Spatiotemporal Evolution Characteristics and Driving Factors of Water Conservation Service in Jiangxi Province from 2001 to 2020

Abstract

Water conservation services are key indicators of ecological services. Against the backdrop of frequent extreme weather events and water scarcity caused by global climate change and intensified human activities, assessing these services and their drivers are crucial tasks for regional ecological security and sustainable development. Jiangxi Province is one of the first national ecological civilization pilot zones in China, representing an important ecological barrier in southern China. Exploring the characteristics of spatial and temporal changes in water conservation and their driving factors can facilitate the rational development and utilization of regional water resources and the construction of ecological civilizations. Therefore, based on long time series data, the InVEST model was used to explore the spatiotemporal evolution characteristics of water conservation services, and to elucidate the trend of their change through the Theil–Sen median trend analysis and the Mann–Kendall test; then, the geographic detector and geographically weighted regression model were used to further analyze the drivers of spatial variability of water conservation services. The results showed the following: (1) The average depth of water conservation was 103.18 mm, showing a spatial pattern of “low in the middle, high in the surroundings, high in the north and low in the south”. (2) Slight improvements were primarily observed (77.49%), with only 1.60% of the area showing significant improvements. (3) Land use was the main driver of the spatial differentiation, and the interaction between precipitation and forestland had a significantly greater effect on spatial heterogeneity than any single factor. (4) Obvious spatial heterogeneity occurred in the driving factor impacts, with natural factors (precipitation, evapotranspiration, forestland, and grassland) having a positive impact on water conservation services, and land-use factors (construction land and cropland) and socioeconomic factors (population density and land area) having a negative effect. This study provides a reference for water-conservation-based ecosystem construction and policy formulation in Jiangxi Province.

1. Introduction

Ecosystem services are the basis for maintaining a balance among the Earth’s living systems and realizing the sustainable development of human society [1], and they have become a topic of interest for multidisciplinary research in ecology [2], economics [3], resources, and environmental studies [4]. Water conservation services entail the unique ecosystem structure’s functions of retaining, infiltrating and storing precipitation at a given spatial and temporal scale [5,6]. As an important component of ecosystem-regulating services [7,8], water conservation services help meet the water demand of vegetation, soil, and other internal ecosystem components [9], provide water resources for external use during dry periods and regulate the uneven seasonal distribution of water resources [10]. In addition, these services have various ecological functions, such as regional climate regulation, water purification, and biodiversity conservation [11,12]. In recent years, severe global climate change and intensified human activity have led to frequent extreme weather events [13], and increased the severity of water shortages [14]. Thus, in-depth research on water conservation is of great significance for regional ecological protection and water resource management.

Water conservation functions were originally discussed in terms of forest hydrology [15,16], and many studies have been conducted in various countries to assess water conservation functions at different regional scales [17,18,19], as well as in different ecosystems [9,20] and land-use types [11,21]. Traditional assessment methods focus on integrated water storage [22], water balance [23], and soil water storage [24]. With the continuous development of remote sensing and geographic information technology, the use of models such as MIKE System Hydrological European (MIKE SHE) [25], Soil and Water Assessment Tool (SWAT) [26], and Integrate Valuation of Ecosystem Services and Tradeoffs tool (InVEST) [27], among others, has become a trend observed for the quantitative and refined assessment of water conservation services. For example, Gajanan et al. [28] used the MIKE SHE model to study the impacts of future climate change protection measures on surface runoff and groundwater recharge in the basin. Li et al. [21] used SWAT as the basis for assessing the water conservation capacity of ecosystems in the Han River Basin of China and the spatiotemporal evolutions of water conservation capacity of different ecosystems, and explored the mechanisms through which land use change and climate change affect water conservation capacity. Hu et al. [11] used the water yield module of the InVEST model to spatially quantify the water yield capacity of Dongting Lake and Poyang Lake wetlands, and found that climate change is the main factor of water yield. Among the numerous model simulation methods, the MIKE SHE model is able to integrate groundwater, surface water simulations, and specific model structure designs according to the needs of the researcher [29]. However, the model requires a high level of input data and a large number of hydraulic parameters, thus increasing the difficulty of identifying a basin with complete operational data. Although the SWAT model is suitable for the assessment of large-scale watersheds and enables long-term continuous simulations, the model requires a great number of input parameters [30,31], and the simulation results in China are not satisfactory at present. The InVEST model, however, combines refinement and quantification and integrates precipitation, evapotranspiration, soil, and land-use factors [32]; thus, it has obvious advantages in terms of data acquisition, parameter setting, and spatial representation [33]. It can also combine GIS technology with ecosystem service assessments, thus laying the foundation for subsequent geospatial research [33,34]. Since its release, the model has been widely used at watershed and regional scales, such as in the Guishui River Basin [35], Beijing–Tianjin–Hebei region in China [36], Kentucky in the US [37] and São Paulo in Brazil [38]. However, most existing studies on the spatiotemporal evolution of water conservation based on the InVEST model have analyzed data from certain periods in a given year or over a long-time span, and studies based on continuous time series are lacking. The spatiotemporal evolution of water conservation services is a complex eco-hydrological process under the coupling of factors such as precipitation, soil, and topography [10,39], and low-density data may ignore the effects of environmental fluctuations [32], thus leading to errors in describing long-term trends, which may result in unscientific perceptions and decision making. Therefore, long time series and high-density data must be introduced to assess the spatiotemporal evolution of water conservation services.

Most existing analyses on the drivers of water conservation services are based on correlation, principal component [39,40], and cluster analyses [41]. However, water conservation services are affected by multiple factors that are often interactive, nonlinear, and complex [42]. Geographic detector (GD) is independent of linear assumptions [43] and employs spatial variance, thereby enabling a direct quantification of the effects of single and multiple drivers on the spatial distribution pattern of water conservation services [44]. Geographically weighted regression (GWR), however, incorporates the spatial location of driver samples into the parameters based on the traditional least-squares model [45], thus allowing for a localized parameter assessment that better reveals regional differences in the impacts of the drivers. Thus, the combination of the two provides an effective quantitative approach to efficiently analyze the combined response of multiple drivers and the spatial differentiation of their effects on changes in water conservation services. For example, Lei et al. [18] studied the spatiotemporal characteristics and drivers of water yield in three major watersheds on Hainan Island and found that land-use type was the dominant factor in the spatial differentiation of water yield using GD. Qiao et al. [46] performed a quantitative assessment of water conservation in Heilongjiang Province, analyzed the dominant influencing factors on the spatial differentiation characteristics of the observed changes over the last two decades using GD and GWR, and captured the spatial differentiation of driving factors.

Jiangxi Province is located on the southern bank of the middle and lower reaches of the Yangtze River and has a diverse ecosystem, and it hosts more than 2400 rivers of different sizes and a total length of 18,400 km, as well as Poyang Lake, which is the largest freshwater lake in China. The province is rich in forest resources, with a forest coverage rate as high as 63.1% [47]. Thus, it represents the initial ecological civilization construction zone in China and provides important ecological barrier functions for the Yangtze River Basin in China. However, in recent years, conservation has degraded under the influence of climate change and human activities and the aquatic ecological environment has deteriorated in the province, which seriously threatens regional water security and sustainable development. Therefore, there is an urgent need to analyze the spatiotemporal evolution and drivers of water conservation services in Jiangxi Province.

Thus, this study extends the research on water conservation services in two major dimensions. First, this study introduces a long time series of high-density data to improve the precision of the model results. Although existing studies have focused on the dynamic characteristics of the spatiotemporal evolution of water conservation, most have used phase data that are sparsely distributed over time. Second, this study used GD and GWR models to explore in depth the explanatory power of the drivers as well as the strength and direction of their spatial roles from multiple aspects of natural and human activities, which will provide references to promote water resource protection in Jiangxi Province and support local policy formulation and sustainable regional development.

This study quantitatively assessed the annual water conservation services in Jiangxi Province from 2001 to 2020 using the water yield module of the InVEST model and analyzed the spatiotemporal change characteristics. Then, the change trends were analyzed by combining the Theil–Sen median trend analysis and Mann–Kendall test. Finally, the driving factors and their impacts on spatial variation were investigated using the GD and GWR models to identify change drivers and spatial differences in their impact.

2. Study Area and Data Collection

2.1. Study Area

Jiangxi Province is located along the south bank of the middle and lower reaches of the Yangtze River in southeastern China (24°29′14″ N~30°4′41″ N, 113°34′36″ E~118°28′58″ E), with a total area of 166,900 km² (Figure 1). The terrain is mainly mountainous and characterized by hills, with high terrain in the east, south, and west of the province; mainly mountainous, with hills and river valley plains interspersed in the middle; and flat and open terrain north of the Poyang Lake Plain. The area has a subtropical monsoon climate with warm winters and hot summers, and the annual average temperature is approximately 16.3–25 °C [47]. The annual precipitation ranges from 1341 to 1943 mm, with large interannual variations in precipitation, frequent droughts, and floods. The entire territory is rich in water systems and includes five major rivers, namely, the Ganjiang, Fuxiang, Xinjiang, Xiu, and Rao Rivers, which converge on Poyang Lake and form a centripetal water system with Poyang Lake as the core, which is transferred to the Yangtze River by Hukou County after being impounded by Poyang Lake.

2.2. Data Collection and Preparation

In this study, basic data, such as land cover data, annual precipitation, annual potential evapotranspiration, normalized difference vegetation index and soil depth, were collected for water conservation and change driver analyses, and the data sources and pre-processing are shown in

Table 1. Data source and collection.

Data Name	Description	Data Source	Data Collection Time
Land-use/ land-cover	Land-use types are divided into nine land types: cropland, forest, shrub, grassland, water, snow/ice, barren, construction land and wetland	Annual China Land Cover Dataset (CLCD), containing year-by-year land cover information for China from 1985 + 1990 to 2020 (https://www.zenodo.org/, accessed on 30 March 2023).	2001–2020
Annual precipitation	Annual precipitation/mm	The Chinese meteorological data network (http://data.cma.cn/, accessed on 30 March 2023) was used to interpolate the station data using ANUSPLIN interpolation.	2001–2020
Potential evapotranspiration	Potential evapotranspiration/mm	NTSG (Numerical Terradyamic Simulation Group) MODIS6 evaporation products (http://files.ntsg.umt.edu/data/NTSG_products, accessed on 30 March 2023)	2001–2020
Annual temperature	Annual temperature/°C	National Earth System Science Data Sharing Service Platform (http://www.geodata.cn/, accessed on 30 March 2023)	2001–2020
NDVI	Normalized difference vegetation index	MOD13Q1 satellite data	2001–2020
Soil properties	Includes soil depth, sand, chalk, clay and soil organic matter content/%	World Soil Database, China Soil Data Set (HWSD, http://webarchive.iiasa.ac.at/Research/LUC/External-World-soil-database, accessed on 30 March 2023)	-
DEM	Elevation	Geospatial Data Cloud (http://www.gscloud.cn, accessed on 27 March 2023)	2020
Sub-watershed	Basin boundaries	Geospatial Data Cloud (http://www.gscloud.cn, accessed on 30 March 2023)	-
Z (Zhang) parameter	Z is an empirical parameter indicating the distribution of regional precipitation and other hydrogeology, and takes values between 1 and 30	Z = 6.28. The optimal results were obtained by repeatedly verifying the simulated water yield with the measured surface water resources in the “Jiangxi Water Resources Bulletin”.	-

Note: DEM, digital elevation model.

.

3. Methodologies

3.1. Technical Route

This study collected data on the topography, ecology, population, and economy of Jiangxi Province and combined this information with findings from the literature to analyze the spatiotemporal evolution characteristics and drivers of water conservation in Jiangxi Province from 2001 to 2020. Study processing consisted of the following four aspects: (1) creation of a basic database. Based on the data described in Section 2.2, the dataset for assessing water conservation in Jiangxi Province was created after processing via ArcGIS 10.6. (2) The amount of conserved water was calculated. Using the water yield module of the InVEST model and the water conservation formula, the annual water conservation depth of the study area was measured, and the spatiotemporal evolution characteristics of water conservation services were analyzed. (3) Trends in the dynamic evolution were analyzed. The Theil–Sen median trend analysis and Mann–Kendall test were used to analyze the 20-year trends of water conservation change in Jiangxi Province and their spatial distribution. (4) Spatial differences in the drivers and their explanatory power were assessed. The GD and GWR models were used to explore the drivers of natural and human activities that influence changes in water conservation in the region and the spatial variation in the role of each factor. The technical route is shown in Figure 2.

3.2. Water Conservation Based on the InVEST Model

3.2.1. Water Yield

The water yield module of the InVEST model is based on the Budyko hydrothermal coupling principle [37], and it is calculated in terms of raster cells. The annual water yield for each grid cell is the difference between the annual precipitation and actual annual evapotranspiration of each grid cell [18]. The specific calculation method is as follows:

$$Y_{\mathit{xj}}=\left(1-\frac{{\mathit{AET}}_{\mathit{xj}}}{P_x}\right){\times P}_x$$

(1)

$$\frac{{\mathit{AET}}_{\mathit{xj}}}{P_x}=\frac{{1+\omega }_x{R}_{\mathit{xj}}}{{1+\omega }_x{R}_{\mathit{xj}}+\frac{1}{R_{\mathit{xj}}}}$$

(2)

$$R_{\mathit{xj}}=\frac{K_c{\times \mathit{ET}}_0}{P_x}$$

(3)

$${\omega }_x=Z\times \frac{{\mathit{AWC}}_x}{P_x}+1.25$$

(4)

$${\mathit{AWC}}_x{=\mathit{min}(\mathit{maxsoil}\_\mathit{depth},\mathit{root}\_\mathit{depth})\times \mathit{PWAC}}_x$$

(5)

$${\mathit{PWAC}}_x=54.509-0.132\times \mathit{sand}-0.003\times {\left(\mathit{sand}\right)}^2-0.055\times \mathit{silt}-0.006\times {\left(\mathit{silt}\right)}^2-0.738\times \mathit{clay}+0.007\times \left({\mathit{clay}}^2\right)-2.688\times \mathit{som}+0.501\times {\left(\mathit{som}\right)}^2$$

(6)

in Formula (1), Y_xj is the annual water yield of raster cell x on land type j (mm); P_x is the annual precipitation of raster cell x (mm); AET_xj is the actual total annual evapotranspiration of raster cell x on land type j (mm); $\frac{{\mathit{AWC}}_x}{P_x}$ is the approximate Budyko curve [48,49], calculated using Formula (2). R_xj is the drying index of raster cell x, which is the ratio of potential evapotranspiration to precipitation; ω_x is the plant water use coefficient, which is the ratio of plant water demand to precipitation. R_xj and ω_x are calculated, respectively, according to Formulae (3) and (4). In Formula (3), K_c is the vegetation evapotranspiration coefficient; ET₀ is the potential evapotranspiration. In Formula (4), Z is an empirical parameter that indicates the distribution of regional precipitation and other hydrogeology and takes values between 1 and 30; AWC_x is the effective soil water conservation (mm), which can be calculated as in Formulae (5) and (6). In Formula (5), max soil_depth is the maximum soil depth (mm); root_depth is the root depth (mm); PAWC_x is the available water for vegetation, calculated using Formula (6). In Formula (6), sand, silt, clay, and som are the content of soil sand, silt, clay and organic carbon (%), respectively.

3.2.2. Water Conservation

After calculating the annual water yield using the InVEST model, the annual water conservation was calculated using the topographic index, flow coefficient, and soil saturation hydraulic conductivity [49]. The specific calculation method is as follows:

$$\mathit{RE}=\mathit{min}(1,\frac{249}{V})\times \mathit{min}(1,\frac{0.9\times \mathit{TI}}{3})\times \mathit{min}(1,\frac{\mathit{Ks}}{300})\times Y$$

(7)

$$\mathit{TI}=\lg \frac{D_{\mathit{area}}}{{\mathit{soil}\_\mathit{depth}\times P}_{\mathit{slope}}}$$

(8)

in Formulae (7) and (8), RE is the annual water conservation depth (mm); V is the flow rate coefficient; TI is the topographic index (dimensionless) and is calculated using Formula (8); Ks is the soil saturation hydraulic conductivity (cm/d); Y is the annual water yield (mm), calculated using Formula (1); D_area is the number of grids in the catchment; soil_depth is the soil depth (mm); P_slope is the percentage slope, which was obtained from DEM elevation data based on an ArcGIS 10.5 software slope analysis.

3.3. Theil–Sen Median Trend Analysis and Mann–Kendall Test

The combination of the Theil–Sen median trend analysis and Mann–Kendall test has been widely used to determine the changes in long time series data trends, and it was used in this study to analyze the spatiotemporal variation characteristics of water conservation in Jiangxi Province and its trends. The Theil–Sen median trend analysis is a robust non-parametric statistical trend calculation method that can reduce the influence of data outliers [50]. The specific calculation method is as follows:

$$S_{\mathit{RE}}=\mathrm{Median}\left(\frac{{\mathit{RE}}_j-{\mathit{RE}}_i}{j-i}\right),2001⩽i<j⩽2020$$

(9)

in Formula (9), RE_j and RE_i denote the water conservation depth, calculated from Formula (7) of year i and year j, respectively; S_RE > 0 reflects an increasing trend in water conservation, while S_RE < 0 reflects a decreasing trend in water conservation.

The Mann–Kendall test is a nonparametric statistical test used to determine the significance of a trend [51]. Setting the water conservation depth from 2001 to 2020 as RE_i (i = 2001, 2002, …, 2020), the statistical test value Z was calculated as follows:

$$Z=\left\{\begin{array}{cc}\frac{S-1}{\sqrt{s\left(S\right)}},& S>0\\ 0,& S=0\\ \frac{S+1}{\sqrt{s\left(S\right)}},& S<0\end{array}\right. \mathrm{which},S=\sum _{j=1}^{n-1}\sum _{i=j+1}^n\mathit{sgn}({RE}_j-{RE}_i)$$

(10)

$$\mathit{sgn}\left({\mathit{RE}}_j-{\mathit{RE}}_i\right)=\left\{\begin{array}{ll}1,& {\mathit{RE}}_j-{\mathit{RE}}_i>0\\ 0,& {\mathit{RE}}_j-{\mathit{RE}}_i=0\\ -1,& {\mathit{RE}}_j-{\mathit{RE}}_i<0\end{array}\right.$$

(11)

$$s\left(S\right)=\frac{n\left(n-1\right)\left(2n+5\right)}{18}$$

(12)

in Formulae (10)–(12), RE_j and RE_i denote water conservation depth, calculated using Formula (7) in years i and j, respectively; n is the length of the time series; sgn is the sign function. The value of the statistic Z presents a range of (−∞, +∞). At a given significance level α, when |Z| > u_1−α/2, significant changes occurred in the study series at the α level. Taking a value of α = 0.05 [51], this study determines the significance of the time series change trend of ecosystem services function at a 0.05 confidence level.

3.4. Geographic Detector

GD was used to reveal the main drivers of spatial heterogeneity in water conservation. This model consists of four modules that explain the drivers of water conservation: factor detector, risk detector, ecological detector, and interaction detector. Factor detector was used to detect the extent to which a driver explained the spatial heterogeneity in the spatiotemporal evolution of water conservation change, as measured via the q-value, which identifies the interaction between different independent variables [44]. The q-value is calculated as follows:

$$q=1-\frac{{\sum }_{h=1}^L{N}_h{\mathsf{\sigma }}_h^2}{{N\mathsf{\sigma }}^2}$$

(13)

in Formula (13), q is the explanatory power of the discretized impact factor; h = 1, 2, 3, …; L is the classification or stratification of the independent variable; N_h and N represent impact factor h and the total number of impact factors, respectively; σ² and ${\mathsf{\sigma }}_h^2$ represent the variance of the overall sample size and stratified h in the study area, respectively.

3.5. Geographically Weighted Regression Model

Based on the GD method, this study used a GWR model to identify the spatial distribution of the impact of different drivers on water conservation. GWR applies the concept of local regression based on ordinary least squares (OLS) to consider the spatial non-homogeneity and spatial non-smoothness of geographical elements. Compared with ordinary linear regression, GWR can better reflect the spatial heterogeneity of parameters across different regions. The specific calculation method is as follows [52]:

$$\begin{array}{c}{y}_i{=\beta }_0\left(u_i{,v}_i\right)+{\displaystyle {\displaystyle \sum }_{j=1}^k}{\beta }_j\left(u_i{,v}_i\right)x_{\mathit{ij}}+{\epsilon }_i\end{array}$$

(14)

in Formula (14), y_i is the dependent variable, that is, the amount of change in water conservation services; β₀ is the intercept; (u_i, vi) is the coordinate of sub-watershed i; β₀(u_i, v_i) is the constant term of sub-watershed i; x_ij is the jth independent variable of sub-watershed i; β_j(u_i, v_i) is the regression coefficient of the j independent variable of sub-watershed i; ε_i is the random error term. In this study, an “adaptive” kernel and modified Akaike information criterion (AIC) were chosen to determine the optimal bandwidth.

Studies on the factors that influence water conservation services have shown that these are complex and diverse, but can be broadly divided into natural environmental factors and human activity factors. Precipitation and evapotranspiration are two important climatic factors that directly affect regional water conservation [36,53,54], while temperature indirectly affects the water conservation process by influencing water evaporation [11]. Forestlands and grasslands also influence water conservation because of the water interception, storage, and evapotranspiration effects of vegetation [17,53]. With population increases and rapid industry and agriculture development, human activities have gradually become an important driving force affecting the global hydrological cycle and evolution of water resources. Therefore, human activities must be clarified to study the impacts of human disturbances on water conservation services. In particular, land-use types are subject to intense human disturbances [55,56], which represent important causes of changes in regional water conservation. Population density and GDP reflect the level of urban socioeconomic development, which is a potential factor in water conservation changes [10,57]. Therefore, the following nine factors were selected as explanatory variables: precipitation, potential evapotranspiration, temperature, percentage of forestland, percentage of grassland, population density, GDP, percentage of construction land, and percentage of cropland. Details are presented in

Table 2. Influence factors on the water conservation function change.

Variable Type		Variable Name	Variable Codes	Description and Source
Natural environment	Climatic factors	Precipitation	X1	Precipitation and evapotranspiration have a direct impact on water conservation. Temperature affects water conservation by influencing evapotranspiration [11,36,53,54].
		Potential evapotranspiration	X2
		Temperature	X3
	Vegetation factors	Percentage of forestland	X4	Forestlands and grasslands influence the hydrological cycle processes through the interception and storage of water by vegetation, evapotranspiration, and heat dissipation, thus affecting water conservation [17,53].
		Percentage of grassland	X5
Human activities	Socioeconomic factors	Population density	X6	Population density and GDP represent potential factors that reflect urban socioeconomic development and changes in water conservation [55,56].
		GDP	X7
	Land-use factors	Percentage of construction land	X8	Different land-use types have a significant impact on changes in water conservation by affecting surface runoff, evaporation, and heat dissipation [10,57].
		Percentage of cropland	X9

.

4. Results

4.1. Z Parameter Values

The accuracy of the Z parameter in the water yield module of the InVEST model significantly influences the calculation results of the water conservation model [27]. In this study, the Z parameter was determined based on the measured data of the multi-year average surface water resources from the Water Resources Bulletin of Jiangxi Province. Through repeated tests, the results show that when the Z parameter was 2.67, the simulated multi-year average water yield was 1545.74 × 10⁸ m³, which is close to the multi-year average surface water resources of Jiangxi Province of 1546.12 × 10⁸ m³. This value achieves the best simulation effect (Figure 3).

4.2. Spatial and Temporal Variability in Water Conservation

The temporal distribution characteristics of water yield and conservation in Jiangxi Province from 2001 to 2020 were essentially identical (Figure 4). Specifically, the average annual depth of the water yield in Jiangxi Province fluctuated between 548.87 mm and 1397.94 mm, while the average depth was 925.88 mm (Figure 4a). The minimum and maximum values were observed in 2011 and 2012, respectively. The water yield in Jiangxi Province initially increased and then decreased with large variations, and it generally showed a slight upward trend. The average annual depth of water conservation and water yield showed a consistent overall small increasing trend, with an initial rise and then a decline. Over the past 20 years, the water conservation depth in Jiangxi Province has fluctuated between 61.55 mm and 155.24 mm, with a multi-year average water conservation depth of 103.18 mm (Figure 4b). The change in water conservation depth was more drastic from 2001 to 2012 and dropped to a minimum value of 61.55 mm in 2011, increasing to a maximum value of 155.24 mm in 2012. Although the depth of water conservation slightly rebounded from 2018 to 2020, an overall downward trend was observed from 2013 to 2020.

As shown in Figure 5, the multi-year average water conservation depth space in Jiangxi Province from 2001 to 2020 presented distribution characteristics of high all around and low in the middle, high in the north, and low in the south, with small changes in the spatial pattern of water conservation depth in different years (Figure 6). The high-value areas were located in the eastern, western, and southern areas dominated by forestland, and the high values in the Wuyi Mountain Range and Huaiyu Mountain area in the northeast were particularly significant. The low-value area was located in the central Poyang Lake Plain, which was associated with the higher precipitation in the east and north, lower precipitation in the west and south, and the effects of land cover type. Jiangxi Province is surrounded by mountains on three sides and dominated by forestland. The Wuyi Mountain Range in the east has a large vegetation cover, and the root system of the vegetation can effectively intercept atmospheric precipitation into terrestrial ecosystems; together with sufficient precipitation, these factors produce a clear water conservation effect. The center is the plain of Poyang Lake, which has a flat terrain, frequent human activities (mainly cropland and construction land), and an insufficient capacity to intercept and store water, resulting in a weak water conservation capacity.

4.3. Characteristics of Spatially Significant Changes in Water Conservation

In this study, the Theil–Sen median trend analysis was combined with the Mann–Kendall test to classify S_RE as stable and unchanged (−0.0005–0.0005), improved (≥0.0005), and degraded (<0.0005). The Mann–Kendall significance test at the 0.05 confidence level was also selected to classify the results as significantly changed (Z ≥ 1.96 or Z ≤ −1.96) and insignificantly changed (−1.96 ≤ Z ≤ 1.96). As shown in Figure 7, significantly more improved areas were observed compared with degraded areas in terms of the depth of inter-annual water conservation in Jiangxi Province. The degraded areas were mainly distributed in the central part of the study area, while the improved areas were mainly distributed in the mountainous and hilly areas in the west, south, and east of the study area; the overall improvement and degradation trends were not significant.

From 2001 to 2020, the water conservation depth in Jiangxi Province showed a slight degradation trend, with only a few areas showing significant improvement. The trend results were obtained by overlaying the Theil–Sen median trend analysis with the graded results of the Mann–Kendall test, and the results were classified into five types (

Table 3. Statistics on trends in water conservation depthS.

Trends in Water Conservation	Range of Values	Area/km2	Percentage/%
Severe degradation	β < −0.0005; Z < −1.96	4581.23	2.74
Slight degradation	β < −0.0005; −1.96 ≤ Z ≤ 1.96	17,644.76	10.57
Stable and unchanging	−0.0005 ≤ β ≤ 0.0005; −1.96 ≤ Z ≤ 1.96	11,893.21	7.12
Slight improvement	β ≥ 0.0005; −1.96 ≤ Z ≤ 1.96	129,362.87	77.49
Significant improvement	β ≥ 0.0005; Z ≥ 1.96	3465.93	2.08

and Figure 7). Table 3 shows that the overall improvement in the water conservation depth was slight. The degraded water conservation depth areas accounted for 13.31% of the total study area. Of these, 2.74% were severely degraded. Slightly improved areas accounted for 77.49% of the total study area, and only 2.08% of the area improved significantly over the 20-year period.

Figure 8 shows the spatial pattern in the change trends of water conservation depth, and it reveals that effective improvements in water conservation services occurred in the east, west, and south of Jiangxi Province from 2001 to 2020, and more pronounced degradation of water conservation services occurred within the Poyang Lake plain area. Specifically, the areas with improved water conservation services were mainly concentrated in mountainous forest areas in the east, south, and west of Jiangxi Province. Among them, Nanfeng County in Fuzhou City has been significantly upgraded over the last 20 years. Nanfeng County is one of the upgraded eco-villages in Jiangxi Province, and the enhanced water conservation is closely related to river and lake environmental management and protection. Moreover, ecological protection and restoration projects have been vigorously carried out in Nanfeng County in recent years. Although the slightly degraded areas were widely distributed in the plain area of Poyang Lake, which is dominated by cropland and construction land, the severely degraded areas were widely distributed in various regions of the study area in the form of points and blocks, such as Yichun City. This change is due to the rapid development of the city in recent years, expansion of the city scale, and continuous increase in construction land, which has resulted in serious degradation of water conservation services in the urban periphery.

4.4. Analysis of the Drivers of Spatial and Temporal Variability in Water Conservation

4.4.1. Geographic Detector Results for Drivers

The factor detection results for water conservation services in Jiangxi Province showed that all factors passed the significance test (p < 0.05), and both the natural environment and human activity factors influenced the spatial layout of water conservation, with the degree of influence of land use being slightly higher than that of the other factors (

Table 4. Detection results of factors affecting water conservation in Jiangxi Province.

	X1	X2	X3	X4	X5	X6	X7	X8	X9
q statistic	0.336	0.080	0.190	0.297	0.197	0.318	0.204	0.379	0.438
p value	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000

). The influencing factors were ranked according to their q-value as follows: percentage of cropland > percentage of construction land > precipitation > population density > percentage of forestland > GDP > percentage of grassland > potential evapotranspiration > temperature. Among them, the percentages of cropland (q = 0.438) and construction land (q = 0.379) had the highest explanatory power for the spatial heterogeneity of water conservation services, indicating that land use was the main driver influencing the spatial pattern of water conservation services in Jiangxi Province. In addition, precipitation was an important factor influencing the change in spatial distribution (q = 0.336).

The spatiotemporal changes in water conservation services were associated with multiple factors, and the interaction detection results are shown in

Table 5. Detection results for the interaction among factors affecting water conservation.

	X1	X2	X3	X4	X5	X6	X7	X8	X9
X1	0.336
X2	0.444 ↑↑	0.080
X3	0.413 ↑	0.327 ↑↑	0.180
X4	0.587 ↑	0.482 ↑↑	0.563 ↑↑	0.297
X5	0.475 ↑	0.322 ↑↑	0.432 ↑↑	0.537 ↑↑	0.197
X6	0.442 ↑	0.370 ↑	0.392 ↑	0.535 ↑	0.433 ↑	0.318
X7	0.441 ↑	0.429 ↑↑	0.355 ↑	0.497 ↑	0.372 ↑	0.356 ↑	0.204
X8	0.465 ↑	0.436 ↑	0.445 ↑	0.569 ↑	0.506 ↑	0.410 ↑	0.414 ↑	0.379
X9	0.557 ↑	0.541 ↑↑	0.512 ↑	0.576 ↑	0.579 ↑	0.517 ↑	0.449 ↑	0.511 ↑	0.438

Note: The symbol “↑” indicates a two-factor enhancement; “↑↑” indicates a non-linear enhancement. Values in the table are q-statistics.

. The interaction between factors was dominated by two-factor enhancement, which indicates that the effects of two-factor interactions are more significant than those of any two single independent factors. Among the contributions of the interaction factors, five groups had values greater than 0.55: precipitation ∩ percentage of forestland (q = 0.587), percentage of cropland ∩ percentage of grassland (q = 0.579), percentage of cultivation land ∩ percentage of forestland (q = 0.576), temperature ∩ percentage of forestland (q = 0.563), and precipitation ∩ percentage of cropland (q = 0.557). This finding shows that the combined effect of precipitation and land use had a significantly higher explanatory power for the spatial differentiation of regional water conservation than the effect of any single factor. The interaction between precipitation and the proportion of forested land area had the strongest explanatory power, and the explanatory power of forestland area increased from 0.279 to 0.587 for a single factor, indicating that the coupling of precipitation and forestland had the most significant effect on the variation in water conservation over the past two decades.

4.4.2. Spatial Variation in the Role of Drivers

The GWR model results were used to reflect the direction and strength of the effects of each influencing factor (Figure 9), and the OLS method was used to test for variable covariance. The variance inflation factor for all variables was below 7.5, indicating the absence of redundant factors. A comparison of the OLS and GWR models showed fit values of 0.637 and 0.638, respectively, indicating that the GWR model had a better fit. Moreover, the AIC values and sigma of the GWR model were smaller than those of the OLS model, indicating that the GWR model had better explanatory power and applicability to the drivers of water conservation services (

Table 6. OLS parameter results and GWR diagnostic coefficient statistics.

Diagnostic Indicators	OLS	GWR
R2	0.644	0.645
Adjusted R2	0.637	0.638
AICc	4451.040	894.890

).

In terms of directionality, precipitation, potential evapotranspiration, percentage of forestland, and percentage of grassland showed significant positive correlations with water conservation services, while the percentage of construction land and percentage of cropland mostly showed negative correlations. The positive and negative effects of temperature, population density, and GDP on water conservation services were spatially staggered (Figure 9).

(1)

Climatic factors

Precipitation and potential evapotranspiration had significant positive effects on the depth distribution of regional water conservation, with precipitation showing a stronger positive effect than potential evapotranspiration. The spatial influence of precipitation showed a gradual increase from the east and west to center, and it had the most significant influence in the central and northern parts of the study area, indicating the dominant role of precipitation in regional water conservation. Notably, in Jiujiang City and Shangrao City, precipitation and water conservation showed a weak negative correlation, with a maximum correlation coefficient of −0.46, which was mainly due to the high precipitation and weak water conservation capacity of the two cities.

Potential evapotranspiration had a predominantly positive effect, with a higher degree of influence in the mountainous hills located at the edge of the study area than that in the central plains. Jiangxi Province is generally surrounded by mountainous forests with high vegetation cover, whereas the central plains are dominated by cropland. Compared to cultivated land, forests have a stronger evapotranspiration effect and greater ability to contain water. Therefore, potential evapotranspiration was positively correlated with water conservation in the forest-rich mountainous areas and negatively correlated in the central region.

The effect of temperature on the water conservation depth was spatially heterogeneous. Temperature had strong negative effects on the cities of Yichun, Jingdezhen, and Shangrao in the northern part of the study area and the city of Ganzhou in the southern part, whereas it had strong positive effects in the central–eastern part of the study area. This was mainly due to the low and high temperatures in the north and south, respectively, and the high and low water conservation volume in the north and south, respectively.

(2)

Vegetation factor

Figure 9d,e shows that the effects of forestland and grassland on water conservation depth were generally positive and the strength of the effect of forestland was slightly higher than that of grassland. In the central and southern areas of the Poyang Lake plain area, such as in Nanchang City, southern Ji’an City, and northern Ganzhou City, the positive coefficient was higher than that for grassland (1.50). Meanwhile, in the southern part of Ji’an City, the positive effect of grassland was stronger than that of forestland, which indicates that increasing grassland relative to forestland may be more beneficial for enhancing the water conservation capacity in this area.

(3)

Socio-economics factor

Both the population density and GDP factors showed a strong negative correlation with the depth of water conservation in the northern part of the study area, indicating that human socioeconomic activities can weaken the water conservation effect of the area. In the southeastern part of the study area, the population size and urban economy are lower than those in the northern part of the study area because of limitations associated with topographic conditions, especially in the southern part of the Wuyi Mountain Range. Therefore, in the localized study area, the amount of water conversion showed a positive correlation with both factors.

(4)

Land-use factor

The proportions of cropland and construction land were significantly negatively correlated with the depth of water conservation in the study area. The proportion of construction land had a negative effect on the depth of water conservation in the entire region, and this effect gradually increased spatially from northwest to southeast, with the high-value area located in Nanfeng County, Fuzhou City. This finding indicated that regional construction land decreased and the depth of water conservation increased. The influence of cropland was the strongest among all the factors, with the highest value of −1.43. In areas with a concentrated distribution of cropland, the proportion of cropland and water conservation services showed a significant negative correlation, whereas in areas with a smaller cropland area, such as Nanchang City, Pingxiang City, and the northern part of Ganzhou City, a weak positive correlation was observed due to the lower depth of water conservation.

5. Discussion

5.1. Model Validation

Enhancing the accuracy of water conservation estimates is a prerequisite for the rational formulation of regional policies and plans for water resource use and ecological protection. This study used the water yield module of the InVEST model to assess the changing trends and spatial layout of water conservation in Jiangxi Province. To calculate the total regional water yield, the model estimates the regional water conservation based on factors such as soil properties, topographic features, and cover types. The Z parameter characterizes the seasonal characteristics of regional climate, rainfall intensity, and topographic features, and its value greatly affects the accuracy of the model’s calculation results [10,58]. Different algorithms have been developed for the Z parameter [27], and previous studies more often use empirical values, which may result in large errors. In this study, the measured annual surface runoff data from the Water Resources Bulletin of Jiangxi Province were used as the standard, and the results of the multi-year average water yield of the InVEST model under different Z values were calculated. These results were compared with the annual surface runoff data, and fitting curves were established by repeating the experiments several times. The optimal water yield simulation results were achieved when Z = 2.67 (Figure 3).

Although different assessment methods may cause large numerical differences owing to differences in their theories and parameter settings, the change trends were almost the same in the study area. This study calculated the water conservation service from 2001 to 2020, which showed a trend of initially increasing and then decreasing. Moreover, it showed spatial characteristics of “high in the middle and north and low in the periphery and south”, which are consistent with the results for the same period in the studies carried out by Zou et al. [59] and Chen et al. [60]. To ensure that the model simulation results objectively reflect the regional spatiotemporal characteristics, this study introduced long time series data for calculating the year-by-year water conservation quantity, and the change trends were consistent with the results of Zou et al. [59,61,62]. Chen et al. estimated the average depth of water yield in the ecological and economic zone of Poyang Lake to be 971.04 mm by using the InVEST model with five phases of data: 2000, 2005, 2010, 2015 and 2019 [60], but it was 925.88 mm in this study. The results of the year-to-year trend of this study showed that 2003 and 2011 were the low value years of water yield (Figure 4a). Therefore, the use of stage data ignored the effect of low value years, resulting in high values of water yield.

Natural and human activity factors jointly influenced the spatiotemporal evolutionary characteristics of water conservation services in Jiangxi Province. Researchers have reached similar conclusions in the Yellow River Basin [63], Danjiang River Basin [49], Tumen River Basin [10], and Taohe River Basin [64]. However, previous studies have used correlation analysis or principal component analysis to explore the relationship between factors and water conservation and thus were not able to elucidate the interaction between factors or reflect regional differences in the influence of driving factors. In this study, we extended the study of the driving mechanisms of natural and human environmental factors to the spatial variation in regional water resources using a combination of geographic probes and geographically weighted regression model (Table 4 and Table 5, and Figure 9). The results of this study provide refined management suggestions for regional water conservation and utilization and sustainable urban development.

5.2. Attribution Analysis of the Spatial and Temporal Evolution of the Water Conservation

In this study, the average annual water yield and average annual water conservation in Jiangxi Province increased in 2001–2012 but fluctuated in 2013–2020 (Figure 4), which was highly consistent with the trend of average annual precipitation. This indicates that precipitation is the main driver affecting the temporal variation in the ecosystem water conservation function, which is in line with the results of existing studies [10,11]. Based on the results of the driving factor analysis, precipitation was also an important factor influencing the spatial differentiation of water conservation services in Jiangxi Province. The regression coefficients showed that precipitation had a significant positive effect on water conservation in almost the entire region (Figure 9a) and that the interaction between precipitation and forestland was much stronger than the effect of any single factor (Table 5). Multi-year trend analysis showed that the slight improvement of water conservation areas and the distribution of forestland highly overlapped during the 20-year period (Figure 8). Forestland in Jiangxi Province accounted for approximately 65% of the total area of the province, which was dominated by mountain forests. Thus, forestland is the main land-cover type in the high value area of water conservation (Figure 5 and Figure 6). Precipitation represents a direct source of water conservation; the forest canopy layer and withered leaf layer of forestland can effectively increase the time over which precipitation is converted into runoff [42,65]. This change promotes the high absorption of precipitation by the soil layer, thus increasing the water conservation capacity in terrestrial ecosystems. The interaction between precipitation and forestland was the main reason for the slight improvement in water conservation in Jiangxi Province.

The influence of human activities on the spatial pattern of water conservation is complex, and in this study, land-use factors (percentage of construction land and percentage of cropland) and socioeconomic factors (population density and GDP) strongly influenced the spatial differentiation of water conservation in Jiangxi Province. The explanatory power of land use as a single factor was higher than that of climatic factors, which is consistent with TU et al.’s [62] study on the amount of water conservation in the Dongjiang source area of Jiangxi Province. Land-use changes affect hydrological processes, such as infiltration, evapotranspiration, soil water retention, and surface runoff, by changing soil conditions, soil erosion, and subsurface conditions, which, in turn, affects water conservation. Cropland and construction land are semi-artificial and artificial surfaces with a low vegetation cover, shallow vegetation rooting depths, high surface flow coefficients, and high evapotranspiration [32,66], which results in a low water retention capacity. Combined with the results of the GWR model, land use and socioeconomic behaviors had a significant negative impact on the water conservation capacity of urban areas (Figure 9f–i). Over the past 20 years, the rapid development of urbanization, agriculture, and industry has led to a rapid decrease in the area of forested land in Jiangxi Province, and the area of cropland and land used for construction has increased significantly, which partially explains the significant degradation in the water conservation capacity of the central plains in Jiangxi Province (Figure 8).

5.3. Proposals for Upgrading Water Conservation Services

This study indicates that the water conservation service in Jiangxi Province from 2001 to 2020 presented an increasing and then decreasing trend and a slight improvement in general. Moreover, precipitation and the interaction between precipitation and forestland played a dominant role in this trend. The significant decline in forestland and the rapid expansion of cropland and construction land are the main reasons for the serious degradation of water conservation in the Poyang Lake Plain area of central Jiangxi Province. The results of this study fill the gap in the literature on the characteristics and influencing factors of water conservation services in Jiangxi Province based on the use of high-density and long time-series data, which will help regional managers implement accurate ecological regulations and sustainable ecological management.

Appropriate measures should be implemented based on these findings. First, precipitation dominates the time course of water conservation services in Jiangxi Province, and its combined effect with forestland is greater than that of any single factor; therefore, forestland should be the preferred land-use type for future ecological improvement measures in Jiangxi Province. For important water conservation areas in the east, west, and south of Jiangxi Province, managers should pay close attention to the potential threats brought about by climate change and human activities and carry out regular service function assessments. Jiangxi Province is a large agricultural province in China, and in the central Poyang Lake Plain area, it is necessary to pay attention to the hydrological effects produced by the joint action of cropland and precipitation. Under the premise of strictly abiding by basic farmland practices, measures, such as returning farmland to forests and grasslands, should be implemented appropriately. However, the construction of farmland protection forests may also be an effective measure of enhancing the water conservation efficiency of croplands, including canals and river belts. Finally, future regional managers should pay attention to the management of urban production and construction in the study area and control the intensity of construction activities of croplands, urban and rural industries, and mines to improve the stability of water conservation services and the level of water resource protection in Jiangxi Province.

5.4. Limitations and Prospects

First, the diverse topography and complex land-use/cover components in Jiangxi Province increase the difficulty of assessing the spatial heterogeneity of water conservation services to a certain extent. For the InVEST model selected in this study area, some input data that require field measurements are not easy to obtain because of the large scope of Jiangxi Province; therefore, we can only obtain them by consulting the literature or calculating them based on existing data, which introduces a certain amount of error between the output results of the resulting model and the actual quantities. Therefore, more time and human resources must be invested in data acquisition to improve data accuracy. The water conservation services in this study were based on the water yield module of the InVEST model, and they were corrected for soil and terrain factors. The water yield module of the InVEST model is based on the principle of water balance, which is essentially the difference between rainfall and evapotranspiration [10]; however, the water supply status of terrestrial ecosystems is also related to regional vegetation cover status, groundwater, and other factors [32], which introduces a degree of inaccuracy into the simulated water yield results. The validation of the water yield function of the InVEST model requires the accurate analysis of a large amount of measured hydrological station data; however, owing to the limited available data, the literature review method was finally adopted to estimate and validate the model. In future studies, runoff data from actual hydrological stations should be analyzed as much as possible to verify the simulation results.

In addition, this study introduces the GD and GWR methods to reveal the response of human activities and climate change to water conservation services in Jiangxi Province. Because of the influence of too many driving factors of water conservation, this study could not include all the factors; therefore, it only selected the variables that scholars often pay attention to in the existing research. The exploration of the driving factors in this study mainly focused on the overall impact over the study period, and an exploration of the driving factors for each year has not been conducted. The main reason for this is that grasping the overall trend can already be an effective way to formulate conservation strategies, and realizing real-time drivers of spatiotemporal changes in water conservation may be difficult because of technical barriers in existing research methodologies. Based on global climate change and the dramatic expansion of human activities, it is recommended to make effective use of advanced technologies, such as big data and artificial intelligence (machine learning), to accurately grasp the impact of real-time detection of fluctuations in drivers on water conservation services, prejudge the spatiotemporal trends of water conservation services, and propose safeguard measures.

6. Conclusions

Studies on the spatiotemporal evolution and drivers of water conservation in Jiangxi Province indicate the importance of regulating human activities and promoting rational water resource allocation and ecosystem protection. Therefore, this study first assessed water conservation services and their spatiotemporal variability in Jiangxi Province based on the water yield module of the InVEST model over a continuous time series from 2001 to 2020. Based on this, trends in water conservation were analyzed using the Theil–Sen median trend analysis and Mann–Kendall test. Finally, a combination of GD and GWR was used to explore the drivers of the spatiotemporal evolution of water conservation services and the spatial variation in their effects. The results show the following:

(1) From 2001 to 2020, the average annual depth of water yield in Jiangxi Province fluctuated between 548.87 mm and 1397.94 mm, and during the same period, the depth of water conservation fluctuated between 61.55 mm and 155.24 mm. Moreover, consistent change trends were observed, with an initial increase followed by a decline. The spatial distribution of water conservation services in Jiangxi Province changed significantly and showed a spatial pattern of “low in the middle and north and high in the periphery and south.”

(2) From 2001 to 2020, the overall water conservation services slightly improved and the change trends in various regions were obvious. The area of improvement was dominated by forestland, which accounted for 79.57% of the total. However, 2.74% of the areas presented serious water conservation degradation over the study period, and they were mainly concentrated in the Poyang Lake Plain area and in the central part of the study area. The degraded areas were consistent with the distribution of construction land and arable land.

(3) Water conservation services were driven by multiple factors, including climate, vegetation, socioeconomics, and land use. Precipitation was the dominant factor underlying temporal changes in water conservation in Jiangxi Province, and it was positively correlated with regional water conservation. Land use was the dominant factor underlying spatial changes in water conservation services and had a significant negative impact on the central and northern parts of the study area. The interaction between precipitation and forestland had a greater effect on the spatial variability of water conservation than any individual factor and was the main reason for the slight improvement in regional water conservation. The rapid expansion of cropland and construction land had a huge impact on water conservation in the central part of the study area.