Environmental Modelling & Software 21 (2006) 1579e1586

www.elsevier.com/locate/envsoft

Database architectures: Current trends and their relationships to

environmental data management
Jaroslav Pokorny*
Charles University, Faculty of Mathematics and Physics, Department of Software Engineering, Malostranske nam. 25, 118 00 Praha, Czech Republic
Received 11 November 2005
Available online 27 June 2006

Abstract

Ever increasing environmental information demands from customers, authorities, and governmental organizations as well as new business
control functions are implemented and integrated to environmental information management systems (EIMSs). These systems are often based
on traditional file techniques or, more recently, on commercial database management systems (DBMSs). With a production of huge data sets and
their processing in real-time applications, the needs for environmental data management have grown significantly. Numerous examples from
practice of EIMSs prove that the architecture of DBMS should be open for a permanent evolution. Current trends in database development
and an associated research meet these challenges. New information and communication technologies and techniques influence todays DBMSs.
They include, among other things, sensor networks, stream processing, processing uncertain and imprecise data, knowledge discovery and in-
telligent data analysis, as well as wireless broadcast and mobile computing. Both research and practice indicate that the traditional universal
DBMS architecture hardly satisfies these trends and new solutions are needed. Rather separate specialized engines connected into networks
are beneficial. The paper discusses recent advances in database technologies and attempts to highlight them with respect to requirements of
EIMSs.
2006 Elsevier Ltd. All rights reserved.

Keywords: Environmental management system; Database management system; Sensor; Sensor network; Stream processing; Uncertain and imprecise data; Knowl-
edge discovery and intelligent data analysis; Wireless broadcast; Mobile computing

1. Introduction from customers, authorities and governmental organizations.

Without doubt the world of data is changing, particularly,
the nature and sources of information. All these changes
have a significant influence on database needs, and conse-
quently, on questions where the database field is and where
it should be going. Abiteboul et al. (2005) in their report em-
phasize two main driving forces today: Internet and particular
sciences, like the physical sciences, biological sciences, med-
icine, and engineering. These sciences produce large and com-
plex data sets that require more advanced database support
than current products provide.
Another trend existing since the 1960s concerns the indus-
tries having faced ever increasing environmental demands
* Fax: 420 221914323.
E-mail address: pokorny@ksi.ms.mff.cuni.cz
Recently, reflecting these demands, new business control func-
tions are integrated to environmental management systems1
(EMS). For their computerized part we can use the term envi-
ronmental information system (EIS), if we address public envi-
ronmental information systems, or environmental management
information system (EMIS), if we deal with industrial environ-
mental information systems. As data or information process-
ing is primarily what we focus on, we will use the term EIS
through the paper.
An important observation is that, similarly to the sciences
mentioned, EISs process also huge data sets, often continually
1
By LCA (2005) an EMS is a part of the overall management system that
includes organisational structure, planning activities, responsibilities, prac-
tices, procedures, processes and resources for developing, implementing,
achieving, reviewing, and maintaining the environmental policy.

1364-8152/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.envsoft.2006.05.004
1580 J. Pokorny / Environmental Modelling & Software 21 (2006) 1579e1586
and with triggering various control actions. Consequently, the
needs for environmental data management have grown
significantly.
Considering environmental data sets combined with, e.g.,
business data, emails, documentations etc., adequate informa-
tion integration mechanisms are needed. Since their beginning
the databases have had an integrative role in the world of data.
Reuter (2005) argues that the technological evolution of data-
base technology makes database systems even the ideal candi-
date for integrating all types of objects that need persistence as
well as for supporting all the different types of execution that
are characteristic of the various application classes.
The most important part of each management system deals
with data through querying. When users want to search and
use environmental information, the following problems occur
(Tomasic and Simon, 1997):
(1) Data do not exist or are insufficient; sometimes this may
require synthesis or reproduction of data.
(2) Data are not referenced by data suppliers and therefore
hard to locate, or data are referenced under specific classi-
fication criteria that are domain-specific.
(3) Data are hard to access; they are either private or of a too
high cost, or requiring costly pre-processing (e.g., data
must be re-entered manually from paper documentation)
or format translation.
(4) Accessed data sets are hard to use because they are incon-
sistent or non-compatible; for example, access to long
time series but standard data collection techniques have
not been applied, thereby making adjacent time series
not compatible.
(5) The quality of retrieved data is hard to assess; it is often
hard to compare data produced using different scientific
models because of a lack of documentation about the un-
derlying computational processes.
The database community focuses on information storage,
organization, management, and access in software architec-
tures called database management systems (DBMSs). Always
it is driven by new applications, technology trends, new syner-
gies with related fields, and innovation within the field itself.
The problems (1)e(5) are a natural part of todays database re-
search and development. A natural idea is that EISs based on
advanced database technologies could help to deal with these
issues.
Several technological aspects influence DBMS develop-
ment. Focusing on the scientific data, it is often coming in
streams. The sensor networks producing the data consist of
very large numbers of low-cost devices, each of which is
a data source, measuring some quantity, e.g. the objects loca-
tion, or the ambient temperature. Processing such data is usu-
ally completely different from the data stored in enterprise
databases. Data arrive in high-speed streams, and queries
over those streams need to be processed in an online fashion
to enable real-time responses. Moreover, in comparison to en-
terprise data processing, these data are uncertain or imprecise.
Other aspects of such data processing include unclear
formulation of queries based on common techniques as they
are used for example in classical databases. Often we are
not able to formulate a query, e.g. in SQL, and despite of
the fact we believe on the other hand that something interest-
ing is hidden in our data. In such situations a lack of semantics
is apparent. To describe data semantics, metadata and its for-
mal description are necessary.
Data in collections considered creates an ideal platform for
using knowledge discovery methods and/or intelligent data
analysis. Also online analytical processing (OLAP), data
warehouses (DW), and data mining (DM) techniques can
help in this context.
The purpose of the paper is to present the main challenges
influencing todays database development with respect to the
processing environmental data. First, in Section 2 we discuss
properties of new data sources. In Section 3 we repeat the con-
cepts of the classical centralized DBMS architecture, as it ex-
ists from early 1980s. According to Harder and Reuter (1983),
the architecture models five-level abstraction hierarchy. Its im-
plementation has five technological layers that allow to sepa-
rate some problems and their solutions in relatively
independent way. The main part of the paper presents five
new technologies influencing database architectures in Section 4.
They include sensor data and sensor networks, stream pro-
cessing, approaching uncertain and imprecise data, knowl-
edge discovery methods and intelligent data analysis, and
wireless broadcast and mobile computing. In Section 5, we
argue that new DBMS architectures are needed, describing
briefly some of their proposals, and give several examples
of their occurrences in practice. In conclusions we summa-
rize the basic ideas given in the paper and add a number
of other issues that can influence processing environmental
data.
2. New data sources
Usual enterprise data stored in databases are structured and
can be described by a so-called (database) schema. Such
a schema is almost fixed or it is changed only rarely. It is
not the case of collections of scientific as well as environmen-
tal data. By Reuter (2005), the key properties of these data col-
lections (irrespective of the many differences) are the
following:

The raw data is written once and never changes again. As
a matter of fact, some scientific organizations require for
all projects they support that any data that influences the
published results of the project be kept available for an ex-
tended period of time, typically around 15 years.

Raw data comes in as streams with high throughput (hun-
dreds of MB/s), depending on the sensor devices. The
streams have to be recorded as they come in, because in
most cases there is no way of repeating the measurement.

For the majority of applications, the raw data is not inter-
esting. What the users need are aggregates, derived values,
or e in case of text fields e some kind of abstract of
what the text says.
 J. Pokorny / Environmental Modelling & Software 21 (2006) 1579e1586 1581

In many cases, the schema has hundreds or thousands of procedural access at L5 level provides tables and statements
attribute types, whereas each instance only has tens of at- for their manipulation, usually formulated in the SQL lan-
tribute values. guage. Among objects at L3 level we can find data structures

The schema of the structured part of the database is not supporting indexing, e.g. well-known B-trees for character
fixed in many cases. As the discipline progresses, new strings and number or R-trees for spatial data. In general, go-
phenomena are discovered, new types of measurements ing upwards the objects and associated operations become
are made, units and dimensions are changed, and once in more complex and some additional integrity constraints can
a while whole new concepts are introduced and/or older occur.
concepts are redefined. All those schema changes have The concept of a multi-layered architecture considers its
to be accommodated dynamically. ideal implementation as a set of abstract machines where a ma-
chine of layer k 1 is implemented via a machine of layer k.
We will consider mainly data representing environmental Although the number five in the architecture considered is
objects and their relationships. Both objects and their relation- a good compromise, in practice performance problems occur.
ships are characterized by attributes. Spatial environmental Simplifying the complexity of layers on the one hand in-
objects (such as lakes, bridges, buildings, clouds, whales, trees creases the run-time overhead on the other hand. Conse-
and cars) have, e.g., a shape, and other attributes that can quently, various ways to optimize DBMS performance are
change over time, e.g. the water temperature in a lake, the po- developed and the number of layers is reduced for some sys-
sition of a whale, etc. That is why time and space are impor- tem functions.
tant components of an environmental system. Recently, The development of the layer L5 during last 10 years re-
environmental data transmission is supported by wireless net- sulted in specification of so-called object-relational (OR)
work technology. data model. Its part is standardized in the standard
Obviously, it is not surprising that in many cases only tra- SQL:1999 (ISO, 1999) and, recently, in SQL:2003 (ISO,
ditional file-oriented solutions are at disposal for collections of 2003). In the OR model tables can have structured components
objects considered. For example, the CORIE (Columbia River of their rows, columns can be even of a user-defined type. Spa-
Estuary) system based on three forms of data (scientific data, tial data, time series or texts belong to this category. Such ex-
catalogue data, and task data) produces in its simulations 5 GB tensible approach resulted in the so-called universal DBMSs
of forecasted data each day (Bright and Maier, 2005). Never- in late 1990s. The core of these engines has been extended by
theless, its Metadata Repository is schema-less, no file for- loosely coupled additional modules (components) for each
mats, database access libraries, or XML schemas need be new data type. The vendors of leading DBMSs call these com-
agreed upon. In connection with Internet, Web services, and ponents extenders, datablades, and cartridges, respectively. For
EISs, such solutions seem to be unsustainable. example, spatial and text components belong to the most suc-
cessful among many others.
3. Layered architecture of DBMS The possibility of user-defined types has introduced a lot of
serious problems into implementation of the DBMS architec-
Everybody using e.g. a relational database is aware of the ture. For some such types, e.g. video, image, text, and audio,
fact, that tables of data occurring on the top of a database sys- there are standardized sets of predicates and functions for ma-
tem are virtual in some sense. More specifically, they provide nipulating their instances, but an open problem remains how to
a logical data structure suitable for user-oriented processing integrate these types into a common framework in DBMS ar-
data in the database. In early 1980s, Harder and Reuter chitecture. The implementation of new access paths, like new
(1983) proposed a mapping model consisting of five layers. types of indexes, usually results in modifications of the DBMS
Table 1 adopted from Harder (2005) shows these five layers kernel, e.g. SQL compiler, queries optimizer, etc. Such
in detail. We can observe objects to be dealt with at each changes are very expensive, time consuming, and error prone
abstraction level and particular functions implementing map- to implement and test new access methods and user-defined
pings between two consecutive layers. For example, the non- types, e.g. contiguous data flow from streaming data sources.
Each vendor uses a different approach to open the host sys-
Table 1 tem architecture to a certain degree. Oracle cartridges are re-
Description of the five-layer DBMS mapping hierarchy stricted to secondary index integration. In IBM DB2
Level of abstraction Objects Auxiliary mapping data extender, there is a framework for indexing data of new data
L5 Non-procedural Tables, views, Logical schema types restricted only to B-trees. It means evaluation of only
access rows description some types of queries can be improved with such indexing.
L4 Record-oriented, Records, sets, Logical and physical In other words, a new functionality is supported, but only
navigational access hierarchies, schema description
for limited class of user requirements.
networks
L3 Record and access Physical records, Free space tables, It seems that contribution of such software is mainly in the
path management access paths DB-key translation cases of requirements that can be decomposed into relatively
tables independent parts evaluated separately in the DBMS core
L2 Propagation control Segments, pages Buffers, page tables and the module implementing a particular data type. In other
L1 File management File, blocks Directories
words, these frameworks are either too complex or not flexible
1582 J. Pokorny / Environmental Modelling & Software 21 (2006) 1579e1586
enough to cope with the wide range of requirements in do-
main-specific access methods. A real, seamless integration
could be hardly achieved with most of these attempts. In other
words, current implementations of layered DBMS architecture
are not sufficient and fail in the case of universal DBMSs.
The other issue of traditional solutions is that they are avail-
able mainly for static applications. Because of the inherent
space and time components of environmental data, an environ-
mental system can be implemented on top of a spatio-temporal
DBMS. Unfortunately, such database software is also under
development today.
4. New technologies influencing database architectures
For data representing environmental objects a number of
new technologies are relevant. Some of them, e.g. sensor
data and sensor networks, stream processing, approaching un-
certain and imprecise data, knowledge discovery methods and
intelligent data analysis, and wireless broadcast and mobile
computing are the same as those influencing the database de-
velopment in general. We discuss them shortly in the follow-
ing subsections.
4.1. Sensor data and sensor networks
A sensor network is designed to transmit the data from an ar-
ray of sensors to a data repository on a server. Sensor networks
are based on inexpensive micro sensor technology, which will
enable most environmental objects to report their temperature,
pressure, state or location, e.g. via a global positioning system,
in real time. These small, battery-powered devices are placed in
areas of interest, e.g., in the soil, across the rain forests, or even
a glacial area to track global warming and climate change. Each
sensor node collects environmental data primarily about its im-
mediate surroundings. These data can support applications
whose main purpose is to monitor the objects attributes. Various
environmental data can be collected, analyzed to forecast the up-
coming phenomenon, and send prompt warnings.
The sensors are generally self-powered, wireless devices
with limited processing speed, storage capacity, and communi-
cation bandwidth. Such a device draws far more power when
communicating than when computing. Thus, when querying
the information in the network as a whole, it is often prefera-
ble to distribute as much of the computation as possible to the
individual nodes. Some system architectures for environmen-
tal monitoring include a base station equipped by a (relational)
DBMS which communicates with wireless sensor networks.
The base station ensures an access and a control from remote
users. With a server behind, a many-layered architecture can
look similarly to that in Fig. 1. In fact, the network becomes
a new kind of database machine, whose optimal use requires
operations to be pushed as close to the data as possible. In
a more complicated case, sensors and/or users of the sensor
networks can be even mobile.
Sensor networks provide important data sources and create
new data management requirements. For example, they do not
necessarily use a simple one-way data stream over
Sensor Network Server
Base Station Base Station
Sensor Nodes
Fig. 1. A multi-layered sensor network architecture.
a communications network. Elements of the network architec-
ture will make decisions about what data to pass on, such as
local area summaries and filtering in order to minimize power
use while maximizing information content.
Sensor information processing raises many of the most in-
teresting database issues in a new environment, with a new set
of constraints and opportunities. Huge data sets of environ-
mental data generated by sensors will be distributed through-
out the world, and can come and go dynamically. For example,
the Earth Observing System (EOS) of NASA is a collection of
satellites producing data regarding atmosphere, oceans, and
land about 1/3 of a petabyte of information per year. Since ter-
abytes of data from individual nodes will soon be the norm,
new requirements on computational and data management
infrastructures appear. For example, DataDirects enterprise-
class S2A 6000 Silicon Storage Appliance directly supports
up to 512 workstations and 180 terabytes of storage.
From one perspective, sensor networks are similar to dis-
tributed databases, but with inherent real-time properties.
One important difference is that the evaluation rate of data
produced in a sensor network is much higher than typically
considered in distributed DBMSs. This breaks the traditional
information integration paradigm, since there is no practical
way to extract and load data into a common database to
each such occurrence. Also strategies of query optimization
and query processing must be redefined.
There are a lot of examples of such networks in practice.
For example, Mainwaring et al. (2002) mention experiments
with environmental monitoring in the context of two wildlife
habitats: Great Duck Island and James Reserve. Based on
the requirements from the researchers studying these habitats,
they propose a sensor network architecture for this class of ap-
plications. On a much larger scale, the development of Envi-
ronmental Observations and Forecasting System (EOFS)
combines real time in situ monitoring with distribution net-
works that carry data to centralized processing sites. One ex-
ample of this is the above cited CORIE project. The
FLOODNET project (Envisense, 2004) plans to provide a flood
warning in the UK.
 J. Pokorny / Environmental Modelling & Software 21 (2006) 1579e1586
In general, there is a need to design flexible, lightweight da-
tabase abstractions that are optimized for data movement as
opposed to data storage (Stonebraker and Cetintemel, 2005).
4.2. Stream processing
As mentioned in Section 4.1, sensors can produce continu-
ous, possibly infinite, streams of data. EISs based exclusively
only on the traditional store-and-query model cannot handle
the volume and velocity of streaming data, whose values might
exist a moment. There are a growing number of applications,
monitoring e.g. an environment, where DBMSs are used to cre-
ate a near-real-time image of some critical parts of the environ-
ment. Comparing to the streams of usual scientific data that
typically run at a constant speed, monitoring applications
must be able to accommodate significant fluctuations in the data.
Traditional DBMSs are unsuited to deal with such streams
for various reasons (Amato et al., 2004):

sensor nodes produce and deliver data continuously with-
out receiving requests for that data,

queries over collected data can be less frequent than data
insertions,

produced data has often to be processed in real-time
because it can represent events that need a rapid answer,

queries run continuously because data streams never ter-
minate, so that they can see system conditions change dur-
ing their execution,

because of storage constraints, an entire stream cannot be
stored in the disk,

because data streams are possibly infinite, only non-block-
ing operators can be used, and

if the data to be processed is not available, then operators
must process data only when nodes make it available.
In consequence, stream processing is not a data management
task; it is a data-filtering task. New architectures, so called data
stream processing systems (DSPS), have emerged; see e.g. Car-
ney et al. (2002). A rather restricted solution, stream-processing
engine (SPE), is an example of a new database architecture that
enables the execution of queries, computations, and actions on
streaming data in real time. Such SPE should accept SQL-like
queries, stream-oriented, continuous queries and execute them
over live event streams with outputting results in real time.
In SPEs most of the data processing is processed in main mem-
ory, read or write operations to storage is optional, and can be
handled asynchronously in many cases.
For example, in a recent pilot program, Streambase devel-
oped by Stonebraker (StreamBase Systems, Inc., 2005) should
be able to analyze 140,000 messages/s, while a leading rela-
tional DBMS could handle only 900 messages/s.
4.3. Approaching uncertain and imprecise data
In addition to data management issues of environmental
data in data streams, many other problems arise. There are dif-
ferent sources that cause information to be uncertain:
1583
incompleteness, inconsistency, vagueness, imprecision, and
error. Worboys (1998) associates these notions typically with
spatial data. Incompleteness is related to totally or partly miss-
ing data: the prototypical situation of this kind is when a data
set is obtained from digitizing paper maps and pieces of lines
are missing. Inconsistency arises when several versions of the
same object exist, due either to different time snapshots, or
data sets of different sources, or different abstraction levels.
Vagueness is an intrinsic property of many natural geographic
features that do not have crisp or well-defined boundaries.
Imprecision is due to a finite representation of spatial entities:
the basic example of this kind is the regular tessellation used in
raster data, where the element of the tessellation is the smallest
unit that represents space. Scientific measurements have stan-
dard errors. Error is everything that is introduced by limited
means of taking measurements. For example, location data
for moving objects involve uncertainty in current position.
Individual sensors are not reliable and, consequently, wire-
less communication is also unreliable. Thus, various approaches
are used to provide more accurate estimation of the environ-
ment. In multisensor data fusion, approaches like fuzzy sets
or DempstereShafer evidential theory are sometimes used
(Ramamritham et al., 2004). Sequences and images require
approximate processing based on similarities, metrics, etc.
Another source of using techniques based on similarities is the
information retrieval area. Considering environmental data
equipped by metadata expressed by text strings, we can take
also these methods into account. An excellent survey of similar-
ity measures used which are applicable on environmental data is
presented in Nunez et al. (2004).
Traditional DBMSs were applied to business data process-
ing, which typically focused on numbers and character strings.
In those application areas, data elements are precise quantities
like address, quantity on hand, balance, status, and delivery
date. As a result, current DBMSs have no facilities for either
approximate data or imprecise queries.
Huge data sets and their imperfect nature produce a number
of direct consequences for computing in general. They include
(Cohen, 2005):

the notion of practical complexity must be revised in the
sense that any above-linear algorithms might be too time
consuming; one may even avoid algorithms having large
coefficients of linearity,

the processed results should reflect existing data
imperfections,

the ability to perform pre-processing and use incremental
algorithms will become essential approaches in reducing
computing times, and

approximate solutions may be the only resort for solving
large complex problems.
4.4. Knowledge discovery and intelligent data analysis
Environmental data often need to be analyzed in order to
obtain information necessary for environmental management
1584 J. Pokorny / Environmental Modelling & Software 21 (2006) 1579e1586
decisions. Environmental Decision Support Systems (EDSS)
are often mentioned in this context. EIS and EDSS are major
building blocks in environmental management and environ-
mental science today. EIS and EDSS are usually said to
have certain characteristics, which distinguish them from stan-
dard information systems, e.g. information complexity in time
and space or incompleteness or fuzziness of data items
(Denzer, 2005). The authors of the project GESCONDA men-
tion in Gibert et al. (2005) the high quantity of information
and knowledge patterns that are implicit in large databases
coming from environmental domains, specially oriented to en-
vironmental databases.
Particularly, DM methods are suitable for this purpose. His-
torically, DM has focused on efficient ways to discover models
of existing data sets. These models must expose some useful
aspects of the data, while obscuring details not useful for the
intended application. In comparison to simple forms of regu-
larities/dependencies treated by statistical methods, DM
methods can find more complex hypotheses that include
both numerical and logical conditions. Algorithms have been
developed by many research communities to perform such op-
erations as classification, clustering, association-rule discov-
ery, and summarization. These techniques are now part of
mainstream products from the major DBMS vendors and
most of them are applicable in EISs.
OLAP or DW techniques are often sufficient. For example,
temperature and pressure trends are required in an environ-
ment. Derivation of such information typically requires past
temperatures and pressures stored in a database and processed
along the time dimension. Often multidimensional data struc-
tures are used in the context of such applications. Data in
OLAP and DW systems are processed by columns rather
than by rows. Data processing uses also special indexing tech-
niques like bitmap indexes and various trees. Although a lot of
suitable data structures have been developed during last 20
years, only few of them, e.g. UB-trees and M-trees, are inte-
grated into commercial DBMSs. Rather specialized engines
absorb them. The architecture with two engines united by
a common parser occurs in practice. Classical transaction da-
tabase and DW database are stored separately and viewed as
one database.
Recent interests in combining DM technology with DBMSs
require new approaches to storage data sets to be mined and to
optimize DM processing. New research directions include:
(1) multi-dimensional OLAP for discovering unusual patterns
in stream data;
(2) mining clusters and outliers in stream data for discovering
unusual patterns; and
(3) single-pass classification methods for stream DM.
4.5. Wireless broadcast and mobile computing
Data broadcast is an attractive alternative to on demand ac-
cess because it can broadcast data simultaneously to a large
number of clients at a fixed cost. It is suitable for location-
based services, which exhibit strong temporal and spatial
locality in that clients within the neighbourhood and a certain
time period tend to seek the same kind of information (Zheng
and Lee, 2005).
The data to be broadcast includes also sensor data. Sensors
deployed in the environment can broadcast their data periodi-
cally or when interesting events happen. Unlike to traditional
computing, client devices cannot make requests to sensors for
the data. Instead, client devices just listen to the broadcast
channels passively. Thus, the sensors have the initiative in
communication. Sensors may broadcast data periodically, if
they are measuring a continuous phenomenon producing envi-
ronmental data, or may broadcast data only when a particular
event occurs, if they are detecting whether an RFID tag2 has
just come into range.
Higher-level sensors in a sensor network can pre-process
low-level sensor data and then broadcast this derived informa-
tion to client devices. Such processing can require modified
database techniques to be successful.
Since environmental data require often to be disseminating
timely to the user anytime and anywhere, a mobile environ-
ment is of increasing importance in this context. Particularly
in periodic broadcast, data are broadcast periodically on a wire-
less channel. A mobile client listens to the broadcast channel
and downloads the desired data from the channel according to
a query issued from the user or a stored profile of interest on
the client. Of course, these networks should be also able re-
spond to aperiodic queries.
Besides this, mobile devices introduce yet another category
of application (Seltzer, 2005): caching relevant portions of
a larger data set on a smaller, low-functionality device. One
can think of a mobile device as a cache of a global data set.
This model has attractive properties e in particular, the ability
to augment the local data set with entries as they are used or
needed. Mobile telephony infrastructure requires similar cach-
ing capabilities to maintain communication channels to the de-
vices. The access pattern observed in these caches is also read
mostly, and the data itself is completely transitory; it can be
lost and regenerated if necessary.
We observe that location becomes a very important prop-
erty of data and introduces a new dimension to data access
methods. Traditional data access methods are not suitable
for such computing and new researches redefine some well-
known techniques, e.g. spatial queries, in the mobile environ-
ment with a particular emphasis on broadcast data.
5. Towards new database architectures
Database technology seems to be fundamental for a deploy-
ment of technologies presented in Section 4 in context of EISs.
Some attempts to influence the development of EISs by database
specialists exist even from the past. For example, the Sequoia
2000 (Stonebraker, 1994) project speaks about collaboration
2
RFID (Radio-frequency identification) is the latest technology based on
radio waves that is useful for precisely identifying objects.
 J. Pokorny / Environmental Modelling & Software 21 (2006) 1579e1586
between computer scientists and environmental researchers to
design a next-generation information system for managing
data for global change research. The primary challenges with
a database approach are flexibility without complexity and
ease of use. Moreover, a database approach brings the opportu-
nity to link all data together on a user level, and it will make all
analysis of the data easier, e.g. via the database technology like
DM.
A common view on issues mentioned in Section 4 concerns the
DBMS architecture. In fact, todays DBMSs provide a universal
architecture applicable to a lot of various types of tasks. By words
of Stonebraker and Cetintemel (2005) one size fits all. In new
architectures of DBMSs, separate engines rather made to mea-
sure are supposed according to requirements of various applica-
tions. Besides rather traditional applications e OLAP, data
warehouses, and text retrieval, another candidates for a separate
engine are:

stream processing,

sensor networks,

scientific data bases,

native XML databases.
We have tried to highlight some characteristics of the first
three technologies with respect to their association to environ-
mental data management.
Considering native XML databases, solutions with separate
engines are popular today. Harder (2005) presents XTC archi-
tecture (XML Transaction Controller) which proves that native
XML DBMS can be implemented along the lines of five-layer
architecture (see Table 1). Also a possibility of a hybrid engine
occurs. To integrate relational and XML data, IBM develops
a new hybrid DB2 DBMS enabling to work on a truly native
XML store that sits side by side with DB20 s relational data re-
pository. On top of both data stores (relational and XML) sits
one hybrid database engine. Similar solution is used by most
vendors combining a data warehouse DBMS and a usual on-
line transaction processing DBMS, which are united by a com-
mon parser. Such architecture can be inspiring for
implementation of other data types too.
Another approach evolves the original idea of DBMS ex-
tensibility. Acker et al. (2005) developed an Access Manager
specification, a new programming interface to several layers
of a DBMS kernel. This enables the programmer to add new
data structures to the DBMS with a minimum of effort.
There is also the third approach to achieve a flexibility of
processing data in a database way: to produce a storage en-
gine that is more configurable so that it can be tuned to the
requirements of individual applications (Seltzer, 2005). There
are fundamentally two properties that a solution must possess
to address the wide range of application needs emerging
today: modularity and configurability. A modular DBMS en-
gine must allow the developer to use or exclude major subsys-
tems depending on whether the application needs them. The
DBMS must be also configurable to its operating environ-
ment: the specific hardware, operating system, and applica-
tion using it.
1585
6. Conclusions
Environmental data management, analysis, and communi-
cation are essential components of environmental character-
ization and decision making. DBMSs, the Internet, and
associated Web technologies have become an integrating force
for these components.
According to Selinger (2005), data research challenges for
the next decade include apart from other things the following
tasks:

re-examine DBMS architecture and invent ways to scale
more and better, without sacrificing user-visible availabil-
ity or performance,

learn what managing content is all about, what is needed
and create new models,

treat metadata as a first class research.
We have focused mainly on the first issue. The others can also
improve accessibility and availability of environmental data.
Approaching uncertain and imprecise data as well as knowledge
discovery and intelligent data analysis requires new models and
semantic annotations of the data. Some specific approaches al-
ready exist. For example, to increase environmental data quality
new information processing occurs that preserves and retrieves
the origins and processing history d that is, the lineage d of
objects and processes (Bose and Frew, 2005). To ensure that
the greatest use is made of environmental data, data producers
should include data lineage (and authenticity information) in
the metadata. On a database level, this requires more sophisti-
cated techniques for metadata processing.
Everything indicates that the development of new database
technologies has and will have consequences which will affect
EISs of the future.
Acknowledgement
This research was supported in part by the National
programme of research (Information society project
1ET100300419).
References
Abiteboul, S., Agrawal, R., Bernstein, P.A., Carey, M.J., Ceri, S., Croft, W.B.,
DeWitt, D.J., Franklin, M.J., Garcia-Molina, H., Gawlick, D., Gray, J.,
Haas, L.M., Halevy, A.Y., Hellerstein, J., Ioannidis, Y.E., Kersten, M.L.,
Pazzani, M.J., Lesk, M., Maier, D., Naughton, J.F., Schek, H.-J.,
Sellis, T.K., Silberschatz, A., Stonebraker, M., Snodgrass, R.T.,
Ullman, J.D., Weikum, G., Widom, J., Zdonik, S.B., May 2005. The Low-
ell Database Research self-assessment. Communications of the ACM 48
(5), 111e118.
Acker, R., Pieringer, R., Bayer, R., 2005. Towards truly extensible database
systems. In: Proceedings of DEXA 2005 Conference, LNCS 3588.
Springer-Verlag, pp. 596e605.
Amato, G., Caruso, A., Chessa, S., Masi V., Urpi, A., 2004. State of the art and
future directions in wireless sensor networks data management, 2004-TR-
16, ISTI.
Bose, R., Frew, J., 2005. Lineage retrieval for scientific data processing: a sur-
vey. ACM Computing Surveys 37 (1), 1e28.
1586 J. Pokorny / Environmental Modelling & Software 21 (2006) 1579e1586
Bright, L., Maier, D., 2005. Deriving and managing data products in an environ-
mental observation and forecasting system. In: Proceedings of Conference
on Innovative Data Systems Research (CIDR), January 2005. pp. 162e173.
Carney, D., Cetintemel, U., Cherniack, M., Convey, C., Lee, S., Seidman, G.,
Stonebraker, M., Tatbul, N., Zdoni, S., 2002. Monitoring streams e a new
class of data management applications. In: Proceedings of the 28th Inter-
national Conference on Very Large Data Bases. Morgan Kaufmann Pub-
lishers, pp. 215e226.
Cohen, J., 2005. Updating computer science education. Communications of
the ACM 48 (6), 29e31.
Denzer, R., 2005. Generic integration of environmental decision support systems e
state-of-the-art. Environmental Modelling & Software 20 (10), 1217e1223.
Envisense, 2004. FloodNet: pervasive computing in the environment. Avail-
able at: <http://envisense.org/floodnet/floodnet.htm>.
Gibert, K., Sanchez-Marre, M., Rodrguez-Roda, I., 2005. GESCONDA: an
intelligent data analysis system for knowledge discovery and management
in environmental databases. Environmental Modelling & Software 21 (1),
115e120.
Harder, T., Reuter, A., 1983. Concepts for implementing a centralized database
management system. In: Proceedings of International Computing Sympo-
sium on Application Systems Development, March 1983. B.G. Teubner-
Verlag, Nurnberg, pp. 28e104.
Harder, T., 2005. DBMS architecture e still an open problem. In: Proceedings
of BTW, Karlsruhe, March 2005. pp. 2e28.
ISO, 1999. Information technology e database languages e SQL e Part 1:
framework (SQL/framework). ISO/IEC 9075-1:1999.
ISO, 2003. Information technology e database languages e SQL e Part 2:
foundation (SQL/foundation). ISO/IEC 9075e2:2003.
LCA. Glossary. Available at: <http://www.lineadecreditoambiental.org/html/
glossary.html>.
Mainwaring, A., Polastre, J., Szewczyk, R., and Culler D., 2002. Wireless sen-
sor networks for habitat monitoring. Intel Research Berkeley, IRB-TR-
02e006.
Nunez, H., Sanchez-Marre, M., Cortes, U., Comas, J., Martnez, M., Rodr-
guez-Roda, I., Poch, M., 2004. A comparative study on the use of similar-
ity measures in case-based reasoning to improve the classification of
environmental system situations. Environmental Modelling & Software
19 (9), 809e819.
Ramamritham, K., Son, S.H., Dipippo, L.C., 2004. Real-time databases and
data services. Real-Time Systems 28, 179e215.
Reuter, A., 2005. Databases: the integrative force in cyberspace. In: Data
Management in a Connected World, LNCS 3551. Springer Verlag,
pp. 3e16.
Selinger, P., 2005. Five data challenges for the next decade. In: Key note of the
Conference ICDE, held in April 2005, Tokyo, Japan.
Seltzer, M.I., 2005. Beyond relational databases. Databases 3 (3), 50e58.
Stonebraker, M., 1994. Sequoia 2000 e a reflection on the first three years.
Sequoia technical report S2K-94-58, Berkeley, CA. Available at: <http://
epoch.cs.berkeley.edu:8000/sequoia/techreports/s2k-93-23/>.
Stonebraker, M., Cetintemel, U., 2005. One Size Fits All an idea whose
time has come and gone. In: Proceedings of the Conference ICDE, April
2005, Tokyo, Japan. pp. 2e11.
StreamBase Systems, Inc., 2005. StreamBase 2.0. Available at: <http://
www.streambase.com/index.html>.
Tomasic, A., Simon, E., 1997. Improving access to environmental data using
context information. ACM SIGMOD Record 26 (1), 11e15.
Worboys, M.F., 1998. Imprecision in finite resolution spatial data. GeoInfor-
matica 2, 257e279.
Zheng, B., Lee, D.L., May 2005. Information dissemination via wireless
broadcast. Communications of the ACM 48 (5), 105e110.