Ecosystem and Ecosystem Services

Evaluating the Trade-offs and Synergies of Ecosystem Services in the Jinsha River Basin

  • CHEN Weiting , 1 ,
  • HU Qiyan 2 ,
  • LIU Fenglian , 1, * ,
  • LIU Yan 1 ,
  • WANG Shu 1
  • 1. Institute of Land & Resources and Sustainable Development, Yunnan University of Finance and Economics, Kunming 650221, China
  • 2. School of International Languages and Cultures, Yunnan University of Finance and Economics, Kunming 650221, China
*LIU Fenglian, E-mail: ; .

Received date: 2023-07-07

  Accepted date: 2023-10-12

  Online published: 2023-12-27

Supported by

The Scientific Research Fund Project of Yunnan Education Department(2021J0592)

The Talent Project of Yunnan University of Finance and Economics(2022D13)


Global climate change and human activities continue to threaten watershed ecosystems. The Jinsha River constitutes the upper reaches of the Yangtze River, so studying its ecosystem services (ES) is of great significance for maintaining ecological security and promoting ecological sustainability in the entire Yangtze River Basin. By using the integrated valuation of ecosystem services and trade-offs (InVEST) models and revised universal soil-loss equation (RUSLE) models, we evaluated five ecosystem services of water yield (WY), habitat quality (HQ), soil retention (SR), food supply (FS), and carbon storage (CS) provided by the Jinsha River Basin ecosystem from 2000 to 2020, as well as their spatial-temporal variations and driving factors. The results show three main features of this system. (1) From 2000 to 2020, each ecosystem service in the Jinsha River Basin exhibited different degrees of fluctuation, except for habitat quality, and each ecosystem service basically showed a spatial distribution pattern of high in the southeast and low in the northwest. (2) There were significant synergistic relationships between CS_SR_HQ and WY_SR_FS, and a significant trade-off between WY_CS. (3) The main driving factors of CS_SR_HQ were net primary productivity (NPP) and land-use type (LU), the main driving factors of WY_SR_FS were annual precipitation (PRE), LU, and rainfall erosivity (R), and the main driving factors of WY_CS varied considerably during the study period.

Cite this article

CHEN Weiting , HU Qiyan , LIU Fenglian , LIU Yan , WANG Shu . Evaluating the Trade-offs and Synergies of Ecosystem Services in the Jinsha River Basin[J]. Journal of Resources and Ecology, 2024 , 15(1) : 15 -32 . DOI: 10.5814/j.issn.1674-764x.2024.01.002

1 Introduction

Ecosystem services (ESs) are the products and services that humans derive directly or indirectly from ecosystems (Costanza et al., 2014), and that are essential for human well-being, health, livelihoods and survival. The United Nations’ Millennium Ecosystem Assessment Plan classifies these services into supply, regulating, cultural, and support services (Reid et al., 2005). There are complex relationships among ecosystem services, which is why the ecosystem services at a particular scale are nonlinear in space and dynamic in time (Nelson et al., 2008; Lee and Lautenbach, 2016). These relationships can be divided into synergistic relationships (synergies) and trade-off relationships (trade- offs). Synergies refer to the simultaneous improvement or weakening of multiple ecosystem services, while trade-offs occur when an improvement of one ecosystem service will weaken other ecosystem services (Foley et al., 2005; Bennett et al., 2009). Precisely because of these relationships, when stakeholders need to improve a specific ecosystem service to meet some demands, they must consider the impacts on other ecosystem services, which makes maximizing the overall benefit of ecosystem services hard to achieve (Vidal-Legaz et al., 2013). In the past half-century, rapid changes in land use/land cover have aggravated the burden on ecosystems, resulting in the degradation of more than 60% of ecosystem services worldwide (Reid et al., 2005). Therefore, analyzing the trade-off and synergistic relationships (trade-offs/synergies) have become critical issues in the field of ecology (Gong et al., 2019).
Existing studies in this field mainly use quantitative modeling, statistical analysis, scenario simulation, and other methods to evaluate the ecosystem service values in specific regions, or to study the correlations between different ecosystem services, their changes on the spatial-temporal scales, and the mechanisms of how different factors influence the trade-offs/synergies of ecosystem services (Wang et al., 2017; Wu et al., 2017; Lyu et al., 2019). Costanza et al. (2014) estimated the global loss of ecosystem services due to land use change and the total value of ecosystem services from 1997 to 2011. El-Hamid et al. (2022) used the CA Markov model to predict land use/land cover (LU/LC) in Damieta, Egypt, and evaluated its impact on ecosystem service value. Jiang et al. (2023) used the production possibility frontier assessment (PPF) model to evaluate the trade-off intensity of ecosystem services in the West Liao River Basin. Zhang et al. (2022) explored the ecosystem services of the Shandong Yellow River Basin by using the InVEST model and the automated linear model (ALM) to identify the key drivers of ecosystem service trade-offs/synergies. These quantitative research methods can directly reflect the correlations between different ecosystem services, but there is a lack of research on the driving factors of such correlations and the consideration of their nonlinear characteristics.
The complex nonlinear relationships of ecosystems and their spatial and temporal heterogeneity greatly hinder the assessment of ecosystem services (Donohue et al., 2013). The unpredictable nature of environmental changes also makes the intrinsic mechanisms of ecosystem services ambiguous (Li and Fang, 2023). Studies have shown that the Bayesian belief network (BBN) model has advantages for studying the nonlinear relationships between multiple factors (Landuyt et al., 2016b; Li et al., 2023), which can facilitate further study on the relationships among ecosystem services. Clarifying and exploring the non-linear relationships and intrinsic mechanisms of ecosystem services can fundamentally improve the function of ecosystem services, make the management of ecosystems more effective, and ultimately realize the sustainable development of human society (Chen and Dou, 2017). The Jinsha River Basin (JRB) is an essential ecological barrier to the Yangtze River Basin, and its ecosystem stability is vital for the ecological security and economic development of the entire Yangtze River Basin (Shang et al., 2021). However, since it is affected by the complex natural conditions, unreasonable land use, and economic development models, the JRB has faced ecological problems such as land degradation and geological disasters for a long time (He et al., 2019; Liu et al., 2021). Therefore, there is an urgent need for research on the ecosystem services in the JRB.
To address this knowledge gap, this study selected the JRB as the study area and analyzed the spatial-temporal distributions and changes of its five ecosystem services: Water Yield (WY), Food Supply (FS), Soil Retention (SR), Habitat Quality (HQ), and Carbon Storage (CS) from 2000 to 2020. In addition, based on the Spearman correlation analysis, the variations in the strength of ES trade-offs/synergies were investigated over a long period. Finally, using the BBN model, the main drivers affecting these trade-offs/ synergies over time were explored. The results revealed how natural, socioeconomic, and other factors influence the strength of ES trade-offs/synergies in the JRB.

2 Materials and methods

2.1 Study area

The Jinsha River (24°N-36°N, 90°E-105°E) is the upper reaches of the Yangtze River (Fig. 1). It originates from the Geladandong Peak of the Tanggula Mountain in Qinghai Province, and flows through Tibet, Yunnan, and Sichuan provinces. The total length of the river’s main stream is 3481 km, with a total basin area of 474600 km², which accounts for about 26% of the Yangtze River Basin area. The terrain of the basin is high in the west and low in the east, with a mean altitude of 2000 m. The climate varies significantly throughout the year, and the spatial-temporal distribution of rainfall is uneven. In the winter half of the year, the basin is affected by the westerly wind flow, and in the summer half of the year, it is affected by the oceanic southwest monsoon and southeast monsoon. The average annual precipitation is 600-800 mm, the average annual temperature is 20-23℃, and the average annual evapotranspiration is about three times the annual precipitation. Due to the complex topography in the JRB, geological hazards such as landslides and debris flows often occur on both sides of the river.
Fig. 1 Location and elevation map of the study area

2.2 Data sources

The data used in this study mainly include land-use data, digital elevation data (DEM), meteorological data (precipitation, evapotranspiration, etc.), socio-economic data, soil characteristics, and other data (Table 1). The land-use data for five periods from 2000 to 2020 were obtained from the Resource and Environmental Science Data Centre of the Chinese Academy of Sciences, with a spatial resolution of 30 m. For the five time periods, Landsat-TM/ETM images were used in 2000, 2005 and 2010, and Landsat 8 OLI images were used in 2015 and 2020. The Kappa coefficient (88.95%) indicates that the interpretation accuracy is high. The DEM data were obtained from the Geospatial Data Cloud platform, and meteorological data were obtained from the National Tibetan Plateau Data Center, with a spatial resolution of 1000 m. The soil data were from the Second National Soil Survey, obtained by the National Tibetan Plateau Data Center (Shangguan et al., 2013).
Table 1 Data sources
Data Resolution Data source Website link
Land-use data 30 m Resource and Environment Science and Data Center (accessed on 2022-10-23)
DEM 90 m Geospatial Data Cloud (accessed on 2022-10-23)
Meteorological data 1000 m National Tibetan Plateau Data Center (including precipitation, evapotranspiration, temperature, etc.) (accessed on 2023-02-15)
Soil data 900 m National Tibetan Plateau Data Center (accessed on 2023-02-15)
NDVI 30 m National Ecosystem Science Data Center (accessed on 2023-01-07)
Population density 1000 m National Earth System Science Data Center (accessed on 2023-01-25)
GDP data 1000 m National Earth System Science Data Center (accessed on 2023-01-25)
Watershed boundary - National Earth System Science Data Center (accessed on 2022-10-19)

2.3 Methods

2.3.1 Ecosystem service evaluation

(1) Water yield
The InVEST Water Yield (WY) model calculates water yield based on a water balance equation that takes into account the effects of many factors, such as surface runoff, soil water content, water holding capacity of litter, and the amount of water received by the tree canopy (Tallis et al., 2011). The formulas are as follows:
${{Y}_{xj}}=\left( 1-\frac{AE{{T}_{xj}}}{{{P}_{x}}} \right)\times {{P}_{x}}$
$\frac{AE{{T}_{xj}}}{{{P}_{x}}}=\frac{1+{{\omega }_{x}}{{R}_{xj}}}{1+{{\omega }_{x}}{{R}_{xj}}+\frac{1}{{{R}_{xj}}}}$
$\begin{align} & VR=V(Q)-V(Q\left| F \right.)=\sum{{}_{q}}p(q)\times \\ & \begin{matrix} {} & {} \\ \end{matrix}{{\left[ {{X}_{q}}-E(Q) \right]}^{2}}-\sum{{}_{q}p(q\left| f \right.)}\times {{\left[ {{X}_{q}}-E(Q\left| F \right.) \right]}^{2}} \\ \end{align}$
${{R}_{xj}}=\frac{{{k}_{xj}}\times E{{T}_{0}}}{{{P}_{x}}}$
$AW{{C}_{x}}=\text{Min}(MS{{D}_{x}},R{{D}_{x}})\times PAW{{C}_{x}}$
where Yxj is the annual water yield (mm) of grid x in land use/land cover (LU/LC) type j; AETxj is the annual actual evapotranspiration (mm) of grid x in LU/LC type j; Px is the annual precipitation (mm) of grid x; ωx is a dimensionless parameter describing climate and soil properties; Rxj is Budyko aridity index; Z is an empirical constant that captures the local precipitation pattern and characteristics; AWCx is the plant available water content (mm) of grid x (determined by soil depth and physicochemical properties); kxj is the vegetation evapotranspiration coefficient of grid x in LU/LC type j; ET0 is the potential evapotranspiration (mm); MSDx and RDx are the maximum soil depth and root depth, respectively; and PAWCx is the plant available water capacity (mm) of grid x.
(2) Carbon storage
The calculation of carbon storage (CS) by the InVEST Carbon Storage and Sequestration model is based on four basic carbon pools: aboveground biomass carbon, belowground biomass carbon, soil carbon, and dead matter carbon. The principle of this model is to multiply the total carbon densities of the different land use/land cover (LU/LC) types by their areas to get the carbon storage of the study area. The carbon contents of the four carbon pools of the different LU/LC types were calculated based on relevant research results. The equations are as follows:
${{C}_{\text{tot}}}=\sum\limits_{i=1}^{n}{{{C}_{i}}}\times {{S}_{i}}$
where Ci is the total carbon density of LU/LC type i (t ha‒1); ${{C}_{i-\text{above}}}$ is the carbon density of aboveground biomass (t ha‒1); ${{C}_{i-\text{below}}}$ is the carbon density of belowground biomass (t ha‒1); ${{C}_{i-\text{soil}}}$ is the carbon density of soil (t ha–1); $E(Q\left| F \right.)$ is the carbon density of dead matter (t ha‒1); ${{C}_{\text{tot}}}$ is the total CS (t); Si is the area of LU/LC type i (ha); and n is the number of LU/LC types.
(3) Habitat quality
The InVEST Habitat Quality (HQ) model evaluates habitat quality by calculating the negative impacts of threat factors on the habitat. The formula is as follows:
${{Q}_{xj}}={{H}_{j}}\times \left[ 1-\left( \frac{D_{xj}^{z}}{D_{xj}^{z}+{{k}^{z}}} \right) \right]$
where Qxj is the habitat quality of grid x in LU/LC type j; Hj is the habitat suitability of LU/LC type j; k is the half-saturation constant; z is the default parameter of the normalization; and Dxj is the degree of habitat degradation of grid x.
(4) Soil retention
Soil Retention (SR) reflects the difference between potential and actual soil erosion (Borselli et al., 2008). The annual SR of the study area is calculated by the RULSE, and the equation is as follows:
$RKLS=R\times K\times LS$
$USLE=R\times K\times LS\times P\times C$
where RKLS denotes the possible soil erosion (t ha–1 yr–1); $USLE$ denotes the actual soil erosion (t ha–1 yr–1); $SR$ denotes soil retention (t ha–1 yr–1); R is the rainfall erosion factor (MJ mm ha–1 h–1 yr–1); K is the soil erosion factor (t h MJ‒1 mm‒1); LS is the slope and length gradient factor; P is the soil and water conservation measure factor; and C is the vegetation cover factor.
(5) Food supply
Food supply (FS) is one of the most essential ecosystem services, which significantly impacts human existence and development. The current studies show a significant linear relationship between food supply and NDVI (Zhao et al., 2012). Based on the statistical yearbooks of the study area, this study calculated the food supply per unit area of agricultural land (including grains, vegetables, and oil-bearing crops) and grassland (including meat and milk), respectively, and multiplied them by the corresponding areas to obtain the total amount of the agricultural and pastoral products. The calculation formula of FS is as follows:
$F{{S}_{i}}=\frac{NDV{{I}_{i}}}{NDV{{I}_{\text{sum}}}}\times {{G}_{\text{sum}}}$
where FSi is the food supply of grid i (t); NDVIi is the NDVI of grid i; $NDV{{I}_{\text{sum}}}$is the sum of the NDVI values of the agricultural land and grassland in the study area; and Gsum is the total amount of the agricultural and pastoral products (t).

2.3.2 Trade-offs and synergistic analyses

Since the correlations between ecosystem services are nonlinear, in order to accurately reflect the relationships between them, we used the Spearman correlation coefficient to analyze the trade-offs/synergies of the five ESs from 2000 to 2020 (Niu et al., 2022; Xu et al., 2023). Firstly, we used the fishnet tool to create a 2 km×2 km sampling grid in ArcGIS 10.6 (using 2-5 times the average LU/LC area as the sample size, which can better reflect the characteristics of the sampled area (Wang et al., 2022)). Then, we used the extract values tool to sample the ES values of different years. The values were recorded in an Excel spreadsheet and imported into the Spearman Analysis Toolkit in R Language for calculation. Finally, we analyzed them and obtained the correlation coefficients and significant values (P values). The equation for the trade-offs/synergies is as follows:
$\rho_{s}=\frac{\sum_{i=1}^{N}\left(R_{i}-\bar{R}\right)\left(S_{i}-\bar{S}\right)}{\left[\sum_{i=1}^{N}\left(R_{i}-\bar{R}\right)^{2} \sum_{i=1}^{N}\left(S_{i}-\bar{S}\right)^{2}\right]^{\frac{1}{2}}}=1-\frac{6 \sum_{i=1}^{N} d_{i}^{2}}{N\left(N^{2}-1\right)}$
where Ri and Si respectively represent the observation ranking of a pair of ES data in grid i; $\overline{R}$ and $\overline{S}$ respectively represent the average ranking of two ES observations; N denotes the total number of observations; and di denotes the ranking difference of a pair of data in grid i.

2.3.3 BBN model

(1) Network development
The Bayesian belief network (BBN) is an extension of the Bayes method, which was proposed by Pearl in 1988 (Pearl, 1988). A BBN consists of two parts.
The directed acyclic graph (DAG) consists of nodes representing variables and arrows connecting these nodes. It determines the dependence and independence of nodes and constructs the possible causalities between them (Hough et al., 2010).
The conditional probability tables (CPTs) represent the strength of the relationship between the parent node and the child node. The states of each node in BBN can be continuous or discrete. The probability of a node in CPT represents the possibility of its specific state. The probability distribution of node X is determined by CPT and the probability distribution of its parent node, and the probability distribution of nodes without parent nodes is its prior probability distribution. The joint probability distribution of BBN can be obtained by multiplying the conditional probabilities of all nodes (Equation (14)). The joint probability distribution can be used to study the probability distribution of a specific node in the network under specific conditions, which can provide a basis for analyzing the driving factors of the trade-offs/synergies of ESs.
$P\left( {{X}_{1}},{{X}_{2}},\cdots,{{X}_{n}} \right)=\prod\limits_{i=1}^{n}{P\left( {{X}_{i}}\left| parenrs({{X}_{i}}) \right. \right)}$
where $P\left( {{X}_{1}},{{X}_{2}},\cdots,{{X}_{n}} \right)$ is a joint probability distribution, $P\left( {{X}_{i}}\left| parenrs({{X}_{i}}) \right. \right)$ is the conditional probability distribution, and $\left( {{X}_{1}},{{X}_{2}},\cdots,{{X}_{n}} \right)$ represent the random variables.
(2) Parameter learning
The trade-offs/synergies of ESs in JRB are affected by many factors. Considering the effects of natural factors and human factors, this study selected 10 variables as the influence factor nodes to create a BBN, including actual evapotranspiration (EVA), annual precipitation (PRE), vegetation coverage (NDVI), net primary productivity (NPP), annual average temperature (TEM), population density (POP), land-use type (LU), slope (SLOP), rainfall erosivity (R), and soil erodibility (K). To ensure the accuracy of the model, except for LU in which its state is still defined by using the classification of data sources, the remaining variables were discretized into four categories (low, medium, high, and ex-high) by the natural breakpoint method in ArcGIS software. In addition, to make the discretization results comparable over the years, we unified the discretization standard of the same type of data according to the results of the natural breaks (Table 2). The discrete variables were input into Netica for parameter learning. This study created a 2 km×2 km fishing net to extract the discrete grid data and imported the 119615 samples that were extracted into the BBN for learning based on Netica.
Table 2 The discretization of network nodes
Node Variable Type States and ranges Unit
WY Water yield Continuous Low: 0-100; medium: 100-200; high: 200-400; ex-high: >400 mm
CS Carbon storage Continuous Low: 0-5; medium: 5-10; high: 10-20; ex-high: 20-40 t
HQ Habitat quality Continuous Low: 0-0.25; medium: 0.25-0.6; high: 0.6-0.8; ex-high: 0.8-1
SR Soil retention Continuous Low: 0-1500; medium: 1500-6000; high: 6000-14000; ex-high: >14000 t
FS Food supply Continuous Low: 0-10; medium: 10-25; high: 25-40; ex-high: >40 t
POP Population density Discrete Low: 0-2000; medium: 2000-10000; high: 10000-40000; ex-high: >40000 person km-2
SLOP Slope Discrete Low: 0-10; medium: 10-20; high: 20-30; ex-high: >30 °
PRE Annual precipitation Discrete Low: 0-400; medium: 400-600; high: 600-800; ex-high: >800 mm
LU Land-use type Discrete Cropland; forest; grassland; water; built-up land; unused land
TEM Annual average temperature Discrete Low: -20-0; medium: 0-5; high: 5-10; ex-high: >10
K Soil erodibility Discrete Low: 0.10-0.24; medium: 0.24-0.26; high: 0.26-0.30; ex-high: >0.30 t h MJ‒1 mm‒1
R Rainfall erosivity Discrete Low: 25-80; medium: 80-130; high: 130-180; ex-high: >180 MJ mm ha‒1 h‒1 yr‒1
EVA Actual evapotranspiration Discrete Low: 0-700; medium: 700-1000; high: 1000-1500; ex-high: >1500 mm
NPP Net primary productivity Discrete Low: 0-200; medium: 200-500; high: 500-800; ex-high: >800 kgC m-2
NDVI Vegetation coverage Discrete Low: 0-0.1; medium: 0.1-0.3; high: 0.3-0.5; ex-high: >0.5
(3) Sensitivity analysis
Sensitivity analysis can be used to compare the effects of different factors on ESs. Based on Netica, the relative importance of different influencing factor nodes to ecosystem service nodes were compared by calculating variance reduction (VR) (Landuyt et al., 2016a). The value of VR reflects the influence of the variable nodes on the ES nodes, so we chose the influence factors that were > 0.5% as the key nodes. The formula is as follows:
$\begin{align} & VR=V(Q)-V(Q\left| F \right.)=\sum{{}_{q}}p(q)\times \\ & \begin{matrix} {} & {} \\ \end{matrix}{{\left[ {{X}_{q}}-E(Q) \right]}^{2}}-\sum{{}_{q}p(q\left| f \right.)}\times {{\left[ {{X}_{q}}-E(Q\left| F \right.) \right]}^{2}} \\ \end{align}$
where VR is variance reduction; V(Q) and E(Q) are the variance and expectation of ecosystem service Q, respectively; $V(Q\left| F \right.)$ and $E(Q\left| F \right.)$ are the variance and expectation of ecosystem service Q when the variables are under condition F; and Xq is the value of the variable X under condition q.
(4) Driving factor analysis
The change in Bayesian probability of an impact factor can infer its driving effect on trade-offs/synergies. A scenario setting was used to observe the changes in the posterior probabilities compared with the prior probabilities of each state of nodes under a specific scenario to determine the key driving factors of the trade-offs/synergies between ESs. According to the correlations between ESs in different years, the scenario will be set to one of four different types: Scenario Ⅰ: For the ES A and B with synergistic relationships, their state will be set to ‘ex-high’ at the same time; Scenario Ⅱ: For the ES A and B with synergistic relationships, their state will be set to ‘low’ at the same time; Scenario Ⅲ: For the ES A and B with trade-off relationships, the state of A will be set to ‘ex-high’ and the state of B will be set to ‘low’; and Scenario IV: For the ES A and B with trade-off relationships, the state of A will be set to ‘low’ and the state of B will be set to ‘ex-high’.

3 Results

3.1 Spatiotemporal evolution of ecosystem services

3.1.1 Water yield

The water yield (WY) of the Jinsha River Basin fluctuates in its temporal distribution and has significant differences in its spatial distribution (Fig. 2). In the five periods from 2000 to 2020, the average WY values were 154.02 mm, 133.99 mm, 114.93 mm, 102.38 mm, and 134.87 mm, respectively. The average WY decreased at first and then increased over time. The total WY decreased by 51.64 mm or 33.53% from 2000 to 2015, and then increased by 32.49 mm or 31.73 % from 2015 to 2020. The spatial distribution of WY shows the characteristics of “high in the southeast and low in the northwest.” The high-value areas are mainly distributed in south Sichuan and northeast Yunnan, and are located at the Sichuan Basin and Yunnan-Guizhou Plateau junction. This area has abundant precipitation, lush vegetation, and good water production and storage capacity.
Fig. 2 Spatial distribution of water yield in Jinsha River Basin (JRB) in 2000, 2005, 2010, 2015 and 2020

Note: The values here represent the water production capacity of a grid rather than the specific amount of water produced.

On the contrary, the water production capacity of southern Qinghai and central Yunnan is weak. The main reason is that southern Qinghai is located in the Qinghai-Tibet Plateau with sparse vegetation and a lack of precipitation, while central Yunnan is located in the western Yunnan-Guizhou Plateau, where there is no shortage of precipitation. However, its high annual average temperature makes the annual evapotranspiration high as well.

3.1.2 Carbon storage

In each period from 2000 to 2020, the total carbon storage (CS) values of JRB were 11.93×109 t, 11.94×109 t, 12.21×109 t, 12.20×109 t, and 12.17×109 t, respectively. In the past two decades, the total CS has increased by 2.4×108 t, for a rate of 2%, showing a fluctuating rising trend (Fig. 3). The spatial distribution of CS changed little, and the high-value areas were concentrated in the central and southern parts of JRB. In addition, the CS values did not change in most areas (90.20%) of the JRB, with only 6.15% of the areas showing an increase in the CS value, and 3.65% showing a decrease. Among them, the areas where CS increased were mainly in the northern and middle parts of JRB, while the areas where CS decreased did not show a significant spatial distribution. The main contribution to the increase in total CS came from changes in the LU/LC because the implementation of soil and water conservation measures in recent years has transformed some bare land into grassland and woodland.
Fig. 3 Spatial distribution of average carbon storage in Jinsha River Basin (JRB) in 2000, 2005, 2010, 2015 and 2020

3.1.3 Habitat quality

The mean habitat quality (HQ) of JRB showed little change from 2000 to 2020, with values of 0.6758, 0.6759, 0.6861, 0.6858, and 0.6849, respectively (Fig. 4). In terms of spatial distribution, HQ showed a gradual increase from west to east. The high-value areas were mainly distributed in the southern part of the JRB (i.e., southern Sichuan and northern Yunnan), where the main LULC type is forest with high biodiversity levels and few threat sources. The low-value areas were mainly concentrated in the northern part of the JRB (i.e., southern Qinghai), which has a high altitude, a vast area of Gobi and bare land, a meager vegetation coverage, and a fragile ecological environment. In addition, by analyzing the areas where HQ has changed, we found that the decrease in HQ mainly occurred in areas with newly reclaimed cultivated land and construction land, while the increase of HQ mainly occurred in the areas where vegetation coverage has risen.
Fig. 4 Spatial distribution of habitat quality in Jinsha River Basin (JRB) in 2000, 2005, 2010, 2015 and 2020

3.1.4 Soil retention

The total soil retention (SR) values of JRB from 2000 to 2020 were 5.90×1010 t, 4.54×1010 t, 6.04×1010 t, 16.82×1010 t, and 7.89×1010 t. The average SR values were 1.4×105 t, 1.08×105 t, 1.44×105 t, 3.98×105 t, and 1.86×105 t (Fig. 5). On the one hand, both the total and average values of SR fluctuate significantly over time, with the fluctuations in some periods, such as 2010 to 2015, even reaching two or three times the previous level. There may be various reasons for these fluctuations. 1) The effect of soil and water conservation projects is affected by the climate in different periods. 2) The expansion of human activities in some specific periods may weaken the effect of soil conservation. 3) The spatial distribution of SR is relatively stable, which is almost highly correlated with the distribution of vegetation cover. The low-value areas were mainly distributed in the lower vegetation coverage area in the upstream of JRB, and the high-value areas were mainly concentrated in the downstream forests.
Fig. 5 Spatial distribution of average soil retention in Jinsha River Basin (JRB) in 2000, 2005, 2010, 2015 and 2020

3.1.5 Food supply

The average food supply (FS) values of JRB from 2000 to 2020 were 17.02 t, 18.00 t, 19.91 t, 21.76 t, and 20.77 t, respectively, showing a slow growth trend in the time domain (Fig. 6). FS increased by 3.75 t in the past two decades, for a growth rate of 22.03%. From the perspective of spatial distribution, the FS in the southeast of JRB was higher than that in the northwest of JRB, and the ratio by which the FS rose with time in the southeast was also greater. These differences are mainly due to regional differences in the food production capacity of JRB. The terrain of the northwest JRB is undulating, and the land cover types are mainly grassland and bare land, which is unsuitable for the growth of food crops. The main food supply here comes from pastoral products. In contrast, there is a vast cultivated land in the southeast of JRB. Compared with the upstream grassland, the food supply from this cultivated land benefits more from the construction of soil and water conservation projects, so the increase in the FS is higher.
Fig. 6 Spatial distribution of average food supply in Jinsha River Basin (JRB) in 2000, 2005, 2010, 2015 and 2020

3.2 Trade-offs and synergies of ecosystem services

The results of the Spearman correlation analysis (Fig. 7) showed that there were different correlations (p<0.01) among Water yield, Soil retention, Habitat quality, Food supply, and Carbon storage in JRB from 2000 to 2020, and their correlations changed with time. In 2000, there were significant trade-offs or synergies between nine pairs of ESs and a neutral relationship between one pair of ESs. Similar figures in the other time periods were eight pairs and two pairs respectively in 2005, nine pairs and one pair in 2010, and nine pairs and one pair in 2015. In 2020, there were significant correlations between each of the ten pairs of ESs.
The WY and SR of JRB showed a significant synergistic relationship over the past two decades, but their correlation fluctuated downward. The correlation between WY and HQ was unstable and showed an alternating pattern. The synergy between WY and FS was significant, but showed the pattern of a fluctuating decrease. WY and CS showed a trade-off relationship, and the degree of trade-off increased at first and then decreased. SR and HQ had a stable synergistic relationship during the study period, and the intensity of synergy did not change significantly. SR and FS showed a synergy in four periods, with a weak correlation. There was a stable synergistic relationship between SR and CS. The correlation between HQ and FS was not significant. HQ and CS showed a synergistic relationship, with the strongest correlation among all the ES pairs. The correlation between FS and CS changed from neutral to a weak trade-off. In conclusion, the correlations between the ESs in JRB were unstable, and there were considerable fluctuations between some of the pairs of ESs.
Fig. 7 Correlations of ecosystem services in Jinsha River Basin (JRB): (a) Correlation index for 2000-2005; (b) Correlation index for 2010-2015; (c) Correlation index for 2020

Note: * p<0.05; ** p<0.01; *** p<0.001.

3.3 Construction of the BBN and accuracy test

After the construction of the BBN (Fig. 8), the accuracy of the BBN model was verified by the error matrix evaluation (Congalton, 1991). A 5 km×5 km fishing net was used for sampling, and 19126 cases were extracted for accuracy testing. Taking the error matrix of WY in 2000 as an example (Table 3), the numbers of correct predictions for states of poor to ex-high are 5201, 4146, 2679 and 1136, respectively, and the prediction accuracy is 68.82%. Furthermore, the prediction accuracies of SR, FS, CS, and HQ in 2000 were 80.47%, 66.00%, 69.25%, and 72.18%, and the overall prediction accuracy was 71.34%. The overall prediction accuracies of the BBN model from 2000 to 2020 were 71.34%, 71.96%, 72.07%, 71.52%, and 70.05%, respectively, indicating that the model has good simulation accuracy in the ES field and can predict the state of ES supply accurately and stably.
Fig. 8 Parameter learning results of the BBN model in 2000
Table 3 Error matrix of the water yield service prediction in 2000
Actual state Prediction state
Low Medium High Ex-high Row sum
Low 5201 1861 52 63 7177
Medium 654 4146 1271 61 6132
High 27 1015 2679 486 4207
Ex-high 4 18 452 1136 1610
Column sum 5886 7040 4454 1746 19126
Water yield accuracy 68.82%

3.4 Factors influencing the ESs

According to the results of Netica sensitivity analysis, the nodes with VR>0.5% were selected as the key influencing factors affecting the ESs. For example, as shown in Table 4, the key nodes of SR in 2000 were R, PRE, K, and LU. The key variables affecting WY were PRE, LU, R, NPP, and EVA. NDVI and LU had considerable influence on FS. The state of CS was affected by NDVI, NPP, TEM, and EVA. The variable nodes that affected HQ were LU, NPP, TEM, and EVA. The key variables affecting the ESs of JRB and their importance orderings are shown in Tables 4 to 6.
Table 4 Results of BBN sensitivity analysis in 2000
Importance ordering SR WY FS CS HQ
Node VR (%) Node VR (%) Node VR (%) Node VR (%) Node VR (%)
1 R 4.11 PRE 10.10 NDVI 7.62 NDVI 10.40 LU 13.30
2 PRE 2.73 R 5.31 LU 3.12 NPP 5.34 NPP 5.16
3 K 0.90 LU 3.76 PRE 0.43 TEM 2.07 TEM 2.28
4 LU 0.88 NPP 0.65 R 0.09 EVA 1.46 EVA 2.23
5 NPP 0.02 EVA 0.64 SLOP 0.05 PRE 0.42 PRE 0.40
6 SLOP 0.01 TEM 0.29 NPP 0.03 R 0.19 R 0.24
7 EVA 0.01 SLOP 0.00 EVA 0.02 K 0.00 POP 0.01
8 POP 0.00 K 0.00 K 0.00 LU 0.00 SLOP 0.00
9 TEM 0.00 POP 0.00 POP 0.00 SLOP 0.00 K 0.00
10 NDVI 0.00 NDVI 0.00 TEM 0.00 POP 0.00 NDVI 0.00
Table 5 Results of BBN sensitivity analysis in 2010
Importance ordering SR WY FS CS HQ
Node VR (%) Node VR (%) Node VR (%) Node VR (%) Node VR (%)
1 R 1.96 PRE 6.93 NDVI 6.22 NDVI 9.21 LU 10.20
2 PRE 1.39 LU 6.38 LU 5.48 NPP 5.28 NPP 5.40
3 LU 0.92 R 4.33 PRE 1.45 TEM 1.92 TEM 2.26
4 K 0.67 EVA 3.92 NPP 0.03 EVA 1.41 EVA 2.22
5 NPP 0.12 TEM 2.34 R 0.21 PRE 0.46 PRE 0.51
6 SLOP 0.03 SLOP 0.12 SLOP 0.11 R 0.22 R 0.26
7 EVA 0.01 NPP 0.10 EVA 0.03 K 0.00 POP 0.01
8 POP 0.00 K 0.01 POP 0.00 SLOP 0.00 NDVI 0.00
9 TEM 0.00 POP 0.00 K 0.00 LU 0.00 K 0.00
10 NDVI 0.00 NDVI 0.00 TEM 0.00 POP 0.00 SLOP 0.00
Table 6 Results of BBN sensitivity analysis in 2020
Importance ordering SR WY FS CS HQ
Node VR (%) Node VR (%) Node VR (%) Node VR (%) Node VR (%)
1 R 3.70 PRE 10.50 NDVI 6.88 NDVI 10.20 LU 9.20
2 PRE 2.90 R 6.72 LU 6.56 NPP 4.05 NPP 6.50
3 LU 1.00 LU 4.22 PRE 1.48 TEM 1.42 TEM 2.75
4 K 0.47 EVA 1.81 R 0.42 EVA 1.06 EVA 2.47
5 NPP 0.30 TEM 1.21 NPP 0.23 PRE 0.40 PRE 0.68
6 SLOP 0.03 NPP 0.37 SLOP 0.06 R 0.18 R 0.35
7 EVA 0.01 SLOP 0.08 EVA 0.01 K 0.00 POP 0.02
8 POP 0.00 POP 0.01 POP 0.00 LU 0.00 K 0.00
9 TEM 0.00 K 0.01 K 0.00 SLOP 0.00 NDVI 0.00
10 NDVI 0.00 NDVI 0.00 TEM 0.00 POP 0.00 SLOP 0.00
In the time domain, SR was greatly affected by PRE, R, and LU in most periods, indicating that rainfall is the main factor causing soil loss in JRB. LU will also have a considerable impact on SR, which proves that human activities, such as forest rehabilitation or urban expansion, will have a positive or negative impact on SR services. For WY, the key influencing factors show that WY is affected not only by rainfall and evapotranspiration but also by vegetation factors in some years. The factors influencing FS are relatively stable in all periods, basically NDVI and LU. The factors influencing CS and HQ are variable, showing that their mechanisms of influence are more complex than those of the other ESs. Besides vegetation factors, they may even be affected by rainfall, temperature, and land use in some years.

3.5 Driving factors affecting changes in trade-offs/synergies

The states of the ES nodes were set according to the rules of scenario setting in 3.3.4 (Table 7), and the probability changes of the 10 influencing factor nodes were observed to determine the key driving factors of trade-offs/synergies of ESs and to analyze their driving mechanisms. The Spearman correlation analysis in 2.2.1 shows that the synergies between CS_SR_HQ and WY_SR_FS, and the trade-off between WY_CS were stable and high intensity during the study period. Therefore, CS_SR_HQ, WY_SR_FS, and WY_CS were selected for the driving factor analysis. Based on the calculation in Netica and the situation of the study area, factors with a sum of probability changes greater than 50 in Scenarios I and II were identified as the driving factors.
Table 7 The setting rules and description of the Bayesian Belief Network in each scenario
Correlation Scenario Description
Synergy Scenario Ⅰ The probability of an ex-high state was set to 100% in the CS, SR and HQ (WY, SR and FS) nodes
Scenario Ⅱ The probability of a low state was set to 100% in the CS, SR and HQ (WY, SR and FS) nodes
Trade-off Scenario Ⅲ The probability of an ex-high state was set to 100% in the WY node, and the probability of a low state was set to 100% in the CS node
Scenario Ⅳ The probability of a low state was set to 100% in the WY node, and the probability of an ex-high state was set to 100% in the CS node

3.5.1 The driving factors of CS_SR_HQ

With the scenario settings of the Netica parameters, the state changes of each node for different variables can be obtained, as shown in Table 8. Since presenting the entire table of probability changes for each year would require a lot of space in the article, they are not all presented here. By summing up the probability changes of all nodes for each variable we can compare the state changes of different variables in a given Scenario (Figs. 9 and 10).
Table 8 The probability change table of the CS_SR_HQ node states in 2000
Scenario Ⅰ (CS_SR_HQ) Scenario Ⅱ (CS_SR_HQ)
A B C D E F Sum of probability changes A B C D E F Sum of probability changes
EVA ‒5.04 ‒15.2 19 1.11 40.35 9.31 13.6 ‒22.9 ‒0.05 45.86
PRE 14 ‒4.6 ‒11.4 2.1 32.1 14.4 4.9 ‒10.6 ‒8.7 38.6
NDVI ‒5.7 ‒12.57 ‒2.7 21 41.97 21.3 ‒12.69 10.3 ‒18.92 63.21*
NPP ‒17.3 ‒10.6 5.1 22.7 55.7* 47.8 ‒26.22 ‒11.58 ‒10.06 95.66*
TEM ‒18.6 ‒2.4 7.5 13.5 42 25.8 ‒6.1 ‒9.48 ‒10.3 51.68
POP ‒0.1 0.01 ‒0.002 0 0.112 ‒0.1 0.05 0.01 0.004 0.164
LU 1.9 31.5 ‒28.2 0.35 ‒0.16 ‒5.4 67.51* ‒1.5 ‒15.2 ‒30.6 ‒0.45 1.13 46.6 95.48*
SLO 1.1 ‒0.8 ‒0.5 0.1 2.5 0.2 0 ‒0.1 ‒0.1 0.4
R 16.6 ‒16.8 ‒7.4 7.6 48.4* 16.1 ‒1.2 ‒8.4 ‒6.5 32.2
K ‒4.9 ‒9.7 1.9 12.66 29.16 1.4 0.1 ‒0.4 ‒1.27 3.17

Note: * represents the top three nodes in the sum of probability changes. In the LU nodes, A represents cropland, B represents forest, C represents grassland, D represents water, E represents built-up land, and F represents unused land. In other nodes, A represents state low, B represents state medium, C represents state high, and D represents state ex-high.

Fig. 9 The sum of the probability changes of each state node of CS_SR_HQ from 2000 to 2020 (Scenario I)

Note: The red font indicates factors with the sum of probability changes greater than 50.

Fig. 10 The sum of the probability changes of each state node of CS_SR_HQ from 2000 to 2020 (Scenario II)

Note: The red font indicates factors with the sum of probability changes greater than 50.

Figures 9 and 10 show that the driving factors of CS_SR_HQ in 2000 were NPP, NDVI, LU, and TEM; in 2005, they were NPP, LU, and EVA; in 2010, they were NPP, LU, PRE, and R; in 2015, they were NPP, R, LU, and PRE; and in 2020, they were NPP, PRE, LU, R, and EVA. During the study period, NPP and LU had high driving effects on CS_SR_HQ. Moreover, PRE, EVA, NDVI, TEM and R also had specific impacts in some research periods.
The results showed that vegetation factors were important for driving the CS_SR_HQ synergy, which was reflected in the state changes of NPP and LU. Regardless of the period, Scenario Ⅰ reflects the increase of State B (forest land), and Scenario Ⅱ reflects the increase of State F (Sandy land, bare land, saline land, etc.). This phenomenon implies that an increase in forest land will simultaneously promote the improvement of CS, SR, and HQ. In contrast, land desertification and salinization will harm all three at the same time. The driving effect of precipitation-related factors on CS_SR_HQ may be due to their positive impact on vegetation factors, thereby indirectly affecting the ES supply.

3.5.2 The driving factors of WY_SR_FS

Figures 11 and 12 show that in Scenario Ⅰ and Scenario Ⅱ, PRE and LU had enormous probability changes in 2000; while in 2005, the probabilities of TEM, NDVI, PRE, EVA and LU changed the most; in 2010, the probabilities of PRE, LU, R changed the most; in 2015, the probabilities of NPP, PRE, R and LU changed the most; and in 2020, the probabilities of PRE, R and LU changed the most. Therefore, WY_SR_FS, PRE, LU, and R are the main driving factors in most periods, and NDVI, EVA, TEM, and NPP will also have certain impacts in some periods.
Fig. 11 The sum of the probability changes of each state node of WY_SR_FS from 2000 to 2020 (Scenario I)

Note: The red font indicates factors with the sum of probability changes greater than 50.

Fig. 12 The sum of the probability changes of each state node of WY_SR_FS from 2000 to 2020 (Scenario II)

Note: The red font indicates factors with the sum of probability changes greater than 50.

These analysis results prove that precipitation and land use changes are the essential factors driving the WY_ SR_FS synergy. Moreover, it is also worth mentioning that in most years of Scenario II, LU reflects the significant increase of State B (forest land), which means that, under some specific conditions, a significant increase in forest land may cause harmful effects on the state of WY_SR_FS. These results show that the impact of forest land on the ecosystem may be two-sided. On the one hand, an increase in forest land positively impacts soil retention, carbon storage, and other aspects. However, on the other hand, a substantial increase in forest land will affect the water content of the region and cause damage to water-sensitive ecosystem services.

3.5.3 The driving factors of WY_CS

Figure 13 and 14 show that the main factors that drove the WY_CS trade-off were PRE and NDVI in 2000; TEM, NPP and EVA in 2005; PRE and EVA in 2010; NPP, NDVI, PRE and R in 2015; and NDVI, PRE and R in 2020. These results show that the driving factors of WY_CS are variable. According to the specific conditions of different periods, temperature and rainfall may affect this group of trade-offs.
Fig. 13 The sum of the probability changes of each state node of WY_CS from 2000 to 2020 (Scenario III)

Note: The red font indicates factors with the sum of probability changes greater than 50.

Fig. 14 The sum of the probability changes of each state node of WY_CS from 2000 to 2020 (Scenario IV)

Note: The red font indicates factors with the sum of probability changes greater than 50.

The trade-off driving mechanism of WY_CS is complicated. Most carbon storage comes from vegetation, which requires heat and water to grow. However, higher temperatures will intensify transpiration, coupled with water uptake by plant growth, thus reducing WY.

4 Discussion

4.1 Precision of ecosystem service simulation

Since JRB flows through several provinces, and the current ES evaluation studies on JRB are not comprehensive, we cannot directly compare our results with existing data or research results. To verify the quality and reliability of the ES evaluation results in this study, the raster data results of WY, CS, HQ, SR, and FS were divided according to provincial administrative divisions and compared with the statistical yearbooks of the provinces through which the JRB flows and the results of related studies. These comparisons indicated no significant differences between them, which proved that the ES simulation results in this study were highly reliable.

4.2 Factors affecting ecosystem services

Frequent human activities have caused severe impacts on the global climate and ecosystem services. Since the catastrophic flood disaster in the Yangtze River in 1998, JRB has carried out many ecological restoration projects (Cui et al., 2004). Nevertheless, from the results of the ES evaluation, except for FS, the other ESs have mostly stayed the same from 2000 to 2020. This consistency is because, on the one hand, people have realized the importance of ecosystem services to the natural environment and have taken many measures to prevent their degradation. On the other hand, population growth has stimulated the demand for food and land. During this period, newly reclaimed cultivated land and newly developed construction land have been increasing. Once these human activities break through the ecosystem’s carrying capacity, they will harm ecosystem services. Land-use type is the key influencing node of all ESs except HQ, which reflects this situation. Extensive research has shown that reasonable human activities can improve the ecological environment (Aneseyee et al., 2022). However, in addition to human activities, ecosystem services are greatly influenced by natural factors such as climate and vegetation cover. Climatic factors such as temperature and rainfall are difficult to change by human activities in specific regions, while vegetation cover factors are easier to change. So, whether in China or other regions, the governments regard increasing vegetation coverage as an important way to improve ecosystem services.

4.3 The relationship between ecosystem services and its driving factors

The interactions between the different ecosystem services of JRB affect their trade-offs/synergies and make these relationships fluctuate in the different periods. During the study period, one pair of ESs in JRB showed synergistic strengthening, while five pairs showed synergistic weakening. The only pair of ESs with a trade-off showed a strengthening trend. The calculation of water yield by the InVEST model includes the consideration of water retention capacity. The ecological restoration project implemented in the JRB in the past few years has enhanced the water retention capacity of plants in the watershed and reduced soil erosion, resulting in a significant synergy between WY_SR_FS. However, the positive impacts of some ecological restoration projects on ecosystem services may be nonlinear. Taking afforestation as an example, too much forest coverage may affect groundwater and thus weaken the supply of WY_SR_FS simultaneously. The synergy among CS_HQ_SR is influenced by the role of vegetation in ecosystems, especially forests. Vegetation is the primary source of carbon storage in the ecosystem, and its roots can prevent soil loss to a certain extent. Moreover, high vegetation coverage can also increase the quality of habitats. The trade-off between WY_CS may be primarily influenced by temperature, and the effects of temperature on them are very different. Studies have shown that an increase in carbon storage is highly related to an increase in temperature (Kong et al., 2020). However, the increase in temperature will also improve evapotranspiration and weaken WY. The trade-off/synergy driving factor analysis results show that LU will significantly change ecosystem service trade-offs/synergies no matter which group of them are analyzed. Therefore, changing the LU is the most direct and effective way to optimize ecosystem services. Specifically, this optimization can be achieved by increasing vegetation coverage to inhibit desertification or reducing the invasion of construction land and farmland into natural habitats. However, the threshold of the vegetation coverage increase needs to be further studied in various regions. Different ecosystem services have different responses to an increase in vegetation cover, and the significantly increasing vegetation cover in some areas may negatively impact some ecosystem services.

4.4 Limitations and future work

Compared with other studies on ESs, this study has some advantages but also some limitations. First, all five ESs evaluated in this study come from the natural supply of ecosystems, ignoring the positive roles of the ecosystems in human culture and entertainment. In future research, cultural and entertainment ecosystem services can be added to make the research between ecosystem services more comprehensive. Secondly, using the natural breakpoint method to classify ESs and influencing factor nodes may lead to some potential uncertainty in running the model (Xue et al., 2017), so there is room for improvement in the variable classification methods. Thirdly, there may be some errors in calculating some of the parameters in the InVEST model. For example, due to the significant relative elevation differences and complex terrain of JRB, the values of some grids will be huge in calculating soil erodibility factor K, and the verification of some model parameters needs to be strengthened in subsequent work. Fourthly, because some InVEST model parameters of JRB cannot be obtained from existing data, some parameters of this study were calculated based on the results of similar studies. For example, the carbon density table from related research was used for the calculation of carbon storage in this study (Jia et al., 2020; Li et al., 2021). In future work, such model parameters could be obtained by field investigation to enhance the reliability of the conclusions. Fifthly, we did not study the relationships between the ESs in the upper and lower watersheds because the contributions of the five ecosystem services studied in this paper to the downstream are difficult to quantify. Finally, this study could benefit from innovation on the spatial scale evaluation of ESs. In the future, ESs could be analyzed based on differences in altitude or land type to deepen the analysis of ES characteristics at different spatial scales.

4.5 Policy implications

From the perspective of spatial scale, the ESs of JRB are significantly different between the east and the west, so different strategies should be adopted for optimizing the ESs. The western region has a worse ecological environment and more serious soil erosion than the eastern region. Thus, the ecological protection and restoration projects should focus on the western part of the basin. However, the local water conditions should be fully investigated before construction, and areas with insufficient groundwater are not suitable for the construction of large-area shelter forests. In the western mountainous areas, steep slopes cause substantial soil loss every year. For these areas, local slope reconstruction or partial grassland coverage can be considered to increase the soil retention capacity.
The forest coverage in the east is significantly higher than in the west, and the ecological conditions are better, but the main urban built-up areas and high-quality cultivated land are also distributed there. Therefore, the key to the ecological protection of the eastern part is to deal with the contradiction between development and protection to prevent the ecosystem degradation caused by urban expansion and unreasonable cultivated land reclamation. Improving the intensive utilization rate of urban land and reducing the amount of steep slope farmland will help to mitigate the deterioration of the ecological environment.

5 Conclusions

In this study, the InVEST model was used to evaluate five ESs (WY, CS, HQ, SR, and FS) in the Jinsha River Basin. Furthermore, we analyzed their temporal-spatial evolution characteristics from 2000 to 2020. The correlations of ESs were analyzed by the Spearman coefficient and visualized by the R language. BBNs were constructed by using Netica, and key nodes affecting ecosystem services were identified through sensitivity analysis. Finally, scenario simulation was used to analyze the driving factors and mechanisms of ES trade-offs/synergies. Based on the analysis in this study, we can suggest several possible ways to improve regional ecosystem services and avoid potential problems. The main conclusions are as follows.
(1) There was significant spatial heterogeneity among the ESs in Jinsha River Basin. The high-value areas of ESs were mainly concentrated in the forests in the southeast of the basin, and the low-value areas of ESs were mainly concentrated in the desert and bare land in the northwest. Temporal heterogeneity was significant in WY, SR, and FS but not in CS and HQ. During 2000-2020, WY decreased by 12.43%; CS increased by 2%; HQ decreased by 1.3%; SR increased by 33.73%; and FS increased by 22%.
(2) Spearman correlation coefficients showed that there were significant synergistic relationships between CS_SR_ HQ and WY_SR_FS, and a significant trade-off between WY_CS.
(3) The results of sensitivity analysis showed that the key variables influencing WY were PRE, EVA, NPP, LU, and R; for CS, they were EVA, TEM, NDVI, and NPP; for HQ, they were EVA, TEM, NPP, and LU; for SR, they were PRE, R, K, and LU; and for FS, they were NDVI and LU.
(4) The scenario simulation results of the BBNs showed that the main factors driving CS_SR_HQ were NPP and LU, the main factors driving WY_SR_FS were PRE, LU, and R, and the main factors driving WY_CS varied considerably during the study periods.
Aneseyee A B, Soromessa T, Elias E, et al. 2022. Evaluation of water provision ecosystem services associated with land use/cover and climate variability in the Winike watershed, Omo Gibe Basin of Ethiopia. Environmental management, 69(2): 367-383.


Bennett E M, Peterson G D, Gordon L J. 2009. Understanding relationships among multiple ecosystem services. Ecology letters, 12(12): 1394-1404.


Borselli L, Cassi P, Torri D. 2008. Prolegomena to sediment and flow connectivity in the landscape: A GIS and field numerical assessment. Catena, 75(3): 268-277.


Chen H P, Dou M. 2017. Study on carbon storage and spatial pattern based on InVEST model in Xiaojiang River Basin of Yunnan Province. Journal of Anhui Agriculture, 45(12): 51-54.

Congalton R G. 1991. A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of Environment, 37(1): 35-46.


Costanza R, D'arge R, De Groot R, et al. 1997. The value of the world’s ecosystem services and natural capital. Nature, 387(6630): 253-260.


Costanza R, De Groot R, Sutton P, et al. 2014. Changes in the global value of ecosystem services. Global Environmental Change, 26: 152-158.


Cui P, Wang D J, Wei F Q. 2004. Study on ecological restoration of Dongchuan Debris Flow Observation Area in Jinsha River dry-hot valley. In:Proceedings of the National Symposium on Ecological Rehabilitation for Soil and Water Conservation, July 2004, Beijing. Beijing, China: Chinese Academy of Sciences. (in Chinese)

Donohue I, Petchey O L, Montoya J M, et al. 2013. On the dimensionality of ecological stability. Ecology Letters, 16(4): 421-429.


El-Hamid H T A, Nour-Eldin H, Rebouh N Y, et al. 2022. Past and future changes of land use/land cover and the potential impact on ecosystem services value of Damietta Governorate, Egypt. Land, 11(12): 2169. DOI: 10.3390/land11122169.

Foley J A, Defries R, Asner G P, et al. 2005. Global consequences of land use. Science, 309(5734): 570-574.


Gong J, Liu D Q, Zhang J X, et al. 2019. Tradeoffs/synergies of multiple ecosystem services based on land use simulation in a mountain-basin area, western China. Ecological Indicators, 99: 283-293.


He J, Xu X P, Zhang Y W, et al. 2019. Spatial-temporal differences of coupling coordination of “Environment-Economy-Society” composite system in river basin: A case study of Jinsha River. Ecological Economy, 35(6): 131-138. (in Chinese)

Hough R L, Towers W, Aalders I. 2010. The risk of peat erosion from climate change: Land management combinations—An assessment with Bayesian belief networks. Human and Ecological Risk Assessment, 16(5): 962-976.


Jia W L, Wu S N, Chen A. 2020. Research on evaluation of ecosystem services in Chishui River Basin based on InVEST. Journal of China Institute of Water Resources and Hydropower Research, 18(4): 313-320. (in Chinese)

Jiang W, Gao G Y, Wu X, et al. 2023. Assessing temporal trade-offs of ecosystem services by production possibility frontiers. Remote Sensing, 15(3): 749. DOI: 10.3390/rs15030749.

Kong R, Zhang Z X, Zhang F Y, et al. 2020. Spatial and temporal dynamics of forest carbon storage and its driving factors in the Yangtze River Basin. Journal of Soil and Water Conservation, 27(4): 60-66.

Landuyt D, Broekx S, Engelen G, et al. 2016a. The importance of uncertainties in scenario analyses—A study on future ecosystem service delivery in Flanders. Science of the Total Environment, 553: 504-518.


Landuyt D, Broekx S, Goethals P L. 2016b. Bayesian belief networks to analyse trade-offs among ecosystem services at the regional scale. Ecological Indicators, 71: 327-335.


Lee H, Lautenbach S. 2016. A quantitative review of relationships between ecosystem services. Ecological Indicators, 66: 340-351.


Li C F, Fang L. 2023. Ecosystem carbon storage assessment based on land use change scenarios: A case study of Yunnan-Guizhou region. Journal of Northwest Forestry University. 38(5): 34-42. (in Chinese)

Li R W, Ye C C, Wang Y, et al. 2021. Carbon storage evaluation and its driving force analysis based on InVEST model in the Tibetan Plateau. Acta Agrestia Sinica, 29(1): 43-51. (in Chinese)

Li T, Liang X Y, Zhang J, et al. 2023. Ecosystem service trade-off and synergy relationship and its driving factor analysis based on Bayesian belief network: A case study of the Loess Plateau in northern Shaanxi Province. Acta Ecologica Sinica, 43(16): 1-14. (in Chinese)


Liu W, Wang M, Zhu S N, et al. 2021. An analysis on chain characteristics of highstand geological disasters in high mountains and extremely high mountains based on optical remote sensing technology: A case study of representative large landslides in upper reach of Jinsha River. The Chinese Journal of Geological Hazard and Control, 32(5): 29-39. (in Chinese)

Lyu R F, Clarke K C, Zhang J M, et al. 2019. Spatial correlations among ecosystem services and their socio-ecological driving factors: A case study in the city belt along the Yellow River in Ningxia, China. Applied Geography, 108: 64-73.


Nelson E, Polasky S, Lewis D J, et al. 2008. Efficiency of incentives to jointly increase carbon sequestration and species conservation on a landscape. Proceedings of the National Academy of Sciences of USA, 105(28): 9471-9476.


Niu H P, Liu M M, Xiao D Y, et al. 2022. Spatio-temporal characteristics of trade-offs and synergies in ecosystem services at watershed and landscape scales: A case analysis of the Yellow River Basin (Henan Section). International Journal of Environmental Research and Public Health, 19(23): 15772. DOI: 10.3390/ijerph192315772.

Pearl J. 1988. Probabilistic reasoning in intelligent systems: Networks of plausible inference. San Francisco, USA: Morgan kaufmann Books.

Reid W V, Mooney H A, Cropper A, et al. 2005. Ecosystems and human well-being-synthesis:A report of the Millennium Ecosystem Assessment. Washington DC, USA: Island Press

Shang J N, Shao H Y, Li F, et al. 2021. Assessment of ecological vulnerability in Jinsha River Basin. Hubei Agricultural Sciences, 60(8): 50-54. (in Chinese)

Shangguan W, Dai Y J, Liu B Y, et al. 2013. A China data set of soil properties for land surface modeling. Journal of Advances in Modeling Earth Systems, 5(2): 212-224.


Tallis H, Ricketts T, Guerry A, et al. 2011. What is InVEST? Palo Alto, California: Nature Capital Project at Stanford University. Viewed on 2023-10-15

Vidal-Legaz B, Martínez-Fernández J, Picón A S, et al. 2013. Trade-offs between maintenance of ecosystem services and socio-economic development in rural mountainous communities in southern Spain: A dynamic simulation approach. Journal of Environmental management, 131: 280-297.


Wang J T, Peng J A, Zhao M Y, et al. 2017. Significant trade-off for the impact of Grain-for-Green Programme on ecosystem services in North- western Yunnan, China. Science of the Total Environment, 574: 57-64.


Wang M, Hu S G, Zhang X B, et al. 2022. Spatio-temporal evolution of landscape ecological risk in oasis cities and towns of arid area: A case study of Zhangye Oasis Township. Acta Ecologica Sinica, 42(14): 5812-5824. (in Chinese)

Wu J X, Zhao Y, Yu C Q, et al. 2017. Land management influences trade-offs and the total supply of ecosystem services in alpine grassland in Tibet, China. Journal of Environmental Management, 193: 70-78.


Xu J, Yang G S, Xu C, et al. 2023. Trade-offs and multi-objective optimization among ecosystem services in headwater catchments: A case study in Tianmu Lake Catchment. Resources and Environment in the Yangtze Basin, 32(1): 62-70. (in Chinese)

Xue J, Gui D W, Lei J Q, et al. 2017. A hybrid Bayesian network approach for trade-offs between environmental flows and agricultural water using dynamic discretization. Advances in Water Resources, 110: 445-458.


Zhang X F, Yang Y, Zhao M H, et al. 2022. Trade-off analyses of multiple ecosystem services and their drivers in the Shandong Yellow River Basin. International Journal of Environmental Research and Public Health, 19(23): 15681. DOI: 10.3390/ijerph192315681.

Zhao W, He Z, He J, et al. 2012. Remote sensing estimation for winter wheat yield in Henan based on the MODIS-NDVI data. Geographical Research, 31(12): 2310-2320. (in Chinese)