Spatial Analysis of Water Conservation and Its Driving Factors in an Urban Citarum Tropical Watershed: Geospatial Approach

Abstract

Water conservation (WC) is a vital ecosystem service (ES) that plays an essential role in the sustainable management of water resources and ensures ecological security. This research examines the WC capacity of the Citarum watershed in West Java Province, Indonesia, from 2010 to 2020. The specific objectives of this research are as follows: (1) to assess the 10-year temporal and spatial variations of WC using the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) and topographic index model; (2) to analyze the temporal interchange between WC and its influencing factors through the Spatial Elastic Coefficient Trajectory Model (SECTM); and (3) to determine the driving factors (DFs) affecting WC by applying Multiscale Geographically Weighted Regression (MGWR). Key findings of this study reveal significant trends in WC from 2010 to 2020: the average WC in the Citarum watershed decreased from 513.96 mm/ha to 453.64 mm/ha (11.74%), indicating a concerning decline in ES capacity. This study also noted that regions implementing comprehensive regulations dominated the study area, covering approximately 72.70% of the total area (502,295 hectares). This illustrates that the implementation of rules plays a crucial role in the WC. Lastly, the MGWR analysis found that WC in the Citarum watershed positively correlated with topography, climate, and vegetation while negatively correlating with socioeconomic factors. This indicates that WC levels are generally lower in areas with higher human activity and economic growth, highlighting the impact of anthropogenic pressures on natural resources. This framework helps stakeholders plan to ensure sustainable development in the area, as it provides valuable insights into the interactions between the ecological and socioeconomic factors affecting WC.

1. Introduction

Water conservation (WC) is a vital regulatory service of watershed ecosystems that reflects their capacity to store and regulate water resources. This indicates watershed efficiency in managing precipitation, controlling surface runoff, and preserving water quality [1]. This function is shaped by environmental factors such as soil, vegetation, topography, and climate [2]. The role of WC is critical for sustaining ecosystem water needs and supporting downstream water supply, making it indispensable for sustainable water resource management [3]. Researchers have emphasized the importance of WC among ecosystem services (ESs) due to its direct implications for ecosystem health and functionality [4].

WC encompasses interception, infiltration, and precipitation retention and functions across spatial and temporal scales [5]. It involves ecological components like forest canopies, soil, lakes, and reservoirs, which collectively influence water availability [6]. Additionally, WC is essential for water resource planning, hydropower development, and key sectors such as agriculture and ecological conservation [7]. However, climate change and land-use dynamics significantly impact WC by altering hydrological cycles and affecting water yield (WY) and soil moisture content [8]. Changes in precipitation and evapotranspiration further modify WC capacity [9], necessitating the development of effective management strategies [10]. In addition to ecological and climate factors, WC is also significantly influenced by socioeconomic factors. Socioeconomic factors relate ecosystems to people and determine the supply and demand for ESs [11]. Population density and Gross Domestic Product (GDP) are two important socioeconomic factors that indicate the level of urban development and control WC dynamics.

Hydrological modeling is essential for assessing WC function, with tools such as the Artificial Intelligence for Ecosystem Services (ARIES), Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST), and Soil and Water Assessment Tool (SWAT) widely used for this purpose [12,13]. However, in this study, we applied the InVEST model because it is more widely used and has proven especially suitable for valuing ESs, including WY, nutrient retention, crop pollination, and sediment retention in areas with limited available data [14]. InVEST also prioritizes identifying areas where natural capital investments enhance WY and reduce runoff [13], using simplified calculations for rapid assessments [15]. Few studies have employed the InVEST WY model and topographic index methods for WC analyses.

Decoupling indicators, such as the Spatial Elastic Coefficient Trajectory Model (SECTM), analyze WC dynamics by assessing spatial variations in response to environmental changes [16]. The SECTM assesses water resource sensitivity to land-use changes and climate variability. SECTM helps quantify water availability elasticities, predict future trends, and optimize sustainable management strategies with an emphasis on analyzing spatial relationships and long-term dynamics [16]. Meanwhile, the MGWR examines spatially varying relationships at different scales, aiding the WC by identifying localized impacts, detecting water stress or pollution hotspots, and supporting region-specific policy decisions [17,18].

These models—InVEST for ESs valuation, SECTM for elasticity analysis, and MGWR for spatial variability—offer complementary approaches to address complex challenges in water conservation. Together, these enable integrated strategies for sustainable resource management under diverse environmental conditions.

Globally, water scarcity is a pressing issue in Indonesia, where limited water availability affects both the industrial and agricultural sectors. Research on WC in Indonesia has explored participatory micro-watershed management and the adaptability of native plants to local conditions [19]. However, studies linking WC to ESs indicators such as WY remain limited. This study addresses this gap by analyzing WC changes from 2010 to 2020 based on ES metrics. These models complement traditional hydrological tools, such as SWAT, by emphasizing spatial and temporal variability. The SECTM focuses on sensitivity analysis and regulation zones with visually represented areas of varying strengths.

Meanwhile, the MGWR provides insights into localized drivers across varying spatial scales. Integrating the InVEST model with SECTM and MGWR holds significant value, as it enables more accurate mapping of WC impacts while also providing deeper insights into the spatial and temporal relationships between environmental and hydrological variables. Additionally, this approach facilitates a systematic examination of the factors influencing WC, from the strongest to the weakest.

The objectives of this study were to (1) examine changes in WC from 2010 to 2020 using the InVEST and topographic index model, (2) investigate the interaction between WC and its DFs over time using the SECTM, and (3) identify factors influencing WC through the MGWR. The findings of this study are expected to have practical implications for policymakers in formulating sustainable WC plans. This framework also provides valuable insights into the interactions between ecological and socioeconomic factors affecting WC.

2. Materials and Methods

2.1. Study Area

This study was conducted in the Citarum watershed, Indonesia, which covers approximately 690,916 ha across eight regencies in West Java Province. The spatial extent and topographic characteristics of the watershed are illustrated in Figure 1.

Geographically, it lies between 7°15′3″ and 5°54′11″ S latitudes and between 106°56′31″ and 107°58′33″ E longitudes. The climate in this region features three months of minimal precipitation and an average annual precipitation of 2358 mm. The Citarum River, vital for providing electricity, supporting agriculture, and supplying fresh water to communities in various West Java districts, runs through this area and is managed by three significant dams: Cirata, Jatiluhur, and Saguling.

The study area is subject to significant anthropogenic and industrial pressures that critically affect its water conservation capacity. Rapid urbanization and economic growth have substantially increased industrial water consumption, particularly in the textile manufacturing sector. As one of the region’s largest water consumers, the textile industry requires considerable water for dyeing, finishing, and cleaning processes [20]. Other major contributors to industrial water demand include the chemical processing and food and beverage sectors [21].

The terrain in this area varies significantly, featuring hills, volcanic formations, and slopes ranging from 5% to 90%. Upstream mountains reach altitudes of around 750 to 2300 m above sea level (m asl), while plains display mild volcanic relief features [22,23]. The Citarum watershed is geomorphologically classified into three distinct zones: upstream, middle, and downstream zones. It encompasses a vast basin with elevations ranging between 625 and 2600 m asl. The region’s geological composition predominantly comprises lapilli, breccia, lava, and tuff. The climate in this watershed exhibits an average minimum temperature of 15.3 °C, accompanied by annual precipitation of approximately 4000 mm. The soil types found in the Citarum watershed are Latosol (35.7%), Andosol (30.76%), Alluvial (24.75%), Red Yellow Podzolic (7.72%), and Regosol (0.86%) [24].

2.2. Data Sources

This study utilized various data types, including remote sensing and secondary data, as presented in

Table 1. The dataset, processing, and source for WC.

Parameter/Data	Description	Data and Value	Source
Water Yield (mm)	Map of average annual WY	Average Annual WY (mm) Raster	Published on [25]
Precipitation (mm)	Map of average annual precipitation	Average Annual Precipitation (mm) Raster	Published on [25]
Reference Evapotranspiration (mm)	The quantity of water that transitions from land to the atmosphere over a specific period is determined by the combined processes of evaporation—from soil, water bodies, and various surfaces—and transpiration through vegetation.	Global Potential Average Evapotranspiration (mm) Raster	Terra Modis Yearly L4 Global (https://earthexplorer.usgs.gov/ (accessed on 12 March 2024))
Land Use/Land Cover (LULC): 2010 and 2020	Defines the physical features of the land and/or the way individuals use it.	Raster–Coded Land Use/Land Cover	Landsat-OLI analysis results
Boundary shapefile (watershed)	Map of watershed boundaries	Integer (ws_id) from one to n. Vector file (.shp)	Ministry of Environment and Forestry, Republic of Indonesia
Biophysical Table		CSV File (Values assigned per Land Use/Land Cover type)
Coefficient of velocity	Coefficient of velocity	Flow velocity coefficient	Refer [16] modification
Elevation	Digital elevation Model	Used to calculate percentage slope, topography index,	Aster DEM-USGS
Soil depth	Soil depth (mm).	A comprehensive soil characteristics dataset	(http://globalchange.bnu.edu.cn/research/soil2 (accessed on 15 March 2024))
Soil saturation hydraulic conductivity	Global soil saturation hydraulic conductivity	Clipping, for calculation of WC	Future Water (http://www.futurewater.eu/hihydrosoil (accessed on 20 March 2024))
Topographic factors
Slope (Variable name: X1)	Shuttle Radar Topography Mission (SRTM) Global (USGS)	Calculation from DEM (X2)	30 m × 30 m
Digital Elevation Model (DEM) (Variable name: X2)	Shuttle Radar Topography Mission (SRTM) Global (USGS)	-	30 m × 30 m
Climate
Annual average temperature (TEM) (Variable name: X3)	Meteorological, climatological and Geophysical Agency	Numerical and tabular data, including geographic coordinates, were analyzed using the spline interpolation technique	30 m × 30 m
Annual average precipitation, PRE (Variable name: X4)	Meteorological, climatological and Geophysical Agency		30 m × 30 m
Vegetation Factor
Net Primary Production (NPP) (Variable name: X5)	MODIS -MOD17A3HGF V6.1 product in 2000, 2010, and 2020 (https://www.usgs.gov/ and Google Earth Engine (accessed on 20 April 2024))	The sum of all 8-day Net Photosynthesis is the difference between Gross Primary Productivity and Maintenance Respiration. Resampling from 500 m × 500 m	30 m × 30 m
Fractional Vegetation Cover (FVC) (Variable name: X6)	MODIS–MOD13Q1 with NDVI value in 2000, 2010, and 2020 (https://www.usgs.gov/ and Google Earth Engine (accessed on 25 April 2024))	FVC = ((NDVI − 0.2)/0.3) × 100. Resampling from 250 m × 250 m	30 m × 30 m
Socioeconomic
Income per capita (X7)	West Java Central Statistics Agency 2010 & 2020	Numerical and tabular data	30 m × 30 m
Population Density (X8)	West Java Central Statistics Agency 2010 & 2020	Numerical and tabular data	30 m × 30 m

. Several software tools were used for data processing and analysis, including ArcGIS 10.8, InVEST 3.14.1, MGWR 2,2, and Python 3.8, to assess WC and its driving factors comprehensively. ArcGIS 10.8 was used to manage, manipulate, and visualize geospatial datasets, facilitating data conversion, reclassification, overlay analysis, and generating thematic maps. It also supports the integration of various spatial layers, streamlining the preprocessing and postprocessing stages required for spatial modeling.

The InVEST model served as the core spatial analysis tool for estimating WY by integrating key biophysical inputs, such as precipitation, evapotranspiration, soil properties, land use/land cover (LULC), and vegetation characteristics. This model enabled the spatial quantification of water-related ESs and the identification of priority areas for conservation.

MGWR was applied to capture the spatially varying relationships between WC outcomes and potential explanatory variables such as topography, land management practices, and socioeconomic indicators. MGWR provided nuanced insights into each factor’s localized influence, allowing the coefficients to vary across space and scale.

Python programming was used to implement the SECTM, which focuses on analyzing the sensitivity of water conservation to changes in its driving variables and delineating regulation zones for targeted interventions. The model focuses on analyzing the sensitivity of water conservation to changes in its driving variables and delineating regulation zones for targeted interventions.

However, one key aspect of the data above is that all datasets were resampled to a uniform resolution of 30 m × 30 m despite originating from various resolution scales. This standardization ensures consistency across spatial analyses, facilitating comparative assessments and integrated modeling of the data. By harmonizing data resolution, this study enhances the compatibility of diverse datasets, allowing for a more cohesive evaluation of water conservation dynamics and ecosystem service indicators.

2.3. Research Methodology

The results of the InVEST WY sub-model, namely WY, were input to calculate WC using a topographic index. Furthermore, the SECTM was used to determine sensitivity and regulation zones based on (1) precipitation, (2) evapotranspiration, and (3) land-use data. Meanwhile, MGWR determines local driving factors (socioeconomic, topography, climate, and vegetation) at various spatial scales. The integration of SECTM and MGWR advances the accuracy of WC strategies by addressing gaps in spatiotemporal dynamics and heterogeneity analysis. The study was conducted in four stages, as shown in Figure 2.

2.3.1. Data Preparation

The first stage is to collect input data for WC analysis (water yield, DEM, and KSat). In addition, input data were gathered for water conservation drivers related to sensitivity and regulation analysis zones (SECTM) and for local drivers influencing water conservation across spatial heterogeneity (MGWR). In the SECTM, WC capacity is influenced by three key factors: precipitation (G₁), potential evapotranspiration (G₂), and land use/land cover (LULC) (G₃). The MGWR model identifies eight DFs affecting WC changes, categorized into four groups: (1) Topography, represented by slope (SLO, X₁) and elevation (ELE, X₂). (2) Climate—including changes in annual average temperature (ΔTemp, X₃) and changes in annual average precipitation (ΔPrecip, X₄). (3) Vegetation, encompassing changes in net primary production (ΔNPP, X₅) and changes in fractional vegetation cover (ΔFVC, X₆). (4) Socioeconomic—considering changes in per capita income (ΔGDRI, X₇) and population density (Pop, X₈).

All input data were uniformly projected to WGS84, and all vectors were converted into 30-m spatial resolution raster data to solve compatibility issues in geospatial analyses and to minimize spatial differences between datasets.

2.3.2. Calculating WC

To calculate WC using the topographic index model, the first step was to determine WY using the InVEST model. The calculations were based on several data, including precipitation, LULC, soil, climate, Z coefficient, and PAWC (Plant Available Water Content) biophysical data. WY calculations for the study area have been performed by previous researchers [25].

WC volume is determined by considering the WY from a previous study [25], topographic features, soil permeability, and surface runoff coefficients [1]. This volume shows how much water seeps into the earth when surface runoff, evapotranspiration, and precipitation are considered. WC was calculated using Equation (1) [26].

$$WC=min\left(1,{\displaystyle \frac{249}{V}}\right) \times min\left(1,{\displaystyle \frac{0.9TI}{3}}\right) \times min\left(1,{\displaystyle \frac{k_{sat}}{300}}\right) \times WY$$

(1)

where: WC is water conservation, i.e., the volume of water conserved (mm), V represents the coefficient of velocity, k_sat denotes the saturated hydraulic conductivity of the soil (mm/day), WY is the annual water yield (mm), and TI is the topographic index (dimensionless). TI was computed using Equation (2) [26].

$$TI=\lg {\displaystyle \frac{Drainage\_area}{Soil\_depth \times Percent\_slope}}$$

(2)

where: $Drainage\_area$ represents the number of grids in the catchment area, $Soil\_depth$ is the soil depth (mm), and $Percent\_slope$ is the percentage slope (%).

The ratio of WC or retention to precipitation is known as the WC coefficient (WCC). The WCC indicates the proportion of precipitation converted into conserved water. It depicts the effects of vegetation, precipitation, and surface features on WC [27]. The WCC values range from 0 to 1, and higher values of WCC indicate a stronger ability of the ecosystem to intercept and store precipitation, demonstrating a higher potential for WC. The WCC was calculated using Equation (3) [28].

$$WCC={\displaystyle \frac{WC }{P}}$$

(3)

where WCC is the water conservation coefficient, WC is the water conservation volume (mm), and P is the precipitation (mm). In this study, WC calculations were performed using Python software presented in Supplementary S1.

2.3.3. The Trajectory of Spatial Elasticity

SECTM was used in this research to analyze WC due to its ability to dynamically assess the responsiveness of WC strategies over time and space. Unlike traditional static models, this approach allows for a nuanced understanding of how various factors, such as precipitation and land use, influence WC efforts. Chen et al. [16]. highlight the effectiveness of this model in capturing spatial and temporal heterogeneity in WC, demonstrating its capability to provide insights that are critical for effective policymaking in urban agglomerations Beijing-Tianjin-Hebei Furthermore, the model’s strength lies in its ability to differentiate between intra- and inter-regional effects, which is essential for understanding the local and regional interactions that impact WC. This is particularly relevant in contexts where water resources are shared across regions, as noted by Fang et al. [29] who discussed the economic implications of water reallocation and its effects on sectoral growth. By quantifying both direct impacts, such as precipitation, and mediated pathways, like vegetation’s role in water retention, the SECTM provides actionable insights for policymakers aiming to enhance WC strategies. Additionally, the model facilitates predictive simulations under various policy scenarios, which is crucial for developing adaptive management strategies. This predictive capability is underscored by Yu et al. [30], who emphasized the importance of understanding the temporal and spatial evolution of WY for effective water resource management. The model’s robustness in accounting for non-stationarity and scale effects further enhances its applicability in diverse and large basins, making it a valuable tool for regional water resource management.

The elasticity coefficient of WC and its temporal DFs were calculated for a 10-year period (2010–2020) based on the decoupling concept using Equation (4) [30].

$$n_i={\displaystyle \frac{u_{wy}}{u_p}}={\displaystyle \frac{{(WC}_{t1}{-WC}_{t0})/{ WC}_{t0} }{{(T}_{t1}{-T}_{t0})/T_{t0}}}$$

(4)

where n_i represents the elasticity coefficient for WC and temporal DFs during the ith year (i ranges from 2010 to 2020). WC_t and WC_t0 denote WC from t₀ to t₁ years (from 2010 to 2020), while T_t1 and T_t0 refer to the temporal DFs over the same period. u_wy and u_p express the change rates of the WC and temporal DFs, respectively.

A Spatial Elastic Coefficient Trajectory Model for WC capacity was developed by combining decoupling indicators and logical coding. The model incorporates three DFs: precipitation, potential evapotranspiration, and LULC type, highlighting the sensitivity of WC capacity to these variables. The logical coding calculation was performed using Equation (5) [16], as follows:

$$T_{ij}={{\left(G1\right)}_{ij} \times {10}^{n-1}+{\left(G2\right)}_{ij} \times {10}^{n-2}+\dots +\left(Gn\right)}_{ij} \times {10}^{n-n}$$

(5)

where T_ij is the logical coding value for the cell at column j and row i in the raster image generated from the logical analysis. The variable n denotes the number of temporal driving factors, which is three in this study. Thus, n = 3, and G1_ij, G2_ij, and G3_ij represent the logical coding for the respective raster types related to precipitation (G₁), potential evapotranspiration (G₂), and LULC (G₃) from 2010 to 2020. A spatial elastic coefficient trajectory code was developed using reclassification and grid calculation techniques. Following the classification results from [30] and considering the natural environmental characteristics of the study area, the calculated elastic coefficients were categorized into three levels: low, medium, and high (1, 2, and 3). A higher elastic coefficient signifies a diminished ability to manage variations in DFs, thus making these areas more sensitive to such variations.

2.3.4. MGWR Model

MGWR simultaneously facilitates the analysis of relationships at multiple spatial scales, which is particularly beneficial in complex environments where different processes may operate at different scales. MGWR often results in better model fit and predictive accuracy compared to traditional regression models, which is crucial in fields like environmental science, where spatial relationships are critical [16]. Lastly, the MGWR provides greater flexibility in model specification, allowing for the inclusion of different predictors at varying scales, which is lacking in traditional models. This flexibility is underscored by the work of [31], who described how the MGWR can explore the spatial heterogeneity of influencing factors, thus enhancing the understanding of complex spatial relationships. These improvements make the MGWR a powerful tool for understanding complex spatial relationships across various fields, including environmental studies, urban planning, and social sciences.

Except for topography factors, all variables were based on changes observed during the study period from 2010 to 2020, calculated as the difference in the WC. The average values of the explanatory districts were calculated using the Zonal Statistics tool in ArcGIS 10.8. Principal Component Analysis (PCA) was applied as a crucial method for reducing data dimensionality. Through PCA, the various DFs for each principal component were identified, with their weightings based on the variance contributions of each component. The comprehensive score for each principal component was calculated to select four primary factors [32].

Geographic detectors are useful for identifying spatial variations and uncovering potential influencing factors, highlighting both similarities within regions and differences between them [33]. Our study primarily examines the interaction and factor detectors within the geographic detector model. The q-value shows how well the independent variable clarifies the variation in the dependent variable and is computed using Equation (6) [34].

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

(6)

where: q denotes the explanatory power of independent variables, which encompass socioeconomic, topography, climate, and vegetation factors concerning the dependent variable, representing various forms of WC on a scale from 0 to 1, h represents the classification or stratification of variables (h = 1, 2, 3, …, L), $N_h$ and ${\sigma }_h^2$ correspond to the sample count and variance within layer h, respectively, while N and ${\sigma }^2$ indicate the total sample size and variance.

MGWR captures local-scale relationships between different variable types while effectively minimizing the errors associated with spatial variability. In this study, the bandwidth selection process follows the gold section search method and utilizes the Gaussian model, with the optimal bandwidth determined based on the corrected Akaike Information Criterion (AICc) [35]. The MGWR model was applied to assess the spatial association between the dependent and explanatory variables. This model incorporates adaptive bandwidth selection, identifies critical influencing factors across varying spatial scales, and is recognized as an advanced form of Geographically Weighted Regression by using Equation (7) [36].

$$y_i={\beta }_0\left({\mu }_i,{\gamma }_i\right)x_{ij}+{\displaystyle \sum _{j=1}^m}{\beta }_{bwj}\left({\mu }_i,{\gamma }_i\right)x_{ij}+{ϵ}_i, i \mathrm{€}\left\{\mathrm{1,2}, \dots , n\right.\}$$

(7)

where: y_i represents the dependent variable of the ith district; x_ij denotes the ith independent variable of the jth district; β₀ $\left({\mu }_i,{\gamma }_i\right)$ is the intercept; β_bwj$\left({\mu }_i,{\gamma }_i\right)$ represents the regression coefficient after correction of the bandwidth for the jth DF in the ith district, and bwj in β_bwj indicates the bandwidth used to calibrate the jth DF.

All model calibrations were conducted using the MGWR 2.2 software. For a more detailed explanation of the MGWR modeling process, refer to the works of [37,38]. Model performance was assessed using metrics such as [39]: Adjusted R²: Indicates the proportion of variance explained by the model while accounting for the number of predictors AICc: lower values indicate better model fit. q-values: assess the statistical significance of spatial relationships.

3. Results

3.1. Analysis of WC

The WC obtained in this study is presented in Figure 3 and

Table 2. Changes in WC Across Sub-Watersheds in the Citarum watershed (2010–2020).

Name Sub Watershed	2010		2020		Change 2010–2020
	Mean (mm)	Total (108 m3)	Mean (mm)	Total (108 m3)	(108 m3)	%
Upstream	397.42	10.6	368.17	9.86	−0.74	−6.98113
Middle	631.22	17.37	548.16	15.12	−2.25	−12.9534
Downstream	508.07	10.3	437.87	8.86	−1.44	−13.9806
Citarum	513.96	38.26	453.64	33.84	−4.42	−11.5525

, based on the calculation using Equation (1). This table represents the total amount of WC in different sub-watersheds within the Citarum watershed in 2010 and 2020. The WC patterns in 2010 and 2020 were relatively similar. In 2010, the average WC ranged from 397.42 to 631.22 mm/ha, with the highest levels in the central sub-basin and the lowest in the upstream sub-basin. By 2020, there was a decline in the WC. The average WC in the Citarum watershed from 2010 to 2020 decreased by 11.55%, from 513.96 mm/ha to 453.64 mm/ha. All sub-watersheds experienced a decline in WC from 2010 to 2020. These results were also reflected in a study conducted by [40], in which four catchment areas in the Citarum watershed experienced a decrease in WC between 2010 and 2020. The downstream sub-watershed saw the largest percentage reduction (−13.98%), while the upstream sub-watershed saw the smallest decrease (−6.98%). The overall Citarum watershed experienced an 11.55% reduction in the total WC volume over the decade, reflecting environmental or land-use challenges affecting water retention.

The WC coefficient was calculated using Equation (3) and is shown in Figure 4. Additionally, using the zonal static WC coefficient with the LULC map, we derived the WC coefficient presented in

Table 3. Average WC Based on LULC.

No	LULC	2010		2020		Change 2010–2020
		mm	%	mm	%	Mm	%
1	Virgin Forest	1145.54	0.50	1060.19	0.48	−85.35	(7.45)
2	Plantation Forest	913.95	0.44	648.69	0.31	−265.26	(29.02)
3	Shrub	295.78	0.13	263.34	0.13	−32.44	(10.97)
4	Estate Crop Plantation	721.75	0.35	680.36	0.31	−41.39	(5.73)
5	Settlement Area	846.19	0.28	741.76	0.20	−104.43	(12.34)
6	Bare Land	366.52	0.21	270.77	0.13	−95.75	(26.12)
7	Lake	84.74	0.04	65.34	0.02	−19.40	(22.89)
8	Dry Farming	902.66	0.41	784.67	0.37	−117.99	(13.07)
9	Paddy Field	742.97	0.43	623.19	0.36	−119.78	(16.12)
10	Fishpond	158.82	0.08	139.54	0.05	−19.28	(12.14)
11	Airport	370.58	0.15	280.96	0.19	−89.62	(24.18)

. The table indicates an average decline in WC across all LULC types, with reductions ranging from 5.73% in estate crop plantations to 29.02% in plantation forests. The WCC values ranged from 0.08 in fishpond areas to 0.5 in virgin forests. Overall, LULC types with more vegetation tended to have higher WCC values than open areas such as bare land and settlements. Table 3 shows the downward trend in WC for most LULC types from 2010 to 2020. The lack of effectiveness of some land cover types in retaining water is potentially due to deforestation, urbanization, and land-use change. Thus, sustainable land and water management strategies are necessary to maintain the ES.

Decreased WC values were observed in several land cover types, such as shrublands, settlements, and bare land, with values of 10.97%, 12.34% (related to urban expansion), and 26.12% (presumably due to soil erosion), respectively. In addition, agriculture, including dry and paddy fields, showed a significant decrease of 13.07% and 16.12%, respectively, indicating that agricultural practices can harm WC. Indications of environmental degradation or a decrease in water levels in aquatic systems are characterized by a reduction in water conservation in lakes and fishponds of 22.89% and 12.14%. In addition, on built-up land such as the airport, there was a decrease in WC of 24.18%, allegedly caused by infrastructure expansion or land-use change.

Figure 4 shows a consistent pattern of WC coefficient values in 2010 and 2020, with range values of 0–0.68. The results indicate that the proportion of rainwater stored as WC in the Citarum watershed area varies from 0% to 68%. In 2020, the WC coefficient in the downstream sub-basin area of the Citarum watershed dropped significantly. Meanwhile, the coefficients in the other sub-basin areas remained relatively stable. In detail, virgin forests experienced a decrease in WC of as much as 85.35 mm, with a 7.45% decline between 2010 and 2020, likely due to deforestation or land degradation. Plantation forests saw a more substantial drop of 265.26 mm (29.02%), reflecting their reduced WC capacity, potentially caused by land conversion or management changes.

3.2. The Temporal Interaction Between WCs

Based on Equations (4) and (5), the grid calculation analysis produced a spatial elastic coefficient trajectory table, which is displayed in

Table 4. Trajectory of Spatial Elasticity Coefficient WC in Citarum watershed.

Regulation Area	Code *	Area
		ha	%
Strong PREC regulation	122, 123, 132, 133	8636.45	1.25
Strong PEVA Regulation	212, 213, 312, 313	-	0
Strong LULC regulation	221, 321, 331, 231	179,983.62	26.05
Strong COMP regulation	111, 112, 113, 121, 131, 211, 311	502,295.93	72.7
Weaks regulation area	222, 223, 232, 233, 322, 323, 332, 333	-	0
No data		-	0

Note: PREC: Precipitation, PEVA, potential evapotranspiration; LULC: Land use Land cover; COMP, comprehensive. * The meaning of “Code” is the identity of an area as either a critical zone for conservation interventions or initiatives. By utilizing the decoupling indicator introduced by the OECD, the model also provides insights into how changes in driving factors affect the dependent variable over time, thus providing a clearer understanding of temporal dynamics [16].

and illustrated in Figure 5. Table 4 outlines the distribution of land regulated by different regulations, showing that comprehensive and land-use regulations dominate the regulated area. The Citarum watershed study area is divided into 27 codes. This area is primarily characterized by strong comprehensive regulations covering 502,295 ha (72.70%). In addition, there are areas under strong land-use regulation, covering 179,983 ha (26.05%), and areas with strong precipitation regulation status, covering 8636.45 ha (1.25%). Most areas, 502,295.93 ha (72.7%), are under strong COMP regulations, reflecting extensive land use and resource management policies.

Strong LULC regulations covered 179,983.62 ha (26.05%), emphasizing land use and cover management. In contrast, strong PREC regulation applies to a much smaller area of 8636.45 ha (1.25%), concentrating on precipitation-related measures. This breakdown shows that most of the Citarum watershed is subject to comprehensive regulations, followed by specific land-use regulations, with little focus on precipitation management. No areas are governed by PEVA regulations or classified as areas with weak regulations, suggesting concentrated efforts on specific aspects of regulations, particularly land and resource management.

3.3. Driving Factors on WC

The correlation values reveal differences in spatial variation, as presented in Figure 6. Overall, the local R² values, which range from 0.50 to 0.95, reflect the explanatory power. The four driving factors analyzed in this study: socioeconomic, topography, climate, and vegetation account for 97.7% of the variations in WC, as shown in Figure 6.

Figure 6 illustrates the relationship between changes in WC and the driving factors in the upstream sub-basin area. A stronger correlation was observed compared to the middle and downstream sub-watershed areas. Conversely, the downstream sub-watershed areas exhibited weaker correlations. The factor detectors were used to rank the relevant factors listed in

Table 5. The q-values of the driving factors influencing changes in WC.

Driving Factors	Socioeconomic	Topography	Climate	Vegetation
Socioeconomic	0.277
Topography	0.496	0.085
Climate	0.497	0.311	0.006
Vegetation	0.527	0.380	0.318	0.114

according to their ability to explain the geographic variation in WC change. The influence of different drivers on WC changes can vary. The highest q-values were found for socioeconomic, topography, climate, and vegetation factors at 0.277, 0.496, 0.497, and 0.527, respectively (Table 5). These values underline the role of these three parameters as major contributors to the spatial variation in WC.

Socioeconomic factors dominated the spatial heterogeneity of WC changes, with q-values of 0.277, followed by vegetation factors at 0.114, topography at 0.0846, and climate at 0.006. The interaction detector results revealed that the combined effects of multiple driving factors on WC are greater than the impact of any individual factor, with varying interactions between factors. The strongest interaction is between socioeconomic and vegetation factors, with a q-value of 0.527. The interaction between socioeconomic and topographical factors has a q-value of 0.496, and the interaction between socioeconomic and climate factors has a q-value of 0.497.

This study applied the MGWR model to examine the spatial influence of multiple factors on variations in WC. The performance of the MGWR model was compared to that of the Ordinary Least Squares (OLS) regression model.

Table 6. Fit metrics for OLS, GWR, and MGWR.

Fit Metrics	Model
	OLS	GWR	MGWR
Fix Model
R2 (adjust)	0.003	0.741	0.977
AICc	470.549	347.870	5.973
Adaptive Model
R2 (adjust)	0.003	0.414	0.565
AICc	470.549	417.120	357.093

presents a comparative analysis of the MGWR, OLS, and Geographically Weighted Regression (GWR) models. The evaluation of the MGWR model with the adaptive bandwidth approach revealed that the four key driving factors considered in this study explained 56.50% of the variation in WC. In contrast, the fixed MGWR model demonstrated a significantly higher explanatory power, accounting for 97.7% of the variation. Furthermore, the fixed MGWR model was employed to maintain a consistent bandwidth across all driving factors, where higher regression coefficients indicate a stronger influence of these factors on WC dynamics.

Higher adjusted R² values suggest stronger explanatory power and better model fit; nevertheless, lower Akaike Information Criterion (AICc) values suggest a more accurate and dependable regression estimate [35]. The MGWR model provides a more accurate depiction of spatial phenomena than the OLS and GWR models. The MGWR model was used to investigate the spatial correlation between the driving forces and WC changes. The correlation coefficient for each driving factor illustrates how changes in WC respond spatially, as presented in Figure 7.

The mean MGWR coefficients for the relationship between WC and its DFs are as follows: socioeconomic (−0.087), topography (0.047), climate (0.002), and vegetation (0.180). Between 2010 and 2020, variations in WC within the Citarum watershed exhibited a negative correlation with socioeconomic factors (Figure 7a). In contrast, WC was positively correlated with topographic (Figure 7b), climatic (Figure 7c), and vegetation factors (Figure 7d). The correlation values highlight spatial variations, as illustrated in Figure 7b. The local R² values, ranging from 0.50 to 0.95, indicate the explanatory power of the MGWR model. The four key driving factors, namely, socioeconomic, topography, climate, and vegetation, collectively explain 97.7% of the observed variations in WC, as depicted in Figure 7b.

As illustrated in Figure 7b, the relationship between changes in WC and DFs exhibits a stronger correlation in the upstream sub-watershed than in the middle and downstream sub-watershed areas. By contrast, the downstream region demonstrates a weaker correlation.

4. Discussion

4.1. Impact of Change in Comprehensive Factors on WC

Changes in WC are significantly influenced by the interaction between multiple factors, as evident in Java, where watershed conditions and community engagement levels vary widely. A national policy promoting Integrated Water Resources Conservation Management emphasizes stakeholder participation and local wisdom to enhance water strategies across diverse regions [41]. This aligns with research from Jilin Province, China, which demonstrates that combined factors have a more significant impact on WC than single factors, often resulting in a non-linear increase [42]. Similar findings from Jiangxi Province highlight that climatic and ecological changes together exert a more substantial influence than isolated factors [43].

WC is shaped by spatial and temporal heterogeneity and is influenced by soil-saturated hydraulic conductivity, precipitation, and vegetation cover. Research indicates that soil permeability is crucial, with higher hydraulic conductivity improving water retention, particularly in mountainous regions like Yanshan and Taihang in northern China, where precipitation is higher than in plains [44]. Variability in soil properties affects water budgets and impacts conservation efforts [45]. Additionally, climate and land-use changes are the primary drivers of temporal fluctuations in WC, with precipitation and evapotranspiration playing key roles [46]. Urbanization further influences WC by altering land-use patterns [47].

The interaction pattern of comprehensive regulation in the Citarum watershed aligns with the findings from the Han and Geum Rivers in South Korea, where precipitation, evapotranspiration, and land use influence WC more when considered together rather than individually. Thus, effective policy development must integrate these interdependencies to enhance regional water resource management [48]. While these findings support the necessity of an integrated management approach, they also highlight the importance of localized strategies that account for specific environmental and socioeconomic conditions. Adaptive management practices tailored to regional characteristics are crucial for sustainable water conservation [49].

4.2. Impact of Climate and Land-Use Changes on WC

This study assessed the WC using the WY approach, in which the balance between precipitation and evapotranspiration determines the conservation capacity. The findings confirm that climate variability significantly impacts WC, as observed in the Citarum watershed from 2010 to 2020. A decline in precipitation over the decade corresponded with fluctuations in WC, reinforcing the regulatory roles of precipitation and evapotranspiration in hydrological dynamics [50]. The relationship between climate and vegetation was also significant, with a q-value of 0.318, highlighting the role of vegetation cover in mediating WC.

WC dynamics in the Citarum watershed reflect the interactions among climate change, land-use modifications, and socioeconomic factors. Climate change affects precipitation and evapotranspiration, while land-use changes alter hydrological processes and local water demand. Future climate projections indicate rising temperatures and shifts in precipitation, potentially exacerbating water scarcity [51]. Research in the Three-Rivers Headwater Region shows that precipitation influences vegetation shifts [34], similar to the findings in the Sundarbans, where climatic and tidal factors shape WC strategies [52].

Land-use changes contribute to decreased annual runoff and base flow, increased erosion, and affect water quality [53]. In the Maying River Basin, land-use changes significantly reduced runoff and base flow, demonstrating their critical role in hydrology [54]. While land use influences WC, climate factors remain the dominant determinants [50].

Precipitation and temperature shape WC, as observed in eco-geographical zones where vegetation interactions play a role [55,56]. Soil properties, particularly hydraulic conductivity, affect water retention [57]. Research in the Three-Rivers Source Area indicates that climate change-induced fluctuations in precipitation and evapotranspiration drive WC disparities [58]. In the Yiluo River Basin, WC declined by 47% from 1966 to 2018 due to meteorological conditions [56], while in Tamil Nadu, a decrease in reference evapotranspiration further highlighted the climate impacts on water resources [59].

WC and land use have a complex relationship. Different land-use types vary in WC capacity due to differences in vegetation structure, root depth, and evapotranspiration rates [60]. Changes in land use modify soil characteristics, thereby affecting infiltration and evaporation rates [61]. Land-use changes have been shown to decrease annual runoff and base flow, increase erosion, and further affect water quality and availability [53]. Research suggests that under changing climatic conditions, land-use modifications may have diminishing impacts, reinforcing the need for integrated management approaches [50].

Converting agricultural land to forest influences WC. In the Hanjiang River Basin, China, forested areas enhance water retention but limit the water supply [62]. Despite prioritizing climate factors, land-use changes have localized impacts on WC dynamics. Sustainable practices and public awareness are essential for effective water management [63]. Understanding these interactions is key to formulating effective WC strategies in the future.

4.3. Socioeconomic Factors and WC Decline

The study’s findings revealed a negative correlation between WC and socioeconomic variables, indicating that economic growth and human activity contribute to reduced WC capacity. Urbanization, population density, and industrial expansion exacerbate this decline by altering the land-use patterns, increasing surface runoff, and reducing infiltration. Additionally, economic growth intensifies water extraction, further depleting the conservation resources. Deforestation and agricultural expansion to support growing populations diminish vegetation cover, limiting the role of evapotranspiration in water retention. These trends align with previous research, which highlights that socioeconomic development increases water demand [64], disrupts hydrological cycles [65], and ultimately reduces WC capacity [66].

Household-level WC behavior is crucial for conservation outcomes, particularly in urban settings. Environmental awareness, financial constraints, and access to water-saving technologies shape the patterns of water consumption. While pro-environmental attitudes promote conservation, practical barriers, such as affordability and infrastructure, often hinder widespread adoption [67]. Individuals concerned about water scarcity are more likely to engage in conservation practices [68], although psychological and economic factors also influence water-saving intentions [69]. For example, strong subjective norms in South Africa created a gap between knowledge and behavior [70]. Additionally, income levels and pricing schemes significantly impact conservation behaviors, with higher water costs encouraging more sustainable usage [55]. Socio-demographic factors, including age, education, gender, and homeownership, influence WC practices [71].

Households with higher incomes are typically better positioned to invest in water-saving technologies like low-flow toilets, efficient irrigation systems, and water meters. Such investments often result in substantial reductions in water usage and increased water efficiency. For example, studies of metering initiatives in the UK reveal that wealthier households gain the most from lower water bills, largely because they are more capable of adopting water-efficient measures than poorer households. Conversely, lower-income households may struggle to afford these technologies or bear greater financial burdens from pricing strategies. Price-based water conservation efforts tend to impact low-income groups more heavily without effectively limiting discretionary water use among affluent households. This underscores the importance of non-price interventions, such as mandatory usage restrictions, which generally offer a fairer approach across different income levels [72].

Socioeconomic factors drive spatial variations in regional WC by altering ESs. Human activities disrupt ecological processes and influence WC and resource availability [73]. Studies on the Qinghai−Tibet Plateau and China’s national key ecological function zones (NKEFZ) have highlighted GDP and population density as the primary drivers of ES fluctuations [2]. Similarly, the increased GDP in the Nanchang metropolitan area has increased water demand, thereby reducing conservation capacity. In the Xiangjiang River Basin, human-induced changes have resulted in a 4.59% decrease in water supply [74].

Regional WC patterns underscore the need for localized management strategies. The MGWR model demonstrated that socioeconomic factors influence WC differently across spatial scales, offering insights into targeted conservation measures.

4.4. Policy and Management Recommendations

Engaging local communities and stakeholders is crucial for effective WC strategies. Their perspectives provide qualitative insights that complement quantitative data and help shape context-specific conservation strategies. Interviews with farmers, surveys targeting residents and business owners, and focus group discussions with policymakers enhance WC efforts by assessing water availability perceptions, conservation practices, and management challenges [75]. Research highlights that structured educational programs improve community engagement in WC, emphasizing public awareness and stakeholder involvement.

Showcasing successful WC initiatives can serve as models for other regions to follow. Community-led rainwater harvesting projects, for example, have improved water availability and conservation, particularly in rural areas [76]. Effective WC strategies should integrate land-use planning, sustainable resource management, and community participation. Policies must balance economic development and environmental sustainability to mitigate the impacts of urbanization and industrialization [77]. Watershed zoning regulations, including buffer zones and restrictions on urban expansion near water bodies, help reduce the adverse effects of deforestation and impermeable surface expansion on WC [78].

Forest conservation is vital for enhancing infiltration rates, reducing soil erosion, and stabilizing hydrological cycles [79]. Research suggests that natural secondary forests significantly improve WC [80]. Large-scale afforestation and reforestation programs in deforested upstream areas help mitigate WC loss while supporting carbon sequestration [81]. Sustainable urban development should incorporate green infrastructure, such as permeable pavements, rainwater harvesting, and bioretention systems, to enhance water retention and mitigate runoff [82].

Integrated Water Resource Management (IWRM) provides a framework for balancing social, economic, and environmental objectives of water resource management. Watershed-level IWRM strengthens water security through decentralized governance and conservation technologies [83]. It promotes the inclusion of industrial stakeholders alongside domestic, agricultural, and environmental users in basin-wide planning to ensure that water use remains sustainable and pollution loads do not exceed ecological thresholds. Public participation in WC policies is crucial for long-term success, as social perceptions and institutional trust influence conservation behavior [80]. Policymakers should integrate these approaches into the Citarum watershed management frameworks, prioritizing long-term monitoring of WC trends under various land-use and climate scenarios to refine conservation strategies.

Future work should include an overlay analysis of the SECTM model output zones with local DFs identified using the MGWR model. This integration will enable a more comprehensive understanding of how spatial variations in DFs influence WC dynamics at different scales and locations. Additionally, the use of medium-resolution imagery is necessary for future research to improve the accuracy of spatial analysis and enhance the assessment of environmental and hydrological patterns. Although this study was conducted in a tropical watershed, i.e., the Citarum watershed, the methodology used can be applied to other watersheds by adjusting the parameters to suit local conditions.

5. Conclusions

This study integrated the InVEST model, topographic index model, SECTM, and MGWR methods to examine changes in WC and the factors influencing these changes. The analysis considers behavioral aspects, where each driving factor dynamically interacts with the spatial distribution of WC over time and the spatial-temporal variation in the Citarum watershed. In the 2010 WC calculation, the average WC value ranged from 397.42 to 631.22 mm/ha, with the highest values found in the middle sub-watershed. However, by 2020, there was an 11.55% decrease in WC across the entire Citarum watershed, with the largest decrease occurring in the downstream sub-watershed and the smallest in the upstream sub-watershed. The analysis of average WC based on land use and land cover (LULC) type revealed that virgin forests exhibited the highest WC values, followed by Plantation Forest and Dry Farming.

The study of temporal interactions between WC was conducted using the SECTM, which analyzes the relationship between land distribution and various implemented regulations. The results indicate that comprehensive regulation (COMP) and LULC regulations dominate the regulated area, with most of the Citarum watershed subject to COMP regulation, covering 502,295.93 hectares (72.7%), followed by LULC regulation (26.05%), and limited focus on precipitation regulation (PREC). Meanwhile, no area is subjected to PEVA regulations or classified as weakly regulated.

Furthermore, this study analyzed the driving factors of WC, including socioeconomic, topography, climate, and vegetation. The MGWR model was used to rank the relevance of these driving factors. Socioeconomic factors emerged as dominant in the spatial heterogeneity of WC changes, with a q-value of 0.277, followed by vegetation factors (0.114), topography (0.0846), and climate (0.006). The performance of the MGWR model was evaluated by comparing it with Ordinary Least Squares (OLS) and Geographically Weighted Regression (GWR) models. The results demonstrate that the MGWR Model exhibits the highest R2 value (97.7%), followed by GWR (74.1%) and OLS, providing a more accurate depiction of spatial phenomena in MGWR than in OLS and GWR.

This study provides novel insights into the evolution of watershed function and its underlying determinants. This study serves as a robust foundation for the scientific community to conduct comprehensive follow-up research on a more intricate scale, utilizing multiple scenarios to forecast future watershed conditions. Furthermore, this research offers valuable perspectives for stakeholders, enabling them to make informed decisions grounded in comprehensive scientific analyses for sustainable watershed planning in diverse critical watersheds within humid tropical regions.