Water conservation function is one of the most important ecosystem service functions in watersheds and is an important indicator of regional ecosystems (Xu et al., 2022). Research on the water conservation functions started from forest ecosystems and has mainly reflected the role of trees, shrubs, litter, and soil in the redistribution of precipitation (Li et al., 2021). In the context of global water scarcity and rapidly decreasing groundwater reserves, the sustainability of the water conservation has become a core component of regional ecological security assessments (Li et al., 2022c). Wetland water conservation of the delta regions in arid inland river basins has been widely studied for its effects on biodiversity maintenance, soil and water conservation, runoff regulation, climate regulation, and freshwater supplies (Hu et al., 2021; Zhang et al., 2022).

Many studies have been conducted on the water conservation function and its spatial and temporal variations at different scales and in different regions (Wang et al., 2022b; Wu et al., 2022; Wang et al., 2023). Several methods have previously been developed to assess the water conservation function of ecosystems, including the water balance method (Hong et al, 2018), canopy residual interception method (Deng et al., 2002), and precipitation storage method (Wu et al., 2016). However, owing to the obvious spatial and temporal scale characteristics of the water conservation function, the econometric models are not only less adaptable to regional scales but also not suitable for long time series applications in regions with large scales, complex topography, and diverse ecosystem types (Wang et al., 2013). The Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model can be used to integrate these analytical features (Yin et al., 2020). This model is a modelling tool for assessing ecosystem services and their economic values to support ecosystem management and decision-making (Liu et al., 2019). In contrast to other hydrological models, the InVEST model is on the basis of a grid scale running at an average annual time step, with spatiotemporal large-scale modelling based on water balance principles; therefore, it is appropriate for assessing the impacts of land use/land cover change (LUCC) on multiple ecosystem services (e.g., water yield, carbon storage, and habitat quality) (Li et al., 2021).

The earliest application of the InVEST model was in the Amazon Basin, where the model was applied to assess ecosystem services in ecologically functional areas (Tallis and Polasky, 2009). The results of that study had a significant effect on international ecological evaluations and greatly facilitated the application of the InVEST model in other regions. Later, Erik et al. (2009) applied the InVEST model to the Willamette Basin in Oregon, USA and designed three plausible LUCC scenarios to analyze the spatial patterns of ecosystem services, such as hydrological services, soil conservation services, carbon storage, and biodiversity, under the three scenarios. Sánchez-Canales et al. (2012) assessed the function of the water conservation service in the Mediterranean watershed. Since 2013, research on ecosystem services based on the InVEST model has increased rapidly, mainly involving habitat quality (Baral et al., 2014), soil erosion (Qiao et al., 2023), carbon storage (Fan et al., 2023), water yield (Jia et al., 2022), and water conservation (Li et al., 2021). For example, Baral et al. (2014) used the InVEST model to evaluate the biodiversity of north-central Victoria, Australia, and analyzed the impact of LUCC, thus evaluating its applicability in habitat quality assessment and ecological conservation planning. Li et al. (2021) analyzed the effects of soil change, land use, and soil and water conservation on the water conservation function in the Danjiang watershed of the Qinling Mountains, China. From the perspective of different climate types, the spatial and temporal variability of the water conservation under alpine climates, monsoon climates at medium latitudes, and subtropical monsoon climates have been extensively assessed (Hu et al., 2023). For example, Wang et al. (2021a) analyzed the water conservation in the eastern part of the Loess Plateau, China and quantified the environmental drivers of the water conservation change; Xue et al. (2022b) studied the water conservation function and spatial and temporal variations in the water conservation in the alpine region of the Tibetan Plateau, China.

The Ili River is an international river that crosses the border between China and Kazakhstan; it is the largest cross-border river in Xinjiang Uygur Autonomous Region, China in terms of water volume. The Ili River Delta (IRD) with fragile arid ecosystem is located in the lower reaches of the Ili-Balkhash Basin (an arid endorheic basin shared by China and Kazakhstan), Central Asia, and is an important water conservation area in this basin. Water resources play an important role in improving the ecological environment of the region. Many previous studies have focused on analyzing the hydrological, vegetation, and wetland changes in the IRD (Xie et al., 2011; Yao et al., 2022). From the 1840s to the late 1950s, the IRD ecosystem remained relatively stable. Beginning in the 1960s, however, because of regional climate change and increasing human activities upstream of the delta, river runoff into the delta was greatly reduced or even cut off until the 1970s, and the IRD ecosystem continued to deteriorate (Cao et al., 2022). As human activities in the middle reaches of the Ili River Basin have decreased and river inflows to the IRD have increased since the 1990s, the ecological environment of the delta has gradually improved (Xie et al., 2011). Deng et al. (2011) noted that the construction of the Kapchagay Reservoir and the overexploitation of water resources in the middle and lower reaches of the Ili River have caused the decrease of the water level of the Lake Balkhash and the deterioration of the ecology in the IRD. Wang and Lu (2009) suggested that climate change is the dominant factor influencing the dynamics of the water level of the Lake Balkhash, whereas human activities are reinforcing factors impacting the lake water level changes. The inflow to the lake is significantly correlated with the water volume of the Lake Balkhash, which is the leading factor affecting the changes in the water volume of the lake (Wang et al., 2022c). Between 1970 and 1990, the wetland area of the IRD reduced from 2607 to 1841 km² (a reduction of 29.38%), with half of the reduced area converted to grassland and one-third to saline alkali land (Xie et al., 2011). Yao et al. (2022) studied the conversion of large areas of bare land to grassland around the IRD from 2000 to 2020 and found that forests were degraded into grassland within the delta. Runoff into the delta and the water level of the Lake Balkhash are significantly and positively correlated with the wetland area, which are the leading factors driving the wetland evolution (Jin et al., 2016; Cao et al., 2022). Since the 1990s, under the influence of climate change and human activities, not only the water level and lake area of the Lake Balkhash have continued to decline, but also the water overflow capacity, wetland area, and ecosystem services of the IRD have been severely weakened (Xie et al., 2011), resulting in significant changes in the water conservation function of the IRD. However, the spatial and temporal variations in the water conservation function of the IRD and their driving factors are still not clear.

In this study, we used the InVEST model with localized parameters to simulate the water yield of the IRD from 1975 to 2020, and evaluated the water conservation function of the ecosystem visually and quantitatively. The effects of local climate change, LUCC, and inflow from the Ili River on the water conservation function in the IRD were also investigated. The results can provide a basis for making scientific and reasonable decisions regarding the water conservation function in the IRD.

The IRD (74°00′-76°30′E，45°20′-46°15′N) ) is located in southeastern Kazakhstan, Central Asia and covers an area of about 8.00×10³ km² in the southwestern part of the Lake Balkhash (Luo et al., 2012) (Fig. 1). The delta is bordered by the Saryesik-Atyrau Desert in the northeast, the Tawkum Desert in the southwest, and the Lake Balkhash in the northwest. The geomorphological unit is a sedimentary-alluvial plain in the lower reaches of the Ili River, and the overall topography is high in the southeast and low in the northwest (Xie et al., 2011). The delta has an extremely dry continental climate, with little precipitation and high evapotranspiration. The average annual precipitation in the study area is 192 mm, the average annual temperature is 7°C, and the average annual evapotranspiration is about 1000 mm (Zhou et al., 2013; Guo and Xia, 2014). The vegetation here mainly consists of false reed fescue (Calamagrostis pseudophragmites) and reeds (Phragmites communis) (Zheng et al., 2010). The main land use types in the delta are water body, grassland, and unused land.

The Lake Balkhash (73°21′-79°30′E, 44°45′-46°44′N) is a typical plain coccyx lake (Yang, 1993) with a wide area of approximately 1.83×10⁴ km²; it is about 71 km wide at its widest point and 600 km long (Gao et al., 2016). As the main water artery of the Lake Balkhash, the Ili River flows into the west of the lake through the IRD, contributing 78.00% of the total runoff to the lake, which is the main source of the water volume of the Lake Balkhash (Xiao et al., 2011). The Ili-Balkhash Basin is one of the largest lake ecosystems in the world. It extends around the Lake Balkhash as the centre, with the highest point at the Khan Tengri Peak in the Tianshan Mountains and the lowest point at the Lake Balkhash, becoming the tail end of the rivers (Long et al., 2011).

In this study, we collected and calculated data on precipitation, potential evapotranspiration, land uses, soil depth, root depth, plant available water content (PAWC), velocity coefficient, topographic index, percentage slope, soil saturation hydraulic conductivity, runoff, Normalized Difference Vegetation Index (NDVI), digital elevation model (DEM), and water level to estimate the water conservation and evaluate the water conservation function of the IRD from 1975 to 2020. The data sources and parameter processing methods of the input data are listed in Table 1. The raster resolution of model inputs is uniformly 30 m×30 m. The biophysical parameters of each land use type are also listed in Table 2.

In addition, we verified the accuracy of the simulated results from the InVEST model by calculating the measured annual water conservation volume from the surface runoff volume (data from the Uskerma hydrological station). The results indicated that the simulated water conservation volume from the InVEST model was smaller than the measured water conservation volume from the surface runoff volume during the study period (1975-2020); however, the trends of the two remained basically the same (Fig. 2).

The land use transfer matrix is a classical method for studying the direction and quantity of transfer between land use types; this method can reflect the structural characteristics of regional LUCC and the evolution of the spatial patterns (Li et al., 2022b). Using the spatial analysis tools in ArcGIS, the land use data during different periods were spatially superimposed to obtain the dynamic changes in land use types in the two adjacent periods. The formula is as follows (Xie, 2015):

where S_ij represents the area of land use type i converted to land use type j during the study period (km²); and n is the number of land use types.

The water yield model in the InVEST model is based on the principle of water balance. The regional water yield can be calculated using the annual precipitation and the Budyko curve. The formulas are as follows:

where Y_xj represents the annual water yield depth at grid unit x under land use type j (mm); AET_xj represents the annual actual evapotranspiration at grid unit x under land use type j (mm); p_x represents the annual precipitation at grid unit x (mm); ω_x is a non-physical parameter of natural climate soil properties; R_xj is the Bydyko dryness index; and Z is the seasonal factor (Zhang coefficient) that represents the seasonal precipitation distribution and depth, with Z-values close to 10 indicating that precipitation is concentrated in winter and close to 1 indicating that precipitation is concentrated in summer or has a more uniform seasonal distribution (Wang et al., 2022a). The Z-value was determined to be 1 in this study because the multi-year precipitation is concentrated in summer in the study area. AWC_x represents the available soil water content (mm), which is determined by the soil texture and effective soil depth; K_xj is the evapotranspiration coefficient of land use type j at grid unit x; and PET₀ represents the potential evapotranspiration (mm) (Yang et al., 2019; Li et al., 2021; Li et al., 2022; Wang et al., 2022).

The calculation of the water conservation is based on the water yield combined with the soil saturation conductivity, runoff coefficient, and topographic index. The formulas are as follows:

where Retention is the annual water conservation depth (mm); Velocity is the runoff coefficient (dimensionless), representing the impact of different land use types on surface runoff; Kast represents the soil saturation conductivity (mm/d); TI is the topographic index; Drainage_Area indicates the number of grids in the catchment area; Soil_Depth is the soil depth (mm); and Percent_Slope is the percentage slope (%) (Bao et al., 2016; Xu et al., 2022a).

To study the dynamic temporal and spatial trends in the water conservation function in the IRD from 1975 to 2020 at the pixel scale, we established a linear regression equation and a piecewise linear regression equation with the year as the independent variable and the water conservation as the dependent variable, based on the study of Hu and Sheng (2022). Piecewise linear regression is a regression estimation method applied when the regression of y on x obeys a certain linear relationship in one range of x and a linear relationship with different slopes in other ranges. This method uses indicator variables to fit a unified regression model to each segment (i.e., each range) of data simultaneously. Pearson's correlation coefficients were used to test the correlation between the annual-scale water conservation and driving factors (Yu et al., 2010). It can be calculated using Equation 8 (Lee Rodgers and Nicewander, 1988):

where r is the correlation coefficient between two variables (correlations between potential evapotranspiration and water conservation, and between precipitation and water conservation), which is generally used to infer the overall correlation coefficient; m is the number of samples; x_i and y_i are sample values of variables x and y, respectively; and $\bar{x}$ and $\bar{y}$ are the average values of variables x and y, respectively.

From 1975 to 2020, the annual average water yield depth of the IRD showed a trend of "M"-shaped fluctuations, with an average value of 62.88 mm over the studied 46 years, a minimum value of 34.76 mm in 2020, and a maximum value of 98.15 mm in 1978 (Fig. 3a). During the study period, the water conservation volume of the IRD changed greatly, with a multi-year average water conservation volume of 9.06×10⁹ m³; the minimum value was 2.59×10⁹ m³ in 2020, and the maximum value was 19.86×10⁹ m³ in 1993 (Fig. 3b). Based on the scope of the study period, this paper used 2000 as the baseline year to compare and analyze the differences in the water yield depth and water conservation volume in the IRD between the end of the 20^th century and the beginning of the 21^st century. Thus, the changes in the water conservation volume of the IRD over the studied 46 years were mainly divided into two stages. Specifically, from 1975 to 2000, the annual water conservation volume showed a decreasing trend, with a change rate of 0.19×10⁹ m³/a, and from 2000 to 2020, it showed a slightly increasing trend, with a change rate of 0.01×10⁹ m³/a (P<0.05). Overall, the water conservation volume in the IRD exhibited a decreasing trend from 1975 to 2020.

From 1975 to 2020, the spatial distribution pattern of the annual average water yield depth in the IRD showed a trend of "high in the east and low in the west" (Fig. 4a1-a6). High-value regions with the water yield depth greater than 62.88 mm were found in the central part of the IRD, and low-value regions with the water yield depth less than 62.88 mm were mainly observed at the end of the delta. The spatial distribution patterns of the water conservation depth in the IRD from 1975 to 2020 did not change markedly and generally showed a regularity consistent with the changes in the water yield depth, which directly influenced the spatial distribution of the water conservation depth in the region. Based on the natural segment point method, we divided the water conservation depth in the IRD into five classes (I (0.00-2.00 mm), II (2.00-4.00 mm), III (4.00-6.00 mm), IV (6.00-8.00 mm, and V (≥8.00 mm)), with high-value regions mainly located in the central part of the IRD and low-value regions mainly around the Lake Balkhash (Fig. 4b1-b6). The combined effects of climate and topography created this spatial distribution pattern. The Ili-Balkhash Basin extends in all directions with the Lake Balkhash at its centre; the highest point is at the Khan Tengri Peak in the Tianshan Mountains and the lowest point is in the Lake Balkhash. The IRD is a low-altitude region with high annual precipitation and low evapotranspiration; large areas of wetland vegetation and wetland water body are located in the middle of the delta, which are conducive to the growth of vegetation and provide a higher water conservation capacity. In contrast, around the Lake Balkhash, there are large areas of water body and unused land, with intense evapotranspiration and a low water conservation capacity.

The spatial distribution of the water conservation volume variation in the IRD from 1975 to 2020 is shown in Figure 5, which was divided into four categories: significantly decreased, slightly decreased, basically unchanged, and significantly increased. From 1975 to 2020, the area of regions with decreased water conservation volume was 1756.20 km², accounting for 5.91% of the total area in the IRD, whereas the area of regions with increased water conservation volume was 743.26 km², accounting for 2.50% of the total area in the IRD; the water conservation volume in most regions was stable and unchanged, accounting for 91.59% of the total area in the IRD. The regions with increased water conservation volume were concentrated at the front of the delta, mainly for two reasons: the conversion of unused land into grassland, and the increase of precipitation. The regions with decreased water conservation volume were mainly distributed in the middle and end of the delta, which are the main parts of the delta wetland. With the increase in runoff into the delta, the gradual recovery of wetland vegetation and the increase of regional evapotranspiration, combined with the decrease of precipitation, have led to a reduction in the water yield depth and water conservation volume in these regions. At the same time, the increase in wetland water body covering the original grassland and other conversions between land cover types has also resulted in a reduction in the water conservation volume. Water conservation volume was largely stable and unchanged in the western part of the Lake Balkhash.

Precipitation and potential evapotranspiration are important climate factors influencing the water conservation function in the IRD, and their temporal and spatial variability drives the changes in the water conservation function. The spatial differences in the average annual precipitation of the IRD during 1975-2020 were significant. The Lake Balkhash area was dry and rainless, and precipitation gradually increased from the centre of the lake to the outside, while the spatial distribution of potential evapotranspiration was opposite to that of precipitation (Fig. 6a and b).

The water conservation depth and precipitation showed similar spatial distribution characteristics. In the centre of the lake, precipitation was low, potential evapotranspiration was high, and the water conservation depth was low. From 1975 to 2020, annual precipitation in the IRD showed a decreasing trend (Fig. 6c), annual potential evapotranspiration showed an increasing trend (Fig. 6d), and the annual water conservation volume showed a decreasing trend (Fig. 3b), indicating that the relative changes in precipitation and potential evapotranspiration determined the general declining trend of the water conservation function. The water conservation depth was significantly positively correlated with precipitation and negatively correlated with potential evapotranspiration (Fig. 6e and f), with correlation coefficients of 0.72 and -0.12, respectively (P<0.05). The temporal and spatial variations in precipitation significantly impacted the water conservation function in the study area. This is mainly because regions with abundant precipitation correspond to areas with good vegetation growth, sufficient water for plants and soil, and a strong water conservation capacity. In contrast, regions with high potential evapotranspiration are heavily depleted of water in vegetation and soil and have a weak capacity to retain water.

The dominant land use types in the IRD were grassland, water body, wetland, and unused land, which accounted for more than 95.00% of the total area in the delta (Fig. 7a1-a4). Cropland comprised a small proportion of the total area in the IRD, never exceeding 1.00%. Forest land area fluctuated more markedly between 1990 and 2020, showing an overall decreasing trend (Fig. 7b). In general, the areas of grassland, wetland, water body, cropland, and construction land all showed increasing trends, while the areas of forest land and unused land decreased (Fig. 7c). From 1990 to 2020, the largest decrease in the rate of change in area was -53.38%, for forest land, while the largest increase in the rate of change in area was 41.07%, for construction land. From 1990 to 2020, the decreased area of unused land was the largest (2673.64 km²), and this area was converted into grassland, wetland, water body, forest land, cropland, and construction land. The area of unused land transformed to grassland was the largest (1959.97 km²). The areas of lost cropland and construction land were the smallest, with the values of 2.09 and 15.34 km², respectively. Cropland was mainly converted to grassland, and construction land was mainly converted to grassland and unused land (Fig. 7d).

The annual water conservation volume of grassland during 1975-2020 was the highest, with an average value of 42.93×10⁸ m³, followed by forest land and cropland (1.96×10⁸ and 0.22×10⁸ m³, respectively) (Fig. 8a1-a3). The canopy and deadfall parts of grassland and forest land can effectively retain water (Xu et al., 2022); therefore, these regions have a stronger water conservation function. In contrast, cropland has a shallow root system and occupies a small area in the IRD; therefore, the water conservation function of cropland is poor. From 1975 to 2020, the annual water conservation volume of grassland, forest land, and cropland showed an overall decreasing trend in the IRD. However, they all exhibited an increasing trend between 2000 and 2020, which is consistent with the trend of the annual average NDVI in the IRD, indicating that vegetation growth in the delta has gradually improved during 2000-2020.

From 1990 to 2020, the annual potential evapotranspiration of grassland, forest land, and cropland showed an increasing trend (Fig. 8b1-b3). Specifically, the annual potential evapotranspiration of grassland increased at the rate of 5.23 mm/10a, and the total area of grassland increased. However, the water yield depth of grassland decreased from 139.63 mm in 1990 to 74.57 mm in 2020 (Fig. 8c1), and the corresponding water conservation volume also decreased from 46.91×10⁸ m³ in 1990 to 14.82×10⁸ m³ in 2020 (Fig. 8a1). The water yield depth of cropland decreased from 119.14 mm in 1990 to 72.27 mm in 2020 (Fig. 8c2), and the corresponding water conservation volume decreased from 0.24×10⁸ m³ in 1990 to 0.07×10⁸ m³ in 2020 (Fig. 8a2). The water yield depth of forest land decreased from 128.22 mm in 1990 to 70.07 mm in 2020 (Fig. 8c3), and the corresponding water conservation volume decreased from 3.28×10⁸ m³ in 1990 to 0.40×10⁸ m³ in 2020 (Fig. 8a3). The main reason for this situation is that an increase in potential evapotranspiration will lead to a significant consumption of water in vegetation and soil and a decrease in the water yield, thus affecting the accumulation of water and the water-holding function of soil, which will lead to a decline in the water conservation function. Grassland was classified as low-coverage (coverage of >50%), medium-coverage (coverage of 20%-50%), and high-coverage (coverage of 5%-20%) grasslands according to the classification system. From 1990 to 2020, although the areas of low-coverage and medium-coverage grasslands increased, the areas of high-coverage grassland and forest land showed a decreasing trend, resulting in the decline of the water conservation in the IRD (Fig. 9).

From 1990 to 2015, the Ili River entered a period of abundant water, and the runoff into the lake at the Uskerma hydrological station increased. After 2015, runoff into the lake at the Uskerma hydrological station decreased, the water level of the Lake Balkhash decreased, precipitation decreased, and the water conservation volume continued to decrease (Figs. 6c and 10). Thus, the water conservation was affected by both the inflow from the Ili River and precipitation.

Generally, the evolution of ecosystem service functions is driven by various natural and human factors (Wang et al., 2021). Human activities directly or indirectly affect regional ecosystem service functions (e.g., water conservation) by changing the type and structure of the underlying surface (Su and Fu, 2013). In terms of natural factors, the spatial distribution of vegetation also has a significant impact on water conservation, mainly because the corresponding soil physical and chemical properties, vegetation coverage, and root systems of vegetated land and non-vegetated land are different, thus affecting the surface evapotranspiration and soil water storage capacity (Li et al., 2022). Different altitudes have different water and heat conditions; thus, the degree of vegetation coverage varies. The higher the vegetation coverage is, the stronger the water conservation function of the plants will be; therefore, low vegetation coverage means a low water conservation function (Bai et al., 2014). Secondly, changes in land use types caused by human activities affect the water conservation capacity. The water conservation volume of grassland is the highest, followed by forest land, cropland, and construction land (Wen and Théau, 2020). This is mainly because the canopy and dead parts of grassland and forest land can effectively retain water, resulting in a strong water conservation function (Xu et al., 2022).

Meanwhile, the development of water resources from the Ili River began in the 1920s, and the largest reservoir in the region, the Kapchagay Reservoir, was built in 1970. The construction of the reservoir has caused changes in the volume and distribution of annual runoff in the lower reaches of the Ili River, which has had a significant impact on the ecological environment of the delta in the lower reaches of the Ili River and the Lake Balkhash (Xie et al., 2011). From 1970 to 1985, the construction of the Kapchagay Reservoir and the diversion of water for irrigation on the left bank of the reservoir significantly exacerbated the decline in the water level of the Lake Balkhash and the ecological shrinkage of the natural oasis. The areas of high-coverage grassland and forest land decreased, and the changes in runoff and precipitation into the lake at the Uskerma hydrological station both showed an increasing trend first followed by a decreasing trend, which affected the water conservation (Wang and Lu, 2009). As the water level of the Lake Balkhash rose, the ecological problems caused by water shortages were alleviated, but the ecological water demand of vegetation increased, leading to a reduction in the water conservation (Guo et al., 2011). The areas with the greatest reduction in the water conservation and the areas with the greatest increase in grassland coincided, as shown in Figures 4 and 7a1-a4. This indicates that the increase in the runoff to the Lake Balkhash resulted in higher vegetation coverage in the IRD, while simultaneously increasing the surface evapotranspiration of the vegetated areas and the ecological water demand of the vegetation, leading to a reduction in the water conservation. Latest, climate factors are the key natural factors driving the changes in the water conservation. Regions with abundant precipitation correspond to the areas with good vegetation growth, adequate water in plants and soils, and strong water conservation capacity (Guo et al., 2011). In contrast, regions with high evapotranspiration are heavily depleted of water in vegetation and soils and have a weak capacity to retain water (Guo et al., 2011).

In addition, there are two main limitations when modelling the water yield. First, because of the specificity of the geographical location of the study area, the biophysical parameters such as Kc and root depth in this study were derived from empirical data in the literature, which will affect the accuracy of the model simulation to a certain extent; however, this will not affect the basic distribution pattern of the water yield. Second, this study used meteorological data of the current year for model estimation, thus failing to consider the time lags of meteorological factors acting on environmental processes. The validation of the actual survey and monitoring data will be further improved in the future, and attention should also be paid to the modelling of inter-annual and seasonal water yields along with the localization of parameters.

The temporal and spatial variations in the water conservation of the IRD from 1975 to 2020 were analyzed using the InVEST model, and the main driving factors were discussed. The main conclusions are as follows:

(1) From 1975 to 2020, the water conservation showed a decreased trend in the IRD; spatially, it exhibited a pattern of "high in the east and low in the west".

(2) The spatial distribution of the water conservation was affected by the land use types. From 1975 to 2020, the water conservation and water yield of each land cover type decreased slightly. The water conservation volume of grassland was the highest among all land use types.

(3) The water conservation depth and precipitation showed the same spatial distribution characteristics. During the study period, precipitation in the delta continued to decrease, and the increase in the inflow of the delta has promoted the extension of the vegetated areas, especially grassland, which has had a positive effect on the water conservation function. However, this effect cannot offset the negative effect of enhanced evapotranspiration in the IRD.

This study used the InVEST model to evaluate the temporal and spatial variations and driving factors of the water conservation function in the IRD. Changes in the water conservation function in the IRD were strongly correlated with climate change and LUCC, with precipitation having a direct effect on the water conservation function. These results provide a scientific background for making informed ecological protection decisions in the IRD under the impacts of climate change and human activities.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

This research was funded by the National Natural Science Foundation of China (42071245), the Xinjiang Uygur Autonomous Region Innovation Environment Construction Special Project & Science and Technology Innovation Base Construction Project (PT2107), the Third Xinjiang Comprehensive Scientific Survey Project Sub-topic (2021xjkk140305), the Tianshan Talent Training Program of Xinjiang Uygur Autonomous Region (2022TSYCLJ0011), and the K. C. Wong Education Foundation (GJTD-2020-14).

Conceptualization: MA Yonggang; Methodology: CAO Yijie; Formal analysis: MA Yonggang, CAO Yijie; Writing - original draft preparation: CAO Yijie; Writing - review and editing: MA Yonggang, CAO Yiiie; Funding acquisition: MA Yonggang, BAO Anming; Resources: MA Yonggang, BAO Anming, CHANG Cun, LIU Tie; Supervision: MA Yonggang, LIU Tie; Visualization: CAO Yijie.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.