediately emitted and thus
1736 deforestation or degradation estimates are under-estimated. Likewise if CO2 emitted from
1737 the soil is excluded as a source of emissions; but as long as this exclusion is constant
1738 between the reference case and later estimations then no exaggeration of emissions occurs.

1739 4.4.1.1 Key categories

1740 The second deciding factor on which carbon pools to include should be the relative
1741 importance of the expected change in each of the carbon pools caused by deforestation and
1742 degradation. The magnitude of the carbon pool basically represents the magnitude of the
1743 emissions for deforestation as it is typically assumed that most of the pool is oxidized, either
1744 on or off site. For degradation the relationship is not as clear as usually only the trees are
1745 affected for most causes of degradation (cf. Section 3.3).
1746 In all cases it will make sense to measure trees, as trees are relatively easy to measure and
1747 will always represent a significant proportion of the total carbon stock. The remaining pools
1748 will represent varying proportions of total carbon depending on local conditions. For
1749 example, belowground biomass carbon (roots) and soil carbon to 30 cm depth represents
1750 26% of total carbon stock in estimates in tropical lowland forests of Bolivia but more than
1751 50 % in the peat forests of Indonesia (Figure 4.4 a & b9). It is also possible that which pools
1752 are included or not varies by forest type/strata within a country. It is possible that say
1753 forest type A in a given country could have relatively high carbon stocks in the dead wood

9
Unpublished data from measurements by Winrock

55
1754 and litter pools, whereas forest type B in the country could have low quantities in these
1755 pools—in this case it might make sense to measure these pools in the forest A but not B as
1756 the emissions from deforestation would be higher in A than in B.
Soil to 30 cm
depth
13%

Litter
2%

Understory
1%

Standing and lying Aboveground

dead wood trees
7% 41%

"Active" peat*
53%
Belowground
Aboveground
13%
trees
64%

Understory
0%

Dead wood
1757 6%

1758 Figure 4.4: LEFT- Proportion of total stock (202 t C/ha) in each carbon pool in Noel Kempff
1759 National Park, Bolivia, and RIGHT- Proportion of total stock (236 t C/ha) in each carbon pool
1760 in peat forest in Central Kalimantan, Indonesia (active peat includes soil organic carbon, live
1761 and dead roots and decomposing materials).
1762 Pools can be divided by ecosystem and land use change type into key categories or minor
1763 categories. Key categories represent pools that could account for more than 10% of the
1764 total emissions resulting from the deforestation or degradation (Table 4.2).
1765 Table 4.2: Broad guidance on key categories of carbon pools for determining assessment
1766 emphasis. Key category defined as pools potentially responsible for more than 25% of total
1767 emission resulting from the deforestation or degradation.

Biomass Dead organic matter Soils

Below- Soil organic
Aboveground Dead wood Litter
ground matter
Deforestation
To cropland KEY KEY (KEY) KEY

To pasture KEY KEY (KEY)

To shifting
KEY KEY (KEY)
cultivation

Degradation
Degradation KEY KEY

1768

1769 Certain pools such as soil carbon or even down dead material tend to be quite variable and
1770 can be relatively time consuming and costly to measure. The decision to include these pools
1771 would therefore be made based on whether they represent a key category and available
1772 financial resources.

56
1773 Soils will represent a key category in peat swamp forests and mangrove forests (cf Figure 4-
1774 4b) and carbon emissions are high when deforested (see Box 4-11). For forests on mineral
1775 soils with high organic carbon content and deforestation is to cropland, as much as 30% of
1776 the total soil organic matter stock will be lost in the top 30 cm or so during the first 5 years.
1777 Where deforestation is to pasture or shifting cultivation, the science does not support a
1778 large drop in soil carbon stocks.
1779 Dead wood is a key category in old growth forest where it can represent more than 10% of
1780 total biomass, in young successional forests, for example, it will not be a key category.
1781 For carbon pools representing a fraction of the total (<5 %) it may be possible to include
1782 them at low cost if good default data are available.
1783 Box 4.6 provides examples that illustrate the scale of potential emissions from just the
1784 aboveground biomass pool following deforestation and degradation in Bolivia, the Republic
1785 of Congo and Indonesia.

1786 Box 4.6: Potential emissions from deforestation and degradation in three
1787 example countries
1788 The following table shows the decreases in the carbon stock of living trees estimated
1789 for both deforestation, and degradation through legal selective logging for three
1790 countries: Republic of Congo, Indonesia, and Bolivia1: The large differences among
1791 the countries for degradation reflects the differences in intensity of timber extraction
1792 (about 3 to 22 m3/ha).

1793

1794 4.4.1.2 Defining carbon measurement pools:

1795 STEP 1: INCLUDE ABOVEGROUND TREE BIOMASS

1796 All assessments should include aboveground tree biomass as this pool is simple to measure
1797 and will almost always dominate carbon stock changes

1798 STEP 2: INCLUDE BELOWGROUND TREE BIOMASS

1799 Belowground tree biomass (roots) is almost never measured, but instead is included
1800 through a relationship to aboveground biomass (usually a root-to-shoot ratio). If the
1801 vegetation strata correspond with tropical or subtropical types listed in Table 4.3 (modified
1802 from Table 4.4 in IPCC GL AFOLU to exclude non-forest or non-tropical values and to
1803 account for incorrect values) then it makes sense to include roots.

1804 STEP 3: ASSESS THE RELATIVE IMPORTANCE OF ADDITIONAL CARBON POOLS

1805 Assessment of whether carbon pools represent key categories can be conducted via a
1806 literature review, discussions with universities or even field measurements from a few pilot
1807 plots following methodological guidance already provided in many of the sources given in
1808 this section.

57
1809 Table 4.3: Root to shoot ratios modified* from Table 4.4. in IPCC GL AFOLU

Above-
Root-to-
Domain Ecological Zone ground Range
shoot ratio
biomass
<125 t.ha-1 0.20 0.09-0.25
Tropical rainforest
>125 t.ha-1 0.24 0.22-0.33
Tropical
<20 t.ha-1 0.56 0.28-0.68
Tropical dry forest
>20 t.ha-1 0.28 0.27-0.28

Subtropical humid <125 t.ha-1 0.20 0.09-0.25

forest
>125 t.ha-1 0.24 0.22-0.33
Subtropical
Subtropical dry <20 t.ha-1 0.56 0.28-0.68
forest
>20 t.ha-1 0.28 0.27-0.28

1810 *the modification corrects an error in the table based on communications with Karel
1811 Mulroney, the lead author of the peer reviewed paper from which the data were extracted.

1812 STEP 4: DETERMINE IF RESOURCES ARE AVAILABLE TO INCLUDE ADDITIONAL

1813 POOLS
1814 When deciding if additional pools should be included or not, it is important to remember that
1815 whichever pools are decided on initially the same pools must be included in all future
1816 monitoring events. Although national or global default values can be used, if they are a key
1817 category they will make the overall emissions estimates more uncertain. However, it is
1818 possible that once a pool is selected for monitoring, default values could be used initially
1819 with the idea of improving these values through time, but even if just a one time
1820 measurement will be the basis of the monitoring scheme, there are costs associated with
1821 including additional pools. For example:
1822 ‰ for soil carbon—soil is collected and then must be analyzed in a laboratory for bulk
1823 density and percent soil carbon
1824 ‰ for non-tree vegetation—destructive sampling is usually employed with samples
1825 collected and dried to determine biomass and from biomass carbon stock
1826 ‰ for down dead wood—stocks are usually assessed along a transect with the
1827 simultaneous collection and subsequent drying of samples for density
1828 If the pool is a significant source of emissions it will be worth including it in the assessment
1829 if it is possible. An alternative to measurement for minor carbon pools (<10% of the total
1830 potential emission) is to include estimates from look-up tables of default data with high
1831 integrity (peer-reviewed)

1832 4.4.2 General approaches to estimation of carbon stocks

1833 4.4.2.1 STEP 1: Identify strata where assessment of carbon stocks is necessary-
1834 Not all forest strata are likely to undergo deforestation or degradation. For example, strata
1835 that are currently distant from existing deforested areas and/or inaccessible from roads or
1836 rivers are unlikely to be under immediate threat. Therefore, a carbon assessment of every

58
1837 forest stratum within a country would not be cost-effective because not all forests will
1838 undergo change.
1839 For stratification approach B (described above), where and when to conduct a carbon
1840 assessment over each monitoring period is defined by the activity data, with measurements
1841 taking place in nearby areas that currently have the same reflectance as the changed pixels
1842 had prior to deforestation or degradation . For stratification approach A, the best strategy
1843 would be to invest in carbon stock assessments for strata where there is a history or future
1844 likelihood of degradation or deforestation, not for strata where there is little deforestation
1845 pressure.
1846 SubStep 1 – For reference case (and future monitoring for approach B): establish sampling
1847 plans in areas representative of the areas with recorded deforestation and/or degradation.
1848 SubStep 2 – For future monitoring: identify strata where deforestation and/or degradation
1849 are likely. These will be strata adjoining existing deforested areas or degraded forest,
1850 and/or strata with human access via roads or easily navigable waterways. Establish
1851 sampling plans for these strata but, for the current period, do not invest in measuring
1852 forests that are hard to access such as areas that are distant to transportation routes,
1853 towns, villages and existing farmland, and/or areas at high elevations or that experience
1854 very heavy rainfall.

1855 4.4.2.2 STEP 2: Assess existing data

1856 It is likely that within most countries there will be some data already collected that could be
1857 used to define the carbon stocks of one or more strata. These data could be derived from a
1858 forest inventory or perhaps from past scientific studies. Proceed with incorporating these
1859 data if the following criteria are fulfilled:
1860 ‰ The data are less than 10 years old
1861 ‰ The data are derived from multiple measurement plots
1862 ‰ All species must be included in the inventories
1863 ‰ The minimum diameter for trees included is 30cm or less at breast height
1864 ‰ Data are sampled from good coverage of the strata over which they will be
1865 extrapolated
1866 Existing data that meet the above criteria should be applied across the strata from which
1867 they were representatively sampled and not beyond that. The existing data will likely be in
1868 one of two forms:
1869 ‰ Forest inventory data
1870 ‰ Data from scientific studies

1871 Forest inventory data

1872 Typically forest inventories have an economic motivation. As a consequence forest
1873 inventories worldwide are derived from good sampling design. If the inventory can be
1874 applied to a stratum, all species are included and the minimum diameter is 30 cm or greater
1875 then the data will be a high enough quality with sufficiently low uncertainty for inclusion.
1876 Inventory data typically comes in two different forms:
1877 Stand tables—these data from an inventory are potentially the most useful from which
1878 estimates of the carbon stock of trees can be calculated. Stand tables generally include a
1879 tally of all trees in a series of diameter classes. The method basically involves estimating the
1880 biomass per average tree of each diameter (diameter at breast height, dbh) class of the
1881 stand table, multiplying by the number of trees in the class, and summing across all classes.

59
1882 The mid-point diameter of the class can be used10 in combination with an allometric biomass
1883 regression equation. Guidance on choice of equation and application of equations is widely
1884 available (for example see sources in Box 4-9). For the open-ended largest diameter classes
1885 it is not obvious what diameter to assign to that class. Sometimes additional information is
1886 included that allows educated estimates to be made, but this is often not the case. The
1887 default assumption should be to assume the same width of the diameter class and take the
1888 midpoint, for example if the highest class is >110 cm and the other class are in 10 cm
1889 bands, then the midpoint to apply to the highest class should be 115 cm.
1890 It is important that the diameter classes are not overly large so as to decrease how
1891 representative the average tree biomass is for that class. Generally the rule should be that
1892 the width of diameter classes should not exceed 15 cm.
1893 Sometimes, the stand tables only include trees with a minimum diameter of 30 cm or more,
1894 which essentially ignores a significant amount of carbon particularly for younger forests or
1895 heavily logged. To overcome the problem of such incomplete stand tables, an approach has
1896 been developed for estimating the number of trees in smaller diameter classes based on
1897 number of trees in larger classes11. It is recommended that the method described here be
1898 used for estimating the number of trees in one to two small classes only to complete a
1899 stand table to a minimum diameter of 10 cm.

1900 Box 4.7: Adding diameter classes to truncated stand tables

1901

1902 dbh class 1= 30-39 cm, and

1903 dbh class 2= 40-49 cm
1904 Ratio = 35.1/11.8
1905 = 2.97
1906 Therefore, the number of trees in the 20-29 cm class is: 2.97 x 35.1 = 104.4
1907 To calculate the 10-19 cm class: 104.4/35.1 = 2.97,
1908 2.97 x 104.4 = 310.6

1909 The method is based on the concept that uneven-aged forest stands have a characteristic
1910 "inverse J-shaped" diameter distribution. These distributions have a large number of trees in

10
If information on the basal area of all the trees in each diameter class is provided, instead of using
the mid point of the diameter class the quadratic mean diameter (QMD) can be used instead—this is
the diameter of the tree with the average basal area (=basal area of trees in class/#trees).
11
Gillespie, A. J. R, S. Brown, and A. E. Lugo. 1992. Tropical forest biomass estimation from truncated
stand tables. Forest Ecology and Management 48:69-88.

60
1911 the small classes and gradually decreasing numbers in medium to large classes. The best
1912 method is the one that estimated the number of trees in the missing smallest class as the
1913 ratio of the number of trees in dbh class 1 (the smallest reported class) to the number in
1914 dbh class 2 (the next smallest class) times the number in dbh class 1. This method is
1915 demonstrated in the Box 4-7.
1916 Stock tables—a table of the merchantable volume, often by diameter class or total per
1917 hectare. If stand tables are not available, it is likely that volume data are if a forestry
1918 inventory has been conducted somewhere in the country. In many cases volumes given will
1919 be of just commercial species. If this is the case then these data can not be used for
1920 estimating carbon stocks, as a large and unknown proportion of total volume and therefore
1921 total biomass is excluded.
1922 Biomass density can be calculated from volume over bark of merchantable growing stock
1923 wood (VOB) by "expanding" this value to take into account the biomass of the other
1924 aboveground components—this is referred to as the biomass conversion and expansion
1925 factor (BCEF). When using this approach and default values of the BCEF provided in the
1926 IPCC AFOLU, it is important that the definitions of VOB match. The values of BCEF for
1927 tropical forests in the AFOLU report are based on a definition of VOB as follows:
1928 Inventoried volume over bark of free bole, i.e. from stump or buttress to crown point or first
1929 main branch. Inventoried volume must include all trees, whether presently commercial or
1930 not, with a minimum diameter of 10 cm at breast height or above buttress if this is higher.
1931 Aboveground biomass (t/ha) is then estimated as follows: = VOB * BCEF12
1932 where:
1933 BCEF t/m³ = biomass conversion and expansion factor (ratio of aboveground oven-dry
1934 biomass of trees [t/ha] to merchantable growing stock volume over bark [m³/ha]).
1935 Values of the BCEF are given in Table 4.5 of the IPCC AFOLU, and those relevant to tropical
1936 humid broadleaf and pine forests are shown in the Table 4.4.
1937 Table 4.4: Values of BCEF for application to volume data. (Modified from Table 4.5 in IPCC
1938 AFOLU.)

Growing stock volume –average and range (VOB, m³/ha)

Forest type
<20 21-40 41-60 61-80 80-120 120-200 >200
Natural 4.0 2.8 2.1 1.7 1.5 1.3 1.0
broadleaf 2.5-12.0 1.8-304 1.2-2.5 1.2-2.2 1.0-1.8 0.9-1.6 0.7-1.1

1.8 1.3 1.0 0.8 0.8 0.7 0.7

Conifer
1.4-2.4 1.0-1.5 0.8-1.2 0.7-1.2 0.6-1.0 1.6-0.9 0.6-0.9

1939

1940 In cases where the definition of VOB does not match exactly the definition given above, a
1941 range of BCEF values are given:
1942 ‰ If the definition of VOB also includes stem tops and large branches then the lower
1943 bound of the range for a given growing stock should be used

12
This method from the IPCC AFOLU replaces the one reported in the IPCC GPG. The GPG method
uses a slightly different equation :AGB = VOB*wood density*BEF; where BEF, the biomass expansion
factor, is the ratio of aboveground biomass to biomass of the merchantable volume in this case.

61
1944 ‰ If the definition of VOB has a large minimum top diameter or the VOB is comprised
1945 of trees with particularly high basic wood density then the upper bound of the range
1946 should be used
1947 Forest inventories often report volumes to a minimum diameter greater than 10 cm. These
1948 inventories may be the only ones available. To allow the inclusion of these inventories,
1949 volume expansion factors (VEF) were developed. After 10 cm, common minimum diameters
1950 for inventoried volumes range between 25 and 30 cm. Due to high uncertainty in
1951 extrapolating inventoried volume based on a minimum diameter of larger than 30 cm,
1952 inventories with a minimum diameter that is higher than 30 cm should not be used. Volume
1953 expansion factors range from about 1.1 to 2.5, and are related to the VOB30 as follows to
1954 allow conversion of VOB30 to a VOB10 equivalent:
1955 VEF = Exp{1.300 - 0.209*Ln(VOB30)} for VOB30 < 250 m3/ha
1956 = 1.13 for VOB30 > 250 m3/ha
1957 See Box 4-8 for a demonstration of the use of the VEF correction factor and BCEF to
1958 estimate biomass density.

1959 Box 4.8: Use of volume expansion factor (VEF) and biomass conversion and
1960 expansion factor (BCEF)
1961 Tropical broadleaf forest with a VOB30 = 100 m³/ha
1962 First: Calculate the VEF
1963 = Exp {1.300 - 0.209*Ln(100)} = 1.40
1964 Second: Calculate VOB10
1965 = 100 m³/ha x 1.40 = 140 m³/ha
1966 Third: Take the BCEF from the table above
1967 = Tropical hardwood with growing stock of 140 m³/ha = 1.3
1968 Fourth: Calculate aboveground biomass density
1969 = 1.3 x 140
1970 = 182 t/ha

1971 Data from scientific studies

1972 Scientific evaluations of biomass, volume or carbon stock are conducted under multiple
1973 motivations that may or may not align with the stratum-based approach required for
1974 deforestation and degradation assessments.
1975 Scientific plots may be used to represent the carbon stock of a stratum as long as there are
1976 multiple plots and the plots are randomly located. Many scientific plots will be in old growth
1977 forest and may provide a good representation of this stratum.
1978 The acceptable level of uncertainty will be defined in the political arena, but quality of
1979 research data could be illustrated by an uncertainty level of 20% or less (95% confidence
1980 equal to 20% of the mean or less). If this level is reached then these data should be
1981 applied.

1982 4.4.2.3 STEP 3: Collect missing data

1983 It is likely that even if data exist they will not cover all strata so in almost all situations a
1984 new measuring and monitoring plan will need to be designed and implemented to achieve a
1985 Tier 2 level. With careful planning this need not be an overly costly proposition.

62
1986 The first step would be a decision on how many strata with deforestation or degradation in
1987 the reference period are at risk of deforestation or degradation in the future but do not have
1988 estimates of carbon stock. These strata should then be the focus of any future monitoring
1989 plan. Many resources are available or becoming available to assist countries in planning and
1990 implementing the collection of new data to enable them to estimate forest carbon stocks
1991 with high confidence (e.g. bilateral and multilateral organizations, FAO etc.), sources of such
1992 information and guidance is given in Box 4.9).

1993 Box 4.9: Guidance on collecting new carbon stock data

1994 Many resources are available to countries and organizations seeking to conduct carbon
1995 assessments of land use strata.
1996 The Food and Agriculture Organization of the United Nations has been supporting
1997 forest inventories for more than 50 years. The FAO National Forest Inventory Field
1998 Manual is available at:
1999 http://www.fao.org/docrep/008/ae578e00.htm
2000 Specific guidance on field measurement of carbon stocks can be found in Chapter 4.3
2001 of GPG LULUCF and also in the World Bank Sourcebook for Land Use, Land-Use
2002 Change and Forestry (available at:
2003 http://carbonfinance.org/doc/LULUCF_sourcebook_compressed.pdf )

2004 Creating a national look-up table

2005 A cost-effective, good practice method for Approach A and Approach B stratifications may
2006 be to create a “national look-up table” for the country that will detail the carbon stock in
2007 each selected pool in each stratum. Look-up tables should ideally be updated periodically to
2008 account for changing mean biomass stocks due to shifts in age distributions, climate, and or
2009 disturbance regimes. The look up table can then be used through time to detail the pre-
2010 deforestation or degradation stocks and estimated stocks after deforestation and
2011 degradation. An example is given in Box 4.10.

63
2012 Box 4.10: A national look up table for deforestation and degradation
2013 The following is a hypothetical strata look-up table for use with approach A or
2014 approach B stratification. We can assume that remote sensing analysis reveals that
2015 800 ha of lowland forest were deforested to shifting agriculture and 500 ha of
2016 montane forest were degraded. Using the national look-up table results in the
2017 following:
2018 The loss for deforestation would be
2019 154 t C/ha – 37 t C/ha = 117 t C/ha x 800 ha =93,600 t C.
2020 The loss for the degradation would be
2021 130 t C/ha – 92 t C/ha = 38 t C/ha x 500 ha =19,000 t C
2022 (Note that degradation will often have been caused by harvest and therefore
2023 emissions will be decreased if storage in long-term wood products was included—that
2024 is the harvested wood did not enter the atmosphere.)

2025

2026 4.4.3 Guidance on carbon in soils

2027 IPCC AFOLU divides soil carbon into three pools: mineral soil organic carbon, organic soil
2028 carbon, and mineral soil inorganic carbon. The focus in this section will be on only the
2029 organic component of soil.

2030 4.4.3.1 Explanation of IPCC Tiers for soil carbon estimates

2031 For estimating emissions from mineral soil organic carbon, the IPCC AFOLU recommends the
2032 stock change approach but for organic soil carbon, an emission factor approach is used
2033 (Table 4.5). For mineral soil organic carbon, departures in carbon stocks from a reference or
2034 base condition are calculated by applying stock change factors (specific to land-use,
2035 management practices, and inputs (e.g. soil amendment, irrigation, etc.)), equal to the
2036 carbon stock in the altered condition as a proportion of the reference carbon stock. Tier 1

64
2037 assumes that a change to a new equilibrium stock occurs at a constant rate over a 20 year
2038 time period. Tiers 2 and 3 may vary these assumptions, in terms of the length of time over
2039 which change takes place, and in terms of how annual rates vary within that period. Tier 1
2040 assumes that the maximum depth beyond which change in soil carbon stocks should not
2041 occur is 30 cm; Tiers 2 and 3 may lower this threshold to a greater depth.
2042 Tier 1 further assumes that there is no change in mineral soil carbon in forests remaining
2043 forests. Hence, estimates of the changes in mineral soil carbon could be made for
2044 deforestation but are not needed for degradation. Tiers 2 and 3 allow this assumption to
2045 change, and the estimates of changes in mineral soil carbon resulting from forest
2046 management are modeled. In the case of degradation, the tier 2 and 3 approaches are only
2047 recommended for intensive practices that involve significant soil disturbance, not typically
2048 encountered in selective logging. In contrast, selective logging of forests growing on organic
2049 carbon soils such as the peat-swamp forests of South East Asia could result in large
2050 emissions caused by practices such as draining to remove the logs from the forest (see Box
2051 4.11 for further details on this topic).
2052 Table 4.5: IPCC guidelines on data and/or analytical needs for the different Tiers for soil
2053 carbon changes in deforested areas.
Soil carbon
Tier 1 Tier 2 Tier 3
pool
Validated model or
Default reference
Country-specific data on direct measures of
Mineral soil C stocks and stock
reference C stocks & stock change
organic carbon change factors
stock change factors through monitoring
from IPCC
networks
Validated model or
Organic soil Default emission Country-specific data on
direct measures of
carbon factor from IPCC emission factors
stock change
2054

2055 Variability in soil carbon stocks can be large; Tier 1 reference stock estimates have
2056 associated errors of +/- 90%. Therefore it is clear that if soil is a key category, Tier 1
2057 estimates should be avoided.

2058 4.4.3.2 When and how to generate a good Tier 2 analysis for soil carbon
2059 Modifying Tier 1 assumptions and replacing default reference stock and stock change
2060 estimates with country-specific values through Tier 2 methods is recommended to reduce
2061 uncertainty for significant sources. Tier 2 provides the option of using a combination of
2062 country-specific data and IPCC default values that allows a country to more efficiently
2063 allocate its limited resources in the development of emission inventories. Assessments of
2064 opportunities to improve on Tier 1 assumptions with a Tier 2 approach are summarized in
2065 Table 4.6.

65
2066 Table 4.6: Opportunities to improve on Tier 1 assumptions using a Tier 2 approach.
Tier 1
Tier 2 options Recommendation
assumptions
Not recommended. There is
seldom any benefit in sampling to
Depth to which deeper depths for tropical forest
change in May report changes to soils because impacts of land
30 cm
stock is deeper depths conversion and management on
reported soil carbon tend to diminish with
depth - most change takes place
in the top 25-30 cm.
May vary the length of Recommended where
time until new chronosequence or long-term
Time until new
equilibrium is achieved, study data are available. Some
equilibrium
20 years referencing country- soils may reach equilibrium in as
stock is
specific little as 5-10 years after
reached
chronosequences or conversion, particularly in the
long-term studies humid tropics13.
Not recommended – best modeled
with Tier 3-type approaches. As
well, a typical 5-year reporting
Rate of change May use non-linear interval effectively “linearizes” a
Linear
in stock models non-linear model and would undo
the benefits of a model with finer
resolution of varying annual
changes.
Develop country-
specific reference
stocks consulting other
available databases or IPCC defaults comprehensive. Not
Reference
IPCC defaults consolidating country recommended unless country-
stocks
soil data from existing specific data are available.
sources (universities,
agricultural extension
services, etc.).
IPCC defaults fairly
Develop country-
comprehensive. Not recommended
specific stock change
Stock change unless significant areas (that can
IPCC defaults factors from
factors be delineated spatially) are
chronosequence or
represented by drainage as a
long-term study.
typical conversion practice.
2067

2068 The IPCC default values for reference soil carbon stocks and stock change factors are
2069 comprehensive and reflect the most recent review of changes in soil carbon with conversion
2070 of native soils. Reference stocks and stock change factors represent average conditions
2071 globally, which means that, in at least half of the cases, use of a more accurate and precise
2072 (higher Tier) approach will not produce a higher estimate of stocks or emissions than the
2073 Tier 1 defaults with respect to the categories covered.

13
Detwiler, R. P. 1986. Land use change and the global carbon cycle: the role of tropical soils.
Biogeochemistry 31: 1-14.

66
2074 Where country-specific data are available from existing sources, Tier 2 reference stocks
2075 should be constructed to replace IPCC default values. Measurements of soil carbon data can
2076 be acquired through consultations with local universities or agricultural departments or
2077 extension agencies, both of which often carry out soil surveying at scales suited to deriving
2078 national or regional level estimates. It should be acknowledged however that because
2079 agricultural extension work is targeted to altered (cultivated) sites, agricultural extension
2080 agencies may have comparatively little information gathered on reference soils under native
2081 vegetation. Where data on reference sites are available, it would be advantageous if the soil
2082 carbon measurements were geo-referenced. Soil carbon data generated through typical
2083 agricultural extension work is often limited to carbon concentrations (i.e. percent carbon)
2084 only, and for this information to be usable, carbon concentrations must be paired with soil
2085 bulk density (mass per unit volume) and volume of fragments > 2 mm to derive a mass C
2086 per unit volume soil (see Ch. 4.3 of the IPCC GPG report for more details about soil
2087 samples).
2088 A spatially-explicit global database of soil carbon is also available from which country-
2089 specific estimates of reference stocks can be sourced. The ISRIC World Inventory of Soil
2090 Emission (WISE) Potential Database offers 5 x 5 minute grid resolution of soil organic
2091 carbon content and bulk density to 30 cm depth, and can be accessed online at:
2092 http://www.isric.org/UK/About+Soils/Soil+data/Geographic+data/Global/WISE5by5minutes.htm
2093

2094 A soil carbon map is also available from the US Department of Agriculture, Natural
2095 Resources Conservation Service (Figure 4.5). This map is based on a reclassification of the
2096 FAO-UNESCO Soil Map of the World combined with a soil climate map. The soil organic
2097 carbon map shows the distribution of the soil organic carbon to 1 meter depth, and can be
2098 downloaded from: http://soils.usda.gov/use/worldsoils/mapindex/soc.html

2099

2100 Figure 4.5: Soil organic carbon map (kg/m2 or x10 t/ha; to 100 cm depth) extracted from
2101 the global map produced by the USDA Natural Resources Conservation Service.
2102 Existing map sources (e.g. Figure 4.5) are particularly useful for developing estimates of
2103 historical values. Moving forward, new country- and strata-specific reference stocks can be
2104 generated from field measurements. Maps such as those described above can assist a
2105 country determine whether changes in soil carbon stocks after deforestation would be a key
2106 category or not. Deforestation on soils with high carbon stocks could emit up to 30-40% of
2107 their stock in the top 30 cm during the first 5 years or so after clearing14. Once strata are

14
Detwiler, R. P. 1986. Land use change and the global carbon cycle: the role of tropical soils.
Biogeochemistry 31: 1-14.

67
2108 identified for estimating reference soil carbon stocks, soil samples could be collected using a
2109 representative sampling approach and used to quantify an average stock value for each
2110 stratum in terms of mass of carbon per unit area (Box 4-9 for sources of additional guidance
2111 on soil carbon sampling). Efficient use of limited resources for generating new local
2112 reference carbon stocks or stock change factors can be improved by targeting areas for
2113 sampling that can serve as proxies for areas where deforestation is more common.
2114 There are two factors not included in the IPCC defaults that can potentially influence carbon
2115 stock changes in soils: soil texture and soil moisture. Soil texture has an acknowledged
2116 effect on soil organic carbon stocks, with coarse sandy soils (e.g. spodosols) having lower
2117 carbon stocks in general than finer texture soils such as loams or clayey soils. Thus the
2118 texture of the soil is a useful indicator to determine the likely quantity of carbon in the soil
2119 and the likely amount emitted as CO2 upon conversion. Specifically, soil carbon in coarse
2120 sandy soils, with less capacity for soil organic matter retention, is expected to oxidize more
2121 rapidly and possibly to a greater degree than in finer soils. However, because coarser soils
2122 also tend to have lower initial (reference) soil carbon stocks, conversion of these soils is
2123 unlikely to be a significant source of emissions and therefore development of a soil texture-
2124 specific stock change factor is not recommended for these soils.
2125 Drainage of a previously inundated mineral soil increases decomposition of soil organic
2126 matter, just as it does in organic soils, and unlike the effect of soil texture, is likely to be
2127 associated with high reference soil carbon stocks. These are reflected in the IPCC default
2128 reference stocks for forests growing on wetland soils, such as floodplain forests. Drainage of
2129 forested wetland soils in combination of deforestation can thus represent a significant
2130 source of emissions. Because this factor is lacking from the IPCC default stock change
2131 factors, its effects would not be discerned using a Tier 1 approach. In other words, IPCC
2132 default stock change factors would underestimate soil carbon emissions where deforestation
2133 followed by drainage of previously inundated soils occurred. Where drainage practices on
2134 wetland soils are representative of national trends and significant areas, and for which
2135 spatial data are available, the Tier 2 approach of deriving a new, country-specific stock
2136 change factor from chronosequences or long-term studies is recommended.
2137 Field measurements can be used to construct chronosequences that represent changes in
2138 land cover and use, management or carbon inputs, from which new stock change factors
2139 can be calculated, and many sources of methods are available as previously mentioned.
2140 Alternatively, stock change factors can be derived from long-term studies that report
2141 measurements collected repeatedly over time at sites where land-use conversion has
2142 occurred. Ideally, multiple paired comparisons or long-term studies would be done over a
2143 geographic range comparable to that over which a resulting stock change factor will be
2144 applied, though they do not require representative sampling as in the development of
2145 average reference stock values.
2146 Deforestation of peat swamp forests (on organic soils) represent a special case and
2147 guidance is given in Box 4.11.

68
2148 Box 4.11. Emissions as a result of land use change in peat swamp forests
2149 Peat swamp forests are found throughout Southeast Asia (Figure A). Under natural
2150 conditions, the water table depth is near the peat surface and dead organic matter
2151 accumulates under these waterlogged conditions. Many of these peat forests have
2152 been destroyed due to degradation from logging pressure, deforestation for
2153 agriculture, and burning from past land use change. In addition to the aboveground
2154 emissions that result from clearing the forest vegetation, emissions from peat continue
2155 through time because drainage causes a lowering of the water table, causing a release
2156 of CO2 into the atmosphere from peat oxidation (Figure B). If the water table is
2157 lowered by of 0.8 meters by draining, CO2 emissions are estimated at 73 tons per
2158 hectare per year. As the peat drains, it dries out and becomes more susceptible to
2159 burning. In the well-publicized 1997 fires in Indonesia, the average depth of peat
2160 burned in Central Kalimantan was 0.5 meters, resulting in a release of approximately
2161 929 t CO2/ha (253 t C/ha)15.

2162

2163 Figure A. Extent of lowland peat forests in Southeast Asia. The Wetlands International
2164 data have higher detail and accuracy than the FAO data.16
140
y = 0.91x
CO2 emissions (t CO2/ha/yr)

120

100

80

60

40

20

0
-20 0 20 40 60 80 100 120
Drainage depth (cm)
2165

15
Page et al. (2002)
16
Hooijer, A., Silvius, M., Wösten, H. and Page, S. (2006): PEAT-CO2, Assessment of CO2 emissions
from drained peatlands in SE Asia. Delft Hydraulics report Q3943 (2006).

69
2166 Figure B. Relation between drainage depth and CO2 emissions from decomposition
2167 (fires excluded) in tropical peatlands17. Note that the average water table depth in a
2168 natural peatland is near the soil surface (by definition, as vegetation matter only
2169 accumulates to form peat under waterlogged conditions).

2170 4.5 Uncertainty

2171 The uncertainty of carbon estimates should be quantified following Chapter 5 of IPCC GPG
2172 LULUCF and briefly described here. Confidence in estimates of emission reductions can only
2173 arise if the uncertainty of the estimates is included.
2174 The uncertainty of separate components of the total carbon is defined relative to the 95 %
2175 confidence interval around the mean. The 95% confidence interval expresses the range in
2176 which the true value will lie with statistical certainty.
2177 The Tier 1 method for combining separate uncertainties to give a total uncertainty is
2178 “Simple Propagation of Errors”. Under this method the total uncertainty is equal to the
2179 square root of the sum of the squares of each of the component uncertainties.
2180 Where the same units are being combined such as when the total uncertainty from the
2181 combined carbon pools are being assessed, then the 95 % confidence interval should be
2182 used. However, where different units are employed such as carbon biomass and forest area,
2183 uncertainty is equal to the 95% confidence interval as a percentage of the mean ((95%
2184 confidence interval/mean) x 100).

U total = U 12 + U 22 + .... + U n2
2185

2186 Where:
2187 Utotal = total uncertainty
2188 Ui = uncertainty associated with each of the component quantities
2189 This method should be used with caution if there is a high level of correlation between
2190 components of the total error or if any of the component uncertainties is high (a standard
2191 deviation greater than 30% of the mean). Even if these tests are failed the equation can still
2192 be used to give approximate results.
2193 The Tier 2 method is a Monte Carlo type analysis. Monte Carlo analyses model uncertainty
2194 through selecting random values from probability distributions for parameters and
2195 measuring the effect on total stocks. Either training in the use of software packages that
2196 automatically provide Monte Carlo type analyses or contracting an expert in Monte Carlo
2197 analysis is required to implement this higher level method.
2198 All assessments should include at least a simple Tier 1-type of analysis of propagation of
2199 uncertainties. An example is shown in Box 4.12.

17
Hooijer, A., Silvius, M., Wösten, H. and Page, S. 2006 PEAT-CO2, Assessment of CO2 emissions
from drained peatlands in SE Asia. Delft Hydraulics report Q3943 (2006).

70
2200 BOX 4.12: Example of a Tier 1 uncertainty analysis

2201

2202 Therefore the total stock is 138 t C/ha and the uncertainty =

2203 112 + 3 2 + 2 2 = 11.6tC / ha

2204

2205 Therefore the total carbon stock over the stratum is:
2206 8564 * 138 = 1,181,832 t C
2207 And the uncertainty =

2208
14 2 + 8 2 = 15.9%
2209 15.9% of 1,181,832 = 188,165 t C

2210

71
2211 5 METHODS FOR ESTIMATING CO2 EMISSIONS FROM
2212 DEFORESTATION AND FOREST DEGRADATION

2213 5.1 Scope of this Chapter

2214 This chapter describes the methodologies that can be used to estimate carbon emissions
2215 from deforestation and forest degradation. It builds on Chapters 3 and 4 of this Sourcebook,
2216 which describe procedures for collecting the input data for these methodologies, namely
2217 areas of land use and land-use change (Chapter 3), and carbon stocks and changes in
2218 carbon stocks (Chapter 4).
2219 The methodologies described here are derived from the 2006 IPCC AFOLU Guidelines and
2220 the 2003 IPCC GPG-LULUCF, and focus on the Tier 2 IPCC methods, as these require
2221 country-specific data but do not require expertise in complex models or detailed national
2222 forest inventories.
2223 The AFOLU Guidelines and GPG-LULUCF define six categories of land use18 that are further
2224 sub-divided into subcategories of land remaining in the same category (e.g., Forest Land
2225 Remaining Forest Land) and of land converted from one category to another (e.g., Land
2226 converted to Cropland). The land conversion subcategories are then divided further based
2227 on initial land use (e.g., Forest Land converted to Cropland, Grassland converted to
2228 Cropland). This structure was designed to be broad enough to classify all land areas in each
2229 country and to accommodate different land classification systems among countries. The
2230 structure allows countries to account for, and track over time, their entire land area, and
2231 enables greenhouse gas estimation and reporting to be consistent and comparable among
2232 countries. For REDD estimation, each subcategory could be further subdivided by climatic,
2233 ecological, soils, and/or anthropogenic disturbance factors, depending upon the level of
2234 stratification chosen for area change detection and carbon stock estimation (see Chapters 3
2235 and 4).
2236 For the purposes of this Sourcebook, five IPCC land-use subcategories are relevant.
2237 Although the term deforestation within the REDD mechanism remains to be defined, it is
2238 likely to be encompassed by the four land-use change subcategories defined for conversion
2239 of forests to non-forests, namely “Forest Land Converted to Cropland,” “Forest Land
2240 Converted to Grassland,” “Forest Land Converted to Settlements,” and “Forest Land
2241 Converted to Other Land.”19 Forest degradation, or the long-term loss of carbon stocks that
2242 does not qualify as deforestation (i.e., carbon stock loss that does not cross the threshold
2243 below which a forest is no longer defined as a forest) is encompassed by the IPCC land-use
2244 subcategory “Forest Land Remaining Forest Land.” The methodologies that are presented
2245 here are based on the sections of the AFOLU Guidelines and the GPG-LULUCF that pertain to
2246 these land-use subcategories.
2247 Within each land-use subcategory, the IPCC methods track changes in carbon stocks in five
2248 pools (see Chapter 4). The IPCC emission/removal estimation methodologies cover all of
2249 these carbon pools. Total net carbon emissions equal the sum of emissions and removals for
2250 each pool. However, as is discussed in Chapter 4, REDD accounting schemes may or may

18
The names of these categories are a mixture of land-cover and land-use classes, but are collectively
referred to as ‘land-use’ categories by the IPCC for convenience.
19
The subcategory “Land Converted to Wetlands” includes the conversion of forest land to flooded
land, but as this land-use change is unlikely to be important in the context of REDD accounting, and
measurements of emissions from flooded forest lands are relatively scarce and highly variable, this
land-use change is not addressed further in this chapter.

72
2251 not include all carbon pools. Which pools to include will depend on decisions by policy
2252 makers the could be driven by such factors as financial resources, availability of existing
2253 data, ease and cost of measurement, and the principle of conservativeness.

2254 5.2 Linkage to 2006 IPCC Guidelines

2255 Table 5-1 lists the sections of the AFOLU Guidelines that describe carbon estimation
2256 methods for each land-use subcategory. This table is provided to facilitate searching for
2257 further information on these methods in the AFOLU Guidelines, which can be difficult given
2258 the complex structure of this volume of the 2006 IPCC Guidelines. To review greenhouse
2259 gas estimation methods for a particular land-use category in the AFOLU Guidelines, one
2260 must refer to two separate chapters: a generic methods chapter (Chapter 2) and the land-
2261 use category chapter specific to that land-use category (i.e., either Chapter 4, 5, 6, 7, 8, or
2262 9). The methods for a particular land-use subcategory are contained in sections in each of
2263 these chapters.
2264 Table 5.1: Locations of Carbon Estimation Methodologies in the 2006 AFOLU Guidelines
Sections in Sections in
Land-Use Category Land-Use
Relevant Land-Use Generic
(Relevant Land-Use Subcategory
Category Chapter Methods
Category Chapter in (Subcategory
(Chapter 4, 5, 6, 8, Chapter
AFOLU Guidelines) Acronym)
or 9) (Chapter 2)
Forest Land Forest Land 4.2.1 2.3.1.1
(Chapter 4) Remaining Forest 4.2.2 2.3.2.1
Land (FF) 4.2.3 2.3.3.1.
Cropland Land Converted to 5.3.1 2.3.1.2
(Chapter 5) Cropland (LC) 5.3.2 2.3.2.2
5.3.3 2.3.3.1
Grassland Land Converted to 6.3.1 2.3.1.2
(Chapter 6) Grassland (LG) 6.3.2 2.3.2.2
6.3.3 2.3.3.1
Settlements Land Converted to 8.3.1 2.3.1.2
(Chapter 8) Settlements (LS) 8.3.2 2.3.2.2
8.3.3 2.3.3.1
Other Land Land Converted to 9.3.1 2.3.1.2
(Chapter 9) Other Land (LO) 9.3.2 2.3.2.2
9.3.3 2.3.3.1
2265

2266 Information and guidance on uncertainties relevant to estimation of emissions from land use
2267 and land-use change are located in various chapters of two separate volumes of the 2006
2268 IPCC Guidelines. Chapter 3 of the General Guidance and Reporting volume (Volume 1) of
2269 the 2006 IPCC Guidelines provides detailed, but non-sector-specific, guidance on sources of
2270 uncertainty and uncertainty estimation methodologies. Land-use subcategory-specific
2271 information about uncertainties for specific carbon pools and land uses is provided in each of
2272 the land-use category chapters (i.e., Chapter 4, 5, 6, 7, 8, or 9) of the AFOLU Guidelines
2273 (Volume 4).

73
2274 5.3 Organization of this Chapter

2275 The remainder of this chapter discusses carbon emission estimation for deforestation and
2276 forest degradation:
2277

2278 ‰ Section 5.4 addresses basic issues related to carbon estimation, including the
2279 concept of carbon transfers among pools, emission units, and fundamental
2280 methodologies for estimating annual changes in carbon stocks.
2281 ‰ Section 5.5 describes methods for estimating carbon emissions from deforestation
2282 based on the generic IPCC methods for land converted to a new land-use category,
2283 and on the IPCC methods specific to types of land-use conversions from forests, i.e.,
2284 “Forest Land Converted to Cropland” (FC), “Forest Land Converted to Grassland”
2285 (FG), “Forest Land Converted to Settlements” (FS), and “Forest Land Converted to
2286 Other Land” (FO).
2287 ‰ Section 5.6 describes methods for estimating carbon emissions from forest
2288 degradation based on the generic IPCC methods for land remaining in a land-use
2289 category, and on the IPCC methods specific to “Forest Land Remaining Forest Land.”
2290 ‰ Section 5.7 describes methods for estimating uncertainties.

2291 5.4 Fundamental Carbon Estimating Issues

2292 The overall carbon estimating method used here is one in which net changes in carbon
2293 stocks in the five terrestrial carbon pools are tracked over time. For each strata or sub-
2294 division of land area within a land-use category, the sum of carbon stock changes in all the
2295 pools equals the total carbon stock change for that stratum. In the REDD context,
2296 discussions center on gross emissions thus estimating the decrease in total carbon stocks,
2297 which is equated with emissions of CO2 to the atmosphere, is all that is needed a this time.
2298 However, a decrease in stocks in an individual pool may or may not represent an emission
2299 to the atmosphere because an individual pool can change due to both carbon transfers to
2300 and from the atmosphere, and carbon transfers to another pool (e.g., the transfer of
2301 biomass to dead wood during logging). Disturbance matrices are discussed below as a
2302 means to track carbon transfers among pools and thereby avoid over- or underestimates of
2303 emissions and improve uncertainty estimation.
2304 In the methods described here, all estimates of changes in carbon stocks (e.g., biomass
2305 growth, carbon transfers among pools) are in mass units of carbon (C) per year, e.g.,
2306 tonnes C/yr. To be consistent with the AFOLU Guidelines, equations are written so that net
2307 carbon emissions (stock decreases) are negative. To be consistent with the national
2308 greenhouse gas inventory reporting tables established by the IPCC, in which emissions are
2309 reported as positive values, emissions would need to be multiplied by negative one (-1).
2310 There are two fundamentally different, but equally valid, approaches to estimating carbon
2311 stock changes: 1) the stock-based (or stock-difference) approach and 2) the process-based
2312 (or gain-loss) approach. These approaches can be used to estimate stock changes in any
2313 carbon pool, although as is explained below, their applicability to soil carbon stocks is
2314 limited. The stock-based approach estimates the difference in carbon stocks in a particular
2315 pool at two points in time (Equation 5-1). This method can be used when carbon stocks in
2316 relevant pools have been measured and estimated over time, such as in national forest
2317 inventories. The process-based (or gain-loss) approach estimates the net balance of
2318 additions to and removals from a carbon pool (Equation 5-2). In the REDD context, gains
2319 only result from carbon transfer from another pool (e.g., transfer from a biomass pool to a

74
2320 dead organic matter pool due to disturbance), and losses result from carbon transfer to
2321 another pool and emissions due to harvesting, decomposition or burning. This type of
2322 method is used when annual data such as biomass growth rates and wood harvests are
2323 available. In reality, countries often use a mix of the stock-difference and gain-loss
2324 approaches for national inventories of carbon stock changes due to data limitations that
2325 preclude the use of only one approach.

2326 Equation 5.1

2327 Annual Carbon Stock Change in a Given Pool as an Annual Average Difference in Stocks
2328 (Stock-Difference Method)

ΔC =
(Ct 2 − Ct1 )
2329
(t2 − t1 )
2330

2331 Where:
2332 ∆C = annual carbon stock change in pool (tonnes C/yr)
2333 Ct1 = carbon stock in pool in at time t1 (tonnes C)
2334 Ct2 = carbon stock in pool in at time t2 (tonnes C)
2335 Note: the carbon stock values for some pools may be in tonnes C/ ha, in which case the
2336 difference in carbon stocks will need to be multiplied by an area.

2337

2338 Equation 5.2

2339 Annual Carbon Stock Change in a Given Pool As a Function of Annual Gains and Losses
2340 (Gain-Loss Method)

Δ C = Δ C G − ΔC L
2341

2342 Where:
2343 ∆C = annual carbon stock change in pool (tonnes C/yr)
2344 ∆CG = annual gain in carbon (tonnes C/yr)
2345 ∆CL = annual loss of carbon (tonnes C/yr)

2346 The stock-difference method is suitable for estimating for emissions caused by both
2347 deforestation and forest degradation, and can apply to all carbon pools.20 The carbon stock
2348 for any pool at time t1 will represent the carbon stock of that pool in the forest of a
2349 particular stratum (see Chapter 4), and the carbon stock of that pool at time t2 will either be
2350 zero (the Tier 1 default value for biomass and dead organic matter immediately after
2351 deforestation) or the value for the pool under the new land use (see section 5.5.2) or the
2352 value for the pool under the resultant degraded forest. If the carbon stock values are
2353 densities (i.e., in units of t C/ha), the change in carbon stocks, ∆C, is then multiplied by the

20
Although in theory the stock-difference approach could be used to estimate stock changes in both
mineral soils and organic soils, this approach is unlikely to be used in practice due to the expense of
measuring soil carbon stocks. The IPCC has adopted different methodologies for soil carbon, which are
described below.

75
2354 area deforested or degraded for that particular stratum, and then divided by the time
2355 interval to give an annual estimate.
2356 Estimating the change in carbon stock using the gain-loss method (Equation 5-2) is not
2357 likely to be useful for deforestation estimating with a Tier 1 or Tier 2 method, but could be
2358 used for Tier 3 approach for biomass and dead organic matter involving detailed forest
2359 inventories and/or simulation models. However, the gain-loss method can be used for forest
2360 degradation to account for the biomass and dead organic matter pools with a Tier 2 or Tier
2361 3 approach. Biomass gains would be accounted for with rates of growth, and biomass losses
2362 would be accounted for with data on timber harvests, fuelwood removals, and transfers to
2363 the dead organic matter pool due to disturbance. Dead organic matter gains would be
2364 accounted for with transfers from the biomass pools and losses would be accounted for with
2365 rates of decomposition.

2366 5.5 Estimation of Emissions from Deforestation

2367 5.5.1 Disturbance Matrix Documentation

2368 Land-use conversion, particularly from forests to non-forests, can involve significant
2369 transfers of carbon among pools. The immediate impacts of land conversion on the carbon
2370 stocks for each forest stratum can be summarized in a matrix, which describes the
2371 retention, transfers, and releases of carbon in and from the pools in the original land-use
2372 due to conversion (Table 5-2). The level of detail on these transfers will depend on the
2373 decision of which carbon pools to include, which in turn will depend on the key category
2374 analysis (see Table 4.2 in Chapter 4 of this volume). The disturbance matrix defines for
2375 each pool the proportion of carbon that remains in the pool and the proportions that are
2376 transferred to other pools. Use of such a matrix in carbon estimating will ensure consistency
2377 of estimating among carbon pools, as well as help to achieve high accuracy in carbon
2378 emissions estimation. Even if all the data in the matrix are not used, the matrix can assist in
2379 estimation of uncertainties.
2380 Table 5.2 Example of a disturbance matrix for the impacts of deforestation on carbon pools
2381 (Table 5.7 in the AFOLU Guidelines). Impossible transfers are blacked out. In each blank
2382 cell, enter the proportion of each pool on the left side of the matrix that is transferred to the
2383 pool at the top of each column. Values in each row must sum to 1.
Above- Below- Soil Harvested Sum of
To Dead Atmo-
ground ground Litter organic wood row (must
From wood sphere
biomass biomass matter products equal 1)
Abovegrou
nd biomass
Belowgroun
d biomass
Dead wood

Litter

Soil organic
matter

2384 5.5.2 Changes in Carbon Stocks of Biomass

2385 The IPCC methods for estimating the annual carbon stock change on land converted to a
2386 new land-use category include two components:

76
2387 ‰ One accounts for the initial change in carbon stocks due to the land conversion, e.g.,
2388 the change in biomass stocks due to forest clearing and conversion to cropland.
2389 ‰ The other component accounts, in the REDD context, only for the gradual carbon loss
2390 during a transition period to a new steady-state system.
2391 For the biomass pools, conversion to annual cropland and settlements generally contain
2392 lower biomass and steady-state is usually reached in a shorter period (e.g., the default
2393 assumption for annual cropland is 1 year). The time period needed to reach steady state in
2394 perennial cropland (e.g., orchards) or even grasslands, however, is typically more than one
2395 year. The inclusion of this second component will likely become more important for future
2396 monitoring of the performance of REDD as countries consider moving into a Tier 3 approach
2397 and implement an annual or bi-annual monitoring system.
2398 The initial change in biomass (live or dead) stocks due to land-use conversion is estimated
2399 using a stock-difference approach in which the difference in stocks before and after
2400 conversion is calculated for each stratum of land converted. Equation 5-3 (below) is the
2401 equation presented in the AFOLU Guidelines for biomass (the carbon fraction [CF] term is
2402 not used in the equation for dead organic matter).

2403 Equation 5.3

2404 Initial Change in Biomass Carbon Stocks on Land Converted to New Land-Use Category
2405 (Stock-Difference Type Method)

ΔC CONV = ∑ [(B AFTERi − B BEFOREi ) ⋅ ΔAi ] ⋅ CF

2406

2407 Where:
2408 ∆CCONV =initial change in biomass carbon stocks on land converted to another
2409 land-use category (tonnes C yr-1)
2410 BAFTERi =biomass stocks on land type i immediately after conversion (tonnes
2411 dry matter/ha)
2412 BBEFOREi =biomass stocks on land type i before conversion (tonnes dry
2413 matter/ha)
2414 ∆Ai = area of land type i converted (ha)
2415 CF = carbon fraction (t C /t dm)
2416 i = stratum of land

2417

2418 The Tier 1 default assumption for biomass and dead organic matter stocks immediately after
2419 conversion of forests to non-forests is that they are zero, whereas the Tier 2 method allows
2420 for the biomass and dead organic matter stocks after conversion to have non-zero values.
2421 Disturbance matrices (e.g., Table 5.2) can be used to summarize the fate of biomass and
2422 DOM stocks, and to ensure consistency among pools.
2423 The biomass stocks immediately after conversion will depend on the amount of live biomass
2424 removed during conversion. During conversion, aboveground biomass may be removed as
2425 wood harvests, burned and the carbon emitted to the atmosphere or transferred to the dead
2426 wood pool, and/or cut and left on the ground as deadwood; and belowground biomass may
2427 be transferred to the soil organic matter pool. Estimates of default values for the biomass
2428 stocks on croplands and grasslands are given in the AFOLU Guidelines in Table 5.9 on page
2429 5.28 (croplands) and Table 6.4 on page 6.27 (grasslands). The dead organic matter (DOM)
2430 stocks immediately after conversion will depend on the amount of live biomass killed and

77
2431 transferred to the DOM pools, and the amount of DOM carbon released to the atmosphere
2432 due to burning and decomposition. In general, croplands (except agroforestry systems) and
2433 settlements will have little or no dead wood and litter so the Tier 1 ‘after conversion’
2434 assumption for these pools may be reasonable for these land uses.
2435 A two-component approach for biomass and DOM may not be necessary in REDD
2436 estimating. If land-use conversions are permanent, and all that one is interested in is the
2437 total change in carbon stocks, then all that is needed is the biomass and DOM stocks prior
2438 to conversion, and the biomass and DOM stocks after conversion once steady state is
2439 reached. These data would be used in a stock difference method (Equation 5.1), with the
2440 time interval the period between land-use conversion and steady-state under the new land
2441 use.

2442 5.5.3 Changes in Soil Carbon Stocks

2443 The AFOLU Guidelines divide soil organic carbon stocks into two types, based on soil type:
2444 mineral soil organic carbon, and organic soil carbon. The emission estimation methodologies
2445 are distinct for each type, but do not vary between subcategories of land remaining in the
2446 same category and subcategories of land converted from one category to another.
2447 The IPCC Tier 2 method for mineral soil organic carbon is basically a combination of a stock-
2448 difference method and a gain-loss method (Equation 5-4). (The first part of Equation 5-4
2449 [for ∆CMineral] is essentially a stock-difference equation, while the second part [for SOC] is
2450 essentially a gain-loss method with the gains and losses derived from the product of
2451 reference carbon stocks and stock change factors). The reference carbon stock is the soil
2452 carbon stock that would have been present under native vegetation on that stratum of land,
2453 given its climate and soil type.

2454 Equation 5.4

2455 Annual Change in Organic Carbon Stocks in Mineral Soils

(SOC − SOC( 0−T ) )

ΔC Mineral =
0

2456 D

2457
(
SOC = ∑C ,S ,i SOC REFC ,S ,i ⋅ FLU C ,S ,i ⋅ FMGC ,S ,i ⋅ FIC ,S ,i ⋅ ΔAC ,S ,i )
2458 Where:
2459 ∆CMineral = annual change in organic carbon stocks in mineral soils (tonnes C yr-1)
2460 SOC0 = soil organic carbon stock in the last year of the inventory time period
2461 (tonnes C)
2462 SOC(0-T) = soil organic carbon stock at the beginning of the inventory time period
2463 (tonnes C)
2464 T = number of years over a single inventory time period (yr)
2465 D = Time dependence of stock change factors which is the default time period for
2466 transition between equilibrium SOC values (yr). 20 years is commonly used, but depends on
2467 assumptions made in computing the factors FLU, FMG, and FI. If T exceeds D, use the value
2468 for T to obtain an annual rate of change over the inventory time period (0-T years).
2469 c represents the climate zones, s the soil types, and i the set of management systems
2470 that are present in a country
2471 SOCREF = the reference carbon stock (tonnes c ha-1)

78
2472 FLU = stock change factor for land-use systems or sub-system for a particular land use
2473 (dimensionless)
2474 FMG = stock change factor for management regime (dimensionless)
2475 FI = stock change factor for input of organic matter (dimensionless)
2476 A = land area of the stratum being estimated (ha)

2477

2478 The land areas in each stratum being estimated should have common biophysical conditions
2479 (i.e., climate and soil type) and management history over the inventory time period. Also
2480 disturbed forest soils can take many years to reach a new steady state (the IPCC default for
2481 conversion to cropland is 20 years).
2482 Countries may not have sufficient country-specific data to fully implement a Tier 2 approach
2483 for mineral soils, in which case a mix of country-specific and default data may be used.
2484 Default data for reference soil organic carbon stocks can be found in Table 2.3 (page 2.31)
2485 of the AFOLU Guidelines; default stock change factors can be found in the land-use category
2486 chapters of the AFOLU Guidelines (Chapter 4, 5, 6, 7, 8, and 9).
2487 The IPCC Tier 2 method for organic soil carbon is an emission factor method that employs
2488 annual emission factor that vary by climate type and possibly by management system
2489 (Equation 5.5).

2490 Equation 5.5

2491 Annual Carbon Loss from Drained Organic Soils

LOrganic = ∑C ( A ⋅ EF ) C
2492

2493 Where:
2494 LOrganic = annual carbon loss from drained organic soils (tonnes C yr-1)
2495 Ac = land area of drained organic soils in climate type c (ha)
2496 EFc = emission factor for climate type c (tonnes C yr-1)

2497

2498 Note that land areas and emission factors can also be disaggregated by management
2499 system, if there are emissions data to support this.
2500 This methodology can be disaggregated further into emissions by management systems in
2501 addition to climate type if appropriate emission factors are available. Default (Tier 1)
2502 emission factors for drained forest, cropland, and grassland soils are found in Tables 4.6
2503 (page 4.53), 5.6 (page 5.19), and 6.3 (page 6.17) of the AFOLU Guidelines.

2504 5.6 Estimation of Emissions from Forest Degradation

2505 5.6.1 Disturbance Matrix Documentation

2506 As with deforestation, forest degradation can involve significant transfers of carbon among
2507 pools, so it is recommended that the impacts of degradation on each carbon pool for each
2508 forest stratum be summarized in a matrix as shown in Table 5.2 above. This matrix
2509 describes the retention, transfers, and releases of carbon in and from the pools in the
2510 original forest due to degradation. Use of such a matrix will ensure accounting consistency

79
2511 among carbon pools, as well as help to achieve high accuracy in carbon emissions
2512 estimation.

2513 5.6.2 Changes in Carbon Stocks

2514 The AFOLU Guidelines recommend either a stock-difference method (Equation 5-1) or a
2515 gain-loss method (Equation 5-2) for estimating the annual carbon stock change in biomass
2516 and dead organic matter (DOM) in “Forests Remaining Forests” (the land-use subcategory
2517 that encompasses forest degradation). In general, the gain-loss method is applicable for all
2518 tiers, while the stock-difference method is more suited to Tiers 2 and 3 assuming its
2519 application involves accurate and complete forest inventories based on sample plots.
2520 While the decision regarding whether a stock-difference method or a gain-loss method is
2521 used will depend largely on the availability of existing data and resources to collect
2522 additional data. Estimating the carbon impacts of logging may lend itself more readily to a
2523 the gain-loss approach, while estimating the carbon impacts of fire may lend itself more
2524 readily to a the stock-difference approach.
2525 With a gain-loss approach for estimating emissions from forest degradation, biomass gains
2526 would be accounted for with rates of growth in trees after logging, and biomass losses
2527 would be accounted for with data on timber harvests, fuelwood removals, and transfers to
2528 the dead organic matter pool due to fire and other disturbance. Dead organic matter gains
2529 would be accounted for with transfers from the biomass pools and losses would be
2530 accounted for with rates of decomposition. With a stock-difference approach, carbon stocks
2531 in each pool would be estimated both before and after degradation (e.g. a timber harvest),
2532 and the difference in carbon stocks in each pool calculated.
2533 For Forests Remaining Forests, the Tier 1 assumption is that net carbon stock changes in
2534 DOM are zero, whereas in reality dead wood can decompose relatively slowly. Even in
2535 tropical humid climates. Both logging and fires can significantly influence stocks in the dead
2536 wood and litter pools, so countries that are experiencing significant changes in their forests
2537 due to degradation are encouraged to develop domestic data to estimate the impact of
2538 these changes on dead organic matter.

2539 5.6.3 Changes in Soil Carbon Stocks

2540 The emission estimation methodologies in the AFOLU Guidelines for soils do not vary
2541 between subcategories of land remaining in the same category and subcategories of land
2542 converted from one category to another. Therefore, the soil carbon methods that should be
2543 used for forest degradation are the same as those for deforestation (see section 5.5.3
2544 above). However, as is discussed in Chapter 4, estimation of soil carbon emissions is only
2545 recommended for intensive practices that involve significant soil disturbance. Selective
2546 logging of forests on mineral soil does not typically disturb soils significantly; however,
2547 selective logging of forests growing on organic soils, particularly peatswamps, could result in
2548 large emissions caused by practices such as draining to remove the logs from the forest.

2549 5.7 Estimation of uncertainties

2550 Estimates of carbon emissions from deforestation and forest degradation need to include
2551 quantitative estimates of uncertainties. Chapters 3 and 4 describe sources of uncertainty,
2552 and approaches for estimating uncertainties, in the activity data and emission factors used
2553 in REDD accounting (i.e., land areas for activity data; and carbon stocks or changes in
2554 carbon stocks, associated parameters, and organic soil emission factors for “emission
2555 factors”). This section presents the IPCC approaches for estimating the combined

80
2556 uncertainties of activity data and emission factors. This will improve confidence in emission
2557 estimates.
2558 The AFOLU Guidelines present two approaches for estimating combined uncertainties:
2559 Approach 1 uses simple error propagation equations, while Approach 2 uses Monte Carlo or
2560 similar techniques.
2561 In the “Propagation of Errors” approach, the total uncertainty is equal to the square root of
2562 the sum of the squares of each of the component uncertainties (Equation 5-6). Where the
2563 same units are being combined such as when the total uncertainty from the combined
2564 carbon pools are being assessed, then the 95% confidence interval should be used.
2565 However, where different units are employed such as carbon biomass and forest area,
2566 uncertainty is equal to the 95% confidence interval as a percentage of the mean ([95%
2567 confidence interval/mean] x 100).

2568 Equation 5.6

2569 Combined Uncertainties – Propagation of Error Approach

U total = U 12 + U 22 + .... + U n2
2570

2571 Where:
2572 Utotal = total uncertainty
2573 Ui = uncertainty associated with each of the component quantities

2574

2575 This method should be used with caution if there is a high level of correlation between
2576 components of the total error or if any of the component uncertainties is high (a standard
2577 deviation greater than 30% of the mean). Even if these tests are failed the equation can still
2578 be used to give approximate results.
2579 The second IPCC approach for estimating combined uncertainties is a Monte Carlo type
2580 analysis. Monte Carlo analyses model uncertainty through selecting random values from
2581 probability distributions for parameters and measuring the effect on total stocks. Either
2582 training in the use of software packages that automatically provide Monte Carlo type
2583 analyses or contracting an expert in Monte Carlo analysis is required to implement this
2584 higher level method.
2585 All assessments should include at least a simple error propagation type of analysis of
2586 uncertainties. An example is shown in Box 5-1.

81
2587 BOX 5.1: Example of a Propagation of Error Uncertainty Analysis

2588

2589 Therefore the total stock is 138 t C/ha and the uncertainty =

2590 112 + 3 2 + 2 2 = 11.6tC / ha

2591

2592 Therefore the total carbon stock over the stratum is:
2593 8564 * 138 = 1,181,832 t C
2594 And the uncertainty =

2595
14 2 + 8 2 = 15.9%
2596 15.9% of 1,181,832 = 188,165 t C
2597

82
2598 6 GUIDANCE ON REPORTING

2599 6.1 Issues and challenges in reporting

2600 6.1.1 The importance of good reporting

2601 Under the UNFCCC, information reported in greenhouse gas (GHG) inventories represents
2602 an essential link between science and policy, providing the means by which the COP can
2603 monitor progress made by Parties in meeting their commitments and in achieving the
2604 Convention's ultimate objectives. In any international system in which an accounting
2605 procedure is foreseen - as in the Kyoto Protocol and likely also in a future REDD mechanism
2606 – the information reported in a Party’s GHG inventory represents the basis for assessing
2607 each Party’s performance as compared to its commitments or reference scenario, and
2608 therefore represents the basis for assigning eventual incentives or penalties.
2609 The quality of GHG inventories relies not only upon the robustness of the science
2610 underpinning the methodologies and the associated credibility of the estimates – but also on
2611 the way this information is compiled and presented. Information must be well documented,
2612 transparent and consistent with the reporting requirements outlined in the UNFCCC
2613 guidelines.

2614 6.1.2 Overview of the Chapter

2615 Section 6.2 gives an overview of the current reporting requirements under UNFCCC,
2616 including the general underlying principles. The typical structure of a GHG inventory is
2617 illustrated, including an exemplificative table for reporting C stock changes from
2618 deforestation and forest degradation.
2619 Section 6.3 outlines the major challenges that developing countries will likely encounter
2620 when implementing the reporting principles described in section 6.2.
2621 Section 6.4 elaborates concepts already agreed upon in a UNFCCC context and describes
2622 how a conservative approach may help to overcome some of the difficulties described in
2623 Section 6.3.

2624 6.2 Overview of reporting principles and procedures

2625 6.2.1 Current reporting requirements under the UNFCCC

2626 Under the UNFCCC, all Parties are required to provide national inventories of anthropogenic
2627 emissions by sources and removals by sinks of all greenhouse gases not controlled by the
2628 Montreal Protocol. To promote the provision of credible and consistent GHG information, the
2629 COP has developed specific reporting guidelines that detail standardized requirements.
2630 Although these requirements differ across Parties, they are similar in that they are based on
2631 IPCC methodologies and aim to produce a full, accurate, transparent, consistent and
2632 comparable reporting of GHG emissions and removals.

83
2633 At present, detailed reporting guidelines exist for the annual GHG inventories of Annex I
2634 Parties (UNFCCC 2004)21, while only generic guidance is available for the preparation of
2635 national communications from non-Annex I Parties22. This difference reflects the fact that
2636 Annex I (AI) Parties are required to report detailed data on an annual basis that are subject
2637 to in-depth review by teams of independent experts, while Non-Annex I Parties (NAI)
2638 currently report less often and in less detail. As a result, their national communications are
2639 not subject to in-depth reviews.
2640 However, given the potential relevance of a future REDD mechanism - and the consequent
2641 need for robust and defensible estimates - the reporting requirements of NAI Parties on
2642 emissions from deforestation will certainly become more stringent and may come close to
2643 the level of detail currently required from AI Parties. Although at present it is not possible to
2644 foresee the exact reporting requirements of a future REDD mechanism, they will likely
2645 follow the general principles and procedures outlined in the following sections.

2646 6.2.2 Inventory and reporting principles

2647 Under the UNFCCC, there are five general principles which should guide the estimation and
2648 the reporting of emissions and removals of GHGs: Transparency, Consistency Comparability
2649 Completeness and Accuracy. Although some of these principles have been already discussed
2650 in previous chapters, below are summarized and their relevance for the reporting is
2651 highlighted:
2652 ‰ Transparency, i.e. all the assumptions and the methodologies used in the inventory
2653 should be clearly explained and appropriately documented, so that anybody could
2654 verify its correctness.
2655 ‰ Consistency, i.e. the same definitions and methodologies (including the same
2656 emission factor for the most disaggregated reported level) should be used along
2657 time. This should ensure that differences between years and categories reflect real
2658 differences in emissions. Under certain circumstances, estimates using different
2659 methodologies for different years can be considered consistent if they have been
2660 calculated in a transparent manner. Recalculations of previously submitted estimates
2661 are possible to improve accuracy and/or completeness, providing that all the relevant
2662 information is properly documented. In a REDD context, consistency also means that
2663 all the lands and all the carbon pools which have been reported in the reference
2664 period must to be tracked in the future (in the Kyoto language it is said “once in,
2665 always in”). Similarly, the inclusion of new sources or sinks which have existed since
2666 the reference period but were not previously reported (e.g., a carbon pool), should
2667 be reported for the reference period and all subsequent years for which a reporting is
2668 required.
2669 ‰ Comparability across countries. For this purpose, Parties should follow the
2670 methodologies and standard formats (including the allocation of different source/sink
2671 category) provided by the IPCC and agreed within the UNFCCC for estimating and
2672 reporting inventories (see also chapter 2.1). It shall be noted that the comparability
2673 principle may be extended also to definitions (e.g. definition of forest) and estimates
2674 (e.g. forest area, average C stock) provided by the same Party to different

21
UNFCCC 2004 Guidelines for the preparation of national communications by Parties included in
Annex I to the Convention, Part I: UNFCCC reporting guidelines on annual inventories
(FCCC/SBSTA/2004/8).
22
UNFCCC 2002 Guidelines for the preparation of national communications from Parties not included
in Annex I to the Convention (FCCC/CP/2002/7/Add.2).

84
2675 international organizations (e.g. UNFCCC, FAO). In that case, any discrepancy should
2676 be adequately justified.
2677 ‰ Completeness, meaning that estimates should include – for all the relevant
2678 geographical coverage – all the agreed categories, gases and pools. When gaps
2679 exist, all the relevant information and justification on these gaps should be
2680 documented in a transparent manner.
2681 ‰ Accuracy, in the sense that estimates should be systematically neither over nor
2682 under the true value, so far as can be judged, and that uncertainties are reduced so
2683 far as is practicable. Quantify the uncertainties is important to prioritize efforts to
2684 improve accuracy of inventories in the future and, likely, to support the
2685 implementation of the conservativeness approach (see Ch. 6.4).

2686 6.2.3 Structure of a GHG inventory

2687 A national inventory of GHG anthropogenic emissions and removals is typically divided into
2688 two parts:
2689 Reporting Tables are a series of standardized data tables that contain mainly quantitative
2690 (numerical) information. Box 6.1 shows an exemplificative table for reporting C stock
2691 changes following deforestation and degradation (modified from Kyoto Protocol LULUCF
2692 tables for illustrative purposes only). Typically, these tables include columns for:
2693 ‰ The initial and final land-use category. Additional stratification is encouraged (in a
2694 separate column for subcategories) according to criteria such as climate zone,
2695 management system, soil type, vegetation type, tree species, ecological zones,
2696 national land classification or other factors.
2697 ‰ The “activity data”, i.e., area of land (in thousands of ha) subject to gross
2698 deforestation and degradation (see Ch. 3).
2699 ‰ The “emission factors”, i.e., the C stock changes per unit area deforested or
2700 degraded, separated for each carbon pool (see Ch. 4). The term “implied factors”
2701 means that the reported values represent an average within the reported category or
2702 subcategory, and serves mainly for comparative purposes.
2703 ‰ The total change in C stock, obtained by multiplying each activity data by the
2704 relevant emission C stock change factor.
2705 ‰ The total emissions (expressed as CO2).

85
2706 Box 6.1: Example of a typical reporting table for reporting C stock changes
2707 following deforestation and degradation.

2708

86
2709 To ensure the completeness of an inventory, it is good practice to fill in information for all
2710 entries of the table. If actual emission and removal quantities have not been estimated or
2711 cannot otherwise be reported in the tables, the inventory compiler should use the following
2712 qualitative “notation keys” (from IPCC 2006 GL) and provide supporting documentation.

Notation key Explanation

NE (Not estimated) Emissions and/or removals occur but have not been
estimated or reported.

IE (Included elsewhere) Emissions and/or removals for this activity or category

are estimated but included elsewhere. In this case, where
they are located should be indicated,

C (Confidential information) Emissions and/or removals are aggregated and included

elsewhere in the inventory because reporting at a
disaggregated level could lead to the disclosure of
confidential information.

NA (Not Applicable) The activity or category exists but relevant emissions and
removals are considered never to occur.

NO (Not Occurring) An activity or process does not exist within a country.

2713

2714 For example, if a country decides that a disproportionate amount of effort would be required
2715 to collect data for a pool from a specific category that is not a key category (see Ch. 4) in
2716 terms of the overall level and trend in national emission, then the country should list all
2717 gases/pools excluded on these grounds, together with a justification for exclusion, and use
2718 the notation key 'NE' in the reporting tables.
2719 Furthermore, the reporting tables are generally complemented by a documentation box
2720 which should be used to provide references to relevant sections of the Inventory Report if
2721 any additional information is needed.
2722 In addition to tables like those illustrated in Box 6-1, other typical tables to be filled in a
2723 comprehensive GHG inventory include:
2724 Tables with emissions from other gases (e.g., CH4 and N2O from biomass burning), to be
2725 expressed both in unit of mass and in CO2 equivalent (using the Global Warming Potential of
2726 each gas provided by the IPCC)
2727 Summary tables (with all the gases and all the emissions/removals)
2728 Tables with emission trends (covering data also from previous submissions)
2729 Tables for illustrating the results of the key category analysis, the completeness of the
2730 reporting, and eventual recalculations.
2731 In the context of REDD, most of these types of tables will likely need to be completed for
2732 the reference period and for the assessment period, although it is not yet clear if non-CO2
2733 gases and all pools will be required.
2734 Inventory Report: The other part of a national inventory is an Inventory Report that
2735 contains comprehensive and transparent information about the inventory, including:
2736 ‰ An overview of trends for aggregated GHG emissions, by gas and by category.
2737 ‰ A description of the methodologies used in compiling the inventory, the assumptions,
2738 the data sources and rationale for their selection, and an indication of the level of
2739 complexity (IPCC tiers) applied. In the context of REDD reporting, appropriate

87
2740 information on land-use definitions, land area representation and land-use databases
2741 are likely to be required.
2742 ‰ A description of the key categories, including information on the level of category
2743 disaggregation used and its rationale, the methodology used for identifying key
2744 categories, and if necessary, explanations for why the IPCC-recommended Tiers have
2745 not been applied.
2746 ‰ Information on uncertainties (i.e., methods used and underlying assumptions), time-
2747 series consistency, recalculations (with justification for providing new estimates),
2748 quality assurance and quality control procedures.
2749 ‰ A description of the institutional arrangements for inventory preparation.
2750 ‰ Information on planned improvements.
2751 Furthermore, all of the relevant inventory information should be compiled and archived,
2752 including all disaggregated emission factors, activity data and documentation on how these
2753 factors and data were generated and aggregated for reporting. This information should
2754 allow, inter alia, reconstruction of the inventory by the expert review teams.

2755 6.3 What are the major challenges for developing countries?

2756 Although the inventory requirements for a REDD mechanism have not yet been designed, it
2757 is possible to foresee some of the major challenges that developing countries will encounter
2758 in estimating and reporting emissions from deforestation and forest degradation. In
2759 particular, what difficulties can be expected if the five principles outlined above are required
2760 for REDD reporting?
2761 While specific countries may encounter difficulties in meeting transparency, consistency and
2762 comparability principles, it is likely that most countries will be able to fulfill these principles
2763 reasonably well after adequate capacity building. In contrast, based on the current
2764 monitoring and reporting capabilities, the principles of completeness and accuracy will likely
2765 represent major challenges for most developing countries.
2766 Achieving the completeness principle will clearly depend on the processes (e.g.
2767 deforestation, forest degradation) involved, the pools and gases that needed to be reported,
2768 and the forest-related definitions that are applied. For example, evidence from official
2769 reports (e.g., NAI national communications to UNFCCC23, FAO’s FRA 200524) suggests that
2770 only a very small fraction of developing countries currently reports data on soil carbon, even
2771 though emissions from soils following deforestation are likely to be significant in many
2772 cases.
2773 If accurate estimates of emissions are to be reported, reliable methodologies are needed as
2774 well as a quantification of their uncertainties. For key categories and significant pools, this
2775 implies the application of higher tiers, i.e. having country-specific data on all the significant
2776 pools stratified by climate, forest, soil and conversion type at a fine to medium spatial scale.
2777 While the capacity for monitoring the amount of deforested area is improving rapidly with
2778 advances in remote sensing technology, in many developing countries reliable data on
2779 carbon stocks are scarce and allocating significant resources for monitoring may be difficult.
2780 This suggests that in many cases the overall emissions estimates for reference scenarios

23
UNFCCC. 2005. Sixth compilation and synthesis of initial national communications from
Parties not included in Annex I to the Convention. FCCC/SBI/2005/18/Add.2
24
Food and Agriculture Organization. 2006. Global Forest Resources Assessment.

88
2781 from developing countries may not be “accurate and precise” at the country level until
2782 additional carbon stock data become available.

2783 6.4 The conservativeness approach

2784 Even if current UNFCCC reporting provisions underline the fact that national circumstances
2785 (including data and resource availability) are a fundamental parameter against which to
2786 evaluate the quality of an inventory, this argument is weak in the context of REDD. If a
2787 Party’s reported estimates will be the basis of an accounting framework (as in the Kyoto
2788 Protocol) with an eventual assignment of economic incentives, then the requirement for
2789 “robust” estimates seems fully justifiable.

2790 Thus, how should the obstacle of potentially incomplete, highly uncertain REDD
2791 reporting be overcome?
2792 The simple and pragmatic approach of conservativeness may simplify the requirements
2793 necessary for obtaining defensible estimates of reduced emissions from deforestation in NAI
2794 countries. In the REDD context, conservativeness means that when completeness, accuracy
2795 and precision cannot be achieved, the reported reduction of emissions (and thus the
2796 incentives claimed by the country) should be underestimated, or at least the risk of
2797 overestimation should be minimized.
2798 Here we suggest two examples in which the conservativeness principle may be applied to a
2799 REDD mechanism when estimates for some pool or category are not as complete, accurate
2800 or precise as the inventory and reporting principles prescribe.
2801 1. If no data are available at the required aggregation level for a carbon pool that is
2802 significant in terms of emissions (e.g. soil), either in the reference or in the future
2803 monitoring period, then the omission of that pool does not necessarily represent a reporting
2804 problem.
2805 Conceptually, this issue has already been addressed in the Kyoto Protocol rules. Under
2806 Articles 3.3 and 3.4, AI Parties “may choose not to account for a given pool if transparent
2807 and verifiable information is provided that the pool is not a source”. A strict application of
2808 this sentence in a REDD context would not help, because all carbon pools are generally
2809 sources during deforestation and forest degradation events. However, being conservative
2810 in a REDD context does not mean “not overestimating the emissions”, but rather
2811 “not overestimating the reduction of emissions”. In practice, if the area deforested in
2812 the monitoring period has been reduced as compared to the reference period for the
2813 required aggregation level, then if emissions from a carbon pool (e.g., soil C stocks in a
2814 particular forest type converted to cropland) are not reported, the resulting estimates of
2815 reduced emissions will be incomplete but conservative (see example in Table 6.1).

89
2816 Table 6.1: Simplified example of how ignoring a carbon pool may produce a conservative
2817 estimate of reduced emissions from deforestation. The reference level might be assessed on
2818 the basis of historical emissions. (a) complete estimate, including the soil pool; (b)
2819 incomplete estimate, as the soil pool is missing. The latter estimate of reduced emissions is
2820 not accurate, but is conservative.

Emissions
Carbon stock change
(area deforested x C stock
Area (t C/ha deforested)
change, t C x 103)
deforested
Above- Only Above-
(ha x 103) Aboveground
ground Soil ground
Biomass + Soil
Biomass Biomass
Reference
10 100 50 1500 1000
level
Assessment
5 100 50 750 500
period
Reduction of emissions
750 (a) 500 (b)
(reference level - assessment period, t C x 103)
2821

2822 2. If a tier 3 method cannot be implemented for a key category or carbon pool due to
2823 lack of resources and/or poor data quality and availability, this does not necessarily exclude
2824 the possibility of reporting to a REDD mechanism.
2825 Tier 1 estimates are expected to have a significantly higher uncertainty than Tier 3
2826 estimates, which implies that a Tier 1 estimate of reduced emissions has many chances of
2827 being higher than the “true” value. The problem, therefore, is how to decrease the risk of
2828 overestimation of the reduced emissions.
2829 During the accounting phase, the issue has been addressed already by the rules agreed
2830 upon for the Kyoto Protocol. If an AI Party reports at a Tier lower than that required (e.g., a
2831 key source category is reported at Tier 1) or reports in a manner that is not consistent with
2832 IPCC methodologies, then this would likely trigger an “adjustment”, i.e., a change applied
2833 by an expert review team (ERT) to the Party’s reported estimates. In this procedure,
2834 conservativeness is ensured by multiplying the ERT’s calculated estimate by a tabulated
2835 category-specific “conservativeness factor”25.
2836 Differences in conservativeness factors between categories reflect typical differences in
2837 uncertainties in data and estimates: the multiplication of the ERT’s calculated estimate by
2838 conservativeness factors have a higher impact for components that are expected to be more
2839 uncertain. This concept is illustrated in Figure 6.1, which shows two estimates of a
2840 hypothetical reduced emission. The value of the two estimates is the same, but one is
2841 obtained using a Tier 3 method (left column) that is likely more precise but also more
2842 complex and expensive, while the other is obtained using a Tier 1 method (center column).
2843 If uncertainty is quantified correctly, then the magnitude of the confidence interval could be
2844 considered as a proxy for the “quality” of the estimate, with a smaller confidence interval
2845 corresponding to a higher “quality” of the estimate. Presumably, the Tier 3 estimate will be
2846 less uncertain and therefore will have a consequentially lower risk of overestimation (see
2847 graph bracket) than the Tier 1 estimate. In that case, applying a conservativeness factor to
2848 the Tier 1 estimate would reduce the possibility of overestimation (see right column),

25
UNFCCC 2006 Good practice guidance and adjustments under Article 5, paragraph 2, of the Kyoto
Protocol. (FCCC/KP/CMP/2005/8/Add.3).

90
2849 The same approach could potentially be applied also during the reporting phase, if a tier
2850 has been applied that is lower than that required, then the Party could choose to report (or
2851 be requested to report) the lowest (most conservative) estimate of reduced emissions
2852 among the range of possible values of the uncertainty band (see Tier 1 in Figure 6.1).
2853

150

125
Reduced emissions

100

75

50

25

0
Tier 3 Tier 1 Tier 1
adjusted
2854

2855 Figure 6.1. Conceptual example of the application of a conservativeness factor during the
2856 adjustment procedure.
2857 The two examples outlined above demonstrate that a Party whose estimate of reduced
2858 emissions is incomplete (in terms of pools) or highly uncertain (in terms of Tier level
2859 chosen) will not necessarily be excluded from joining a REDD mechanism. When
2860 completeness, accuracy and precision cannot be achieved, reporting a defensible estimate
2861 of reduced emissions is still possible in the framework of the already agreed UNFCCC/Kyoto
2862 Protocol reporting rules.
2863 Furthermore, implementation of the conservativeness approach would help create a win-win
2864 scenario. On one hand, the conservativeness approach preserves the “climate integrity” of
2865 any REDD mechanism by guaranteeing that economic incentives would not lead to "hot air"
2866 and convincing policymakers and investors in industrialized countries that a REDD
2867 mechanism is scientifically valid. On the other hand, the conservativeness approach helps to
2868 ensure broad participation by allowing developing countries to join the mechanism even if
2869 they cannot provide complete estimates for all carbon pools or precise estimates of all key
2870 categories.
2871 Finally, such an approach would provide a clear incentive for increasing the quality of
2872 reporting by developing countries, because more complete reporting (e.g. including all
2873 carbon pools) would likely increase reduced emissions estimates and allow countries to
2874 claim more incentives. Similarly, estimates with an appropriate level of uncertainty (e.g.,
2875 derived from Tier 3 approach for a key category) will likely avoid a downward revision
2876 during the review process.
2877 In future REDD reporting rules, a system can be envisaged in which Parties are allowed to
2878 choose what to estimate/report and at which Tier based on their own cost-benefit analysis,
2879 provided that the conservativeness principle is upheld. If a REDD mechanism begins with
2880 conservativeness, then accuracy will follow.
2881

91