1. Introduction
Land use and land cover (LULC) changes are major factors of global environmental change, with potentially severe impacts on human livelihoods [
1]. LULC changes are usually caused by the expansion of agriculture, urbanization, deforestation, and the day-to-day activities of mankind. At different spatial and temporal scales, LULC affects hydrology [
2,
3,
4]. Assessing the impacts of LULC changes on hydrological characteristics is vital for both understanding the effects of LULC changes on hydrological processes over the earth surface [
5], and managing and developing watersheds [
6]. In addition, it is important for flood potential prediction and mitigation of hazard and sustainable developments, to link the effects of land use change on the hydrological cycle [
7,
8].
LULC changes are one of the main factors influencing regional hydrological characteristics, which lead to changes in water distribution in hydrological processes such as evaporation, runoff, lateral flow, infiltration, and groundwater. Because of the transpiration of vegetation, the conversion of woodland to other types of land is often accompanied by a decrease in evapotranspiration (ET) [
9,
10]. Some studies suggest that grassland and lateral flow are positively correlated [
11]. Urbanization and agricultural land expansion will lead to an increase in impervious surface area, which may lead to an increase in surface runoff, and a decrease in infiltration and groundwater recharge [
12,
13]. Differences in watershed development and landscape patterns make hydrological processes respond differently to LULC change. For example, detection thresholds for urbanized impervious surfaces that begin to have a significant influence on hydrological characteristics are widely distributed between 5% and 20% [
14,
15,
16]. At the same time, studies in urbanization areas show that the increase of streamflow is closely related to urbanization [
10,
17,
18], and regions with high urbanization have a greater impact on streamflow and peak values [
19].
At present, the methods for studying the hydrological response of LULC changes include statistical analysis and model simulation. The statistical analysis methods include trend analysis of long time series [
20], and the paired catchment approach [
21]. The trend analysis method is usually performed by comparing the catchment in which LULC changes in the region, with a catchment without significant changes, and is generally applicable to areas with long-term monitoring data. The advantage of the paired catchments approach is that it can remove the influence of meteorological factors, but is generally only applicable to small watersheds, and it is difficult to find control watersheds in urbanized areas. Model simulation methods can be divided into conceptual model simulation [
22] and distributed (semi-distributed) physical model simulation [
23,
24,
25] from the perspective of model selection. Conceptual models are generally used in large watersheds where it is difficult to find control catchments, but often there is no clear relationship between their parameters and land use characteristics. Therefore, the variation of the model parameters caused by LULC change have great uncertainty [
26]. The distributed (semi-distributed) physical model is capable of describing the various components of the hydrological process, and it is considered to be the most rational way to investigate the hydrological response to LULC change [
27]. The distributed (semi-distributed) physical model is generally simulated by a grid unit [
19] or the watershed division method [
11]. Compared with the sub-watershed division method, grid unit simulation can output more detailed spatial information.
Scenario simulation is a common method of using a model to study the hydrological responses of LULC changes. According to the difference of hypothesis, it can be divided into the artificial hypothesis scenario [
28], the historical scenario [
12,
24], and the future scenario [
19,
29]. Although the artificial hypothesis can reflect the influence of LULC changes on hydrological characteristics, its assumptions are often inconsistent with the actual development of the basin, and the future scenarios are more biased towards prediction. The historical scenarios can better reflect the real impacts of the basin, but they need a large amount of historical data support. Hydrologic responses to LULC changes were studied at the various spatial scales from plot [
30], small watershed [
14,
24], medium watershed [
19], and large watershed [
12]. As the scale of the study area increases, the landscape distribution pattern of the basin and the meteorological factors will cause more spatial differences. Therefore, for large basins, the spatial analysis results will be more conducive to water resource management and decision-making process [
10,
18].
The Taihu Lake Basin (TLB) is one of the most well-developed regions in China. The population of the basin accounts for about 3% of the country’s total, and the gross domestic product (GDP) accounts for about 12%of the total GDP. With the development of the economy, the natural ecological environment of the TLB has been greatly damaged, resulting in frequent flood events in the basin. In the past 30 years, large-scale flood events (in 1991, 1999 and 2016) have occurred in the TLB. Local flood events are more frequent, and some studies have indicated that the frequency and risk of future floods in the TLB will increase further [
31,
32]. Therefore, clarifying the impact of LULC change on hydrological processes in the TLB will help with water resource management in the basin, thereby reducing the risk of flooding. Existing research in the TLB shows that the conversion of other types of land to construction land will cause the largest change in the runoff coefficient, and afforestation will help to reduce flood disasters [
28]. Runoff and peak flow change are most significant in sub-basin areas near the urban center [
7,
33]. The urbanization process has led to an overall increase in the water level of the river network in the basin, and the increase rate is consistent with the urbanization rate, and the water level rise is more obvious in areas with high degrees of urbanization [
34]. Since about 80% of the TLB is plain area with dense river networks, and it is subject to complex artificial regulation and control, it is difficult to validate the streamflow in the whole basin. At present, there are a few studies focusing on the hydrological responses to LULC changes in the entire TLB. In existing research works, some of them calibrate and validate the model parameters through the monitoring results of flood years [
32,
35], while others directly apply the calibrated and validated parameters of the Xitiaoxi small watershed to other parts of the TLB [
31]. Therefore, in order to improve the reliability of the model as much as possible, we combine the above two methods when calibrating and validating the model parameters. On the other hand, in the study of the hydrological response of LULC change in the TLB, only a few studies have analyzed the spatial differences of runoff or water yield [
35,
36], and the scale is limited to large-scale hydrological zoning. It is also not directly related to administrative zoning, which limits further construction of the link between LULC change, hydrological response, and regional development (e.g., GDP, population, urban area, etc.).
Our research objective was to assess the change of water yield and its spatial distribution in rapidly urbanized areas at different scales, analyze the internal causes of frequent local floods and urban waterlogging in the TLB, and to establish a link between land use change, hydrological response, and regional development. In order to achieve this goal, we used the Spatial Tools for River basin Environmental Analysis and Management (STREAM) model (version 2.0, Institute for Environmental Studies Vrije Universiteit De Boelelaan, Amsterdam, the Netherlands) to simulate the water yield under different land use conditions in the TLB in 1985, 1995, 2000, 2005, and 2010. In order to eliminate the influence of meteorological factors as far as possible, the average water yields of different weather scenarios in all land use conditions were calculated to evaluate the LULC change effect on water yield. The average water yields were counted at the hydrological sub-region scale and the township scale in ArcGIS (version 10.1, Esri China Information Technology Co. Ltd., Beijing, China) and used to study the relationship between water yield distribution and regional development. Finally, we explored the relationship between urban development and urban waterlogging through the spatial distribution difference of water production, and analyzed the causes of frequent local floods in the TLB.
2. Materials and Methods
2.1. The Study Area
Taihu Lake Basin lies in the southern part of the Yangtze Delta, which is located at 30°28′–32°15′ N, 119°11′–121°53′ E. It faces Yangtze River on the north, reaches Hangzhou Bay on the south, neighbors the East China Sea on the east, and is bounded on the west by the Tianmu Mountain Range and the Maoshan Mountain Range. The basin (
Figure 1) covers 36 895 km
2, accounting for 0.4% of China’s territory. Seven large and medium cities, including Shanghai, Hangzhou, Suzhou, Wuxi, Changzhou, Jiaxing, and Huzhou, and 31 counties, are distributed in the basin, with a population of 36 million, accounting for 3% of China’s entire population. It is one of the most developed areas in China, and its GDP occupies about 12% of the total GDP of China.
The study area is composed of the mountain region in the west, and plain land in the central and east. The elevations in the basin range from 2 m to 1587 m above sea level. The mountain land and plain land is about 25% and 75%, respectively, of the total basin area. Taihu Lake is located in the middle of the basin, with a water surface area of 2238 km2, and an average depth of 2 m. The plains surrounding Taihu Lake, which extend to the Yangtze River, as well as to the sea, are covered by a dense network of natural and man-made waterways, with a total length of more than 12,000 km. The inflow to the Taihu Lake is mainly from the Tiaoxi river system and the Nanxi river system. The Grand Canal also has a small volume of water interflow with the Taihu Lake. A total of 70% of the lake flow discharge is through the Taipu River to the Huangpu River, and then into the Yangtze River, and 20% is through the Wangyu Rivers to the Yangtze River. The remaining 10% of lake flow discharge is to the south into Hangzhou Bay.
The basin belongs to the subtropical monsoon region, with warm humid summers and cold dry winters. The rainy season is from April to September. The annual average rainfall is about 1010–1400 mm, and high rainfall occurs in the southwest of the basin. The dominant soil types are clay loam (51.00%) and loam (27.27%). The clay loam soil is distributed in the plain land with paddy fields and loam soil in the higher terrains with forest land. The basin has undergone rapid urbanization and extensive cultivation, due to its long history of development. Urban areas cover about 24.24% (in 2010). Cultivated lands occupy about 47.90% of the total basin area. Forest land and grassland are mainly distributed in river valleys and high terrain areas.
2.2. The STREAM Model
STREAM is a raster-based hydrological rainfall–runoff model. Originally, the water balance model was developed for the Rhine Basin, and is called RHINEFLOW. This model runs in MS-DOS mode, and uses a GIS-based programming language called PC-Raster [
37]. The STREAM has been developed for a Windows environment, and has been extended to a river basin management instrument. The idea of developing the STREAM was to simulate the hydrology of large river basins in a simplified way with minimum data requirements, which at the same time is sufficient to gain insight into the major processes of the hydrological cycle. The model requires monthly precipitation, monthly temperature, digital elevation model (DEM, 90 m, Lakes Basin Data Integration and Simulation Center, Institute of Geography and Lakes, Chinese Academy of Sciences, Nanjing, China), soil (water holding capacity), and land cover as inputs (
Figure 2). The water flow direction is determined by the DEM. The model enables the analysis of the impacts of climate and land use changes on the hydrological characteristics of a river basin, and many studies have used the STREAM model to simulate hydrological components successfully [
38,
39,
40,
41]. STREAM enables rapid assessment of water balance, and can be used to estimate the long-term hydrological impacts of land use change, climate change, river basin management, population pressure, economic development, and so on. Another advantage of this model is that the Blaise script language used is very clear, and therefore is easy to be changed by the user to meet special objectives.
The hydrological approach that is used in STREAM is based on the water balance. The potential and actual evapotranspiration is calculated using the Thornthwaite & Mather equation [
37], which uses temperature and precipitation as the major input parameters. The land use map is used as a base map to prepare the crop factor, which plays an important role in determining the actual evapotranspiration. When precipitation is higher than the potential evapotranspiration, the soil will remain full of water, and actual evapotranspiration will be equal to potential evapotranspiration. However, in conditions where the precipitation drops to below potential evapotranspiration, the soil begins to dry out, and actual evapotranspiration is now less than the potential. Actual evapotranspiration is calculated using the moisture-related method with antecedent soil water content and water holding capacity. The water that enters the grid cell can be stored in deep groundwater (deep storage) or shallow storage. The output of water occurs through flow discharge and evapotranspiration (from shallow storage). The model runs on a monthly basis, generating direct runoff, delayed runoff, groundwater storage (shallow and deep), snow water equivalents, and snow melt. STREAM describes the water balance at location (x, y) in month (t) by:
where R = runoff (mm/month); P = precipitation (mm/month); S = water volume stored in the soil, snow and groundwater (mm/month); AE = water loss due to actual evapotranspiration (mm/month); dS = change in water volume stored (mm/month); SS = water stored in the soil and as shallow groundwater (mm/month); GWS = the water stored in aquifers and as deeper groundwater (mm/month); SNS = the amount of water stored in the snow cover.
2.3. Data Collection and Preparation
The spatial input data necessary for running STREAM includes a land use map, a soil type map, monthly precipitation maps, monthly temperature maps, and DEM. As STREAM is a cell-based model, all of the input data should be raster maps, in the same format, and having the same geo-reference and spatial resolution. In this study, the Albers Conical Equal Area Projection was applied, and spatial resolution was established at 250 m. A DEM with a 25 × 25m grid size for the basin was acquired from the National Geomatics Center of China. A DEM for the model was then constructed by resampling to a lower resolution map at a pixel resolution of 250 × 250 m. The flow direction map and slope map were calculated by IDRISI software (version 16.1, Clark University, Wooster, MA, USA), which is an integrated GIS and image processing software solution developed by Clark Labs, providing over 250 modules for the analysis and display of digital spatial information. Information on the soil types was obtained from a digital 1:500,000 soil map of the local Bureaus of Agriculture, and converted to 12 USDA (United States Department of Agriculture) soil texture classes based on textural properties. According to the STREAM model manual, water holding capacity can be determined by soil textural properties. The land use information at a 1:10,000 scale was obtained from the Data Center for Resources & Environmental Sciences, Chinese Academy of Sciences (RESDC, CAS) derived from Landsat TM/ETM
+ remote sensing image (30m, National Aeronautics and Space Administration,
https://www.nasa.gov/) in 1985, 1995, 2000, 2005, and 2010. The land cover was classified as four classes, including farmland, forest, build-up, and surface water. The crop factor could be determined by land use type according to the STREAM model manual.
Monthly mean temperature and precipitation data of 13 meteorological stations from 1979 to 2011, which were distributed in and around the total basin, were acquired in this study. The station data had to be interpolated into grids with a 250 × 250 m grid size by the inverse distance-weighted interpolation (IDW) method, and 120 precipitation and temperature raster maps from 2002 to 2011 were acquired. Due to the flat terrains and complex river network, most of the river system is not independent, and it exchanges water with Taihu Lake and other river systems in the TLB by man-made water courses. Most of the gauge stations are affected by the return flow of Taihu Lake and hydrologic engineering facilities. In this study, the Xitiaoxi Watershed, with its independent natural river system, which lies in the southwest of Taihu Lake Basin and is the largest watershed in the up-river area of Taihu Lake, was selected for model calibration and verification. The monthly flow discharge data of the Fanjiachun gauge station, located downstream of the Xitiaoxi river system, were obtained from Huzhou Bureaus of Water Resource.
The STREAM model was calibrated and validated on the basis of observation data over the stages 1980–1999 and 2002–2011 in Xitiaoxi watershed (
Figure 3). The calibrated parameters included CROPFcal, HEATcal, WHOLDNcal, Ccal and TOGWcal (
Table 1). The initial parameters were adjusted automatically to reflect more realistic values using the monthly input data. The flow discharge dataset was split into two time slices in each phase. The first phase was divided into two periods: 1980–1988 and 1989–1999, and the second phase was divided into two periods, 2002–2006 and 2007–2010. The first period was used to calibrate the model, and the second period was used to validate the model. These parameters were then used to simulate the flow discharge of the entire TLB in 1991 (land use 1995) and 1999 (land use 2000), and the parameters of Ccal and TOGWcal were further adjusted by comparing the flow discharge of the TLB with the measured value in the flood periods of 1991 and 1999.
The model performance was evaluated using two objective functions, namely: the coefficient of determination (R
2) and the percent bias (PBIAS). The two parameters were calculated as follows:
where
is the observed variable,
is the mean of the observed variable,
is the simulated variable,
is the mean of the simulated variable, n is the number of observations.
2.4. Moran’s I
Moran’s I is widely used in the study of spatial distribution patterns, where the global Moran’s I is used to evaluate the spatial correlation and the degree of difference, which is generally between −1 and 1. The Moran’s I is greater than zero, indicating that variables are clustered in space; on the contrary, the Moran’s I is smaller than zero, indicating that variables are discrete in space; the Moran’s I equals zero, indicating that variables are randomly distributed in space. The formula is as follows:
In the formula, sat and are the values of the statistic and the mean of the statistic, respectively, Wij is the spatial weight of factor i and factor j, and n are the sum of factors.
The local indicators of spatial association (LISA) was used to make up for the local instability of the global spatial autocorrelation, and to analyze the contribution of regional units to the global space self-correlation. The formula is as follows:
In the formula, sat and are the values of the statistic and the mean of n locations, respectively, and Wij is the weighted value of the weight matrix at position (i, j).
In our study, based on the calculated results of the average water yields under the 2002–2011 meteorological scenarios, we calculated the Moran’s I (MI) and LISA of the growth rate of water yields in towns by using the GeoDa software (
http://www.csiss.org/); empirical Bayesian adjustment was used to consider the variance instability of the ratio. The weighting factors were calculated by the Euclidean distances of towns.
5. Conclusions
Urbanization was the main driving force in LUCC change in the TLB. From 1985 to 2010, the area of construction land increased by 150.41%, and the area of construction land expansion mainly occurred from farmland around the city. The changes in LUCC patterns significantly affected the hydrological characteristics of the basin. Compared with 1985, the water yield of the basin increased by about 12.85% until 2010; the increase of water yield in the basin was not uniform. In the Pudong sub-region, the Puxi sub-region, the Yangchengdianmao sub-region, and the Wuchengxiyu sub-region, where urban development was more rapid, the water yield increased by a larger proportion. The local water yield increase caused by urbanization was even more significant. In Shanghai, Suzhou, Wuxi, and Changzhou, cities which were relatively developed in the basin, the water yield increase rates caused by construction land expansion exceeded 40%, far exceeding the average level of the basin.
During the study period, the growth rate of water yields in towns showed a spatial clustering feature. MI increased from 0.22 to 0.38, indicating an increasing trend. The results of LISA analysis show that there is a significant spatial difference in the growth rate of water yield in the TLB. The high growth centers are mainly located in the north of the basin, while the low growth centers are mainly located in the southwest of the basin. At the same time, the center of high growth rate of water yield showed a certain trend of expansion and transfer.
Regression analysis was used to analyze the growth of water yield, and the increment of construction land from 1985 to 2010 in both sub-region scales and town scales. The results showed that urban development has a significant impact on water yield; every increase of 1 km2 of construction land in the TLB will lead to an increase of water yield by more than 300,000 m3.
Taking the construction land area in 2010 as the research scope, the analysis showed that the growth of local water production in TLB is much higher than the average value of the basin, and the results show that the growth of local water yield is significantly related to the GDP per capita.