Catena 170 (2018) 305–315

Contents lists available at ScienceDirect

Catena
journal homepage: www.elsevier.com/locate/catena

Spatial hydrological responses to land use and land cover changes in a T

typical catchment of the Yangtze River Delta region
⁎
Qiang Wanga, Youpeng Xua, , Yu Xua, Lei Wua, Yuefeng Wanga, Longfei Hanb
a
School of Geography and Ocean Science, Nanjing University, Nanjing 210023, China
b
College of Resources and Environmental Sciences, Hunan Normal University, Changsha 410081, China

A R T I C LE I N FO A B S T R A C T

Keywords: This study assessed the individual and combined effects of land-use and land-cover changes (LUCCs) on spatial
LUCCs hydrological responses by using an integrated approach involving the Soil and Water Assessment Tool (SWAT)
Spatial hydrological responses and geographically weighted regression (GWR) models in the Xitiaoxi River Basin (XRB). The LUCCs and their
SWAT spatial patterns from 1985 to 2008 were evaluated in the XRB. The hydrological processes during the period
Geographically weighted regression
from 1980–2015 were then generated by the SWAT model under the 1985 and 2008 land-use scenarios. GWR
Xitiaoxi River Basin
models were constructed to quantify the spatial impacts of LUCCs at the sub-basin scale. The results showed that
the predominant trend of land-use conversion was between forest-grass land and agricultural land, and the
diminishing portion of forest-grass land (25.93 km2) and agricultural land (46.77 km2) contributed to the ex-
pansion of urban land during the period 1985–2008. Moreover, the urban area increased from 5.6% to 17.05%,
and the change ratio progressed towards the centre of the XRB. These changes in land use caused the average
annual water yield and surface runoff to increase by 1.09% and 11.87%, respectively, and the average annual
evapotranspiration (ET), percolation and baseflow to decrease by 0.73%, 16.53% and 17.26%, respectively. In
addition, the results of the GWR models indicated that the responses of each hydrological component exhibited
spatial differences. The comprehensive land-use intensification index (La), selected to reflect the combined ef-
fects of LUCCs, was positively correlated with water yield and surface runoff but negatively correlated with ET,
percolation and baseflow. La had a greater effect on water yield, surface runoff, percolation and baseflow in rural
areas than in the urbanized region. The combined effects of LUCCs in certain local areas increased water yield
and surface runoff by up to 13.7% and 114.2%, respectively. The local coefficient indicated that rural regions
might be exposed to greater effects of LUCCs. The results of this study could be useful to understand the effects of
LUCCs on the spatial patterns of each hydrological component and to help improve flood control and water
resource management.

1. Introduction et al., 2014).

Variation of hydrological components caused by climate changes
and anthropogenic activities, such as land-use and land-cover changes
(LUCCs), has resulted in numerous environmental problems (Mittal
et al., 2016; Woldesenbet et al., 2016; Zhang et al., 2016; Rose and
Peters, 2001; Wang et al., 2014). In particular, in the process of rapid
urbanization, regional hydrological characteristics have changed sig-
nificantly because a large portion of the permeable land surface has
been replaced by impervious surfaces (Zhou et al., 2013). Quantitative
assessment of LUCCs and their hydrological impacts has been one of the
hotspots of current hydrological researches, which could facilitate the
development of sustainable water resource strategies and the im-
provement of land management options (Ghaffari et al., 2010; Tekleab
Many studies have investigated the effects of LUCCs on hydrological
processes in different geographical regions (Ye et al., 2009; Chen et al.,
2014; Zuo et al., 2016; Ghaffari et al., 2010; Nadal-Romero et al., 2016;
Woldesenbet et al., 2016; Zhang and Schilling, 2006). These studies
examined the hydrological response of an entire basin, which was based
mainly on a hydrological model only (Sajikumar and Remya, 2015;
Fossey et al., 2016) or a hydrological model and traditional statistical
methods (Nie et al., 2011; Shi et al., 2014; Woldesenbet et al., 2016;
Yan et al., 2013; Zhang et al., 2012; Yan et al., 2018; Wagner and
Waske, 2016). These results reflected the average effects of LUCCs on
the hydrological processes of the entire basin. However, the relation-
ship between LUCCs and hydrological components could not be quan-
titatively evaluated exclusively based on hydrological models (Yan

⁎
Corresponding author at: No. 163 Xianlin Avenue, Nanjing, Jiangsu Province, China.
E-mail address: xypnju@163.com (Y. Xu).

https://doi.org/10.1016/j.catena.2018.06.022
Received 31 October 2017; Received in revised form 31 May 2018; Accepted 22 June 2018
0341-8162/ © 2018 Elsevier B.V. All rights reserved.
Q. Wang et al.
et al., 2018; Zuo et al., 2016). In addition, the traditional statistical
methods, such as linear and multiple linear regression models, logistic
regression models, partial least squares regression models and principal
components analysis, assume the relationships between variables are
constant over the entire drainage area. Therefore, these results show the
average of existing relationships, which may neglect some significant
spatial characteristics and may hide local variations (Chen et al., 2016;
Tu and Xia, 2008).
Some recent studies showed that the hydrological impacts of LUCCs
in different regions differed due to the spatial heterogeneity of physical
geographical components, such as land use, topography, and soil, and
so on (Chang et al., 2014; Zhou et al., 2013; Wang et al., 2017). For
instance, Beighley et al. (2003) and Brun and Band (2000) found that
urbanization increased peak discharges and runoff volume but de-
creased streamflow variability and baseflow, whereas David et al.
(2005), Jennings and Jarnagin (2002) and Kim et al. (2002) suggested
that an increase in the impervious area led to contrasting effects on
baseflow and streamflow. Additionally, Shi et al. (2014) found that an
increase in grassland had a positive relationship with surface runoff in
the upstream region of the Luanhe River Basin but a negative re-
lationship in the downstream region. Similarly, Yan et al. (2018) and
Zuo et al. (2016) also found that the responses of hydrological com-
ponents to LUCCs exhibited spatial differences at the sub-basin scale in
the Loess Plateau of China. Hence, the hydrological effects of LUCCs
may differ between local and global scales.
Recently, a spatial analysis method called geographically weighted
regression (GWR) has increasingly been used to detect the causes of
spatial variation in geographical entities, such as river discharge
(Rennermalm et al., 2012), mean annual precipitation (Yue et al.,
2013), annual runoff (Chang et al., 2014), and surface water quality
(Chen et al., 2016; Tu and Xia, 2008). The model can effectively solve
the spatial non-stationarity issue by allowing regression model para-
meters to change over space. In addition, GWR considers the spatial
autocorrelation of variables, which is difficult to address in traditional
statistical models such as ordinary least squares (OLS) regression (Chen
et al., 2016; Fotheringham et al., 1996). Nevertheless, only a few stu-
dies have investigated spatial hydrological responses to individual and
combined effects of LUCCs by combining a distributed hydrological
model and spatial analysis model.
Therefore, the objectives of this research are to identify the most
suitable statistical model to describe the spatial relationship between
hydrological fluxes and land-use variables and to explore the spatial
patterns of hydrological responses to individual and combined effects of
LUCCs. To achieve these objectives, LUCCs and their spatial patterns
from 1985 to 2008 were evaluated in the study area. Then, the water
balance changes of the basin were estimated based on the calibrated
SWAT model, and finally, the spatial hydrological responses to the in-
dividual and combined effects of LUCCs were assessed through GWR
models, which were compared with OLS models.
2. Materials and methods
2.1. Study area
The study area is in the Xitiaoxi river watershed (1371 km2), where
located in the northwestern region of the Yangtze River Delta between
longitudes 119°14′E–119°45′E and latitudes 30°23′N–31°11′N (Fig. 1).
Xitiaoxi River, one of the main tributaries in the upstream region of
Taihu Lake, comprises five main branches: the Dipu stream, the Hu
stream, the Dragon stream, the South stream and the West stream. The
area is characterized by mountains, and its elevation ranges between 5
and 1580 m above sea level. With a humid subtropical monsoon cli-
mate, the average annual precipitation and temperature were
1584.01 mm and 15.5 °C, respectively, during the study period from
1978 to 2015.
There are two large flood-proofing reservoirs (i.e., Fushi and
306
Catena 170 (2018) 305–315
Laoshikan reservoirs) in the headwaters of Xitiaoxi River, and the main
land-use type in the upstream region of these reservoirs is forest that
has experienced little change. In this paper, the Xitiaoxi River Basin
(XRB, 661.28 km2), which extends between the two reservoirs and the
basin outlet (i.e. Hengtangcun station), had undergone large urbani-
zation and was selected as the study area. The actual outputs of the two
reservoirs (i.e., actual daily streamflow) were input as the upstream
boundary conditions of the hydrological model.
Allowing for the XRB's elevation and ongoing urban expansion, we
divided the XRB into eight zones: the urban area (UA, 69.46 km2), three
suburban areas, i.e., SADX (53.67 km2), SAMX (79.21 km2) and
SAMHD (64.65 km2), and four mountainous areas, i.e., LMUD
(63.38 km2), LMUW (80.08 km2), LMUS (60.6 km2) and HMUHD
(190.77 km2) (Table 1).
2.2. Datasets
The datasets of this study mainly included spatial data, such as to-
pography, soils and land use, and hydrometeorological data.
Topography was presented by a digital elevation model that was de-
rived from 1:10,000 topographic contour data. Soil spatial data with a
scale of 1:10,000 were obtained from the National Second Soil General
Survey data, and soil property information was provided by the Anji
Bureau of Agriculture. The land-use maps for 1985 were obtained by
digitizing land survey data (1:10,000) that had high accuracy and re-
liability. Landsat TM images were selected as the data source to extract
land-use data for 2008 using methods of supervised classification and
manual interpretation, and the results showed an acceptable precision
(Zhou et al., 2013). Therefore, the accuracy of these spatial data met
the requirement for the research. In addition, the above spatial data had
raster-based formats and all are unified into 30 m × 30 m resolution.
The hydrometeorological data from 1977 to 2015 at a daily interval
included streamflow from three stations for two reservoirs and the
outlet of the basin, precipitation from 16 rain gauge stations and
weather data from a meteorological station (Fig. 1). Weather data
consisted of daily precipitation, maximum and minimum temperature,
relative humidity, mean wind speed and daily solar radiation. These
data were supplied by the local Hydrology and Water Resources In-
vestigation Bureau and were corrected before publication with a very
small percentage of missing data. The few missing data points were
filled using calculated data based on neighbouring stations through a
simple linear regression method.
2.3. Methodology
In this study, the spatial response patterns of hydrological compo-
nents to LUCCs in the XRB were detected through the coupling of hy-
drological modelling and a spatial regression model. First, the spatial
patterns of response and explanatory variables from 1985 to 2008 were
evaluated in the XRB; then, the water balance changes in the basin were
estimated based on the calibrated SWAT model; and finally, the spatial
hydrological responses to the individual and combined effects of LUCCs
were assessed through spatial regression models.
2.3.1. Land-use variables and hydrological variables
To analyze the effects of LUCCs on regional hydrological process,
this study selected hydrological components (i.e. water yield, surface
runoff, actual evapotranspiration (ET), percolation and baseflow) as
response variables. The explanatory variables include four land-use
variables (i.e. forest-grass land, agricultural land, urban land and water
bodies) and a comprehensive index of land-use intensification degree
(La).
La is used as an indicator of urbanization and land-use intensity
(Zhuang and Liu, 1997). This index can reflect the impact of human
factors on the land system and can quantitatively measure the level of
intensive land use (Gao et al., 2015). La is calculated as follows (Hu
Q. Wang et al. Catena 170 (2018) 305–315

Fig. 1. The location and topography map of the Xitiaoxi River Basin. The catchment (i.e. the grey areas) between basin outlet and two reservoirs is the study area.
Red solid lines denote the boundaries of the eight areas. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of
this article.)

Table 1
Information for eight subregions in the Xitiaoxi River Basin.
Sub regions Description Total area (km2) Proportion of urban land in Elevation range (m) Mean elevation (m)
2008

UA The urban area 69.46 35.87% 9.21–490 65.70

SADX The suburb area in downstream of Xitiaoxi River 53.67 16.14% 5–451.56 47.08
SAMX The suburb area in midstream of Xitiaoxi River 79.21 18.99% 15–330 50.569
SAMHD The suburb area in midstream of Hu and Dragon stream 64.65 18.43% 17.44–509.38 116.61
LMUD The low mountainous area in upper regions of Dipu stream 63.38 8.61% 5–590 172.18
LMUW The low mountain area in upper West stream 80.08 11.79% 25–630 140.86
LMUS The low mountain area in upper South stream 60.60 19.24% 35–430 113.03
HMUHD The high mountain area in upper regions of Hu stream and 190.77 13.39% 60–1190 390.47
Dragon stream

307
Q. Wang et al.
Table 2
Land-use intensity grade of each land use type.
Land use types Forest-grass Agriculture land Urban land
land
Classification 1 2 3 4
index
Grading index 2 3 4 1
et al., 2015; Zhuang and Liu, 1997):
n (Abbaspour et al., 2007) to calibrate and validate the parameters.
La = 100 × ∑i =1 Ai × Ci
where La is the land-use intensification degree in each sub-basin and its
value varies from 100 to 400; Ai is the grading index of ith land-use type
(Table 2); and Ci is the percentage of ith land-use types in each sub-
basin.
Based on the well-calibrated SWAT model, this study obtained the
change rates of each hydrological component in each sub-basin (Eq.
(2)) for 1985 and 2008 land-use scenarios under fixed 1980–2015 cli-
mate conditions.
Hi, year 2 − Hi, year1
Ri, H = × 100%
Hi, year1
where year1 and year2 represent the years pertaining to the first and
second land-use scenarios, respectively; Ri, H is the change rate of hy-
drological component H (H = water yield, surface runoff, actual ET,
percolation and baseflow) in sub-basin i; and Hi, year1 and Hi, year2 re-
present the amount of hydrological component H in sub-basin i in year1
and year2, respectively.
Then, we determined the spatial responses of each hydrological
component to the changes in individual land-use types and La (Eq. (3)).
Si, j, year 2 − Si, j, year1
L i, j = × 100%
Si
where Li, j is the area change rate of land-use variables j (j = 1, 2, 3, 4, 5
refers to forest-grass land, agricultural land, urban land, water bodies
and La); Si is the area of sub-basin i; Si, j, year1 and Si, j, year2 represent the
area of land-use variable j in sub-basin i in year1 and year2, respec-
tively.
2.3.2. SWAT model
The Soil and Water Assessment Tool (SWAT) is a physically based,
semi-distributed, continuous daily time-step model developed for the
U.S. Department of Agriculture (USDA) Agricultural Research Service
by Backland Research Center in Texas (Arnold et al., 1998). The model
has been widely used to simulate hydrological processes in complex and
various watersheds around the world (see the SWAT literature data-
base: https://www.card.iastate.edu/swat_articles/). In the SWAT
model, the watershed is divided into sub-basins connected to the river
network based on a digital elevation model, and each sub-basin is
further divided into the Hydrologic Response Units (HRUs) on the basis
of a unique combination of land use, soil and slope (Arnold et al., 2012;
Neitsch et al., 2011). Hydrological components are simulated from each
HRU to the sub-basin level and then aggregated to the basin outlet
through the stream network.
The hydrological cycle simulated by SWAT is based on the following
water balance equation (Eq. (4)), in daily time steps for each HRU, and
the net water yield (WYLD) (mm H2O) to the stream channel was es-
timated by Eq. (5) (Neitsch et al., 2011).
t models with different explanatory variables as long as they apply to the
SWt = SW0 + ∑ (Rday − SURQ − ET − PERC − GWQ)
i=1
WYLD = SURQ + LATQ + GWQ − TLOSS
where SWt and SW0 are the final and initial soil water contents (mm
Catena 170 (2018) 305–315
H2O), respectively; t is the time (days); and Rday, SURQ, ET, PERC and
GWQ are the respective precipitation, surface runoff, ET, percolation
Water and groundwater flow (baseflow) amounts on day i (mm H2O); LATQ is
bodies the lateral flow contribution to stream discharge (mm H2O); and TLOSS
is the transmission loss from the system (mm H2O).
SWAT calibration and uncertainty programs (SWAT-CUP) provide
parameter calibration and verification procedures and enable sensi-
tivity analysis, calibration, validation and uncertainty analysis of the
SWAT model (Abbaspour, 2014). Therefore, based on SWAT-CUP, this
study used the Sequential Uncertainty Fitting algorithm (SUFI-2)
(1) Parameter sensitivity was determined using t-statistics and the p-value
from SWAT-CUP. Parameters with a larger absolute value of the t-sta-
tistic are more sensitive; the p-value determines the significance of the
sensitivity, and a value closer to zero indicates greater significance. The
coefficient of determination (R2), the Nash-Sutcliffe coefficient (ENS)
and percent bias (PBIAS) were used to describe the SWAT model per-
formance (Moriasi et al., 2007).
2.3.3. Geographically weighted regression
GWR was used to quantify the spatial impacts of LUCCs on the
variation of each hydrological component at the sub-basin scale. As an
(2) extension of the OLS regression model, GWR can explore local varying
relationships by embedding localized data into the regression para-
meters, which would present spatial non-stationarity within the tar-
geted geographic space (Fotheringham et al., 1996). Generated by a
local regression estimation using sub-samples from the nearest neigh-
bouring observation, the regression coefficients in GWR vary with lo-
cation (Lv and Zhou, 2016). The principle of GWR is as follows:
m
yj = β0 (uj , vj ) + ∑ βj (uj , vj ) xij + εj, j = 1, 2, …, n
i=1 (6)
where yj is the observation of the dependent variable at location j,
(3) (uj, vj) represents the spatial location of each observation, βj(uj, vj) re-
presents the local regression coefficient for independent variable xi at
location j, and β0(uj, vj) and εj represent the intercept and error term,
respectively. βj(uj, vj) is estimated by the following equation:
n m 2
β0 (uj , vj ) = ∑ wjk ⎛⎜yk − β0 (uj , vj ) − ∑ βi (uj , vj ) xij ⎞⎟
k=1 ⎝ i=1 ⎠ (7)
where wjk represents the distance decay function for location j and k,
with the basic assumption that observations closer to sample point j
have a higher impact on local regression parameters. The weighting
function can be stated using the exponential distance decay form:
wij = exp(−dij2/ b2) (8)
where wij represents the weight of observation j for observation i, and dij
and b are the distance and kernel bandwidth, respectively. ArcGIS 10.3
provides both fixed and adaptive bandwidths and calculates the optimal
distance for a fixed kernel or the optimal number of neighbours for the
adaptive kernel. This study chose the adaptive kernel bandwidth be-
cause the density of sub-basins varied over the study area.
The performance of the OLS and GWR models was evaluated using
the adjusted R2 value, the Akaike Information Criterion (AICc) and the
spatial autocorrelation index (Global Moran's I) of regression residuals.
The adjusted R2 value optimizes the R-squared value by normalizing the
numerator and denominator by their degrees of freedom; its value
ranges from 0 to 1.0, with a higher value being preferable. AICc is not
an absolute measure of goodness of fit but is useful for comparing
same dependent variable. If the AICc values for two models differ
(4) by > 3, the model with the lower AICc value is considered to be better.
A comparison of AICc values provides a way to assess the benefits of a
(5)
global model (OLS) relative to a local regression model (GWR). Global
Moran's I evaluates whether the pattern of spatial data is clustered,
308
Q. Wang et al. Catena 170 (2018) 305–315

Table 3 was the main type of land use and represented approximately two-
The land use conversion matrix between the years 1985 and 2008 in XRB (km2). thirds of the total area of the XRB, despite a continuous decrease since
1985 2008 1985. Agricultural land occupied proportions of 24.8% and 18.46% in
1985 and 2008, respectively. In addition, the urban land area exhibited
Forest- Agriculture land Urban land Water Total an obvious increase from 5.6% to 17.05%, which indicated that the XRB
grass land bodies experienced rapid urbanization since 1985. The water bodies occupied
Forest-grass land 369.61 47.07 25.93 1.33 443.94 a small proportion and exhibited a slight change for the whole area.
Agriculture land 44.15 69.77 46.77 0.54 161.23 Table 3 showed that the predominant trend of land-use conversion
Urban land 0.91 0 35.5 0.04 36.45 was the transformation between forest-grass land and agricultural land,
Water bodies 2.2 3.22 2.69 0.75 8.86 and the diminishing area of forest-grass land (25.93 km2) and agri-
Total 416.87 120.06 110.89 2.66 650.48
cultural land (46.77 km2) contributed to the expansion of urban land.
In addition, we computed the spatial distribution of each land-use
variable (i.e., forest-grass land, agricultural land, urban land, water
dispersed, or random. Over- and under-predictions for a well-specified bodies and La) between 1985 and 2008 based on Eq. (3) for all sub-
regression model will be randomly distributed. Moran's I indicates basins (Fig. 2). Deforestation occurred in the majority of the sub-basins
whether the explanatory variable is appropriate by examining the of the XRB, while there was an increase in forest-grass land in some
patterns of residuals of OLS and GWR models. parts of SAMX, SADX and LMUD. The area of agricultural land ex-
panded in the northern part of the LMUW and SAMX regions due to
3. Results and discussion deforestation. Urban expansion occurred in all sub-basins but to dif-
ferent degrees, and the change ratio progressively increased towards
As the aim of the current study is to assess the spatial patterns of the centre. The largest increase in the urban area occurred in the UA
hydrological responses to the individual and combined effects of region, which is the location of the only county town (Anji County) in
LUCCs, the model parameters were estimated and the application of the the XRB. The area of water bodies was very small and changed little at
SWAT model was assessed. Then, the effects of LUCCs on hydrological each sub-basin, i.e., within a range from −10.0% to 0. The spatial
components in the XRB were investigated. Finally, the spatial hydro- distribution of the La change showed that the land-use intensification
logical responses to the individual and combined effects of LUCCs were degree increased in most regions.
assessed through GWR models.
3.2. SWAT model performance
3.1. LUCCs
SWAT model calibration and validation were carried out after
The area of each land-use type and the land-use conversion matrix parameter sensitivity analysis, and the years 1977–1979, 1980–1997
between the years 1985 and 2008 was calculated using the Raster and 1998–2015 were treated as the warm-up, calibration and validation
Calculator tool in ArcGIS 10.3, as shown in Table 3. Forest-grass land periods, respectively. The land use/cover in 1985 was used for the

Fig. 2. Spatial distribution of changes in each land use variable between the years 2008 and 1985 in XRB. (a) Forest-grass land, (b) Agriculture land, (c) Urban land,
(d) Water bodies and (e) La.

309
Q. Wang et al. Catena 170 (2018) 305–315

Table 4
List of parameters and their ranking that produced the 12 highest relative sensitivity and optimal values for SWAT calibration.
Parameters Rank Definition Range of values in SWAT-CUP t-Statistics p-Value Fitted value

V__ALPHA_BNK.rte 1 Baseflow alpha factor for bank storage 0–1 22.84 0.00 0.68
V__CH_N2.rte 2 Manning's “n” value for the main channel 0.01–0.2 −14.35 0.00 0.05
V__CH_K2.rte 3 Effective hydraulic conductivity in main channel alluvium 5–99 −10.98 0.00 57.17
V__CH_L1.sub 4 Longest tributary channel length in subbasin 0.05–20 −5.36 0.00 0.3
V__CH_K1.sub 5 Effective hydraulic conductivity in tributary channel alluvium 0–300 −3.89 0.00 7.7
R__CN2.mgt 6 SCS runoff curve number −0.05–0.05 3.03 0.00 0.04
V__GW_REVAP.gw 7 Groundwater “revap” coefficient 0.02–0.2 −2.50 0.01 0.06
V__ESCO.hru 8 Soil evaporation compensation factor 0–1 2.48 0.01 0.97
V__EPCO.hru 9 Plant uptake compensation factor 0–1 −2.00 0.05 0.50
V__CANMX.hru 10 Maximum canopy storage 0–95 −1.67 0.10 18.31
R__SOL_K(1).sol 11 Saturated hydraulic conductivity −0.3–0.3 1.65 0.10 0.13
V__REVAPMN.gw 12 Threshold depth of water in the shallow aquifer for “revap” to occur (mm) 0–300 −1.54 0.12 207.11

Table 5 Table 6
Evaluation of SWAT model performance in monthly time step. Average annual values of hydrological components and its' changes under two
different land use scenarios during the period of 1980–2015 in XRB.
Calibration (1980–1997) Validation (1998–2015)
Water Surface ET (mm) Percolation Baseflow
2
R 0.89 0.89 yield runoff (mm) (mm)
ENS 0.89 0.87 (mm) (mm)
PBIAS (%) −1.66 −0.07
1985 land use 849.56 451.65 568.03 294.13 226.79
2008 land use 858.78 505.28 563.91 245.50 187.64
Changes 9.22 53.63 −4.12 −48.63 −39.15
Change rate (%) 1.09 11.87 −0.73 −16.53 −17.26
initial set up, calibration of the SWAT model, and the land use/cover in
2008 was used for validation of the model. Model pre-processing and
setup were performed using the ArcSWAT extension for ArcGIS 10.3.
This study used SWAT-CUP to calibrate and validate the parameters. 3.3. The effects of LUCCs on hydrological components
We chose 25 parameters that might be sensitive in the XRB in ac-
cordance with the relevant literature and then selected the top 12 3.3.1. Changes in the hydrological components at basin and sub-basin scales
parameters with the most sensitive values to calibrate the SWAT model Based on the calibrated SWAT model, the hydrological processes of
(Table 4). XRB were simulated under 1985 and 2008 land-use scenarios during
The performance of SWAT model in XRB is shown in Table 5 and the period of 1980–2015 (Table 6). Compared to the land-use condition
Fig. 3. For both the calibration and validation periods at monthly level, in year 1985, the average annual water yield value increased slightly
R2 and ENS are higher than 0.85. The absolute values of PBIAS are all (1.09%) under the land-use scenario of 2008. The average annual basin
lower than 10. The results indicate a good relationship between ob- surface runoff had a larger increase from 451.65 mm for land use in
served and simulated streamflow values. According to the evaluation 1985 to 505.28 mm for land use in 2008, increased 11.87%. On the
criteria of the SWAT model (Moriasi et al., 2007), the model perfor- contrary, average annual value of ET, percolation and baseflow de-
mance in the XRB could be considered very good. Therefore, SWAT is creased 0.73%, 16.53% and 17.26%, respectively. These results is
capable of simulating hydrological processes in the study area. consistent the results reported by Zhou et al. (2013), and they mainly

Fig. 3. Comparison between observed and simulated monthly streamflow data at Hengtangcun station during calibration (1980–1997) and validation (1998–2015)
periods.

310
Q. Wang et al. Catena 170 (2018) 305–315

Fig. 4. Spatial distribution of changes in each simulated hydrological component for land use maps between 2008 and 1985 in XRB. (a) Water yield, (b) Surface
runoff, (c) ET, (d) Percolation and (e) Baseflow.

Table 7
Pairwise Pearson correlation for the changes of land use variables and hydrological components between land use maps in 1985 and 2008.
Forest-grass land Agriculture land Urban land Water bodies La Water yield Surface runoff ET Percolation Baseflow

Forest-grass land 1.000

Agriculture land −0.579⁎⁎ 1.000
Urban land −0.376⁎⁎ −0.474⁎⁎ 1.000
Water bodies 0.141 −0.436⁎⁎ −0.013 1.000
La −0.807⁎⁎ 0.424⁎⁎ 0.555⁎⁎ −0.587⁎⁎ 1.000
Water yield −0.809⁎⁎ 0.081 0.805⁎⁎ −0.139 0.849⁎⁎ 1.000
Surface runoff −0.854⁎⁎ 0.201⁎ 0.689⁎⁎ −0.109 0.847⁎⁎ 0.967⁎⁎ 1.000
ET 0.337⁎⁎ −0.505⁎⁎ −0.111 0.950⁎⁎ −0.701⁎⁎ −0.307⁎⁎ −0.279⁎⁎ 1.000
Percolation 0.591⁎⁎ 0.210⁎ −0.910⁎⁎ 0.060 −0.674⁎⁎ −0.894⁎⁎ −0.789⁎⁎ 0.193 1.000
Baseflow 0.607⁎⁎ 0.187 −0.899⁎⁎ 0.062 −0.681⁎⁎ −0.897⁎⁎ −0.792⁎⁎ 0.198⁎ 0.999⁎⁎ 1.000

n = 103.
⁎⁎
Are for p < 0.01.
⁎
Are for p < 0.05.

attributed these changes in hydrologic components to the process of UA region followed by the southwestern part of the research region
rapid urbanization. Similarly, in the Upper Du watershed, China, Yan (i.e., the SAMHD, LMUS and HMUHD regions). This pattern is asso-
et al. (2013) found the expansion of urban area was also the main in- ciated with the forest-grass land degradation and reclamation of agri-
fluence factors on the variation of streamflow. Construction land ex- cultural land.
pending with the increase in impervious surface during the course of
urbanization decreased the ET, percolation and baseflow and increased
3.3.2. Correlations between hydrological components and land-use
surface runoff accordingly. What's more, the local changes of each
variables
hydrological component may be different with overall average changes
Pairwise regression correlation analyses were performed using IBM
due to spatial difference of LUCCs.
SPSS Statistics 22 software to assess the relationship between land-use
Therefore, the variation of five simulated hydrological components
variables and hydrological components (Table 7). For the XRB, the
was further calculated according to Eq. (2) (Fig. 4). The figure showed a
change in the area of forest-grass land was the dominant factor that
spatial aggregating feature at the sub-basin scale. Compared with the
affected the variation of hydrological processes and correlated sig-
spatial patterns of five land-use variables (Fig. 2), we found that the
nificantly with all variations in hydrological components. The change in
most significant increases in water yield and surface runoff occurred
forest-grass land had a positive correlation with the changes in water
mainly in the UA region, largely matching the spatial distribution
yield and surface runoff and a negative correlation with the changes in
pattern of urban expansion. Spatial pattern of percolation were similar
ET, percolation and baseflow. Agricultural land was negatively corre-
to the distribution of baseflow, with the most significant decrease in the
lated with ET, which means the decrease in agricultural land increased

311
Q. Wang et al. Catena 170 (2018) 305–315

ET at the basin scale. The change in urban land had significant positive

Moran's I

−0.081
−0.09
correlations with the variation in water yield and surface runoff and

0.225

0.097
0.082

0.238
0.095
0.097
0.002
0.04
negative correlations with the variation in percolation and baseflow,
indicating an association between urban expansion and an increase in
886.214
874.836
869.551
893.859
763.202

933.236
924.061
869.551
846.973
water yield/surface runoff and a decrease in percolation/baseflow. The

764.51
AICc

change in the water bodies was only correlated with the ET component
(Pearson correlation coefficient = 0.95), which indicates that the re-
duction in the water bodies had a positive effect on actual ET.
Adjusted R2

The pairwise Pearson correlation analysis results also showed that

Baseflow

−0.006
0.363
0.497
0.458
0.393
0.807
0.806

0.168
0.458
0.638
some land-use types were co-linear (Table 7); for example, significant
negative correlations were observed between forest-grass land and both
agricultural land and urban land, indicating a conversion from forest-
Moran's I

grass land to agricultural land and urban land. Therefore, we in-

−0.096
−0.105
0.223
0.038
0.241
0.087

0.242
0.089
0.091
0.001

troduced La to analyse the contribution of the combined effects of

LUCCs on hydrological components. Moreover, La was significantly
correlated with all land-use types, indicating that La can be used as an
886.504
874.848
926.161
888.885
749.136
750.412
930.418
920.068
868.353
846.336

index of the comprehensive pattern of land use at the sub-basin scale.

AICc

3.4. Spatial hydrological responses to the individual and combined effects of

Adjusted R2
Percolation

LUCCs
−0.006
0.343
0.483
0.034
0.406
0.827
0.826

0.188
0.449
0.63

Although the SWAT model results showed the changes in the hy-
drological components at the basin and sub-basin scales under the
different land-use scenarios, the quantitative hydrological impacts of
Moran's I

the individual and combined effects of LUCCs need to be further de-

0.059

0.082

0.049
0.035
0.049
0.029
0.01

0.24
0.14

0.06

termined. Therefore, GWR models were applied in this study to explore

the spatial relationships between the changes in hydrological compo-
581.232
580.241
563.335
550.889
592.421
583.345
354.539
352.604
523.941
523.722

nents and land-use variables, which were constructed by using ArcGIS

AICc

10.3.

3.4.1. Performance of GWR and OLS models

Adjusted R2

The performances of the GWR and OLS models are presented in

Table 8. Most of the adjusted R2 values for the GWR models are higher
0.105
0.123
0.248
0.376
0.002
0.216
0.901
0.907
0.487
0.493
ET

than those for the OLS models, and most of the AICc values for the GWR
models are lower than those for the OLS models and differ by more than
Moran's I

three, indicating that the GWR models had a better prediction accuracy
−0.018

−0.083

−0.036
−0.051
0.125

0.201
0.037
0.019

0.166

than the OLS models. We used the global Moran's I value to evaluate
0.11

whether the pattern of spatial data was clustered, dispersed, or random.

The majority of the Moran's I values did not exhibit a statistically sig-
769.233
751.588
899.833
868.497
837.785
819.552
902.821
897.403
773.676
781.312

nificant clustering of high and/or low residuals (p > 0.05) for the
Comparison of Adjusted R2, AICc and residuals' Moran's I from GWR and OLS models.

AICc

GWR models, which suggests a spatially random pattern of the re-

gression residuals. However, the Moran's I values of both the OLS and
Surface runoff

Adjusted R2

GWR models for the water bodies were statistically significant for all
hydrological components except ET, which indicates that the change in
0.727
0.784
0.031
0.368
0.469
0.602
0.002
0.072
0.715
0.711

water bodies was the main factors effecting the process of ET. Com-
parison with the OLS model, the results of GWR model could better
reflect the actual spatial characteristics of hydrological responses.
Moran's I

−0.001

−0.017
−0.091

−0.023
−0.039
0.217

0.246
0.065

0.208
0.104

3.4.2. The spatial hydrological responses to individual LUCCs

The local coefficient distributions of the GWR models, which had
289.535

398.297
362.148
291.267

396.949
389.662
267.573
261.001

highly significant correlations and shown in bold numbers in Table 7,

274.99

279.32

n = 103, underlined numbers are for p > 0.05.

AICc

are shown in Fig. 5 and Fig. 6. In contrast to the OLS models, re-
lationships in the GWR models had complex local characteristics at
spatial scales, which indicated that the responses of each hydrological
Adjusted R2
Water yield

component to land-use variables were spatially different.

−0.003

The local coefficient distributions of the changes in forest-grass and

0.651
0.733

0.376
0.645
0.717

0.125
0.718
0.743
0.01

urban land for water yield in the GWR models are shown in Fig. 5a and
Fig. 5b, respectively. Forest-grass land had negative regression coeffi-
GWR

GWR

GWR

GWR

GWR

cients (−0.087–−0.002), which indicated that a decrease in forest-

OLS

OLS

OLS

OLS

OLS

grass could result in a water yield increase of 0.2%–8.7%, and the in-
fluence was higher in the downstream regions of the research area,
Forest-grass land

Agriculture land

namely, UA, SADX and LMUD (Fig. 5a). The increase in urban land
Water bodies
Urban land

affected the water yield more strongly in the middle region of the XRB
(i.e., SAMX) with local regression coefficient of 0.11–0.139, which
Table 8

means the change of urban land could increase the water yield up to a
La

maximum of 11%–13.9% in local areas. Urban land change affected the

312
Q. Wang et al.
Fig. 5. Local coefficients of individual land use types for each hydrological component in the GWR model.
process of water yield more than the change in forest-grass land. Si-
milarly, the dominant factors that influenced surface runoff were the
changes in forest-grass and urban land (Fig. 5c and Fig. 5d). Forest-
grass land was negatively correlated with surface runoff in all sub-ba-
sins, and the biggest absolute value of coefficient was observed in the
northern area of the XRB. By contrast, urban land exhibited positive
coefficients across the entire watershed, and most affected areas were
regions of SAMX. The value of the coefficient suggested that the de-
crease in forest-grass land could increase the surface runoff by up to
106.2%, and the expansion of urban land could locally increase surface
runoff up to a maximum of 173.1%. These results showed the re-
lationships in GWR models were divergent from OLS results and other
studies using traditional global statistical models (e.g., Nie et al., 2011;
Woldesenbet et al., 2016). In these studies, the relationships between
water yield and surface runoff and the changes in forest-grass land and
urban land were constant across the entire basin. The GWR model re-
sults showed the relationships varied spatially. Moreover, the coeffi-
cients reflect the regional responses of each hydrological component to
changes in individual land-use types.
The GWR ET model exhibited negative coefficients for agricultural
land throughout the entire study area, whereas positive coefficients
(0.001–0.678) were observed for the water bodies in the majority of the
regions (Fig. 5e and Fig. 5f). Negative coefficients (−0.171–−0.013)
for agricultural land in the GWR ET model were higher in the down-
stream region of the XRB. Urban land and forest-grass land were the
main factors that influenced percolation and baseflow (Fig. 5g–j). The
313
Catena 170 (2018) 305–315
spatial patterns of the coefficients between urban land and percolation
and baseflow in the GWR models were almost the same, with increasing
impacts from south to north, which indicates the effects of urban ex-
pansion were greater in the northern regions of the XRB. In contrast to
the positive correlation in the OLS models, percolation and baseflow
had both positive and negative relationships with forest-grass land in
the GWR models (Fig. 5g and Fig. 5i). The coefficients switched from
positive to negative in the southeastern part of the SAMHD region. It is
likely that this is mainly due to the implementation of the Conversion of
Cropland into Forest Project and the expansion of urban land, which
caused the increase in forest-grass land because of the transformation of
agricultural land. However, the expansion of urban land, with an in-
crease in the impervious surface, was also one of the primary factors
involved in the local hydrological process, which reduced the percola-
tion and baseflow. Thus, the relationships between forest-grass land and
percolation and baseflow were negative. Many previous studies also
showed that the relationships between forest-grass land and percolation
and baseflow were inconsistent at the basin scale. For instance, Nie
et al. (2011) reported positive relationships between forest-grass land
and percolation and baseflow in the upper San Pedro watershed, while
Woldesenbet et al. (2016) found negative relationships in the source
region of the Upper Blue Nile Basin, Ethiopia. Hence, an analysis at the
basin scale may ignore the details of the local influence of LUCCs, and
their relationships were the average results for the entire area.
Q. Wang et al.
Fig. 6. Local coefficient values of La for each hydrological component in GWR models. (a) Water yield, (b) Surface runoff, (c) ET, (d) Percolation and (e) Baseflow.
3.4.3. The spatial hydrological responses to the combined effects of LUCC
To spatially analyse the combined effects of LUCCs on each hydro-
logical process, GWR models were constructed between each hydro-
logical component and La, and the local coefficients are shown in Fig. 6.
The results show that the land-use intensification degree was positively
correlated with water yield and surface runoff but negatively correlated
with ET, percolation and baseflow. In other words, urbanization caused
an increase in water yield and surface runoff but a decrease in ET,
percolation and baseflow in the XRB.
In addition, the responses of each hydrological component to La also
varied spatially. For instance, the coefficients of La for the water yield
and surface runoff increased gradually from the centre, where the only
county seat (i.e., the UA region) is located, to the upper sub-basins,
indicating that water yield and surface runoff in rural areas were more
sensitive to LUCCs than in the urbanized region. The value of the
coefficient also indicated greater combined effects of LUCCs on surface
runoff than on water yield. Although LUCCs caused the average annual
water yield and surface runoff to increase by 1.09% and 11.87%
(Table 6), respectively, their combined effects could result in corre-
sponding increases of 13.7% and 114.2% at a local scale. The responses
of ET to La decreased gradually from downstream to upstream. The
spatial response patterns of percolation and baseflow to La were almost
the same, which show that the intensification of land-use had a lower
impact in the sub-basins near the central county.
Fig. 6 shows that the absolute values of local coefficient of La for
water yield, surface runoff, percolation and baseflow were greater in
rural areas than in the urbanized region. The results indicate that the
same degree of urbanization in rural areas had a greater influence on
these hydrological processes. Thus, rural areas should receive more
attention under the process of urbanization, which has practical sig-
nificance to establish measures for flood control and water resource
protection.
314
Catena 170 (2018) 305–315
3.5. Prospects
Although many previous studies researched the impacts of LUCCs
on each hydrological component in SWAT combined with traditional
statistical models at the basin scale, we adopted GWR models to con-
sider the effects of the spatial differences in LUCCs on hydrological
components using a spatial regression statistical model. In addition, we
evaluated the combined effects of LUCCs using a comprehensive index,
i.e., the land-use intensification degree. The various regional physical
geographic (such as land use, topography, soil and other factors) and
climate conditions lead to spatial differences in hydrological processes,
and the degree of hydrological alteration is controlled by climate
change and anthropogenic interventions such as LUCCs and the con-
struction of water projects. However, this study considered only the
influence of LUCCs on hydrological components in the typical region of
the Yangtze River Delta by controlling variables in the SWAT model.
Thus, further research is necessary to determine the effect of other
factors, such as topography, soil and climate conditions, on the tem-
poral and spatial changes in hydrological processes. Moreover, com-
parative experimental observations on hydrological responses might
help to obtain more details of spatial effects of climate and underlying
surface conditions.
4. Conclusion
This study investigated the spatial response patterns of hydrological
components to the individual and combined effects of LUCCs in a ty-
pical region of the Yangtze River Delta, i.e., the Xitiaoxi River Basin,
using a combination of SWAT and GWR models. The good performance
of the SWAT model in the calibration and validation periods indicated
that SWAT is capable of simulating hydrological processes in the study
basin. Subsequently, the hydrological consequences simulated by the
SWAT model under two land-use scenarios that represent rapid urba-
nization were used to evaluate the spatial influence of LUCCs by
Q. Wang et al.
coupling the GWR models. Compared with the OLS models, the GWR
models had a better prediction accuracy and was more suitable to
analyse the spatial relationship between hydrological components and
land-use variables.
A major finding of this study is that the hydrological impacts of
LUCCs exhibited local spatial differences in the XRB. The predominant
trend of land-use conversion was between forest-grass land and agri-
cultural land, and the diminishing portion of forest-grass land
(25.93 km2) and agricultural land (46.77 km2) contributed to the ex-
pansion of urban land during the period 1985–2008. Moreover, the
urban area increased from 5.6% to 17.05%, and the change ratio pro-
gressed towards the centre of the XRB. At the basin scale, LUCCs in-
creased the average annual water yield and surface runoff by 1.09%
and 11.87%, respectively, and decreased the average annual ET, per-
colation and baseflow by 0.73%, 16.53% and 17.26%, respectively,
under the 1985 and 2008 land-use scenarios during the 1980–2015
period. However, the response relationships between LUCCs and hy-
drological components were not constant at the sub-basin scale. With
respect to water yield and surface runoff, the greatest impact of forest-
grass land changes occurred downstream of the research region, and the
greatest impact of urban land occurred in the middle region of the
basin. Urban land and forest-grass land were also the main factors that
affected percolation and baseflow, and the local regression coefficients
of urban land increased gradually from south to north. In contrast to the
positive correlation in the OLS models, relationships between forest-
grass land and percolation and baseflow in the GWR models had both
positive and negative effects depending on the spatial scale.
Furthermore, La, which reflects the combined effects of LUCCs, was
positively correlated with water yield and surface runoff but negatively
correlated with ET, percolation and baseflow. Local regression coeffi-
cients suggested that La could increase water yield and surface runoff
by up to 13.7% and 114.2%, respectively, in some local areas. The
spatial patterns of the local regression coefficients indicated that rural
regions might experience greater effects of LUCCs in the XRB. Thus,
rural areas should receive more attention when establishing measures
for flood control, disaster reduction and water resource protection.
The results increase the understanding of the impact of the in-
dividual and combined effects of LUCCs on hydrological processes on a
spatial basis, which can assist in establishing flood control and water
resource protection measures in detail. Further, due to the deficiency
and uncertainty of the data, it is necessary to detect the effects of other
factors, such as topography, soil and climate conditions, on the tem-
poral and spatial changes of hydrological processes.
Acknowledgements
Financial support for this study was provided by the National Key
Research and Development Program of China (No. 2016YFC0401502),
the National Natural Science Foundation of China (No. 41371046 and
41771032), and the Water Conservancy Science and Technology
Foundation of Jiangsu Province (No. 2015003).
References
Abbaspour, K.C., 2014. SWAT-CUP 2012: SWAT Calibration and Uncertainty Programs - A User
Manual.
Abbaspour, K.C., et al., 2007. Modelling hydrology and water quality in the pre-alpine/alpine
Thur watershed using SWAT. J. Hydrol. 333 (2), 413–430.
Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologic modeling
and assessment part I: model development. J. Am. Water Resour. Assoc. 34 (1), 73–89.
Arnold, J.G., Kiniry, J.R., Srinivasan, R., Williams, J.R., Neitsch, S.L., 2012. Soil and Water
Assessment Tool Theoretical Documentation, Version 2012.
Beighley, R.E., Melack, J.M., Dunne, T., 2003. Impacts of California's climatic regimes and
coastal land use change on streamflow characteristics. J. Am. Water Resour. Assoc. 39 (6),
1419–1433.
Brun, S.E., Band, L.E., 2000. Simulating runoff behavior in an urbanizing watershed. Comput.
Environ. Urban. Syst. 24 (1), 5–22.
Chang, H., Johnson, G., Hinkley, T., Jung, I.W., 2014. Spatial analysis of annual runoff ratios
315
Catena 170 (2018) 305–315
and their variability across the contiguous U.S. J. Hydrol. 511 (7), 387–402.
Chen, C.C., Zhang, Y.Q., Xiang, Y., et al., 2014. Study on runoff responses to land use change in
Ganjiang Basin [J]. J. Nat. Resour. 29 (10), 1758–1769.
Chen, Q., et al., 2016. Impacts of land use and population density on seasonal surface water
quality using a modified geographically weighted regression. Sci. Total Environ. 572, 450.
David, B., Cavallo, G.J., Nilson, M.L., 2005. Base flow trends in urbanizing watersheds of the
Delaware River Basin. J. Am. Water Resour. Assoc. 41 (6), 1377–1391.
Fossey, M., Rousseau, A.N., Savary, S., 2016. Assessment of the impact of spatio-temporal at-
tributes of wetlands on stream flows using a hydrological modelling framework: a theo-
retical case study of a watershed under temperate climatic conditions [J]. Hydrol. Process.
30 (11), 1768–1781.
Fotheringham, A.S., Charlton, M., Brunsdon, C., 1996. The geography of parameter space: an
investigation of spatial non-stationarity. Int. J. Geogr. Inf. Sci. 10 (5), 605–627.
Gao, P., Niu, X., Wang, B., Zheng, Y., 2015. Land use changes and its driving forces in hilly
ecological restoration area based on gis and rs of northern China. Sci. Rep. 5, 11038.
Ghaffari, G., Keesstra, S., Ghodousi, J., Ahmadi, H., 2010. SWAT-simulated hydrological impact
of land-use change in the Zanjanrood basin, Northwest Iran. Hydrol. Process. 24 (7),
892–903.
Hu, W., Zhou, W., He, H., 2015. The effect of land-use intensity on surface temperature in the
Dongting Lake Area, China. Adv. Meteorol. 2015, 1–11.
Jennings, D.B., Jarnagin, S.T., 2002. Changes in anthropogenic impervious surfaces, pre-
cipitation and daily streamflow discharge: a historical perspective in a mid-Atlantic sub-
watershed. Landsc. Ecol. 17 (5), 471–489.
Kim, Y., Engel, B.A., Lim, K.J., Larson, V., Duncan, B., 2002. Runoff impacts of land-use change
in Indian River Lagoon Watershed. J. Hydrol. Eng. 7 (3), 245–251.
Lv, A., Zhou, L., 2016. A Rainfall model based on a geographically weighted regression algo-
rithm for rainfall estimations over the Arid Qaidam Basin in China. Remote Sens. 8 (4),
311.
Mittal, N., Bhave, A.G., Mishra, A., Singh, R., 2016. Impact of human intervention and climate
change on natural flow regime. Water Resour. Manag. 30 (2), 685–699.
Moriasi, D.N., et al., 2007. Model evaluation guidelines for systematic quantification of accu-
racy in watershed simulations. Trans. ASABE 50 (3), 885–900.
Nadal-Romero, E., Cammeraat, E., Serrano-Muela, M.P., Lana-Renault, N., Regüés, D., 2016.
Hydrological response of an afforested catchment in a Mediterranean humid mountain
area: a comparative study with a natural forest. Hydrol. Process. 30 (15), 2717–2733.
Neitsch, S., Arnold, J., Kiniry, J., Williams, J., 2011. Soil and Water Assessment Tool
Theoretical Documentation, Version 2009, Texas Water Resources Institute Technical
Report No.406.
Nie, W., et al., 2011. Assessing impacts of land use and land cover changes on hydrology for the
upper San Pedro watershed. J. Hydrol. 407 (1–4), 105–114.
Rennermalm, A.K., Bring, A., Mote, T.L., 2012. Spatial and scale-dependent controls on North
American Pan-Arctic minimum river discharge. Geogr. Anal. 44 (3), 202–218.
Rose, S., Peters, N.E., 2001. Effects of urbanization on streamflow in the Atlanta area (Georgia,
USA): a comparative hydrological approach. Hydrol. Process. 15 (8), 1441–1457.
Sajikumar, N., Remya, R.S., 2015. Impact of land cover and land use change on runoff char-
acteristics. J. Environ. Manag. 161, 460–468.
Shi, X., Yang, Z., Yan, D., Ying, L.I., Zhe, Y., 2014. On hydrological response to land-use/cover
change in Luanhe River basin. Adv. Water Sci. 25 (1), 21–27.
Tekleab, S., Mohamed, Y., Uhlenbrook, S., Wenninger, J., 2014. Hydrologic responses to land
cover change: the case of Jedeb mesoscale catchment, Abay/Upper Blue Nile basin,
Ethiopia. Hydrol. Process. 28 (20), 5149–5161.
Tu, J., Xia, Z.G., 2008. Examining spatially varying relationships between land use and water
quality using geographically weighted regression I: model design and evaluation. Sci. Total
Environ. 407 (1), 358–378.
Wagner, P.D., Waske, B., 2016. Importance of spatially distributed hydrologic variables for land
use change modeling [J]. Environ. Model. Softw. 83, 245–254.
Wang, R., Kalin, L., Kuang, W., Tian, H., 2014. Individual and combined effects of land use/
cover and climate change on Wolf Bay watershed streamflow in southern Alabama. Hydrol.
Process. 28 (22), 5530–5546.
Wang, Q., Xu, Y.P., Gao, B., et al., 2017. Runoff spatial responses to land use change in Xitiaoxi
River Basin[J]. J. Nat. Resour. 32 (4), 632–641.
Woldesenbet, T.A., Elagib, N.A., Ribbe, L., Heinrich, J., 2016. Hydrological responses to land
use/cover changes in the source region of the Upper Blue Nile Basin, Ethiopia. Sci. Total
Environ. 575 (C), 724.
Yan, B., Fang, N.F., Zhang, P.C., Shi, Z.H., 2013. Impacts of land use change on watershed
streamflow and sediment yield: an assessment using hydrologic modelling and partial least
squares regression. J. Hydrol. 484 (6), 26–37.
Yan, R., Zhang, X., Yan, S., et al., 2018. Spatial patterns of hydrological responses to land use/
cover change in a catchment on the Loess Plateau, China. Ecol. Indic. 92, 151–160.
Ye, X.C., Zhang, Q., Li, J., et al., 2009. Impacts of climate change and human activities on runoff
of Poyang Lake Catchment [J]. J. Glaciol. Geocryol. 31 (5), 835–842.
Yue, T.X., et al., 2013. Climate change trend in China, with improved accuracy. Clim. Chang.
120 (1), 137–151.
Zhang, Y.K., Schilling, K.E., 2006. Increasing streamflow and baseflow in Mississippi River since
the 1940s: effect of land use change. J. Hydrol. 324 (1–4), 412–422.
Zhang, C., Zhang, B., Li, W., et al., 2012. Response of streamflow to climate change and human
activity in Xitiaoxi river basin in China [J]. Hydrol. Process. 28 (1), 43–50.
Zhang, Y., You, Q., Chen, C., Ge, J., 2016. Impacts of climate change on streamflows under RCP
scenarios: a case study in Xin River Basin, China. Atmos. Res. 178–179, 521–534.
Zhou, F., et al., 2013. Hydrological response to urbanization at different spatio-temporal scales
simulated by coupling of CLUE-S and the SWAT model in the Yangtze River Delta region. J.
Hydrol. 485 (485), 113–125.
Zhuang, D., Liu, J.Y., 1997. Study on the model of rehional differentiation of land use degree in
China. J. Nat. Resour. 2, 105–111.
Zuo, D., Xu, Z., Yao, W., et al., 2016. Assessing the effects of changes in land use and climate on
runoff and sediment yields from a watershed in the Loess Plateau of China [J]. Sci. Total
Environ. 544, 238–250.