Situated at 110°04°59E-111°17°35E and 24°06°39N- 25°54°56N, the study area is located in the northeast of Guangxi Zhuang Autonomous Region, which is an essential link between the Yangtze River water system and the Pearl River water system (Fig. 1). It is a belt-shaped area along its north and south orientation, spanning 16 districts and counties in Guangxi Zhuang Autonomous Region and one county in Hunan Province, with a total area of 13000 km², accounting for 46.76% of the total area of Guilin City. The site is rich in mountain and water resources and has nurtured a world-class valuable resource, the karst-style Guilin landscape, of which the Li River is the lifeline. The Li River (also called the Lijiang River) is one of the national key protected rivers and a natural heritage shared by all humanity. It originates in Mao’er Mountain at the junction of Xing’an and Lingchuan counties, flows through Guilin City and Yangshuo County, and finally reaches near Pingle County where it meets with Gongcheng River and Lipu River to become the upstream of the Guijiang River. Given the complexity of the river confluence, the Li River Basin studied in this paper includes parts of the Gongcheng River and the Lipu River.

The topography of the Li River Basin is high in the north and low in the south, high in the east and low in the west, with the highest elevation in the territory of 2110 m and the lowest is at 9 m (Fig. 2a). The basin is located at a low latitude. It belongs to the typical mid-subtropical monsoon climate, with high temperatures and humidity in summer and cold and dry weather in winter. Its annual average temperature is 18 ℃ and annual average precipitation is 1688 mm. There are many rivers in the basin, mainly the mainstream of Li River, including tributaries such as Lingqu River, Gantang River, Taohua River, Liangfeng River, and Yulong River. Since precipitation in the basin is greatly affected by monsoon activities, more than 80% of the rainfall is concentrated in March to August, resulting in significant differences in the runoff volumes of the tributaries during the rainy and dry seasons (Duan et al., 2014).

According to factors such as land use/land cover, climate, and soil, the water yield model of the InVEST model realizes the quantitative, dynamic, and visual evaluation of the water production function of the ecosystem. It is based on the Budyko curve and the principle of water balance, quantitatively analyzes abstract ecosystem services on a unified pixel scale, and uses the precipitation of each pixel minus the actual evapotranspiration to estimate the water production of the pixel, including surface runoff, soil water content, litter water holding capacity, and canopy interception. The equation is given as:

where Y_xj is the annual water yield for pixel x on the land use type j (mm), P_x is the average yearly precipitation for pixel x (mm), and AET_xj is the actual annual evapotranspiration for pixel x on land use type j (mm) which is calculated by the following formula:

where PET_xj is the potential evapotranspiration for pixel x on land use type j (mm), which is defined by Equation (3) andω_x is an empirical parameter that can be expressed as a linear function, as shown in Equation (4). Their computation formulas are as follows:

where ET_0xj is the reference evapotranspiration of pixel x on land use type j (mm) that can be calculated according to the Penman-Monteith Equation, and Kc_xj is the plant evapotranspiration coefficient associated with the land use type j on pixel x. Z is an empirical constant, also known as the “seasonality factor”, which characterizes regional precipitation patterns and local hydrogeological features. AWC_x is the effective soil water content (mm), which is used to determine the total amount of water provided and stored by the soil for vegetation growth and can be determined by the minimum value of the maximum soil root burial depth and plant root depth in conjunction with the vegetation available water content. The formula is as follows:

where MaxSoildepth is the maximum root burial depth of the soil, Rootdepth is the plant root depth, and PAWC is the plant available water content, usually expressed as the difference between the field water holding capacity and the wilting point, here using the Zhou et al. (2005) nonlinear fitted soil PAWC estimation model:

where the sand%, silt%, clay%, and OM% refer to the specific gravities of sand, silt, clay, and organic matter in the soil texture, respectively.

The datasets required for the InVEST water yield model include watershed boundaries, land use/land cover, precipitation, average annual reference evapotranspiration, root restricting layer depth, and plant available water content data. The projection coordinates of all GIS data were unified to Krasovsky_1940_Albers, with linear units of meters and a uniform spatial resolution of 30 m.

Water yield and water supply are the important indicators of water supply services in a basin, with the former expressed through unit water production depth (mm) and the latter characterized by total basin water production (m³). The average water yields in the Li River Basin were 1069 mm, 1024 mm, 859 mm, 1512 mm, and 840 mm in 2000, 2005, 2010, 2015, and 2018, respectively, with a multi-year average yield of 1061 mm (Fig. 4). In terms of overall trends, the water supply services fluctuated widely from year to year and varied significantly from period to period. The water yield in the basin showed a slight downward trend from 2000 to 2010, with a rate of decline of 21 mm yr^-1; a more pronounced upward trend from 2010 to 2015, with a rate of increase of 130.6 mm yr^-1; and a significant downward trend from 2015 to 2018, with a rate of decline of 67.2 mm yr^-1. The total water supply in the whole basin showed the same trend directions, with the highest total water supply in 2015 of 196.84×10⁸ m³, the lowest total water supply in 2018 of 109.44×10⁸ m³, and the multi-year average total water supply of 138.14×10⁸ m³. From the water balance perspective, precipitation and actual evapotranspiration are the two key factors that determine water production. As shown in Fig. 4, precipitation over the same period also showed a trend of decreasing, then increasing and then decreasing again, basically maintaining a level between 1400-2100 mm, with the highest average precipitation of 2127 mm in 2015. The average potential evapotranspiration and actual evapotranspiration showed a decreasing trend, decreasing by 23 mm from 2000 to 2018.

The spatial distribution pattern of water yield in the Li River Basin was generally high in the north and low in the south, and showed a gradual decrease from north to south with noticeable gradient changes (Fig. 5). The high-value areas were mainly concentrated in the northern part of the basin, with the most concentrated areas in the low elevation plains centered in Guilin City. The low-value areas were distributed primarily in the south of the basin, with the most significant concentration in the high elevation areas. This distribution pattern was directly related to average annual precipitation (PRE) and actual evapotranspiration (AET). Areas with high PET and low AET had a substantial water production capacity, and conversely, areas with low PET and high AET had a weak water production capacity. Specifically, PRE gradually decreased from northwest to southeast, showing a high north to low south pattern, similar to the water yield (Fig. 5). This distribution was caused by the Mao’er Mountains in the northern part of the basin blocking the southward airflow, forcing the air masses to lift and form heavy rainfall, resulting in the upper reaches becoming one of the high-value rainfall areas in China. In this case, the PRE in the storm center reached 2600 mm, making the moisture input in the northern part of the basin much higher than in the southern region. AET was another critical factor affecting water yield. Its high values were distributed at high elevations, while the low values were concentrated in the central plains of the basin, especially in urban settlements (Fig. 5). The former had higher AET values due to higher vegetation cover at higher altitudes, while the latter was dominated by construction land and agricultural cropland and had lower AET values.

Water yield was also directly related to the distribution of land use types, and there are significant differences in water yield among the land use types in the study area (Table 2). The primary land use types in descending order of water yield were artificial surface (1559 mm), unused land (1349 mm), grassland (1169 mm), forest land (1050 mm), cropland (1036 mm), and wetland (912 mm). The top six secondary land use types in terms of water yield were urban land, other construction land, bare land, rural settlement, middle coverage grassland, and swamp land. These differences are largely due to impervious surfaces increasing the surface temperature by absorbing solar radiation energy. Thus, the sensible heat is high and the latent heat is low, which ultimately leads to low evapotranspiration. Therefore, these types had a higher water supply capacity under the same precipitation conditions. In comparison, the water yields of lake, beaches, canals, woodland, shrubland, and paddy field occupied the last six positions. The lowest water yield of the first three land use categories was due to the large and rapid evaporation from the water surface. The lower water yield in the last three land use categories was due to high evapotranspiration caused by vegetation transpiration.

Sub-basins are the basic units of water production formation, so it is more meaningful to explore the spatial and temporal differences of water supply services at the sub-basin level. Using the average water yield (mm) of each sub-basin in 2000, 2005, 2010, 2015, and 2018 as an indicator to assess the regional water supply services, the results showed that the water supply services in the Li River Basin were predominantly upstream and the yields decreased from upstream to downstream (Fig. 6). Overall, the average multi-year water yields of the five major regions fell in the following order: upstream region (1253.2 mm), midstream region (1132.6 mm), Gongcheng River region (928 mm), Lipu River region (925.2 mm), and downstream region (823.4 mm). Specifically, sub-basins 1-4 had higher water yields, with multi-year averages above 1200 mm; sub-basins 5-8 and 13-16 had the next highest water yields, fluctuating between 900 mm and 1200 mm; and sub-basins 9-12 and 17-18 had the lowest water yields, all below 900 mm (Fig. 6).

In terms of the magnitude of changes in water yield in the sub-basins, the overall trend of water supply services in the sub-basins from 2000 to 2010 was decreasing (Fig. 7). The sub-basins with larger decreases include 1, 2, 4, 12, and 18, where the yearly declines were higher than 30 mm yr^-1; the sub-basins with smaller decreases include 3, 5, 14, 15, and 16, where the annual decreases were lower than 20 mm yr^-1. From 2010 to 2015, the overall water supply services in the sub-basins increased significantly, with all sub-basins producing more than 1000 mm, which was related to the increase in precipitation during this period. The sub-basins with higher annual increases include 5, 7, 8, 13, 15, and 17, with annual increases higher than 150 mm yr^-1; while the sub-basins with smaller increases include 2, 4, 9, 10, 11, and 12, with annual increases lower than 100 mm yr^-1. The water yields of the sub-basins declined significantly from 2015 to 2018, and the overall water yield was lower than the multi-year average. Sub-basins with significant decreases include 3, 5, 7, 15, and 16 with annual decreases above 150 mm yr^-1 and the sub-basins with smaller decreases were 11, 12, and 14 with annual reductions below 80 mm yr^-1.

In terms of the fluctuations of water yield in the sub-basins (Fig. 8), the sub-basins with large changes in the water yield between the first stage (2000-2005) and the second stage (2005-2010) include 4, 7, 13, 14, and 15, which correspond to the midstream, Lipu River and the northern part of Gongcheng River region. The sub-basins with more minor fluctuations were 1, 9, 11, and 18, corresponding to the upstream, downstream, and south of the Gongcheng River region (Fig. 8). The sub-basins with more significant changes in water yield between the third stage (2010-2015) and the fourth stage (2015-2018) include 5, 7, 15, and 17, which correspond to the midstream and the Gongcheng River region; while the sub-basins with more minor fluctuations were 4, 10, 11, and 12, which correspond to the midstream region and the downstream region.

The Li River Basin has always been a critical area of national concern. On the one hand, the Li River provides abundant water resources for the Xijiang River system. On the other hand, the basin has nurtured the world-class landscape Guilin landscape, which possesses high landscape aesthetic and cultural heritage value. As such, it provides numerous ecosystem services to human society, including but not limited to freshwater supply, food production, atmospheric regulation, hydrological regulation, recreational tourism, and landscape aesthetics (Scholte et al., 2015; Meng et al., 2020). Among them, the water supply services are the most basic but essential services. Water supply services are usually measured by average water yield and entire water supply, which are related to the average water yield and area of each land use type. Therefore, the contribution of each region to the entire water supply depends on the regional area and land use composition (Table 3). The area of each region and the entire water supply were calculated and showed that the regional areas decrease in the order of Gongcheng River, midstream, upstream, Lipu River, and downstream. The entire water supply decreases in the order of upstream, midstream, Gongcheng River, Lipu River, and downstream (Table 4). These results imply that the spatial variation of the water supply volume is largely due to the different areas of the regions.

However, the differences in water supply volume due to area differences are not a good representation of the actual water supply of the region. Therefore, we attempted to quantify the areas and entire water supply in other regions, using the downstream region as the smallest unit, to analyze and compare their water supply capacities. Firstly, using the downstream region as 1 area unit, and upstream, midstream, downstream, Lipu River, and Gongcheng River regions correspond to 5.70, 5.72, 1, 3.25, and 6.81 area units, respectively. Secondly, using the entire downstream water supply as 1 supply unit, the upstream, midstream, downstream, Lipu River, and Gongcheng River regions correspond to 8.66, 7.76, 1, 3.58, and 7.63 water supply units, respectively. Finally, the difference between the above two numbers was used to characterize the water supply capacity, with a larger difference implying a higher water supply capacity. These results show that the regional water supply decreased for the same area scenario: Upstream, midstream, Gongcheng River, Lipu River, and downstream. The area difference between the upstream and the midstream was small, but the water supply capacity in the midstream was lower than that in the upstream. This difference may be due to the sum of the shares of cropland and forest land in the upstream region being slightly higher than that in the midstream. In addition, the water supply capacity of the midstream was significantly higher than that of the Gongcheng River region, even though the Gongcheng River region has a large area. This probably due to the aggregate of the shares of grassland and industrial and mining land in the Gongcheng River region was much lower than that in the midstream. In summary, the water supply services in the upstream and midstream regions play an indispensable role in the water supply service function of the whole Li River Basin, so the utilization and management of water resources in these two regions should be strengthened. In addition, emphasis should be placed on enhancing the Gongcheng River and Lipu River region water supply service capacities.

There may be some shortcomings in our analysis. For example, the study area is a typical karst landscape, implying that part of the water loss is groundwater rather than surface water. However, the InVEST model does not consider the loss of groundwater, which leads to some bias in the simulation results of water yield. In addition, the precipitation and evapotranspiration required for the model operation are spatial data obtained by interpolation, and the number and distribution of meteorological stations affect the interpolation accuracy of precipitation and evapotranspiration. In particular, the Li River Basin is strongly influenced by the climate, which further increases the uncertainty of meteorological interpolation.

Meanwhile, some parameters such as those in the biophysical parameter table, root restricting layer depth, plant available water content, and seasonal factor are derived from empirical data in the literature or the distribution of example parameters, which may also lead to lower simulation accuracy. Therefore, subsequent studies should optimize these parameters, increase the amount of measured data, and validate the yield results in situ by field sampling to improve the accuracy of the water supply service calculations. In addition, the flow paths and flows of water should be identified (Bagstad et al., 2013; Comberti et al., 2015), which will help to accurately quantify the physical and value flows of water supply services and provide a scientific reference basis for the reverse construction of ecological compensation mechanisms.