Professional Documents
Culture Documents
Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol
known respectively as suprapermafrost, intrapermafrost, and sub- permafrost groundwater (Zhou et al., 2002; Woo, 2012).
Supraper- mafrost groundwater occurs mostly in the active layer. As this layer is subject to seasonal freezing and thawing, the
active soil freezing–thawing processes play a decisive role in the recharge, movement, spatial distribution, and cycle of
suprapermafrost groundwater in cold areas (Quinton and Marsh, 1999; Woo, 2012; Chang et al., 2015). The circulation of the
suprapermafrost groundwater is important to the hydrological cycle in the per- mafrost region because it is strongly linked to the
processes of infiltration, evaporation, redistribution of water in the soil, and exfiltration in support of runoff (Callegary et al.,
2013). Only the suprapermafrost groundwater supplies water to the base flow for the continuous permafrost area (Woo, 2012;
Ligotin et al., 2010). Therefore, understanding the dynamics and variation of the suprapermafrost groundwater under climate
change is crucial in revealing the hydrological cycle processes of cold regions.
Mathematical modeling is generally used as a powerful tool to improve our understanding of groundwater systems. In non-
permafrost regions, physically based numerical models have been used in recent years for the simulation and analysis of
groundwa- ter systems and have been applied to problems ranging from aquifer safe yield analysis to groundwater remediation
and quality issues. However, these modeling techniques are very data- intensive, labor-intensive and expensive. The
hydrogeological theory and models that have widely been used in non- permafrost regions, such as Darcy’s flow law based
differential equation systems of groundwater dynamics, are not suitable for permafrost groundwater flows that are controlled by
thermody- namics (Rushton, 2003; Bense et al., 2009; Ge et al., 2011). Further- more, field observation data of groundwater
systems are very sparse in permafrost regions. In the permafrost region of QTP, for example, there were no field groundwater
monitoring sites before 2008 (Cheng and Jin, 2013; Chang et al., 2015). Under data-scarce conditions, the use of physical-based
models is highly restricted. Therefore, in such cases, empirical models serve as an attractive alternative, as they can provide
useful results using less data, are less laborious and are more cost-effective. Artificial neural network (ANN) models are one such
type of model, which can serve as universal approximators and are very much suited to dynamic nonlinear system modeling
(ASCE, 2000a).
The applications of the ANN technique in hydrology range from real-time modeling to event-based modeling. It has been
used for rainfall–runoff modeling and precipitation forecasting as well as for the modeling of stream flows, evapotranspiration
and water quality (ASCE, 2000a,b; Garcia and Shigdi, 2006). In particular, ANN models have been successfully used in the
prediction of groundwater level changes (Daliakopoulos et al., 2005; Dillip et al., 2010; Heesung et al., 2011). However, most
research areas are arid or semi-arid areas without permafrost; very little research has been conducted on groundwater systems in
permafrost regions using artificial neural network models. Due to the cold climate and harsh conditions in the region, obtaining
the measured data is rel- atively difficult. Using an ANN forecasting model could provide a way to reveal the suprapermafrost
groundwater level fluctuations, which will help us to better understand the influence of supraper- mafrost groundwater on the
hydrological cycle in permafrost regions.
The Qinghai-Tibet Plateau in western China, a source area for several rivers in Asia such as the Yellow River and the Yangtze
River, embraces a variety of hydrologic processes. Water cycling plays an unequivocal role in buffering or intensifying the
climate impact on water resources and ecosystems (Wang et al., 2012). The warming climate has attracted researchers’ attention
to shrinking glaciers, permafrost degradation, and the deterioration of ecosystems on the Plateau (Cheng and Wu, 2007; Yang et
al.,
1212 J. Chang et al. / Journal of Hydrology 529 (2015) 1211–1220
2007; Wang et al., 2011). For insight into the impacts of climate change and permafrost degradation on the regional water cycle,
it is imperative to understand the suprapermafrost groundwater dynamics in QTP under these conditions. Consequently, using
extensive field monitoring data from the center of the QTP per- mafrost region in this research, we want to (1) construct a suprap-
ermafrost groundwater level model by using an ANN framework and examine the validity of the model and (2) investigate the
impacts of air temperature and precipitation variation on the suprapermafrost groundwater dynamics and its spatial differences
along a hill slope.
2. Materials and methods
2.1. Site description and data collection
The study area is located in the Zuomaoxikong watershed of the Fenghuo Mountains. The Zuomaoxikong watershed is an
important anabranch of the Beilu River in the source area of the Yangtze River on the Qinghai-Tibet Plateau. The observation
sites of the suprap- ermafrost groundwater were located in a typical alpine meadow area on the left bank of the Zuomaoxikong
River (Fig. 1), where the permafrost is well developed. The elevation ranges from 4680 to 5360 m a.s.l., and it belongs to the arid
plateau climate region. The annual average temperature is À5.2 °C, and the annual average precipitation is 290.9 mm, but the
annual average evapo- transpiration is as much as 1316.9 mm, and the annual relative atmospheric humidity is 57%. The depth of
the permafrost layer is approximately 50–120 m, with the active layer’s depth varying from 0.8 to 2.5 m. The main type of
vegetation is alpine meadow, and the dominant vegetation community consists of Kobresiapyg- maea, K. humilis, and K.
capilifolia.
In a typical alpine meadow (site B in Fig. 1), a slope groundwa- ter flow field that was cut by the river (the underground ice
exposed at the river edge, which has an obvious exchange relation- ship between the underground water and the surface river)
was chosen to build suprapermafrost groundwater observation plots at an elevation of 4766 m. In the observation field, two
observation holes were constructed at a depth of 160 cm, of which one was located on the upper section of the slope and the other
on the lower section, with a distance of 100 m between them (Fig. 1). HOBO U20-001-04 water level data loggers (ONSET Co.,
USA) were used to monitor the suprapermafrost groundwater level. This built- in pressure type groundwater level sensor with a
fully enclosed titanium alloy shell is suitable for use in cold, harsh environments, with the measurement accuracy and resolution
being less than 0.014 kPa and 0.14 cm. The groundwater level was observed twice every day, and the average value of all
observations every day was taken as the suprapermafrost groundwater level depth. As shown in Fig. 1, two portable
micro-meteorological stations (HOBO Weather Station, ONSET Co., USA) were established near the experimental fields to
measure the air temperature, precipitation, wind velocity and direction, and total radiation. All the data were collected from July
2009 to December 2012.
2.2. Artificial neural network methods
An ANN is a large parallelly distributed information processing system that has certain performance characteristics
resembling the biological neural networks of the human brain (Haykin, 1999). A neural network is characterized by its
architecture that represents the pattern of connection between nodes, its method of determin- ing the connection weights and the
activation function. Unlike physically based numerical models, ANNs do not require explicit characterization and quantification
of the physical properties and
conditions of the system under investigation. ANNs learn the system’s behavior from representative data. The ability to learn and
generalize from sufficient data pairs makes it possible for ANNs to solve large-scale complex problems (ASCE, 2000a; Haykin,
1999). The most commonly used neural network structure is the feed-forward hierarchical architecture. A typical three-layered
feed-forward neural network is comprised of multi- ple elements called nodes and connection pathways that link them (Haykin,
1999). A neuron receives an input signal, processes it, and transmits an output signal to the other interconnected neurons. The
output of a node in a layer is only dependent on the input it receives from the previous layers and the corresponding weights. Fig.
2 shows the feed-forward network for the current study, hav- ing one hidden layer with several nodes between the input and
output layers. Information passes from the input to the output side. The nodes in one layer are connected to those in the next but
not to those in the same layer.
A training algorithm is needed to solve a neural network problem. Because there are so many types of algorithms available for
training a network, the selection of an algorithm that provides the best fit to the data is required. In the current research, the ANN
models were trained using the Levenberg–Marquardt training algorithm. The sigmoid and linear activation functions were used
for the hidden and output node(s), respectively. The optimal number of neurons in the hidden layer was identified using a trial
and error procedure by varying the number of hidden neurons
J. Chang et al. / Journal of Hydrology 529 (2015) 1211–1220 1213
Fig. 1. Location of the study area and the sketch of observations. Groundwater monitoring sites presented in this paper were
made at site B.
from 2 to 25. Furthermore, the optimal network architecture was selected as the one with a minimum mean squared error (MSE).
2.3. Model development
An artificial neural network model is developed in this study to reproduce the dynamic processes of the suprapermafrost
ground- water level for a period using the observed time series of the exter- nal variables. To perform the system identification,
the neural network model is first trained to perform predictions of the suprapermafrost groundwater level depth using past
observed groundwater level depths. Once this autoregressive model has been developed, simulations are produced by feeding
forward its output as the simulation time increases.
The selection of appropriate input parameters is a very important aspect of ANN modeling because it provides the basic
information about the system being modeled. Despite decades of research in hydrological modeling, no specific guidelines exist
on how to solve those problems. Statistical procedures were sug- gested for appropriate input vectors for a model (Sudheer et al.,
2002; Lin et al., 2006). In this study, a partial autocorrelation func- tion (PACF) was employed to determine the number of
parameters corresponding to different values of the suprapermafrost ground- water level depth antecedents. PACFs are generally
used to diag- nose the order of an autoregressive process and can be employed in prediction modeling (Sudheer et al., 2002). The
PACF of the daily
V1
V2
V3
V
9
Input layer Hidden layer Output layer
Groundwater level
Fig. 2. Architecture of the feed-forward three-layer ANN for the study area.
data is shown in Fig. 3. Considering the difficulty of data acquisi- tion in this study region, three previous lags were considered as
inputs to the models in this study (Fig. 3). The inputs present the previous suprapermafrost groundwater level depth (t, t À 1, t À
2), and the output layer node corresponds to the supraper- mafrost groundwater level depth at time t + 1. Thus, the following
combinations of output data of the suprapermafrost groundwater level depth were evaluated: (1) GW
t+1
; (2) GW
t+3
; (3) GW
t+5
; (4) GW
t+7
. The other input parameters for the ANN model were selected by considering the parameters that have the potential to affect
the groundwater level. To examine the effect of precipitation and tem- perature on groundwater, we plotted the daily
suprapermafrost groundwater level depth along with the precipitation and temper- ature. Fig. 4 shows the daily variation of
suprapermafrost ground- water level depth at two sites, along with the daily precipitation, indicating that the suprapermafrost
groundwater level depth is generally higher on high precipitation days. Hence, precipitation is one potential input parameter that
influences the supraper- mafrost groundwater level depth of the study area. Similarly, the relationship between the daily
temperature and the daily variation of the suprapermafrost groundwater level depth at the two sites is shown in Fig. 5. Due to the
good correlation between the daily tem- perature and the suprapermafrost groundwater level depth, air temperature is considered
another potential input parameter that influences the groundwater of the study area. To correspond to the suprapermafrost
groundwater level data, the input variables of precipitation (P) and temperature (T) are as follows: P
t
, P
tÀ1
, P
tÀ2
and T
t
, T
tÀ1
, T
tÀ2
. We knew that the more variables of the ANN model utilization, the better prediction capability of the
model performance. However, there are two main factors restrict- ing the availability of more variables in this study. Firstly, it is
very difficult to obtain some more monitoring data in the serious harsh conditions over 4500 m a.s.l. altitude with periglacial
environment, which limited our effort to acquire more field data. Secondly, it is possible that add other parameters can raise the
simulation preci- sion of the model, but the significant of accuracy improvement is relatively small or weak after considering
more than the three parameters, and while the efficiency of the ANN model was signif- icantly reduced. Therefore, in this study,
we developed the ANN model in two ways, one using the three driving factors of the air temperature, precipitation and the
previous suprapermafrost groundwater level depth and another using only the two driving factors of the air temperature and
precipitation, for application in the absence of prior suprapermafrost groundwater level data.
1214 J. Chang et al. / Journal of Hydrology 529 (2015) 1211–1220
) F C A P ( s n o i t a l e r r o c o t u A l a i t r a P
To eliminate dimensional differences, before the training pro- cess, the inputs and target data were normalized by the
following equation:
X
norm
1⁄4
X
X À X
min
ð1Þ max
where X
norm
À X
min
, X, X
min
and X
max
stand for the normalized value, real value, minimum value, and maximum value, respectively.
The performances of the models developed in this study were assessed using standard statistical performance evaluation
criteria, including the coefficient of correlation (R), root mean squared error (RMSE), mean absolute error (MAE) and
Nash–Sutcliffe efficiency coefficient (NS). R measures the degree to which two variables are linearly related. RMSE and MAE
provide different types of
8#
1
0.8
0.6
0 2 4 6 8 10 12 14 16 18 20 Lag (day)
7#
) F C A P (
0.8
s n o i t a
0.6
l e r r o c
0.4 0.4
o t u A
0.2
0.2
l a i t r a
0
0
P
-0.2
-0.2
-0.4
0 2 4 6 8 10 12 14 16 18 20
Lag (day)
Fig. 3. Partial autocorrelation function of suprapermafrost groundwater level depth time series at 7# and 8#.
)
P
50
P ) m
40
7# 8#
0
m ( n
30
-0.5
o i t a t i
20
-1 p i c e r
10
-1.5
0
-2
m ( h t p e d l e v e l r e t a w d n u o r 2-Oct 1-Dec 30-Jan 31-Mar 30-May 29-Jul 27-Sep Date
Fig. 4. Daily suprapermafrost groundwater level depth fluctuations at sites 7# and 8# with daily precipitation.
-1.5
g
15
T
0 7# 8#
2-Oct 1-Dec 30-Jan 31-Mar 30-May 29-Jul 27-Sep
g -25
Fig. 5. Daily suprapermafrost groundwater level depth fluctuations at sites 7# and 8# with daily temperature.
)
T
10 ) ( e
5
r u t a
0
-0.5
r e p m e
-5
-10
-1 -15
m ( h t p e d l e v e l r e t a w d
-20
Date
n u o r
information about the predictive capabilities of the model. The RMSE measures the goodness-of-fit, especially that of the high
suprapermafrost groundwater level values, whereas the MAE yields a more balanced perspective of the goodness-of-fit at mod-
erate value distributions of the estimation errors. The following equations were used for the computation of these parameters.
R 1⁄4
P
n
Q
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P
n
i1⁄41
i1⁄41 r Q
o i
À Q
o
p i
À Q
p
ð2Þ
RMSE 1⁄4
Q
o i
À Q
o
2P
n i1⁄41
Q
p i
À Q
p
2
r
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 n
ð3Þ
MAE 1⁄4
X
n
i1⁄41
À Q
p i
À Q
o i
Á
2
1
Xn
n
i1⁄41
ð4Þ
NS 1⁄4 1 À
Qp i
À Qo i
P P
n
n i1⁄41
À Qo i
o
Á
2
i1⁄41
Q
i
À À Qp Q
i
o
2
ð5Þ
where n is the number of input samples, Q
o i
and Q
p i
are the observed and predicted suprapermafrost groundwater level depths at
time t, and Qo and Qp are the means of the observed and predicted suprap- ermafrost groundwater level values, respectively. The
best fit between the observed and predicted values would have R = 1, RMSE = 0, MAE = 0 and NS = 1.
2.4. Data analysis
All data types (including the daily average precipitation, daily average temperature and daily average suprapermafrost ground-
water level depth data) collected for predicting the groundwater level depth cover 653 days, to be regarded as 653 sets of data.
All data were divided into two parts, as training and testing periods. The first 364 sets of data (2010/10/2–2011/9/30) were used
for training of the ANN models, and the last 289 sets of data (2011/10/1–2012/7/15) were employed for testing purposes. Table 1
shows the statistics of the training, testing and total data for the daily suprapermafrost groundwater level depths of the 7# and 8#
sites. The recorded daily maximum suprapermafrost groundwater level depths were À1.21 and À1.05 m for the 7# and 8# sites,
respectively. The daily minimum suprapermafrost groundwater level depths were À0.049 and À0.072 m for the 7# and 8# sites,
respectively. The observed daily suprapermafrost groundwater level values show low positive skewness for all stations,
indicating that the data has a less scattered distribution.
J. Chang et al. / Journal of Hydrology 529 (2015) 1211–1220 1215
3. Results
3.1. ANN model with three input factors
By using the temperature, precipitation and previous supraper- mafrost groundwater level as driving factors, the performance
of the ANN models for forecasting the 1, 3, 5 and 7-day-ahead suprap- ermafrost groundwater level variations at the 7# and 8#
sites were summarized in Table 2. During the testing periods, the perfor- mance of the forecasting model for 1-day ahead is better
than those of the 3, 5, 7-day ahead time forecasts. The accuracy of the forecasted results decreases as the lead time of the forecast
increases. However, the performance of the ANN model also has excellent validity for 3-day, 5-day and 7-day ahead forecasting,
with the R values in the testing period of the two sites almost all above 0.95. The RMSE and MAE values, less than 0.081m and
0.055m, respectively, indicate the valuable and appropriate simulation capacity of the ANN model. Normally, a model can be
claimed to produce a perfect and accurate estimate if the NS crite- rion is greater than 0.8 (Shu and Ouarda, 2008). The NS values
for the forecasting ANN models in this study are almost all over 0.8, which indicates that the models can produce acceptable
results (Table 2). However, there were differences between the upper and downslope sites in the forecasting performance of the
ANN model. On the whole, the ANN model with the three driving factors of temperature, precipitation and previous
suprapermafrost groundwater level has more forecasting validity at the downslope 7# site than at the upper slope 8# site.
The suprapermafrost groundwater level depths predicted using the ANN model developed above were compared with field-
observed data of sites 7# and 8# for 1, 3 and 5-days ahead, as shown in Fig. 6. The correlation coefficient R ranged from 0.92 to
0.98 for the 7# site and from 0.85 to 0.97 for the 8# site, which indicates that the suprapermafrost groundwater level values
estimated by the ANN forecasting models were closely matched with the observed values and followed the same dynamic trend
in the shorter term. The developed ANN model could exactly identify the varying pro- cesses of the suprapermafrost groundwater
level that fit better with the measured variation of the groundwater level during the freezing and thawing periods (Fig. 7).
Although the prediction accuracy of the high suprapermafrost groundwater levels during the summer season was not better than
that of the low suprapermafrost ground- water levels during the autumn and winter seasons, the ANN fore- casting model with
three driving factors developed in this study could provide accurate and reliable daily suprapermafrost ground- water level depth
predictions for a short time of 1 week.
3.2. ANN model with the two factors of temperature and precipitation
The above results indicate that the ANN model with three driv- ing factors (temperature, precipitation and previous supraper-
mafrost groundwater level) developed in this study has validity and reliability in simulating and predicting the suprapermafrost
groundwater level variation. However, there is a lack of observed groundwater level data in most permafrost regions, while the
cli- mate factors of temperature and precipitation are relatively easy to obtain. Thus, if the ANN model developed above could be
exe- cuted using only climate factors to predicate the dynamic trend of the suprapermafrost groundwater level, the model should
have application value in permafrost regions. Based on the ANN model developed above, the previous suprapermafrost
groundwater level factor was removed from the original three driving factors, so that the input variables are precipitation (P) and
temperature (T).
When the input variables are the temperature (T) and precipita- tion (P), the performance of the ANN models for forecasting
the 1,
Table 1 Statistical parameters of suprapermafrost groundwater level depth data in each data set.
Plots Min Max Std Mean Cs Cv
7# All À0.05 À1.21 0.33 À0.82 0.82 À0.40 Training À0.05 À1.15 0.36 À0.75 0.51 À0.48 Testing À0.30 À1.21 0.26 À0.91 1.13
À0.28
8# All À0.07 À1.05 0.23 À0.79 1.48 À0.29 Training À0.07 À1.01 0.25 À0.77 1.40 À0.32 Testing À0.1 À1.05 0.20 À0.83 1.47
À0.24
Note: Min is the minimum suprapermafrost groundwater level depth, Max is the maximum suprapermafrost groundwater level
depth, Std is the standard deviation, Cs is the coefficient of skew, and Cv is the coefficient of variation.
3, 5 and 7-day-ahead suprapermafrost groundwater level dynamics at the 7# and 8# sites are shown in Table 3. The R values in
the testing period of the two sites were almost all above 0.72,
1216 J. Chang et al. / Journal of Hydrology 529 (2015) 1211–1220
Table 2 Performance statistics of the ANN model with three input variables (groundwater, precipitation and air temperature)
during the training and testing periods for 1, 3, 5, and 7-day ahead suprapermafrost groundwater level forecasting.
Training periods Structure Testing periods
R RMSE MAE NS R RMSE MAE NS
1-day ahead 7# 0.9948 0.0366 0.0213 0.9896 9-6-1 0.9918 0.0344 0.0248 0.9817 8# 0.9864 0.0400 0.0226 0.9729 9-5-1 0.9860
0.0344 0.0218 0.9713
3-day ahead 7# 0.9747 0.0812 0.0503 0.9485 9-2-1 0.9715 0.0607 0.0389 0.9418 8# 0.9264 0.0919 0.0540 0.8579 9-2-1 0.9574
0.0556 0.0401 0.9130
5-day ahead 7# 0.9615 0.0980 0.0632 0.9245 9-2-1 0.9577 0.0714 0.0495 0.9140 8# 0.8857 0.1137 0.0670 0.7837 9-3-1 0.9236
0.0699 0.0466 0.8529
7-day ahead 7# 0.9688 0.0983 0.0699 0.9235 9-6-1 0.9559 0.0723 0.0476 0.9109 8# 0.8437 0.1318 0.0846 0.7115 9-2-1 0.8897
0.0806 0.0552 0.7901
-0.1
a
8#
y = 0.972 x - 0.034 R2 = 0.984
F
-1.3
y = 0.9839x - 0.0251 R2 = 0.9438
-1.3
a
7# 0
-0.3 ) m ( d e t s a
-0.2
y = 1.0044x - 0.0021
-0.5
R2 = 0.9722
c
-0.7
F
-1.2
measured (m)
) m ( d e t s a
-0.4
e
c e
-0.6 r o
-0.9
r o
-0.8
-1.1
-1
-1.3 -1.1 -0.9 -0.7 -0.5 -0.3 -0.1 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0
measured (m)
-0.1
b
7#
0
b
8#
) m (
-0.3
)
-0.2
y = 1.0215x + 0.0063 R2 = 0.9187
d e t s a
-0.5
F
F
-1.2 -1.3 -1.1 -0.9 -0.7 -0.5 -0.3 -0.1 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0
measured (m)
measured (m)
y = 0.9607x - 0.0533 R2 = 0.9171
-1.3
-1.2 -1 -0.8 -0.6 -0.4 -0.2
m ( d e t s a c
-0.4
c e r o
-0.7
e r o
-0.6
-0.9
-0.8
-1.1
-1
c
7#
y = 0.9975x + 1E-05 R2 = 0.8531
-1.1
c
8#
Fig. 6. Comparison of the measured suprapermafrost groundwater values those forecasted 1 day, 3 days and 5 day (corresponding
to a, b, and c, respectively) ahead by the ANN model.
-0.3
-0.1
) m
-0.5
) m ( d
-0.3
( d e t s
-0.7
e t s a c e
-0.5 a c e r o F
-0.9
r o F
-0.7
-1.1
-0.9
-1.1 -0.9 -0.7 -0.5 -0.3 -0.1 measured (m)
measured (m)
indicating the good correlation between the observed and fore- casted suprapermafrost groundwater levels using the ANN model
derived from only the temperature and precipitation. The RMSE
and MAE were less than 0.176m and 0.126m, respectively. Meanwhile, the NS are all over 0.51. Those statistical indices indi-
cated the forecasting validity and reliability of the ANN model with two input variables. In addition, the performance of the
7-day ahead forecast is better than that of the 5, 3, and 1-day ahead fore- casts both for upper slope and down slope sites (Table
3). This is contrary to the results of the ANN model with three input factors mentioned above. The accuracy of the three input
variables ANN model decreased with the lead time of the forecast increase. These results indicate that the longer term dynamics
of the supraper- mafrost groundwater level were affected by climate factors to a larger extent. The effects of air temperature and
precipitation together could explain more than 77% of the variation of the suprapermafrost groundwater level in 7 days ahead.
Furthermore, it can be observed from Table 3 that the statistical indices of the forecasting results at the upper slope 8# site were
better than that at the downslope 7# site, which indicates that the dynamics of the suprapermafrost groundwater level on the
upper slope were more sensitive to the climate variation than that in downslope. On the upper slope, the effects of air temperature
and precipitation explain more than 82% of the variation in the suprapermafrost groundwater level 7 days ahead. Therefore, the
ANN forecasting model developed in this study using easily obtainable climate data could provide valid and reliable predications
for the daily suprapermafrost groundwater level variation.
3.3. Response of suprapermafrost groundwater level to climate change and spatial variation on the hill slope scale
Over the last 10 years, the air temperature has increased by 0.47 °C/10 years, while the precipitation has increased slightly at
the headwaters of the Yangtze and Yellow Rivers. According to this
J. Chang et al. / Journal of Hydrology 529 (2015) 1211–1220 1217
-0.3
G
7#
-0.1
8# ) m ( l
-0.5
Forcasred measured
) m ( l
-0.3
Forcasred
e v
e v
measured e l r e
-0.7
e l r e
-0.5 t a w d n u
-0.9
t a w d n u
-0.7
o r
-1.1
o r G
-0.9
-1.3
-1.1 8-Oct 23-Nov 8-Jan 23-Feb 9-Apr 25-May 10-Jul 8-Oct 23-Nov 8-Jan 23-Feb 9-Apr 25-May 10-Jul
Date
Date
Fig. 7. Comparison of the measured suprapermafrost groundwater values and those forecasted 7 days ahead by the ANN model
for the 7# and 8# sites.
Table 3 Performance of the ANN model with two input variables (precipitation and air temperature) during the training and
testing periods for 1, 3, 5, and 7-day ahead suprapermafrost groundwater level forecasting.
Training period Testing period
R RMSE MAE NS R RMSE MAE NS
1-day ahead 7# 0.7388 0.2535 0.1762 0.5011 0.7253 0.1764 0.1262 0.5193 8# 0.6364 0.1878 0.1336 0.4036 0.7661 0.1291
0.0860 0.5709 3-day ahead 7# 0.7188 0.2639 0.1849 0.4557 0.7168 0.1734 0.1175 0.5123 8# 0.5857 0.1977 0.1420 0.3417
0.7797 0.1206 0.0827 0.5914 5-day ahead 7# 0.7184 0.2635 0.1901 0.4534 0.7483 0.1645 0.1115 0.5534 8# 0.5959 0.1976
0.1385 0.3463 0.7968 0.1120 0.0749 0.6226 7-day ahead 7# 0.7625 0.2407 0.1698 0.5413 0.7682 0.1562 0.1062 0.5846 8#
0.6201 0.1925 0.1373 0.3843 0.8181 0.1030 0.0690 0.6572
actual climate change background of the research region in the pase 10 years and the prediction trends of climate change in future
100 years by IPCC (AR5) (IPCC, 2013), we assume that the two par- ticular scenarios of future climate change in a year as
follows: 1 Precipitation increases by 10%, while the air temperature increases by 0.2 °C, 0.5 °C or 1 °C; 2 Precipitation increases
by 20%, while the air temperature increases by 0.2 °C, 0.5 °C or 1 °C. The response of the suprapermafrost groundwater level to
climate change are then forecasted by using the ANN model set up above based on T and P, and the results are listed in Table 4.
Future climate change with warming and slight wetting could greatly increase the annual suprapermafrost groundwater level.
In the climate change scenarios of a precipitation increase of 10% and 20%, with an air temperature increase of 1 °C, the average
annual suprapermafrost groundwater level would increase by 2.5% (2.35 cm) and 2.6% (2.44 cm) per year, respectively. There
are spatial variations in the response of the groundwater level to climate change at slope scale. The variation of the amplitude of
the groundwater level at the downslope site is larger than that at the upper slope site, with an air temperature gradient from
approximately 1.0 cm at 0.2 °C to 1.3 cm at 1.0 °C (Table 4). This means that the response of the suprapermafrost groundwater to
climate change is more intense and fluctuant at the downslope site than at the upper slope site. Under the effects of climate
change in the future, the varying amplitude of the minimum groundwater level in the summer will be obviously larger than that in
the winter (Table 4).
Different from non-permafrost regions, the effects of the precipitation change on the suprapermafrost groundwater level were
less than that of the air temperature change. With the same temperature increase, the variation of the suprapermafrost
groundwater level from 10% to 20% precipitation increase is only
Table 4 The response of suprapermafrost groundwater level to climate change (cm).
Site Index Normal P10% P20%
T0.2 T0.5 T1.0 T0.2 T0.5 T1.0
7# in down slope Average depth À92.39 À91.94 À91.24 À90.04 À91.84 À91.14 À89.95 Min À39.55 À38.67 À38.24 À37.50
À38.26 À37.87 À37.04 Max À115.19 À115.10 À114.96 À114.67 À115.09 À114.95 À114.67 Average change 0 2.85 3.04 3.56
2.89 3.09 3.62
8# in upper slope Average depth À83.03 À82.64 À82.15 À81.41 À82.62 À82.16 À81.40 Min À61.18 À61.18 À61.18 À61.18
À61.18 À61.18 À61.18 Max À94.63 À94.65 À94.68 À94.71 À94.66 À94.68 À94.71 Average change 0 1.82 2.02 2.31 1.85 2.01
2.33
Note: (1) T0.2, T0.5 and T1.0 means air temperature increased by 0.2 °C, 0.5 °C and 1.0 °C, respectively. P10% and P20%
means the precipitation increased by 10% and 20%, respectively. (2) The minimum groundwater level is the shallowest
groundwater level during the active soil thaw period in the summer, and the maximum groundwater level is the deepest
groundwater level during the active soil freezing in the winter.
0.02–0.05 cm. However, with the same precipitation increase, the variation in the suprapermafrost groundwater level ranges from
0.55 cm to 2.44 cm with air temperature increase from 0.2 °C to 1.0 °C. The air temperature change controls the inter-annual
variations of the suprapermafrost groundwater system and the expected trend in the future under the effects of climate change. In
fact, it is shown in Fig. 5 that the actual amplitude of annual groundwater level variation ranged from 15 to 20cm with 7–11 °C of
temperature increase during thawing season (from May to August), which means the variation ratio of groundwater level is
approximately 2.0 cm per 1 °C for increase in temperature. Therefore, the forecasting result of groundwater level rise 2.44 cm for
1 °C increase in temperature under same precipitation scene is approximately consistent with the actual variation amplitude per
year in present.
4. Discussion
4.1. Validity of ANN model with single and multi-variables in permafrost groundwater
A physical mechanism based mathematical model of hydrogeol- ogy is certainly a powerful and valid tool to simulate and
forecast variations in groundwater systems. The applicability of a ground- water model to a real situation depends on the
accuracy of the input data and the parameters (Rushton, 2003). Groundwater is difficult to observe and measure because it resides
below the land surface, and aquifer systems are complex due to flows into and out of the aquifer, the interaction between the
surface water and groundwater, and the uncertainty of aquifer properties. Thus, data-scarce conditions and parameter uncertainty
present limita- tions and formidable challenges to numerical groundwater model applicability. Seeking to overcome the data
limitations and param- eter uncertainty and to simplify the physical assumptions inherent to numerical models, artificial neural
networks (ANNs) have been considered a good alternative groundwater modeling and predic- tion approach (Rogers and Dowla,
1994; Coppola et al., 2005; Szidarovszky et al., 2007; Mohanty et al., 2010; Sheelabhadra et al., 2010). Our results confirmed
that the ANN model and predic- tion approach can be successfully used to simulate and predicate the suprapermafrost
groundwater level variation and can improve our understanding of the response of permafrost groundwater to climate changes.
To enable the ANN model to achieve excellent predictive accu-
Table 5 racy over specific field locations in complex
ground water systems,
The weight of three parameters in the prediction result
of the 3 input ANN model. it is encouraged to use as many input variables as possible. Chitsazan et al. (2013) developed an ANN
model with 7 input variables: rainfall, evaporation, relative humidity, temperature (maximum and minimum), discharge of
irrigation canals, and
7# 8# GW% 0.34364 0.342058 P% T%
0.314891 0.341468 0.311886 0.346056
1218 J. Chang et al. / Journal of Hydrology 529 (2015) 1211–1220
groundwater recharge from the plain boundary. Nonetheless, most researchers found that three input variables of accessible field
data (e.g., water levels, precipitation and temperature or evaporation) were adequate to develop an excellent ANN model
(Daliakopoulos et al., 2005; Szidarovszky et al., 2007; Mohanty et al., 2010). In this study, we developed two ANN models of
three input variables (e.g., water levels, precipitation and temperature) and two input variables (e.g., precipitation and
temperature). In comparing Table 2 with Table 3, it is found that the evaluation cri- teria (e.g., R, NS and RMSE) of model
performance were signifi- cantly superior for the three input variable model to that with two variables. Taking the previous
suprapermafrost groundwater level as one of the three input variables, the ANN model achieved excellent predictive accuracy
over the slope area in permafrost ground water systems, with R > 0.89, NS > 0.79 and RMSE < 0.081 for the 1 week ahead time
prediction (Table 3). The results were consistent with the most findings on ANN technology applications worldwide (Coppola et
al., 2005; Mohanty et al., 2010; Banerjee et al., 2009; Sheelabhadra et al., 2010). However, the ANN model using the two input
variables of precipitation and air temperature had also reasonably good prediction of suprapermafrost ground- water levels for
both the upper and downslope sites, with R >0.76, NS>0.58 and RMSE<0.16 for the 1week ahead time prediction (Table 3).
The principal component analysis (PCA) method was used to calculate the weight or contribution of the three factors in the
pre- diction result of the three variables input ANN model. The calcula- tion results, that are listed in Table 5, show that the
weights of the previous of suprapermafrost groundwater level and temperature are approximately similar with 34%, while the
weight of precipita- tion is slightly less with 31%. Due to the previous of supraper- mafrost groundwater level data is the initial
conditions of the model, it occupies a relative large weight in the model prediction (contribution of 34%). Thus, when removing
the previous of suprapermafrost groundwater level in ANN model, the forecasting accuracy would decrease. However, the
consolidated contribution of the temperature and precipitation together reached to 66%, that implied the two factors of
temperature and precipitation could explain most variation of suprapermafrost groundwater level. This is why we can develop the
two input variables ANN model to effec- tively simulate and predict the suprapermafrost groundwater level with satisfactory
results.
In the permafrost region, however, there were difference per- formances of different ANN models at the upper slope and down-
slope sites. When the input variables included the previous suprapermafrost groundwater level, downslope site (7# site) pre-
diction was superior, with a higher R (0.96) and NS (0.91), to that at the upper slope (8# site), with a lower R (0.89) and NS
(0.79). In contrast, when the input variables excluded the previous suprapermafrost groundwater level, as shown in Table 3, the
site-specified prediction of the groundwater level was better at the upper slope site (8#), with a higher R (0.82) and NS (0.66),
than at the downslope site (7#), with a lower R (0.77) and NS (0.58). These results indicate that the previous suprapermafrost
ground- water level, which includes interflow recharge and the same suprapermafrost groundwater flow recharge from the upper
slope (Chang et al., 2015), has more significant effects on the predicted downslope suprapermafrost groundwater level than that
in the upper slope region, while the climate factors had more intense effects on the groundwater level in the upper slope region
than that downslope. In fact, due to the sensitivity and vulnerability of the suprapermafrost groundwater to climate changes in the
permafrost region, the simulation of the suprapermafrost ground- water movement and discharge must consider the heat transmis-
sion and coupled regulation of heat and water flow (Bense et al., 2009; Quinton and Marsh, 1999).
4.2. Impacts of climate changes on suprapermafrost groundwater level
The suprapermafrost groundwater level is intensely affected by the freezing–thawing process of the active layer, which
directly controls the formation and development of the suprapermafrost groundwater system (Walker et al., 2003; Woo, 2012;
Cheng and Jin, 2013). Different from non-frozen regions, the soil freezing process not only confines the water interchange
between the land surface, active soil and permafrost and impedes water movement in the impervious frozen layers, but it also
results in an entropy change process of the phase transition of frozen soil moisture. Therefore, there are unique behaviors of the
hydrological cycle in the groundwater of the permafrost region (Cheng and Jin, 2013; Chang et al., 2015), where the temperature
has crucial impacts on the suprapermafrost groundwater system.
The sensitivity of the suprapermafrost groundwater level to cli- mate change was analyzed by using the ANN model
developed in this study, and the results are shown in Table 4. It is found that there was almost no effect of precipitation increases
of 10% and 20% on the suprapermafrost groundwater level at either the upper or downslope sites. Contrarily, when the air
temperature increased by 0.5 °C, the forecasted suprapermafrost groundwater level increased by an average of 1.15cm (1.24%) at
the 7# site and 0.88 cm (1.06%) at the 8# site. If the air temperature increased by 1.0 °C, the forecasted suprapermafrost
groundwater level increased by an average of 2.24cm at the downslope 7# and 1.62 cm at the upper slope 8# site, respectively.
Even if the air tem- perature increased by only 0.2 °C, the response of the forecasted suprapermafrost groundwater level was
more significant than that with a precipitation increase of 20%. These results were consistent with the findings in other arctic
permafrost regions (Woo et al., 2008; Shiklomanov and Lammers, 2009). Due to the climate warm- ing, the mean annual
groundwater levels in the taiga zone of the West Siberian artesian basins rose on the order of 10–20 cm over the period from
1965 to 2005 (Ligotin et al., 2010). There was a positive correlation among climate warming, a change in the groundwater level
and the discharge, and it was suggested that the accelerated increase in the groundwater discharge from permafrost degradation
led to the runoff increase in the northern Eurasian and northwestern North American rivers in recent decades (Ye et al., 2009;
Smith et al., 2007; Shiklomanov et al.,
J. Chang et al. / Journal of Hydrology 529 (2015) 1211–1220 1219
2013). In the permafrost regions of the Qinghai-Tibet Plateau, Ge et al. (2011) also found that with increased warming, there is
more groundwater flow in the active layer and thereby increased groundwater discharge to rivers. The response of groundwater
sys- tem to climate change in permafrost region is a cumulative pro- cess. According to the forecasting results of this research,
the suprapermafrost groundwater level would raise to 23–36 cm for 10 years under the hypothetical scenarios of climate change
(tem- perature increase 1 °C and precipitation increase 20%), which is a great change in suprapermafrost groundwater system
including the groundwater distribution and discharge in the research region. These results demonstrate that warming has a great
effect on the suprapermafrost groundwater system in the permafrost region, while a variation in the precipitation within 20% has
no obvious effects. Although the suprapermafrost groundwater dynamics on the upper slope have a more significant correlation
with climate change than that downslope, the variation ratio and amplitude of the suprapermafrost groundwater level at the
downslope site are larger than that at the upper slope site under climate warming. The reason for the spatial difference is that the
variation in the suprapermafrost groundwater level at the downslope results from not only the site self-response to climate change
but also the delivery-responses of the upper and lateral sites to climate change through groundwater movement (Chang et al.,
2015; Woo, 2012).
5. Conclusions
In the permafrost region, suprapermafrost groundwater occurs mostly in the active layer. Thus, the active soil
freezing–thawing processes play a decisive role in the recharge, movement, spatial distribution, and groundwater level dynamics.
The hydraulic con- ductivity and liquid water storage capacity are therefore tempera- ture dependent and vary sharply as the soil
freezes and thaws. The heterogeneity of the thermal distribution on the slope scale accounts for the spatial heterogeneity of the
freezing–thawing pro- cess, which resulted in the groundwater systems exhibiting a greater spatial heterogeneity than in
non-permafrost areas. The ANN technology and prediction approach is useful and valid in simulating and forecasting
suprapermafrost groundwater level variations to overcome data limitations, parameter uncertainty and the formidable challenge
of numerical groundwater model applicability.
For the site-specific prediction of the suprapermafrost ground- water levels on the slope scale, two ANN models were
developed in this study. If there were some field observations of the suprap- ermafrost groundwater level, the ANN model with
three input vari- ables (groundwater level, precipitation and temperature) is suggested to be developed and applied in the
permafrost region. The three input variable ANN model has a significantly superior real-time site-specific prediction capability
and produces an excel- lent accuracy performance in the simulation and forecasting of the suprapermafrost groundwater level
dynamics. When there are no field observations of the suprapermafrost groundwater level, the ANN model developed using only
two input variables of accessible climate data (precipitation and air temperature) also has a good accuracy for simulation and
prediction of the variation of suprap- ermafrost groundwater levels.
The suprapermafrost groundwater level is intensively affected by climate warming changes and has more sensitivity to
tempera- ture variation than to precipitation. Under a scenario of the tem- perature increasing by 1.0 °C per 10 years, the
suprapermafrost groundwater level is predicted to increase by 2.5% and 2.6% per 1 year with precipitation increases of 10% and
20%, respectively. On the slope scale, there were spatial variations in the responses of the suprapermafrost groundwater level to
climate change.
Generally, the variation ratio and amplitude of the supraper- mafrost groundwater level downslope is larger than that at the upper
slope under climate warming. The obvious vulnerability and spatial variability of the suprapermafrost groundwater to cli- mate
change will drive intensive effects on understanding the water cycle and alpine ecosystems in permafrost regions.
Acknowledgments
This study was funded by the Natural Science Foundation of China (No. 41301024) and the National Basic Research Program
of China (973, No. 2013CBA01807).
References
ASCE Task Committee, 2000a. Artificial neural networks in hydrology – I:
preliminary concepts. J. Hydrol. Eng. ASCE 5 (2), 115–123. ASCE Task Committee, 2000b. Artificial neural networks in
hydrology – II:
hydrologic applications. J. Hydrol. Eng. ASCE 5 (2), 124–137. Banerjee, P., Prasad, R.K., Singh, V.S., 2009. Forecasting of
groundwater level in hard rock region using artificial neural network. Environ. Geol. 58, 1239–1246. Bense, V.F., Ferguson, G.,
Kooi, H., 2009. Evolution of shallow groundwater flow systems in areas of degrading permafrost. Geophys. Res. Lett. 36,
L23401. http:// dx.doi.org/10.1029/2009GL039225. Callegary, J.B., Kikuchi, C.P., Koch, J.C., Lilly, M.R., Leake, S.A., 2013.
Review:
Groundwater in Alaska (USA). J. Hydrol. 21, 25–39. Chang, J., Wang, G.X., Li, C.J., Mao, T.X., 2015. Seasonal dynamics of
suprapermafrost groundwater and its response to the freeing-thawing processes of soil in the permafrost region of Qinghai-Tibet
Plateau. Sci. China: Earth Sci. 58, 727–738. Cheng, G.D., Wu, T., 2007. Responses of permafrost to climate change and their
environmental significance, Qinghai-Tibet Plateau. J. Geophys. Res. 112, F02S03. http://dx.doi.org/10.1029/2006JF000631.
Cheng, G.D., Jin, H.J., 2013. Groundwater in the permafrost regions on the Qinghai-
Tibet Plateau and its changes. Hydrogeol. Eng. Geol. 40, 1–11. Chitsazan, M., Rahmani, G., Neyamadpour, A., 2013.
Groundwater level simulation using artificial neural network: a case study from Aghili plain, urban area of Gotvand, south-west
Iran. Geopersia 3 (1), 35–46. Coppola, E., Rana, A., Poulton, M., Szidarovszky, F., Uhl, V., 2005. A neural network
model for predicting water table elevations. Ground Water 43 (2), 231–241. Daliakopoulos, I.N., Coulibaly, P., Tsanis, I.K.,
2005. Groundwater level forecasting
using artificial neural network. J. Hydrol. 309, 229–240. Dillip, K.G., Sudhansu, S.P., Prakash, C.S., 2010. Prediction of water
table depth in western region, Orissa using BPNN and RBFN neural networks. J. Hydrol. 394, 296–304. Garcia, L.A., Shigdi, A.,
2006. Using neural networks for parameter estimation in
ground water. J. Hydrol. 318 (1–4), 215–231. Ge, S., McKenzie, J., Voss, C., Wu, Q., 2011. Exchange of groundwater and
surface- water mediated by permafrost response to seasonal and long term air temperature variation. Geophys. Res. Lett. 38,
L14402. http://dx.doi.org/ 10.1029/2011GL047911. Haykin, S., 1999. Neural Networks: A Comprehensive Foundation, second
ed.
Prentice-Hall, Englewood Cliffs. Heesung, Y., Seong-Chun, J., Yunjung, H., Gwang-Ok, B., Kang-Kun, L., 2011. A
comparative study of artificial neural networks and support vector machines for predicting groundwater levels in a coastal
aquifer. J. Hydrol. 396, 128–138. IPCC, 2013. The IPCC Fifth Assessment Report (ar5), Climate Change 2013: The
Physical Science Basis, Stockholm, September 2013. Ligotin, V.A., Savichev, O.G., Makushin, J.V., 2010. The long-term
changes of seasonal and annual levels and temperature of ground waters of the top hydro dynamical zone in Tomsk area.
Geoecology 13 (1), 23–29 (in Russian). Lin, J.Y., Cheng, C.T., Chau, K.W., 2006. Using support vector machines for long-term
discharge prediction. Hydrol. Sci. J. 51, 599–612.
1220 J. Chang et al. / Journal of Hydrology 529 (2015) 1211–1220
McKenzie, J., Voss, C., Seigel, D., 2007. Groundwater flow with energy transport and water–ice phase change: Numerical
simulations, benchmarks, and application to freezing in peat bogs. Adv. Water Resour. 30, 966–983. http://dx.doi.org/
10.1016/j.advwatres.2006.08.008. Mohanty, S., Jha, M.K., Kumar, A., Sudheer, K.P., 2010. Artificial neural network modeling
for groundwater level forecasting in a river island of eastern India. Water Resour. Manage. 24, 1845–1865. Peterson, B.J.,
Holmes, R.M., McClelland, J.W., Vörösmarty, C.J., Lammers, R.B., Shiklomanov, A.I., Rahmsterf, S., 2002. Increasing river
discharge to the Arctic Ocean. Science 298, 2171–2173. Quinton, W.L., Marsh, P., 1999. A conceptual framework for runoff
generation in a
permafrost environment. Hydrol. Process. 13, 2563–2581. Rogers, L.L., Dowla, F.U., 1994. Optimization of groundwater
remediation using artificial neural networks with parallel solute transport modeling. Water Resour. Res. 30 (2), 457–481.
Rushton, K.R., 2003. Groundwater Hydrology: Conceptual and Computational
Models. John Wiley and Sons Ltd., ISBN: 0-470-85004-3. Sheelabhadra, M., Madan, K.J., Ashwani, K., Sudheer, K.P., 2010.
Artificial neural network modeling for groundwater level forecasting in a rover island of eastern India. Water Resour. Manage.
24, 1845–1865. Shen, Y.P., Wang, G.X., Pu, J.C., Wang, X., 2009. Impacts of climate change on glacial water resources and
hydrological cycles in the Yangtze River source region, Qinghai-Tibetan Plateau, China: A progress report. Sci. Cold Arid Reg. 1
(6), 475– 495. Shiklomanov, A.I., Lammers, R.B., Lettenmaier, D.P., Polischuk, Y.M., Savichev, O.G., Smith, L.C.,
Chernokulsky, A.V., 2013. Hydrological changes: historical analysis, contemporary status, and future projections. In: Groisman,
P.Ya., Gutman, G. (Eds.), Regional Environmental Changes in Siberia and Their Global Consequences, Springer Environmental
Science and Engineering. Springer Science + Business Media, Dordrecht. Shiklomanov, A.I., Lammers, R.B., 2009. Record
Russian river discharge in 2007 and the limits of analysis. Environ. Res. Lett. 4. http://dx.doi.org/10.1088/1748–
9326/4/4/045015, 045015, 9 pp. Shu, C., Ouarda, T.B.M.J., 2008. Regional flood frequency analysis at ungauged sites
using the adaptive neuro-fuzzy inference system. J. Hydrol. 349, 31–43. Smith, L.C., Pavelsky, T.M., MacDonald, G.M.,
Shiklomanov, A.I., Lammers, R.B., 2007. Rising minimum daily flows in northern Eurasian rivers suggest a growing influence of
groundwater in the high-latitude water cycle. J. Geophys. Res. Biogeosci. 112, G04S47. Sudheer, K.P., Gosain, A.K., Ramasastri,
K.S., 2002. A data-driven algorithm for constructing artificial neural network rainfall-runoff model. Hydrol. Process. 16,
1325–1330. Szidarovszky, F., Coppola, E.A., Long, J., Hall, A.D., Poulton, M.M., 2007. A hybrid artificial neural
network-numerical model for ground water problems. Ground Water 45 (5), 590–600. Walker, D.A., Jia, G.J., Epstein, H.E.,
2003. Vegetation-soil-thawing-depth relationships along a low-arctic bioclimate gradient. Alaska: synthesis of information from
the ATLAS studies. Permafrost Periglac. Process. 14, 103–123. Wang, G.X., Wang, Y.B., Shen, Y.P., 2007. Impacts of alpine
ecosystem and climate changes on surface runoff in the source region of Yangtze River. J. Glaciol. Geocryol. 29 (2), 159–168.
Wang, G.X., Bai, W., Li, N., Hu, H.C., 2011. Climate changes and its impact on tundra
ecosystem in Qinghai-Tibet Plateau, China. Clim. Change 106, 463–482. Wang, G.X., Liu, G.S., Li, C.J., Yang, Y., 2012. The
variability of soil thermal and hydrological dynamics with vegetation cover in a permafrost region. Agric. For. Meteorol.
162–163, 44–57. Woo, M.K., Kane, D.L., Carey, S.K., Yang, D., 2008. Progress in permafrost hydrology
in the new millennium. Permafrost Periglac. Process. 19, 237–254. Woo, M.K., 2012. Permafrost Hydrology.
Springer-Verlag, Berlin Heidelberg. Yang, J.P., Ding, Y.J., Chen, R.S., 2007. Climatic causes of ecological and environmental
variations in the source regions of the Yangtze and Yellow rivers of China. Environ. Geol. 53, 113–121. Ye, B.S., Yang, D.,
Zhang, Z., Kane, D.L., 2009. Variation of hydrological regime with permafrost coverage over Lena Basin in Siberia. J. Geophys.
Res. 114, D07102. http://dx.doi.org/10.1029/2008JD010537. Zhou, Y.W., Guo, D.X., Qiu, G.Q., 2002. Geocryology in China.
Science Press, Beijing,
450.