Spatial–Temporal Variations in Water Use Efficiency and Its Influencing Factors in the Li River Basin, China

Abstract

As a vital indicator for measuring the coupled carbon–water cycle of an ecosystem, water use efficiency (WUE) can also reflect the adaptive capacity of plants in different ecosystems. Located in Southwest China, the Li River Basin has a representative karst landform, and the uneven rainfall in the region leads to severe water shortage. In this study, we analyzed the spatial–temporal transformation characteristics of the WUE of the basin and its relationship with different influencing factors from 2001 to 2020 based on a correlation analysis and trend analysis. The main conclusions are as follows: (1) The average value of WUE in the Li River Basin was 1.8251 gC· mm⁻¹·m⁻², and it kept decreasing at a rate of 0.0072 gC· mm⁻¹·m⁻²·a⁻¹ in the past 20 years. With respect to the spatial distribution of the multi-year average of WUE, it exhibits a gradual increasing trend from west to east. (2) Between gross primary productivity (GPP) and evapotranspiration (ET), it was found that ET was the primary influencing factor of WUE. Precipitation was positively correlated with WUE in the Li River Basin, accounting for 67.22% of the total area of the basin. The air temperature was negatively correlated with WUE, and the area was negatively correlated with WUE, accounting for 92.67% of the basin area. (3) The normalized difference vegetation index (NDVI) and leaf area index (LAI) were negatively correlated with WUE, and the proportions of negatively correlated areas to the total area of the basin were similar; both were between 60 and 70%. The growth of vegetation inhibited the increase in WUE in the basin to a certain extent. Regarding Vapor Pressure Deficit (VPD), the proportions of positive and negative correlation areas with WUE were similar, accounting for 49.58% and 50.42%, respectively. (4) The occurrence of drought events and the enhancement in its degree led to a continuous increase in WUE in the basin; for different land cover types, the correlation of the standardized precipitation evapotranspiration index (SPEI) was in the following order from strongest to weakest: grassland > cropland > forest > shrubland.

1. Introduction

Water use efficiency (WUE), as an important indicator of carbon and water coupling in terrestrial ecosystems [1], is highly sensitive to climate change, and the study of WUE can help to better understand the carbon and water cycling of ecosystems and reveal the mechanism of ecosystem response to climate change [2]. WUE has different definitions in different research backgrounds and scales, and initially, WUE studies were applied to plant leaves or individual scales, and with the updating and iteration of remote sensing technology, WUE studies at different scales have attracted the attention of many scholars [3,4,5]. At the ecosystem scale, WUE is defined as the ratio of gross primary productivity (GPP) or net primary productivity (NPP) to evapotranspiration (ET) [6], indicating the amount of dry matter produced per unit mass of water consumed by a plant (or the amount of CO₂ fixed).

In recent years, different scholars have conducted targeted studies on WUE in various regions of the world. Globally, WUE is positively correlated with temperature changes below 10 °C, but WUE shows a decreasing trend with further increases in temperature [7,8]. We studied regional WUE in China and found that WUE showed a continuous increasing trend, the northeastern region had a higher WUE, and different regions had different factors with large contributions to the change in WUE [9,10]. The WUE of the southwestern region also showed an increasing trend, in which the annual average WUE of the karst landforms was lower than that of the non-karst landforms [11]. Furthermore, the changes caused by influence factors in the two landforms were different, and the temperature was positively correlated with WUE in the karstic region but negatively correlated with WUE in the non-karstic region [12,13,14]. Gao [15] studied the WUE in Xinjiang region in northwest China and found that vegetation WUE showed an overall decreasing trend, and the plains were the main concentration areas for high WUE values, while low values were mostly distributed in mountainous areas.

In the current context of global warming, extreme climate events occur frequently [16], and many experts and scholars have made consistent judgments through statistical predictions of data models [17,18], and drought, as one of the extreme climate events, influences the carbon and water cycling of terrestrial ecosystems to a certain extent. Some earlier studies implemented the response analysis of WUE to drought based on different drought indices (standardized precipitation evapotranspiration index, palmer drought severity index, temperature vegetation dryness index) and achieved some results [13,19,20]. In Central Asia, generally, WUE is negatively correlated with drought, and the response of WUE to drought changes in various areas and for different vegetation types [21]. Ma [22] studied young forests in northern China and found that drought events increased autumn WUE and decreased summer WUE, and the effect of long-term drought impacts was greater than that of short-term drought.

With a low latitude, Guangxi is located in southwestern China, having a meso-subtropical monsoon climate [23]. As one of the areas with more concentrated karst landforms in China, the karst land areas in Guangxi total 83,500 km², accounting for about one-third of its total area. However, the soil in karst areas is relatively loose and weak in water retention capacity, coupled with the uneven spatial and temporal distribution of rainfall in the region, leading to a number of problems in the utilization of water resources. The Li River is located in the northeastern part of Guangxi, and it belongs to the upper reach of the Gui River in the Xijiang River system of the Pearl River Basin. The Li River Basin has a typical karst landform, which is one of the representative regions of global karst landforms [12]. In view of the special characteristics of the karst landform in the Li River Basin, combined with the background of the global climate anomalies and the frequent occurrence of extreme events, it is vital to analyze the spatial and temporal characteristics of the ecosystem WUE using meteorological factors (air temperature and precipitation) and remote sensing indices, for example, the NDVI, LAI, and SPEI.

2. Study Area

The Li River Basin is located in the northeastern part of Guangxi, geographically 109°45′ E~111°02′ E, 24°16′ N~25°55′ N (Figure 1a). The source of the Li River is located on the south side of the Laoshan Boundary in the northeast of the Kitten Mountain. The basin borders Longsheng, Ziyuan, and Quanzhou in the north; Wuzhou in Guangxi in the south; Yongfu and Rong’an in the west; and Gongcheng, Guanyang, and Zhongshan in the east. The total area of the basin is about 10,803.59 km². The basin is low in the south and center and high in the north, west, and east. Climatically, the basin belongs to the middle-subtropical monsoon climate zone, with a high temperature throughout the year, 1614.7 h of average annual sunshine, an average annual temperature of 19.2 °C, and 1813.5 mm of average annual rainfall. Spatial distribution disparities exist between precipitation and temperature values, with the regions of extremely high and low precipitation located in the central and northern parts of the basin, respectively. Meanwhile, the areas of extremely high and low temperature are found in the southern and northwestern regions of the basin, respectively. The basin has favorable climatic conditions with distinct seasonal turnover and basically the same periods of rain and heat. The land cover types are shown in Figure 1b, which can be categorized into six types, i.e., cropland, forest, shrubland, grassland, water, and impervious surfaces, among which the forest and cropland areas account for a relatively large proportion, (71.26% and 25.6%), and the proportion of grassland area is the lowest, only 0.03%.

3. Data Sources and Research Methodology

3.1. Data Sources and Processing

Various remote sensing data were used in this study, including digital elevation model data, precipitation and air temperature, the leaf area index (LAI), the normalized difference vegetation index (NDVI), GPP, and ET. Among them, the elevation data were obtained from the NASA Earth Science Data Web site (https://nasadaacs.eos.nasa.gov/; accessed on 20 September 2023), and they are high-precision topographic data acquired by ALOSs.

Some data came from https://appeears.earthdatacloud.nasa.gov/, including data on GPP, ET, NDVI, and LAI spanning the years 2001–2020, and the corresponding product names are MOD17A2, MOD16A2, MOD13Q1, and MOD15A2H. Data on GPP, ET, and the LAI are 8-day data with a spatial resolution of 500 m, and the NDVI dataset consists of 16-day data with a spatial resolution of 250 m. The GPP and ET data were synthesized into annual-scale 8-day data using the maximum value synthesis method by Arcgis 10.8, while the NDVI and LAI data were synthesized and processed into annual-scale 8-day data. As the ET data correspond to MOD16A2 in MODIS, the values of some special areas were replaced by 32,762–32,766, and there were no actual ET values, so some areas took a null value.

The VPD data share the same time series as the aforementioned datasets and were derived from the Terra Climate, with a spatial resolution of 0.05°.

Temperature and precipitation data were provided by the Tibetan Plateau Research Institute of the Chinese Academy of Sciences (https://www.tpdc.ac.cn/; accessed on 20 September 2023) with a spatial resolution of 1 km. For the comparative analysis, this research also utilized annual precipitation and temperature data from six meteorological stations in the Li River Basin, spanning the years 2001 to 2020. The data were sourced from actual measurements.

The land cover data were obtained from the annual land cover data with a spatial resolution of 30 m of China from 1990 to 2022 published by Yang [24]. In this study, based on the land cover data in 2020, using Arcgis 10.8, we chose four main land cover types (cropland, forests, shrubland, and grasslands) as the main objects from the seven land cover types.

3.2. Research Methodology

3.2.1. WUE

In this study, WUE is expressed as the ratio of GPP to ET:

$$WUE=GPP/ET$$

(1)

where WUE is water use efficiency, in gC·mm⁻¹·m⁻²; GPP is the gross primary productivity of terrestrial ecosystems, in gC·m⁻²; and ET is ecosystem ET, in mm.

3.2.2. Trend Analysis

The Theil–Sen Median method [25], which is a robust non-parametric statistical method for trend calculation, was used to calculate the trend in WUE in the Li River Basin from 2001 to 2020.

$$\beta =Median{\displaystyle \frac{x_j-x_i}{j-i}}$$

(2)

where β is the trend value of WUE, β > 0 indicates that WUE shows an increasing trend, and β < 0 indicates that vegetation WUE shows a decreasing trend. Median represents the median value; j and i represent the number of time series, and the range of values is 2001 ≤ i < j ≤ 2020. x_j and x_i are the WUE values of the corresponding time series.

3.2.3. Significance Test

The Theil–Sen Median method can be used to calculate the trend in the time series. However, it cannot present the significance level of the trend. Therefore, the Mann–Kendall test [26,27] was used to judge the significance level of the trend in the time series. For the time series {x_i} (i = 1, 2,..., n), the statistic of the trend in change can be defined as Z.

$$V=\left\{\begin{array}{c}{\displaystyle \frac{\mathbf{S}-\mathbf{1}}{{\rho }_s}} \mathbf{S}>\mathbf{0}\\ \mathbf{0} \mathbf{S}=\mathbf{0}\\ {\displaystyle \frac{\mathbf{S}+\mathbf{1}}{{\rho }_s}} \mathbf{S}<\mathbf{0}\end{array}\right.$$

(3)

$$S={{\displaystyle \sum }}_{k=\mathbf{1}}^{n-\mathbf{1}}{{\displaystyle \sum }}_{j=k+\mathbf{1}}^{n}Sgn\left(X_j-X_k\right)$$

(4)

$$Sgn\left(X_j-X_k\right)=\left\{ \begin{array}{c}+\mathbf{1} \left(X_j-X_k\right)>\mathbf{0}\\ \mathbf{0} \left(X_j-X_k\right)=\mathbf{0}\\ -\mathbf{1} \left(X_j-X_k\right)<\mathbf{0}\end{array}\right.$$

(5)

$${\rho }_s=\sqrt{\left(n\left(n-\mathbf{1}\right)\left(\mathbf{2}n+\mathbf{5}\right)-{{\displaystyle \sum }}_{i=\mathbf{1}}^m{t}_i\left(t_i-\mathbf{1}\right)\left(\mathbf{2}{t}_i+5\right)\right)/\mathbf{18}}$$

(6)

where n is the number of elements in the time series, m is the number of repetitive data in the time series, and ti is the number of repetitive data in the ith group. At a given significance level α (99%, 95%, 90%), if |V| ≥ V1 − α/2, this means that the trend is significant; otherwise, the change trend is not significant.

3.2.4. Correlation Analysis

Pearson’s coefficients were used to analyze the correlation between WUE and climate factors as follows.

$$R={\displaystyle \frac{\sum \left(x-\overline{\mathbf{x}}\right)\left(y-\overline{\mathbf{y}}\right)}{\sqrt{\sum {\left(x-\overline{\mathbf{x}}\right)}^{\mathbf{2}}}\sqrt{\sum {\left(\mathbf{y}-\overline{\mathbf{y}}\right)}^{\mathbf{2}}}}}$$

(7)

In the above formula, R is the Pearson coefficient, x represents the climate factors, and y is the WUE.

Partial correlation analysis was used to calculate the partial correlation coefficient between two variables while eliminating the influence of other variables. This analysis can make one of the three variables fixed so as to explore the correlation between the other two variables [28,29]. In this paper, image element-based partial correlation analysis can further investigate the classification of influence of temperature and precipitation on ecosystem WUE, and the calculation formula is as follows:

$${\mathit{R}}_{xyz}={\displaystyle \frac{{\mathit{R}}_{xy}-{\mathit{R}}_{xz}{\mathit{R}}_{yz}}{\sqrt{\left(\mathbf{1}-{\mathit{R}}_{xz}^{\mathbf{2}}\right)\left(\mathbf{1}-{\mathit{R}}_{yz}^{\mathbf{2}}\right)}}}$$

(8)

In Formula (8), x represents the dependent variable (WUE), y and z are the independent variables (climate factors), and R_xyz is the partial correlation coefficient between x and y. In order to calculate the significance level of the correlation, a t-test was used as follows.

$$t={\displaystyle \frac{R\sqrt{n-\mathbf{1}}}{\sqrt{\mathbf{1}-R^{\mathbf{2}}}}}$$

(9)

where n is the number of samples. At given significance levels α1 (p < 0.05) and α2 (p < 0.01), the correlation is significant if t > t_α1, and the correlation is extremely significant if t > t_α2. Otherwise, the correlation is not significant.

4. Results and Analysis

4.1. Characteristics of Spatial Distribution of GPP, ET, and WUE

Figure 2a shows the spatial distribution characteristics of GPP, ET, and WUE in the Li River Basin, and obviously, their spatial distributions are different. The annual average value of GPP was 1655.02 gC·m⁻², and the spatial distribution of GPP showed a decreasing and then increasing trend from the north to the south; among them, the high values were predominantly concentrated in the southeastern of the study area, except for a few distributed in the northern part of the basin, and the low values were mostly distributed in the central part of the basin, except for a few distributed in the southern part of the basin in Pingle.

Figure 2b shows the spatial distribution of ET values, the multi-year average value of ET was 908.25 mm, and the high values were widely and evenly distributed, while the low values were predominantly distributed in the central and western parts of the basin. The ET change trend had a certain similarity with that of GPP, and it also decreased and then increased from the south to the north, but after passing the middle of the basin, the ET values in the south part grew more rapidly.

The average value of WUE in the basin in the past 20 years was 1.8253 gC· mm⁻¹ ·m⁻²; Figure 2c further shows that the spatial distribution of WUE was quite different from that of GPP and ET, and its overall distribution was characterized by high values in the south and east parts and low values in the central part. The high values were concentrated in Pingle on the south side of the basin, with a few distributed in the northern part of the basin. The low values transitioned from the central part to the southeast part, but they were predominantly concentrated in the middle part around the bordering areas of Guilin and Lingui. Regarding GPP, combined with the WUE formula, the increase in ET or the decrease in GPP led to the decrease in WUE.

4.2. Temporal and Spatial Trend Variations of WUE and Various Factors

The Theil–Sen Median method in combination with the Mann–Kendall test can be used for analyzing the trend in the WUE, GPP, and ET values in the basin from 2001 to 2020. The results are shown in Figure 3. The types of changes in the WUE, GPP, and ET values in the basin were classified into nine categories against the trend values and confidence intervals in

Table 1. Classification of WUE trend types.

Trends	M-K Test Confidence Level	Type of Change
	V > 2.58	Extremely significant increase
β > 0	2.58 ≥ V > 1.96	Significant increase
	1.96 ≥ V > 1.65	Slightly significant increase
	V ≥ 1.65	Insignificant increase
β	V	No change
	V ≥ 1.65	Insignificant decrease
	1.96 ≥ V > 1.65	Slightly significant decrease
β < 0	2.58 ≥ V > 1.96	Significant decrease
	V > 2.58	Extremely significant decrease

. Figure 3a shows the trend in spatial and temporal changes in GPP; the overall trend in GPP values in the basin was increasing, with the areas having an increase accounting for about 86.12% of the total area of the basin. The proportions of insignificant, slightly significant, significant, and extremely significant increases were 33.74%, 8.17%, 16.13%, and 28.06%, respectively; the proportion of the areas with almost no change accounted for about 0.1%; and the remaining 13.78% consisted of the areas with significant decreases, with the area proportion of insignificant decrease > significant decrease > extremely significant decrease > slightly significant decrease. In terms of spatial distribution, most of the non-significantly increasing and decreasing areas of GPP were located in the northern part of the basin, with a few areas scattered in the southern part; the significantly increasing areas showed an increasing trend from north to south, which was opposite to that of the significantly decreasing areas.

Figure 3b shows the trend in ET values in the basin over the past 20 years. Similar to the GPP values, the ET values in the basin mostly showed a continuous increase, but the difference was that the area with an extremely significant increase in the ET values accounted for more than that of the same type of GPP values, which accounted for 48.81% of the whole study area, followed by the area with a significant increase, which accounted for 22.79%; the areas with almost no change in ET values in the basin were similar to that of GPP values, accounting for the lowest percentage. The percentage of areas with a decreasing trend were lower than that of GPP, which was only 2.1%. The spatial ET values showed a continuous increasing trend from north to south and reached the highest percentage in the part in Pingle. The areas with significantly decreasing values were mainly concentrated in Lingui, with a few distributed in the urban areas of Guilin and the neighboring junctions.

From Figure 3c, it can be seen that 9.27% of the area had an increasing trend in WUE, about 90.23% of the area had a decreasing trend, and the remaining 0.5% was the unchanged area. A further analysis of Figure 3c showed that the area with an extremely significant decrease was the largest, accounting for 45.8%, while the area with an extremely significant increase was the smallest, accounting for about 0.22%. The areas with decreasing WUE were mainly distributed in the northeast and southeast of the basin, and most of the increasing areas were in the middle and western regions.

Different land cover types in the basin were superimposed on the WUE data of the past years, and the results are shown in Figure 4. The WUE of different land cover types had a decreasing trend. Among them, the decreasing trend in WUE in shrubland was the most significant at 0.008 gC·mm⁻¹ ·m⁻²·a⁻¹, the rate of change in cropland was smaller compared with other land cover types at 0.0049 gC·mm⁻¹ ·m⁻² ·a⁻¹, and the overall WUE of the basin continued to decrease at a rate of 0.0072 gC·mm⁻¹ ·m⁻² ·a⁻¹. The very low WUE values for each land cover type all occurred in 2020; the peak WUE values all occurred in 2007 except for forest WUE in 2006.

4.3. Analysis of the Impact of Various Factors on WUE

4.3.1. Primary Influencing Factor Analysis

The correlation coefficients of WUE with GPP and ET and their spatial distribution characteristics were obtained using partial correlation analysis, as shown in Figure 5a,b. The correlation coefficients of GPP and WUE were positively correlated with those of GPP, ranging from 0.8125 to 0.9996; the correlation coefficients of ET and WUE were the opposite of that of GPP, which were negatively correlated, with correlation coefficients ranging from −0.91014 to −0.9997.

Based on the definition of WUE, the contributions of GPP and ET to the change in WUE were explored to determine which one is the primary influencing factor. The partial correlation coefficients of GPP, ET, and WUE can indicate the relative importance of the two in influencing the change in WUE, and the partial correlation coefficients were further utilized, and the ratios were calculated [30,31] using the following equation:

$$X=\left|{\displaystyle \frac{\mathit{G}\mathit{P}{\mathit{P}}_c}{\mathit{E}{\mathit{T}}_c}}\right|$$

(10)

where GPPc and ETc are the partial correlation coefficients of GPP and ET, respectively, and X is the ratio of the partial correlation coefficients, and it is the absolute value because the partial correlation coefficients are negative in the case of negative correlation. If X > 1, GPP is the primary influencing factor with a larger contribution to WUE; if X < 1, ET is the primary influencing factor influencing the change in WUE.

Based on the spatial distribution of the correlation coefficients of GPP, ET, and WUE in Figure 5a,b, the raster calculator in ArcGIS 10.8 was used, raster calculations were performed on the data results from Figure 5a,b, the absolute value was taken, and the ratio of partial correlation coefficients was obtained, as shown in Figure 5c. In this case, the percentage of the area with the absolute value of the ratio of GPP to ET correlation coefficients higher than 1 was 18.83%, and the percentage of the area lower than 1 was 81.17%. From this, it can be judged that the primary influencing factor of WUE change in the basin was ET, which is consistent with the findings of some scholars [32].

4.3.2. Partial Correlation Analysis of Climate Factors

Figure 6a,b present the meteorological station data and raster data for precipitation and temperature, respectively, in the Li River Basin from 2001 to 2020. For precipitation, the change in meteorological station data is similar with the raster data’s change basically. The precipitation data from meteorological stations exhibit an annual increase at a rate of 56.996 mm/a, while the raster data also show an upward trend at a rate of 19.547 mm/a. A notable difference is observed in 2015 and 2019, where the station data show substantial increases in precipitation, whereas the raster data’s values remain relatively lower. Regarding temperature, the station data for temperature have higher overall values than the raster data. From 2001 to 2018, both datasets exhibit similar trends; however, in 2019, the meteorological station data experience a sudden significant decrease, leading to a downward trend, whereas the raster temperature data maintain an overall upward trend at a rate of 0.0165 °C/a.

The correlation between WUE and climate factors in the basin from 2001 to 2020 is shown in Figure 6. For precipitation, the correlation coefficients of WUE with it ranged from −0.7032 to 0.7234. The proportions of significant positive and insignificant positive correlation areas were 3.67% and 67.22%, respectively, totaling 66.90%; the significant positive correlation areas were predominantly concentrated in the northern part of Xing’an as well as the junction of Linggui and Yangshuo, and the remaining parts were distributed in the southern part of Pingle and the southeastern part of Lingchuan. The proportion of negative correlation area was about 33.10%; almost all of them had an insignificant negative correlation, as the significant negative correlation was only 0.34%. The significant negative correlation areas were small and scattered, and most of them were located in Yangshuo, while the insignificant negative correlation areas were relatively evenly distributed.

Under the condition of controlling the effect of precipitation, Figure 7(b1,b2) show the spatial distribution of correlation coefficients between air temperature and WUE, which are more differentiated than precipitation, with values ranging from −0.809437 to 0.623538. The areas with positive correlation accounted for about 7.33% of the total area, while the percentage of negatively correlated areas was 92.67%, and overall, air temperature was negatively correlated with WUE. Further analysis showed that the percentages of areas with significant and insignificant negative correlations between temperature and WUE were 13.46% and 79.22%, respectively; the proportion of areas with significant negative correlations gradually increased from north to south; and the proportion of areas with insignificant negative correlations were higher and evenly distributed within the basin. Among the positive correlations, the area of significant positive correlation was 0.02%, which only appeared in Lingui and Pingle; the proportion of areas with insignificant positive correlation was relatively high (about 7.30%); and they were predominantly concentrated in Xing’an and Pingle. There was also a certain distribution at the junction of Lingui and Guilin urban areas.

4.3.3. Vegetation Index Correlation Analysis

As an indispensable component of terrestrial ecosystems, surface vegetation is highly sensitive to climate change and plays an important role in regulating biospheric and atmospheric changes, and changes in vegetation phenology can be a good response to the dynamic relationship between terrestrial ecosystems and climate change [33,34,35]. In this paper, the LAI and NDVI from MODIS data were used to analyze and understand their driving role and influence on WUE. The LAI, as an important structural parameter of vegetation, is also one of the indispensable parameters of plant respiration, transpiration, and evaporation in the terrestrial ecological cycling system [36,37], and the magnitude of the LAI influences plant transpiration and soil evaporation, which indirectly leads to fluctuating changes in ecosystem WUE.

Figure 8a,b show the spatial distribution of NDVI and LAI trends, and according to the analysis of the figure, it can be seen that the trends in the NDVI and LAI were more or less the same. Based on the NDVI, 74.13% of the total basin indicated an uptrend, 21.43% of the area showed negative growth, and the area with no change was 4.44%. Compared to the NDVI, the LAI had a higher proportion of area with an increasing trend, which was 80.53%, the area with a decreasing trend was about 19.27% of the total area, and the area with no change was 0.2%. Spatially, the NDVI and LAI showed a decreasing to increasing trend from north to south.

Using correlation analysis in combination with a significance test, the corresponding spatial distribution of correlation coefficients was obtained, as shown in Figure 9. From Figure 9a, it can be seen that the correlation coefficient intervals of WUE and the NDVI ranged from −0.870821 to 0.955953; the negative correlation areas were relatively large, accounting for 66.89% of the total area; and the positive correlation areas accounted for about 33.11%. Among them, the areas with positive correlation and negative correlation through a p < 0.05 significance level test accounted for 2.72% and 7.95% of the total basin, respectively, totaling 10.67%. A further analysis of the correlation coefficients by the p < 0.05 significance level test showed that the areas with negative correlation were mainly concentrated in Pingle in the southern part of the basin, while the areas with positive correlation were concentrated in the central and western part of the basin, with a few distributed in Xing’an on the eastern side of the basin. Figure 9b shows the spatial distribution of correlation coefficients between WUE and the LAI, with values ranging from −0.870821 to 0.955953, and the proportions of positively and negatively correlated areas were 39.11% and 60.89%, respectively. Using the significance test for further analysis, the proportion of areas with p < 0.05 was 14.56% of the total area. Among them, the area with a significant negative correlation was about 8.28%, which was similar to the spatial distribution of the NDVI, and the areas with a significant negative correlation in Pingle accounted for most of the area; the areas with positive correlation at p < 0.05 accounted for about 6.28% of the total area of the basin, which was an increase in comparison with the NDVI, and they were spatially distributed mainly in the middle of the basin.

4.3.4. Vapor Pressure Deficit Correlation Analysis

In the context of current global warming, VPD is also gradually increasing. An excessively high VPD can lead to ecosystem water stress and drought, thereby threatening plant survival. Analyzing the relationship between VPD and WUE is conducive to the long-term sustainability of various ecosystems [38]. Figure 10 illustrates the spatial characteristics of the correlation between annual-scale Vapor Pressure Deficit (VPD) and water use efficiency (WUE) in the Li River Basin from 2001 to 2020. As indicated by this figure, the correlation coefficients between WUE and VPD range from −0.7092 to 0.7341. The areas where VPD and WUE exhibit a positive correlation account for 49.58% of the total, with significant positive correlation and non-significant positive correlation constituting 2.55% and 47.03%, respectively; regions with a significant positive correlation are primarily concentrated in the central part of the basin. The proportion of areas showing a negative correlation is approximately 50.42%, with significant negative correlation slightly lower than significant positive correlation, accounting for about 2.17% of the basin’s area; these regions are mostly located within Xing’an and Pingle counties. Non-significant negative correlation areas are distributed around the perimeter of the basin, predominantly on its northern side.

4.4. Impact of Drought Events on WUE

Based on the annual precipitation and air temperature raster data of the Li River Basin from 2001 to 2020, the SPEI values at the spatial scale were calculated by using the Climta_Indices database in Spyder (Python 3.11) [39]. Figure 11 shows the mean values of the annual-scale SPEI in the Li River Basin for the last 20 years. The analysis shows that the linear tendency rate of the annual SPEI was 0.0326/a, showing a weak upward trend, i.e., the basin tends to become wetter. In terms of drought type and frequency, the total number of years with annual SPEI values lower than −0.5 in the basin was 4 years, which were 2003, 2009, 2011, and 2013; among them, the index value of 2011 was −1.3906, which was a moderate drought within the interval of −1.5 to −1.0. As a result, the total number of annual-scale drought events in the basin from 2001 to 2020 was four, including three mild droughts and one moderate drought.

To further understand the effect of drought events on WUE, correlation analyses were used to obtain the spatial distribution of correlation coefficients between WUE and the SPEI, as shown in Figure 12a,b. The correlation coefficients between them ranged from −0.7333 to 0.6264, and the proportions of positively and negatively correlated areas were 31.69% and 68.31%, respectively, while the proportions of significantly positively and negatively correlated areas were 2.19% and 2.59%. That is, WUE and the SPEI in the basin were mostly significantly negatively correlated, and the WUE value tended to increase with a decrease in the SPEI, and a decrease in the SPEI value showed the occurrence of drought events or the enhancement in the degree of drought, which is in line with the results of the research of some scholars [40,41]. Spatially, the correlation coefficients showed an increasing–decreasing–increasing trend from north to south, with positive correlation areas mainly distributed in the north and south, while the Guilin urban area and its surroundings in the central part of the county were the main concentration areas with negative correlation coefficients.

For different land cover types, the relationship between the SPEI and WUE value points based on the annual scale was plotted and linearly fitted, as shown in Figure 13. According to the analysis of the graph, the SPEI and WUE of different land cover types were negatively correlated, in which grassland WUE showed the strongest negative correlation with the SPEI, with a coefficient of about −0.30, followed by cropland WUE, with a correlation coefficient of −0.23; forest and shrubland WUE decreased with an increase in the SPEI, but the correlation was not significant, and it showed a weak negative correlation, with a coefficient of −0.04.

5. Discussion

The Li River Basin is located in southwest China and possesses a typical karst landscape, and WUE is particularly important for vegetation growth. In this study, the spatial and temporal variations in WUE in the basin from 2001 to 2020 were characterized, and their relationships with precipitation, air temperature, NDVI, LAI, and SPEI were investigated. This study showed that the WUE in the basin showed an overall downtrend at a rate of 0.0072 gC·mm⁻¹ ·m⁻² ·a⁻¹. Based on the analysis of meteorological factors, the effects of air temperature and precipitation on WUE were polarized. The air temperature was positively correlated with WUE, while precipitation was negatively correlated with WUE.

In the past 20 years, the distribution and areas of vegetation in the basin showed a continuous growth trend, which was related to the implementation of local ecological management policies and projects. During the period from the “10th Five-Year Plan” to the “14th Five-Year Plan”, the forest cover of the basin increased from 62.06% in 2000 to 71.87% in 2020. In the case of ET as the dominant factor of WUE in the basin, the ecological management of the basin increased vegetation cover and promoted the transpiration of vegetation, and an increase in ET further promoted a decrease in WUE, which was corroborated by the negative correlation between the NDVI, LAI, and WUE.

In this study, MODIS data were used for estimating the WUE of the Li River Basin from 2001 to 2020. A limitation of our study is that this paper only analyzed the correlation between WUE and air temperature, precipitation, NDVI, LAI, SPEI, etc. However, in addition to the above data, solar radiation, carbon dioxide concentration, etc., also have some influence on WUE. In addition, some researchers conducted a comparative analysis using nine GPP and evapotranspiration ET products against observational data from eight flux towers, revealing significant variations in data outputs among different products when studying diverse ecosystem types [42]. In this study, the data were not measured by us, so there may be some errors, which may lead to some uncertainties in the results of this study. To ensure a further refinement of the experimental results regarding data accuracy, subsequent measurements of raw data and a comparative analysis need to be conducted.

6. Conclusions

In this study, we explored the spatial and temporal evolution of ecosystem WUE in the Li River Basin, 2001–2020, in combination with different analytical tests, to understand the inter-annual trend in WUE and the impacts of different driving factors on the change in WUE. We found the following:

(1)

The average value of WUE in the Li River Basin from 2001 to 2020 was 1.8251 gC·mm⁻¹·m⁻² and showed a decreasing trend at a rate of 0.0072 gC·mm⁻¹·m⁻²·a⁻¹. The WUE of the various land cover types from high to low were forest > shrubland > grassland > cropland; spatially, the south side of the basin was the main distribution area of the high values of WUE, while the low values were mostly located in the central part of the basin.

(2)

By using the correlation analysis method to determine the relationship between different influencing factors and WUE, it was found that ET was the primary influencing factor in the change in WUE; in 67.22% of the total area of the basin, precipitation was positively correlated with WUE; in 92.67% of the total study area, air temperature was negatively correlated with WUE.

(3)

The analysis showed that the NDVI, LAI, and WUE were all negatively correlated with WUE, and the percentages of negatively correlated areas were similar, both between 60 and 70%, and the growth of vegetation inhibited the increase in WUE in the basin to a certain extent. Regarding VPD, the proportions of positive and negative correlation areas with WUE were similar, accounting for 49.58% and 50.42%, respectively.

(4)

Based on the SPEI, we found that the occurrence of drought events and their enhancement resulted in a continuous increase in WUE in the basin; for different vegetation types, the correlation of the SPEI and WUE from strongest to weakest was grassland > cropland > forest > shrubland.