Assessment of the Spatiotemporal Impact of Water Conservation on the Qinghai–Tibet Plateau

Abstract

The Qinghai–Tibet Plateau is a proven essential water conservation region in Asia. However, various factors, such as anthropogenic activities, climate, and vegetation significantly affect its water conservation. Along these lines, a deep understanding of the spatiotemporal patterns of water conservation for this plateau and relevant influencing elements is considered of great importance. This paper calculates the water conservation on the Qinghai–Tibet Plateau based on the InVEST model, and given that the evapotranspiration data are an important parameter of the InVEST model, this study selects the mainstream evapotranspiration data to compare the accuracy of the simulated water yield, and also selects the most accurate remote sensing evapotranspiration data examined in the study to carry out the study of water conservation on the Qinghai–Tibet Plateau. Due to the large area of the Qinghai–Tibet Plateau and the various types of climate and ecological zones, this paper analyzes the spatial and temporal variations of water conservation on the Qinghai–Tibet Plateau in each ecological zone and climate zone division and detects the factors affecting water conservation on the Qinghai–Tibet Plateau by using the geo-detector method. From our analysis, the following outcomes are proven: on the Qinghai–Tibet Plateau, (1) the overall water conservation decreased from southeast to northwest; (2) the water conservation of the studied plateau in 1990, 2000, 2010, and 2020 was 656.56, 590.85, 597.4, and 651.85 mm, respectively; (3) precipitation, evapotranspiration, and NDVI exhibited a positive relationship with water conservation; (4) the precipitation factor had the biggest impact on the spatial distinctions of the water resource governance; (5) the above factors are combined with the slope factor and the interaction of each factor to improve water conservation. Our work provides valuable insights for the further implementation of ecological projects with a view to enhancing water resource management methods.

1. Introduction

As water conservation is regarded as one of the key service functions in the ecosystem, it reflects the conditions and capacity of ecosystems to reserve water resources within a spatial and temporal range [1,2]. Moreover, the Qinghai–Tibet Plateau [3], as an important water-conserving area in Asia, has bred several major rivers and is vital for soil and water conservation, wind-breaking, and sand-fixing [4,5]. Due to the arid, anoxic, and alpine climatic conditions, the ecosystem in this region is rather delicate with a poor self-regulatory capacity [6,7,8]. Furthermore, due to climate change and artificial destruction, such as overgrazing, indiscriminate logging, and inappropriate resource extraction methods [9,10,11], the water-conserving capacity of lands has been negatively affected. To preserve the plateau, the Chinese government has performed many ecological projects including returning grazing land to pasture, sand control, and reforestation [12,13,14,15]. Therefore, a scientific understanding of the spatiotemporal patterns of water conservation is a key step toward the enhancement of the water conservation function on the Qinghai–Tibet Plateau.

The hydrological model [16,17], annual runoff [18,19], overall water storage [20,21], and water balance methods [22,23] are widely employed to assess water conservation. Relative to the Soil and Water Assessment Tool (SWAT) [24] and the hydrological model (TerrainLab model) [25], the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST), which originated from the Budyko hydrological framework, is more often employed to evaluate the performance of various environmental protection methods. More specifically, the InVEST model can operate at different levels of complexity with fewer data inputs and higher sensitivity to changes in data availability and system dynamics [26,27]. Quantitative and visual results generated from the InVEST model can also make the decision-making processes associated with natural resources more efficient and reliable. As an important climatic factor related to the regional water and heat balance [28], evapotranspiration data are a vital input parameter during the InVEST model calculation process. Different remote sensing evapotranspiration product data have different levels of accuracy in the alpine grassland-dominated land type in this region [29,30]. Most of the evapotranspiration data used in existing studies were based on data obtained by interpolation of meteorological data [2,31,32]. Nonetheless, the link between the remote sensing evapotranspiration data and the accuracy of water production has yet to be thoroughly studied.

The adoption of reasonable water conservation and protection measures by conducting a scientific analysis of the influencing factors that induce spatiotemporal changes is a top priority in water conservation for government departments. The origins of the spatiotemporal changes in water conservation on the Qinghai–Tibet Plateau were previously reported in the literature by largely examining the aspect of natural and anthropogenic [33,34,35] impacts. As far as the natural aspects are concerned, the majority of the reported works were mostly conducted with univariates, such as climate (precipitation, temperature, and evapotranspiration) [3,32,36,37] or vegetation [38,39]. The studied plateau, however, has a variety of factors that directly affect water conservation [40,41]. On top of that, various climatic factors including precipitation and evapotranspiration directly influence the water-conserving capacity by affecting the water stored in the ecosystem, while vegetation affects climate change through material and energy exchange within the atmosphere [42,43]. Thus, it is vital to explore the integrated influence of both climate and vegetation on the spatial variability of water conservation on the Qinghai–Tibet Plateau. Studies have been conducted to analyze the correlation between single influencing factors such as precipitation, evapotranspiration, soil type, and water conservation [2,31,32]. Nonetheless, quantitative investigation of the interaction of factors influencing water conservation on the Qinghai–Tibet Plateau has yet to be thoroughly studied.

In this paper, the accuracy of three mainstream remote sensing evapotranspiration products is verified, and the remote sensing evapotranspiration data with the highest accuracy are selected for the study. Since the Qinghai–Tibet Plateau is a vast area, this paper analyzes the interdecadal variation of overall water conservation on the Qinghai–Tibet Plateau from 1990 to 2020 based on the InVEST model in each ecological zone and each climatic zone of the plateau. Then the factors influencing water conservation on the Qinghai-Tibet Plateau and the interactions among the influencing factors were detected by using geo-detector method in combination with slope factor and DEM. The goal of this work is to determine the key conservation areas and provide an effective basis for further reaching ecological conservation and construction decisions.

2. Materials and Methods

2.1. Study Area

Located between 73°19′ and 104°47′ east longitude and 26°00′ to 39°47′ north latitude, the Qinghai–Tibet Plateau covers approximately 2.5 million square kilometers, spanning six domestic provinces including Tibet, Sichuan, Yunnan, Qinghai, Gansu, and Xinjiang, and many other lands in neighboring countries, with an average altitude of over 4000 m. As the highest plateau in the globe, it is ranked as the “Roof of the World” (Figure 1). The Tibetan Plateau borders the Pamir Plateau in the west, the Transverse Range in the east, the Himalayas in the south, as well as the Aerjin, Qilian, and Kunlun Mountains in the north.

Figure 1. The position of the Qinghai-Tibetan Plateau and the spatial distribution of main rivers, lakes, glaciers, meteorological, and hydrological stations: (a) climatic regionalization; (b) ecological regionalization.

The ecological and climatic zones of the Qinghai-Tibet Plateau and their corresponding codes are shown in Table 1 and Table 2, respectively.

Table 1. Name and corresponding code of each ecological zone.

Code	Ecological Zone
I12	Agricultural and grassland zone of the Loess Plateau
I15	Ecological area of deciduous and evergreen broad-leaved forest in the Qinba Mountains
I25	Ecological area of evergreen broad-leaved forest in southwest Sichuan and north-central Yunnan
II03	Grassland desertification ecological area in the middle of the Inner Mongolia Plateau
II08	Tarim Basin–eastern Xinjiang desert ecological area
III01	Qilian mountain forest and alpine grassland ecological area
III02	Desert ecological area of the Qaidam Basin
III03	Pamir-Kunlun–Altun alpine desert grassland ecological area
III04	River source area–Gannan alpine meadow grassland ecological area
III05	Alpine desert grassland ecological area of the northern Tibetan Plateau
III06	Ali Mountain warm arid desert ecological area
III07	Cold temperate coniferous forest ecological area in eastern Tibet–western Sichuan
III08	Alpine meadow grassland ecological area in southern Tibet
III09	Seasonal rainforest ecological area of tropical rainforest in southeast Tibet

Note: See this table for the ecoregion codes in the results section.

Table 2. Name and corresponding code of each climate zone.

First-Level Zone Code	First-Level Climate Zone	Second-Level Zone Code	Second-Level Climate Zone
II	Middle Temperate Zone	IIC2	Central Mongolia
		IID1	Menggan
III	South Temperate Zone	IIID1	Nanjiang
		IIIB3	Weihe
IV	North Subtropical Zone	IVA2	Qinba
V	Middle Subtropical Zone	VA3	Sichuan
		VA5	Northern Yunnan
H	Plateau Climate Zone	HD2	Northern Tibet
		HC3	Southern Tibet
		HC2	Central Tibet
		HB2	Changdu
		HA1	Bomi–Western Sichuan
		HVVIVIIA1	Dawang–Chayu
		HC1	Qilian–Qinghai Lake
		HB1	Southern Qinghai
		HD1	Qaidam

The water resources in this region are extremely abundant and include rivers, glaciers, lakes, and other forms of water, which serve as vital sources of moisture for the ecosystem. According to the literature, in recent decades, the annual precipitation has gradually increased throughout this region, with an overall significant rise in temperature and humidification [44,45,46]. As a result, the actual evapotranspiration has increased while potential evapotranspiration has gradually decreased [47]. The Qinghai–Tibet Plateau has a cold climate, large diurnal temperature differences, strong solar radiation, as well as long daylight hours [48]. The seasonal trends are evident in the warm/humid southeast and cold/dry northwest of the region [49,50]. Seasonal permafrost is also widespread, and the most abundant soil type is alpine meadow soil [51,52]. Vegetation types, such as alpine meadows, alpine grasslands, and alpine shrubs, are also distributed over a large area, ranging from southeast to northwest according to the thermal and water conditions, slope orientation, and other factors, with significant zonal differences. On this Plateau, the variation of water conservation has vital implications for climate regulation, ecosystems, and hydrology in China and the world.

2.2. Materials

Based on water balance, various parameters including root depth, potential evapotranspiration, land use, precipitation, and available moisture content of vegetation were combined in this work, and the InVEST model was used to calculate the water yield of the watershed. To achieve water conservation, the water yield was subsequently adjusted using the water conservation formula, which involved the flow rate coefficient, soil-saturated hydraulic conductivity, and topographic index. In the water yield module of the InVEST model, three different types of remote sensing evapotranspiration products for 2020 were selected. Additionally, to verify the accuracy of these products, the accuracy of the water yield was verified. The remote sensing evapotranspiration data with the highest accuracy were also chosen for calculating water conservation. It is worth noting that even if the MOD16A2 product was more accurate, it was not possible to use it during the study period. This is because MOD16A2 has existed since 2001. To maintain accuracy and consistency, the raster data required in the study were sampled at a spatial resolution of 1 km × 1 km. MOD16A2 and China’s 1 km month-by-month potential evapotranspiration dataset were synthesized using matlab2022a software for annual data. The soil-saturated hydraulic conductivity was calculated with SPAW Hydrology software 6.02.0070, which is an ArcSWAT soil database aid allowing analysis of soil moisture characteristics [53,54]. Table 3 presents the specific data sources and descriptions.

Table 3. The dataset source, spatial resolution, and description of this work.

Data Usage Module	Data	Source	Spatial Resolution	Description
InVEST model input parameters	Precipitation	CAS (https://www.resdc.cn/, accessed on 15 January 2023)	1 km × 1 km	With the daily data of meteorological elements at more than 2400 stations nationwide, the spatial interpolation data of meteorological elements for each year from 1960 to 2021 were generated based on the calculation of annual values of each meteorological element based on Anuspl interpolation software
InVEST model input parameters	China-1 km-monthly potential evapotranspiration dataset	(https://data.tpdc.ac.cn/, accessed on 20 January 2023)	0.0083333 (About 1 km)	Hargreaves Potential Evapotranspiration Calculator was used based on the 1 km monthly average, minimum, and maximum temperature data in China
InVEST model input parameters	Average annual PET	TerraClimate (https://www.nature.com/, accessed on 16 January 2023)	5 km × 5 km	Monthly surface water balance dataset generated using water balance model
InVEST model input parameters	MOD16A2	Google Earth Engine (https://code.earthengine.google.com, accessed on 20 January 2023)	500 m × 500 m	Data collected based on the Penman-Monteith equation
InVEST model input parameters	Land use (LULC)	CAS (https://www.resdc.cn/, accessed on 15 January 2023)	1 km × 1 km	Secondary classification of land resources according to their natural attributes
InVEST model input parameters	Soil root depth	HWSD (https://data.apps.fao.org/, accessed on 20 January 2023)		Contains detailed data on maximum root depth (mm), clay content (%), meal content (%), sand content (%), soil capacity (g/cm³), organic matter content (%), etc.
Calculation of the water conservation	Velocity coefficient	USDA-NRCS		It was obtained by multiplying the flow–slope–landscape table from the National Engineering Handbook provided by the USDA-NRCS by 1000
Calculation of the water conservation	DEM	CAS (https://www.resdc.cn/, accessed on 15 January 2023)	90 m × 90 m	TIt is based on the latest SRTM V4.1 data, which is collated and spliced to generate 90 m of sub-provincial data
Influencing factors	Actual evaporation	InVEST model	1 km × 1 km	Derived from the calculation results of the InVEST model water production module
Influencing factors	NDVI	NESDC (http://www.nesdc.org.cn/, accessed on 8 February 2023)	5 km × 5 km	Based on NOAA CDR NDVI data, monthly mean NDVI data for the 1982–2020 growing season (April–October) in the Chinese regions were obtained by averaging the first 15 and last 15 days of each month and then reconstructed using the maximum value synthesis method (MVC)
Partition Data	Climate zone data	CAS (https://www.resdc.cn/, accessed on 7 February 2023)	(Vector data)	It was compiled by the National Meteorological Administration of China in 1978 using climate data from 1951 to 1970
Partition Data	Ecological zone data	(http://www.ecosystem.csdb.cn/, accessed on 7 February 2023)	(Vector data)	Zoning by ecosystem type and natural conditions such as structural feature and geological feature

2.3. Methods

2.3.1. Water Yield

The assessment of the ecosystem services can be quantitatively evaluated through the Integrated Assessment of Ecosystem Services and Tradeoffs (InVEST) model under various scenarios [32,55]. Using the Budyko water–heat balanced framework, the water yield module of the InVEST model can calculate the annual water yield on a regional scale following the water balance principle with input parameters, such as precipitation, potential evapotranspiration, and soil root depth [56]. The formulas can be written as follows:

$${\mathrm{Y}}_{\mathrm{xi}}=\left(1-\frac{{\mathrm{AET}}_{\mathrm{xi}}}{{\mathrm{P}}_{\mathrm{x}}}\right)\times {\mathrm{P}}_{\mathrm{x}}$$

(1)

$$\frac{{\mathrm{AET}}_{\mathrm{x}}}{{\mathrm{P}}_{\mathrm{x}}}=\frac{1+{\mathrm{w}}_{\mathrm{x}}{\mathrm{R}}_{\mathrm{xj}}}{1+{\mathrm{w}}_{\mathrm{x}}{\mathrm{R}}_{\mathrm{xj}}+\left(\frac{1}{{\mathrm{R}}_{\mathrm{xj}}}\right)}$$

(2)

$${\mathrm{w}}_{\mathrm{x}}=\mathrm{Z} \times \frac{{\mathrm{AWC}}_{\mathrm{x}}}{{\mathrm{P}}_{\mathrm{x}}}$$

(3)

where $Y_{xi}$ represents grid x’s water yield depending on the type of land utilization i, $P_x$ refers to unit x’s annual precipitation, ${AET}_{xi}$ denotes grid x’s actual yearly evapotranspiration depending on the type of land utilization i, $R_{xj}$ stands for the Budyko dryness index, $w_x$ states a non-physical variable indicative of soil and climate traits, $w_x$ stands for a linear function ${AWC}_x$ × $N_p$, $N_p$ represents the yearly rainfall event counts, ${AWC}_x$ indicates the available soil moisture content (mm), and Z refers to the Zhang coefficient that displays seasonal precipitation traits, with a value scope of 1–30. The Z-value is regional, and the model parameters are localized based on the study area data and model runs and a comparison with the actual measured data [57].

Soil type is an important factor affecting water balance [58,59]. In the northwestern part of the Qinghai–Tibet Plateau, there are a large number of cold calcium soils (alpine steppe soils) with thin layers and coarse grains. In the southeastern part of the plateau, there are a large number of grass-felt soils (alpine meadow soils) and black-felt soils (subalpine meadow soils). Grass felt soils are generally moist, with dense alpine dwarf grass meadows, which can effectively contain water [60,61]. Sub-alpine meadow soil is slightly milder than alpine meadow soil, rich in nutrients, and has strong water-holding capacity [62,63]. Yellow brown loam is distributed in the southeast edge of Qinghai–Tibet Plateau, with certain water-holding power but weaker than subalpine meadow soil [64,65]. The spatial distribution of soil types on the Qinghai-Tibet Plateau is shown in Figure 2.

Figure 2. Spatial distribution of soil types on the Qinghai–Tibet Plateau.

Plant available water content (PAWC) refers to a non-linear fitted model, taking into account the soil’s texture and organic content [66]. It is one of the input parameters for the InVEST model. The following formula can be written:

$$\begin{array}{c}\mathrm{P}\mathrm{A}\mathrm{W}\mathrm{C}=54.509-0.132\times \left(\mathrm{sand}\right)-0.003\times {\left(\mathrm{sand}\right)}^2-0.055\times \left(\mathrm{silt}\right)-0.006\times {\left(\mathrm{silt}\right)}^2\\ -0.738\times \left(\mathrm{clay}\right)+0.007\times {\left(\mathrm{clay}\right)}^2-2.668\times \left(\mathrm{OM}\right)+0.501\times {\left(\mathrm{OM}\right)}^2\end{array}$$

(4)

In the equation, PAWC indicates the plant available water content, sand represents sand content in the soil (%), silt denotes meal content in the soil (%), clay represents clay content in the soil (%), and OM signifies the organic matter content in the soil (%).

2.3.2. Calculation of Water Conservation

Next, the water yield obtained by the InVEST model was corrected, where the parameters including flow coefficient (velocity), topographic index (TI), and soil-saturated hydraulic conductivity (${\mathrm{K}}_{\mathrm{sat}}$) were selected following the water conservation formula [31]. The relevant computational equation is shown below:

$$\mathrm{Retention}=\mathrm{Min}\left(1,\frac{249}{\mathrm{Velocity}}\right)\times \mathrm{Min}\left(1,\frac{0.9\times \mathrm{TI}}{3}\right)\times \mathrm{Min}\left(1,\frac{{\mathrm{K}}_{\mathrm{sat}}}{300}\right)\times \mathrm{Yield}$$

(5)

In the equation, Retention refers to the water conservation (mm), Velocity represents the flow velocity coefficient, ${\mathrm{K}}_{\mathrm{sat}}$ stands for the soil-saturated hydraulic conductivity (cm/d), Yield means water yield (mm), which results from InVEST model water yield module runs, and TI denotes the topographic index, which was dimensionless and acquired by Formula (6):

$$\mathrm{TI}=\lg \left(\frac{{\mathrm{Drainage}}_{\mathrm{Area}}}{\mathrm{S}{\mathrm{oil}}_{\mathrm{Depth}}\times {\mathrm{Percent}}_{\mathrm{slope}}}\right)$$

(6)

In the equation, ${\mathrm{Drainage}}_{\mathrm{Area}}$ refers to the number of grids in the catchment area, dimensionless, Soil_Depth signifies the soil depth (mm), and Percent_Slope refers to the percentage slope.

2.3.3. Geo-Detector Model and GWR

The geo-detector model is a statistical method for measuring spatial variability and quantifying the driving forces behind it [66]. This work concentrated on the use of ecological detection, factor detection, and interaction to investigate the interdecadal change in water conservation on the Qinghai–Tibet Plateau from 1990 to 2020.

Ecological detection: compare whether the impact of two independent factors (${\mathrm{X}}_1$, ${\mathrm{X}}_2$) on the dependent variable (Y) show notable differences in the spatial distribution, which is represented by the F statistic, as the following expressions were used:

$$F=\frac{N_{X1}\left(N_{X2}-1\right){\mathit{SSW}}_{X1}}{N_{X2}\left(N_{X1}-1\right){\mathit{SSW}}_{X2}}$$

(7)

$${\mathrm{SSW}}_{\mathrm{X}1}=\sum _{\mathrm{h}=1}^{\mathrm{L}1}{\mathrm{N}}_{\mathrm{h}}{\mathsf{\sigma }}_{\mathrm{h}}^2,{\mathrm{SSW}}_{\mathrm{X}2}=\sum _{\mathrm{h}=1}^{\mathrm{L}2}{\mathrm{N}}_{\mathrm{h}}{\mathsf{\sigma }}_{\mathrm{h}}^2$$

(8)

where $N_{X1}$ and $N_{X2}$ stand, respectively, for the sample sizes of variables ${\mathrm{X}}_1$ and ${\mathrm{X}}_2$; ${\mathrm{SSW}}_{\mathrm{X}1}$ and ${\mathrm{SSW}}_{\mathrm{X}2}$ refer to the sums of the intra-stratum variability established by the individual stratification of ${\mathrm{X}}_1$ and ${\mathrm{X}}_2$; and L1 and L2 are indicative of the respective stratification quantities for ${\mathrm{X}}_1$ and ${\mathrm{X}}_2$. Where the null hypothesis $H_0$: ${\mathrm{SSW}}_{\mathrm{X}1}={\mathrm{SSW}}_{\mathrm{X}2}$. If $H_0$ is rejected at a significance level of α, this suggests that there is a significant difference in the effect of the two factors ${\mathrm{X}}_1$ and ${\mathrm{X}}_2$ on the spatial distribution of attribute Y. Factor detection: the main goal is to detect the spatial heterogeneity of the dependent variable (Y) and the influence of the independent variable factor (X) on it. It can be represented by the q value as follows:

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

(9)

$$\mathrm{SSW}=\sum _{\mathrm{h}=1}^{\mathrm{L}}{\mathrm{N}}_{\mathrm{h}}{\mathsf{\sigma }}_{\mathrm{h}}^2,\mathrm{SST}=\mathrm{N}{\mathsf{\sigma }}^2$$

(10)

where h = l, …, L refers to the stratification (Strata) of factor X or variable Y, i.e., partitioning or classification; ${\mathrm{N}}_{\mathrm{h}}$ and N stand individually for the cell quantities in stratum h and the entire region; ${\sigma }_h^2$ and ${\sigma }^2$ represent the respective Y-value variances in stratum h and the entire region; SSW and SST individually indicate the sum of variance within the strata and the entire region; and q takes values from 0 to 1. A larger q value indicates a more significant spatial heterogeneity of Y; if the stratification is generated by the independent variable X, a larger q value indicates that the independent variable quantity X has a stronger explanatory power for attribute Y, and vice versa. In the extreme case, a q value of 1 indicates that factor X completely controls the spatial distribution of Y of the spatial distribution, while a q value of 0 indicates that factor X has no relationship with Y.

Interaction: the interaction between the factors is detected by comparing the univariate q value with the interaction q value to determine the impact on the dependent variable (Y), that is, to assess whether factors ${\mathrm{X}}_1$ and ${\mathrm{X}}_2$ together increase or decrease the explanatory power of the dependent variable Y, or whether the effects of these factors on Y are independent of each other. Table 4 presents the details of these interactions.

Table 4. Types of geo-detection interactions.

Judgment	Interaction
q(${\mathrm{X}}_1\bigcap {\mathrm{X}}_2$) < min [q(${\mathrm{X}}_1$), q(${\mathrm{X}}_2$)]	Non-linear attenuation
min [q(${\mathrm{X}}_1$), q(${\mathrm{X}}_2$)] < q(${\mathrm{X}}_1\bigcap {\mathrm{X}}_2$) < max [q(${\mathrm{X}}_1$), q(${\mathrm{X}}_2$)]	Single-factor non-linear weakening
q(${\mathrm{X}}_1\bigcap {\mathrm{X}}_2$) > max [q(${\mathrm{X}}_1$), q(${\mathrm{X}}_2$)]	Two-factor enhancement
q(${\mathrm{X}}_1\bigcap {\mathrm{X}}_2$) = q(${\mathrm{X}}_1$) + q(${\mathrm{X}}_2$)	Independent
q(${\mathrm{X}}_1\bigcap {\mathrm{X}}_2$) > q(${\mathrm{X}}_1$) + q(${\mathrm{X}}_2$)	Non-linear enhancement

The geographically weighted regression model (GWR) is a local spatial analysis method [67]. It was used in the study to discuss the spatial heterogeneity between the water congeniality and its main drivers [67,68,69]. The model is expressed as follows:

$${\mathrm{y}}_{\mathrm{i}}=\mathsf{\beta }\left({\mathsf{\mu }}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)+{\sum }_{\mathrm{k}=1}^{\mathrm{p}}{\mathsf{\beta }}_{\mathrm{k}}\left({\mathsf{\mu }}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right){\mathrm{x}}_{\mathrm{ik}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(11)

where $\left({\mathsf{\mu }}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$ refers to the spatial position of point i; p represents the number of independent variables; ${\mathrm{y}}_{\mathrm{i}}$, ${\mathrm{x}}_{\mathrm{ik}}$, and ${\mathsf{\epsilon }}_{\mathrm{i}}$ are the dependent variables, independent variables, and random errors, respectively; $\mathsf{\beta }\left({\mathsf{\mu }}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$ is the intercept at point i; and $\mathsf{\beta }\left({\mathsf{\mu }}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$ is the regression coefficient. The flow chart is shown in Figure 3.

Figure 3. Flow chart of the technical approach of this study.

3. Results

3.1. Results Verification

Three types of remote sensing evapotranspiration product data validation in 2020 were used as the dataset for the assessment, whereas the water yield of the Qinghai and Tibet areas was selected for accuracy validation. With reference to the real statistics of total water resources in the China Water Resources Bulletin 2020, the sum of total water resources in Qinghai and Tibet was 560.92 billion cubic meters. Table 5 lists the accuracy of water yield based on the InVEST model, inputting the data of different remote sensing evapotranspiration products. The annual remote sensing evapotranspiration data from the 1 km monthly potential evapotranspiration data in China were the most accurate, with a relative error of 1.57%. Therefore, these remote sensing evapotranspiration data were selected for research on water conservation in the studied plateau.

Table 5. Validation of the water yield accuracy of each type of remote sensing evapotranspiration data in 2020.

Data	Water Yield (108 m3)	Relative Error
China-1 km-monthly potential evapotranspiration dataset	5697.40	1.57%
MOD16A2	5399.48	3.74%
PET dataset TerraClimate	4970.31	11.39%

3.2. Spatiotemporal Variation

3.2.1. Spatial Distribution

The amount of water conservation on the Qinghai–Tibet Plateau decreased from southeast to northwest (Figure 4). In the southeast, the Sanjiangyuan Reserve in particular, which is known to be the “Chinese Water Tower”, is a vital water-conserving area. In the north, the Qaida Basin has extensive desert distribution and low precipitation, while in the west, glaciers and snow cover high-altitude peaks all year round, thus producing very little water and water-conserving capacity, with insignificant variation in spatial distribution. The main climate in this area is characterized by cold, dry, and low temperatures. The 0–20 mm water conservation areas on this plateau were widely distributed, concentrated in the Qaidam Basin desert ecological zone (III02), the southern Tibet alpine meadow-steppe ecological zone (III08), the southwestern part of the tropical rainforest monsoon forest ecotope of southeastern Tibet (III09), part of the geothermal arid desert ecotope of Alishan (III06), and the alpine desert grassland ecological area of northern Tibetan Plateau (III05). Areas with greater than 100 mm water conservation were also mostly distributed in the cold-temperate coniferous forest ecotope of western Sichuan (III07) and the alpine meadow grassland ecotone of Gannan (III04). The spatial distribution was significantly different.

Figure 4. Spatial distribution of water conservation in varying periods: (a) 1990; (b) 2000; (c) 2010; (d) 2020.

3.2.2. Interdecadal Variation

The water conservation of the studied plateau in 1990, 2000, 2010, and 2020 was 656.56, 590.85, 597.4, and 651.85 mm, respectively. The interdecadal variation in the water cover of the Qinghai–Tibet Plateau showed a decreasing and then increasing trend from 1990 to 2020, and then gradually increased after a decline of 65.71 mm in 2000. The interdecadal variation of water conservation in 2020 was roughly the same as that in 1990, with a difference of only 4.71 mm. In addition, the average water content increased by about 2 mm per decade. In different ecological and climatic zones, the water conservation condition exhibited a fluctuating trend (Figure 5). Under the ecoregion division, the ecoregions with greater water conservation as a whole were I25, III07, and III09. Ecoregions I25 and III04 showed a significant increase in water conservation in 2020, reaching 638.91 mm and 485.06 mm, respectively, with a greater water conservation potential. The maximum water conservation in 2000, 2010, and 2020 occurred in III07. However, the maximum water conservation in 1990 occurred in III09, which might be related to the fact that evaporation was less than rainfall [70]. Under the climatic zoning, the overall water conservation is as follows: central subtropical (V), highland climate zone (H), northern subtropical (IV), middle temperate (II), and southern temperate (III). The plateau climate zone (H) spans a wide area, so the water conservation was larger, and its change trend conformed to the overall change in water conservation in the studied region. Although the area of central subtropical (V) was small, the average value was the highest, and the water conservation capacity was relatively large. The maximum value of water conservation was in the central subtropical (V) climate zone.

Figure 5. Water conservation in different periods in each ecoregion and climate zone. (a) Water conservation in each ecoregion; (b) water conservation in each climate zone.

Under the ecoregion division, the precipitation, actual evapotranspiration, NDVI, and water conservation in different time periods are displayed in Figure 6. Except for III04, where actual evapotranspiration and precipitation increased in 2020, reaching 900 mm and 1453 mm, respectively, the trends of precipitation and actual evapotranspiration did not show any significant changes in 1990, 2000, and 2010. NDVI in some ecoregions exhibited significant fluctuations: NDVI in ecoregion III04 showed an increasing trend in interdecadal variation, reaching a maximum value of 0.81 in 2020. In addition, NDVI reached a maximum value in 2020 in ecoregions I25, III01, III03, and III05, with 0.78, 0.73, 0.61, and 0.92, respectively. On the contrary, NDVI in ecoregions III02 and III08 reached the maximum in 2000, with 0.50 and 0.60, respectively. From 2010 to 2020, the water conservation in III04 increased by 76.69 mm, which was the most significant variation, followed by water conservation in the I25 ecological zone which increased by 60.78 mm.

Figure 6. Precipitation, actual evapotranspiration, NDVI, and water conservation in various ecological zones of the Qinghai–Tibet Plateau in different periods.

The precipitation, actual evapotranspiration, NDVI, and water conservation in different periods under climatic zones are illustrated in Figure 7. In the interdecadal variation, in terms of precipitation, climatic zone II increased, reaching its maximum value in 2020 (1027.18 mm). The rest of the climatic zones showed a reduction from 1990 to 2000 and an increase from 2000 to 2020. The maximum value of precipitation was 2385.99 mm in climate zone V. A large amount of precipitation is considered one of the important reasons for the high conservation potential of water in climate zone V. With respect to the actual evapotranspiration, both climate zones II and III showed an increasing trend, reaching a maximum of 708.51 mm and 878.61 mm, respectively, in 2020. In addition, rainfall increased by 222.72 mm in climate zone III compared to 2010, which was the largest increase. Although the actual evapotranspiration increased, water conservation still increased because the rainfall also increased. In 2000, the NDVI maximum occurred in climate zone V, which reached 0.76. In 1990, 2010, and 2020, the NDVI maximum occurred in climate zone H, which reached a maximum of 0.92 in 2020, while climate zones II and III showed a slight decrease in NDVI in 2020. In view of precipitation, actual evapotranspiration, and NDVI factors, precipitation and actual evapotranspiration fluctuated less in climatic zones H, IV, and V. Nonetheless, 2020 witnessed an upward trend in water-conserving capacity, which was mainly due to the improvement of NDVI.

Figure 7. Precipitation, actual evapotranspiration, NDVI, and water conservation in various climatic zones of the Qinghai–Tibet Plateau in different periods.

3.3. Influencing Factors

In recent years, the Qinghai–Tibet Plateau has been experiencing warming. The warming climate has intensified the hydrological cycle, and thus changes in precipitation have important implications for regional ecosystem services [71]. However, while some regions of the Plateau are becoming wetter, some regions are becoming drier, as reflected by the significant increase in actual evapotranspiration on the Plateau since the 1960s [47]. Moreover, actual evapotranspiration plays a key role in the water cycle and is linked to a variety of surface processes [72]. In addition, the vegetation on the Qinghai–Tibet Plateau is very sensitive to climate change [73], and NDVI can reflect the surface vegetation cover status [74,75]. Therefore, in this study, precipitation and actual evapotranspiration were selected as factors for climatic factors, and NDVI was selected as a factor for vegetation factors. A geo-detector model was adopted for detecting the influence factors and their interactions in combination with topographic slope and elevation factors. The findings of the ecological detection revealed that there existed obvious differences in the spatial distribution between the two factors on water conservation in this region, which were further studied in terms of factor detection and interaction.

3.3.1. Single Influencing Factor

The results of the quantitative analysis of the water conservation factors were based on factor detection, and the five factors, including precipitation, actual evapotranspiration, NDVI, slope, and elevation, all passed the significance test, with all p values being zero. Overall, the dominant factor influencing water conservation was precipitation, followed by NDVI, and actual evapotranspiration (Figure 8). Precipitation was also found to have a positive correlation with water conservation [57]. Precipitation gradually decreased from southeast to northwest, coinciding with water-conserving capacity distribution.

Figure 8. Statistical values of the geo-detector factor detection q for different periods.

3.3.2. Interaction

The textual results of the interaction geo-probing demonstrate that the interaction of both factors non-linearly enhanced the influence on water conservation. As illustrated in Figure 9, the darker red color indicates a stronger explanatory power of the correlation between the two factors. Instead, the dark blue color stands for a weak explanatory power. The explanatory power of slope as a univariate is less than 0.1 and is thus neglected and assigned a null value in Figure 9.

Figure 9. Magnitude of the explanatory power of the interaction of the factors in different periods: (a) 1990; (b) 2000; (c) 2010; (d) 2020.

In this work, the correlations between precipitation and the other factors were proved to be all greater than the interactions between the other factors. Although precipitation interacted with other factors nonlinearly to improve water-conserving capacity in this region, the improvement was minimal. The correlation between NDVI and actual evapotranspiration was roughly 2 times the univariate explanatory power of actual evapotranspiration, and roughly 1.3–1.5 times the univariate explanatory power of NDVI. The correlation between the actual evapotranspiration and slope was roughly 1.5 times the univariate explanatory power of actual evapotranspiration. Additionally, the correlation with DEM was roughly 2 times the univariate explanatory power of actual evapotranspiration. The correlation between NDVI and slope was roughly 1.5 times the univariate explanatory power of NDVI. Moreover, the correlation with DEM was roughly 3 times the univariate explanatory power of NDVI. There was a significant variation. Potential vegetation types distributed at high altitudes are more sensitive to climate change on the Qinghai–Tibet Plateau [76,77].

4. Discussion

4.1. Comparison of Remote Sensing Evaporation Data

PET is a very important parameter in both hydrological and climate studies [78]. To this end, in this work, three remote sensing evapotranspiration product data were compared based on the InVEST model. The Chinese 1 km monthly potential evapotranspiration dataset was obtained based on the Chinese 1 km monthly mean, minimum, and maximum temperature datasets using the Hargreaves potential evapotranspiration formulation; the TerraClimate annual potential evapotranspiration is a dataset obtained using a water balance model derived from monthly surface water balance; and the MOD16A2 data product is based on the Penman–Monteith model-based data collection. Among them, China’s 1 km month-by-month potential evapotranspiration dataset provided the highest accuracy in water yield.

The Penman–Monteith model combines energy balance and mass transfer methods, and takes into account the influence of sunlight, air temperature, humidity, and wind speed on evapotranspiration, which is suitable for various climatic environments; the Hargreaves model is a PET estimation method based on temperature and solar radiation, which requires less raw observation data (observation data can also be applied to estimate potential evapotranspiration in areas where data are scarce, such as the Qinghai–Tibet Plateau region, etc.) and only requires temperature measurements (average, minimum, and maximum values) [79]. The PET estimated by the Penman–Monteith model and the Hargreaves model exhibit a large difference in spatial distribution. The two PET estimates are consistent in the lowest PET region but vary considerably in regions with relatively high PET magnitude, while the Hargreaves model performs better than the Penman–Monteith model in estimating PET [80,81]. The incident solar radiation at the Earth’s surface is an important determinant of PET [82]. In general, clearer skies lead to large diurnal variations in temperature, and cloudy skies lead to smaller diurnal variations in temperature. The Hargreaves model, on the other hand, has the advantage of being more sensitive to daily differences in temperature when cloud cover is overhead. The water balance model used for the TerraClimate annual potential evapotranspiration is very simple, and although the overall mean absolute error is improved compared to the coarser resolution dataset, the dataset underestimates data for temperature and evapotranspiration and overestimates data for rainfall in some areas [83], and limited validation in data-sparse areas. Therefore, the Hargreaves model-based estimates of PET for the Tibetan Plateau are more accurate.

4.2. Influence of the Main Driving Forces on the Spatial Heterogeneity of Water Conservation

Precipitation and NDVI are the most dominant drivers of water conservation, which was consistent with the results of the existing works in the literature. Among the climatic factors, precipitation and evapotranspiration can directly influence the water-conserving capacity in a certain region [84,85,86]. Under warm and humid conditions, convective precipitation can increase, bringing up the surface runoff, while the weakening of solar radiation reduces evapotranspiration. Thus, its water-conserving capacity can be improved [87,88]. The Qinghai–Tibet Plateau has diverse vegetation including alpine grasslands, meadows, and so on, all having a vital impact on water conservation on the plateau. In terms of NDVI, vegetation degradation was effectively controlled [89], and the vegetation cover was elevated notably over the past decade, which could be linked to vigorous environmentally friendly actions including the return of grazing on grass [90,91]. This conformed to the findings of the previously reported works in the literature on the vegetation’s upward trend on the Qinghai–Tibet Plateau [91,92]. In areas with identical vegetation coverage, precipitation had the greatest influence on the spatial distribution of water conservation. In addition, the coefficients between the summer NDVI and contemporaneous precipitation showed a positive, negative, and positive band from south to north, which presented a relationship to the improvement of vegetation cover, and vegetation might have changed atmospheric circulation through transpiration and thus produced feedback in the climate [93].

Although the interaction of the precipitation and NDVI can enhance water conservation, there were differences in the impact of the two on the spatial heterogeneity of water conservation in the region. Since the plateau climate zone of the Qinghai–Tibet Plateau accounts for 96.79% of the whole Qinghai–Tibet Plateau, this work mainly examined ecological sub-regions. The influence of precipitation and NDVI on the spatial heterogeneity of water conservation in the agricultural and grassland zone of the Loess Plateau (I12), the ecological area of evergreen broad-leaved forest in southwest Sichuan and north-central Yunnan (I25), and the alpine desert grassland ecological area of the northern Tibetan Plateau (III05) differ greatly, with spatially significant differences in the vegetation status in the I12 ecoregion, a fragile ecological environment, and a more prominent influence of climate on water conservation [94,95]. The I25 ecological zone belongs to the plateau-type semi-humid subtropical monsoon climate, with obvious differentiation of water and heat conditions and an obvious vertical band spectrum of zonal vegetation distribution. Therefore, the NDVI influence on water conservation is more significant, and the water conservation power is stronger. The water conservation area of the III05 ecoregion is sparse and belongs to the highland cold and sub-cold arid climate zone, which is cold and dry, whereas the climatic factors play a dominant role in the water conservation of this ecoregion [96]. The influence of precipitation and NDVI on the spatial heterogeneity of water conservation is the opposite in the alpine meadow grassland ecological area in southern Tibet (III08) and the seasonal rainforest ecological area of the tropical rainforest in southeast Tibet (III09). The climate of ecological zone III08 is complex, although vegetation is abundant. However, the water conservation of this ecological zone is more influenced by climatic factors. There are large areas of primary forests in the III09 ecoregion, with high vegetation diversity and a complete vertical band spectrum, and there is little anthropogenic interference in this ecoregion. Therefore, the water conservation capacity is strong, and NDVI plays a dominant role in the water conservation of this ecoregion. The influence of precipitation and NDVI on the spatial heterogeneity of the water conservation in the desert ecological area of Qaidam Basin (III02), the river source area–Gannan alpine meadow grassland ecological area (III04), and the cold temperate coniferous forest ecological area in the eastern Tibet–western Sichuan (III07) ecological zones are all greater, which further improves water conservation under the interaction effect. Ecological zone III02 belongs to the highland continental desert climate, with low precipitation and sparse vegetation. Although both precipitation and NDVI have a greater influence on water conservation, their conditions are not conducive to water conservation. The large area of the Gannan grassland in ecological zone III04 and the high annual rainfall make it a humid zone [97], so the water conservation amount is higher. The overall spatial variability of water conservation in ecological zone III07 is significant, and the climate varies vertically. Coniferous forests are also weak in evaporation, highly adaptable to wet, dry, or cold climatic conditions, and have a strong ability to trap precipitation [98]. Consequently, the water conservation capacity is strong. Geographically weighted regression coefficients for rainfall and NDVI versus water conservation are shown in Figure 10.

Figure 10. Geographically weighted regression coefficients for rainfall and NDVI versus water conservation.

4.3. Limitations of the InVEST Model and Perspectives

The InVEST model used in this work provides a feasible method for quantifying water conservation on the Qinghai–Tibet Plateau [19], which is suitable for large-scale studies, takes into account natural factors such as climate and vegetation, and has the advantage of spatial visualization [99]. However, the InVEST model simplifies the simulation of hydrological models and ignores the interaction between surface water and groundwater and the influence of human activities [99,100]. Since the InVEST model calculates a value of 0 for water production in the lake area [101] and does not consider glacial meltwater and freeze-thaw factors, which are widely distributed in the Qinghai–Tibet Plateau [32,102], it also cannot accurately simulate the effects of freeze-thaw on water conservation and the water conservation in the lake area. In future studies, other variables affecting hydrological processes (land cover [103], soil type [104], etc.) need to be further considered to establish a more accurate water conservation assessment system. Therefore, the model may need to be improved or extended in future studies [31,105].

5. Conclusions

To conclude, in this work, a foundation for ecosystem enhancement on the Qinghai–Tibet Plateau was provided by analyzing the spatiotemporal difference in water conservation on the plateau, as well as the associated influential factors. In this paper, we compared the mainstream remote sensing evapotranspiration products based on the InVEST model according to the division of the different ecological and climatic zones of the plateau and selected the data with the highest accuracy of simulated water yield to quantitatively study the spatial and temporal evolution patterns of water conservation on the Qinghai–Tibet Plateau in 1990, 2000, 2010, and 2020. The interaction study was carried out and compared with the single-influence study, and the ecological and climatic zones with high water content were identified. From our analysis, it was proven that (1) the water conservation of the Qinghai–Tibet Plateau decreases from southeast to northwest, and the interdecadal variation showed a reduction in water conservation from 1990 to 2000, and an increase from 2000 to 2020. (2) Precipitation greatly affects water conservation in this area, and the interaction with NDVI and the actual evapotranspiration factor non-linearly enhances water conservation. (3) The northern mountainous indeciduous broad-leaved forest ecozone (I25) in central Yunnan, cold-temperate coniferous forest ecozone (III07) in western Sichuan, and southeast Tibetan tropical rainforest monsoon forest ecoregion (III09) exhibited high water conservation. (4) The central subtropical (V) climate zone exhibited the highest water conservation capacity, followed by the Plateau climate zone (H), which covers the widest region.

To achieve effective restoration of the studied grasslands, greater stability of the ecosystem, and improved water-conserving capacity, it is essential to thoroughly investigate the correlation between the Qinghai–Tibet Plateau and nature. Moreover, the water-conserving function of the Qinghai–Tibet Plateau should be maximized under the conditions that the natural environment can bear. An important contribution to global climate change and water conservation was made by our work by focusing on achieving a win-win situation for both sustainable economic and ecological development.