Exploring the Impact of Climate Variables and Scenario Simulation on Ecosystem Service Value Profits and Losses in China

Abstract

Ecosystem restoration can yield multiple benefits, and the quantitative accounting of ecosystem service value (ESV) profits and losses is of significant importance to the economic benefits of ecosystem restoration. This study reveals the dynamic impacts of climate change on ESVs by analyzing the effects of climate variables on ESV profits and losses across different periods and scenarios. The research findings are as follows: (1) From 1990 to 2020, and extending to simulated projections for 2030, China’s ESV exhibits a high distribution pattern in the southern regions. In 2030, under the natural development scenario (NDS), the southwestern region shows a coexistence of high and low ESVs. Under the ecological protection scenario (EPS), ESV in the southwestern region increases, whereas under the urban development scenario (UDS), ESV in the southwest decreases. (2) In both the NDS and UDS, the trends in ESV profits and losses continue from 2010 to 2020. Under the EPS, there is a significant increase in ESV in the southwestern region. The largest contributors to ESV loss are the conversion of grassland to unused land and forest to farmland. The southwestern region shows the most significant spatial differences in ESV profits and losses, with an increase in ESV profits in the northeastern region. In contrast, other regions show no significant spatial differences in ESV profits and losses. (3) From 1990 to 2000, Bio13 (the precipitation of the wettest month) and Bio12 (annual precipitation) had a significant positive impact on ESV profits and losses, indicating that increased precipitation promotes the functioning of ESVs. This study indicates that fluctuations in precipitation and temperature are significant climate factors influencing the value of ESV. Due to climate change, precipitation patterns and temperature swings are now key determinants of ESV changes. By carefully studying ESV profits and losses and their driving factors, this research can serve as the scientific basis for ecosystem restoration and management strategies.

1. Introduction

Ecosystem services (ESs) serve as an essential link between human society and the natural environment, acting as a catalyst to promote sustainable development [1]. One key indicator used to assess ecosystem health and function is its ecosystem service value (ESV) [2]. ESV refers to all tangible and intangible benefits that ecosystems provide to human society, such as cleaner air and water purification systems, recreation opportunities, and recreational services [3]. Their supply is essential to economic and societal progress [4]. Climate change and pollution have greatly diminished ES supplies; human activities now exceed ecosystem carrying capacities [5], necessitating the quantification of ESV profits and losses as an essential step toward effective ecosystem restoration and management [6]. Environmental degradation, pollution, climate change, and ecological degradation have had a devastating impact on ESV’s profits and losses, with climate change being particularly significant [7]. Climate variables such as precipitation and temperature exert considerable influence on the functionality of ESs [8]. Climate change can lead to notable changes in ESV across different periods and scenarios [9]. For instance, increased precipitation may enhance ecosystem functionality by boosting water resource availability and promoting plant growth, thereby improving both productivity and stability in an ecosystem [10]. Conversely, temperature fluctuations can negatively impact ESV by increasing plant transpiration and soil moisture evaporation rates, reducing water resource availability [11]. Climate change is widely recognized as a major determinant of ESV changes, as the stability and variability of precipitation and temperature are crucial factors [12]. Moreover, the effects of climate change are not limited to these variables; other environmental factors, such as wind speed and humidity levels, also play a role [13]. Extreme weather events like flooding, drought, and heatwaves have downstream and lasting impacts on ecosystems, significantly altering their structure and function and subsequently affecting ESV [14]. For example, flooding can cause soil erosion and deteriorate water quality, threatening purification services; droughts reduce water supply, negatively impacting agricultural production and drinking water availability; heat waves affect both human health and biodiversity. These events, driven by climate change, necessitate further research to develop effective adaptation and mitigation strategies [15].

To better comprehend and predict the impact of climate change on ESV, many scholars have conducted extensive research [16]. Watson et al. (2020) used general circulation models to simulate 20 climate scenarios with projected shifts in Koppen–Geiger climate classifications. They combined global ESVs with 20 climate scenarios and utilized these global ESVs as indicators to track ESV distribution during transitions among climate categories [17]. Yang et al. (2021) also explored this subject matter extensively [18]. Wilkes et al. (2024) assessed climate change’s impacts on biodiversity and ecosystem services in China, concluding that climate change is having a substantial effect on productivity, species interactions, biological invasions in northern agropastoral ecotones, as well as fragile ecosystems of China [19]. Asmus et al. (2019) created a framework to investigate species’ vulnerability to climate change, their roles in food webs, contributions to ESs, and overall persistence despite climate-induced species loss [20]. Leal Filho et al. (2021) studied the effects of climate change on African ESV through case studies from different African countries in order to gain clarity on its destructive impact [21]. Furthermore, Zhang et al. (2022) created a network between climate forcings and ecological systems in order to analyze how climate change is driving change within ecologies [22]. Sharma et al. (2023) pointed out that climate change and biodiversity loss pose significant systemic risks to humanity and proposed an integrated approach to addressing climate change through the circular economy [23]. Rahim et al. (2024) studied potential ESs for climate change adaptation [24]. However, these studies mainly focus on climate scenario simulations and biodiversity assessments on a global or specific regional scale, lacking detailed dynamic analysis of ESV profits and losses in China. In contrast, our study not only reveals the dynamic impacts of climate change on ESV within China but also analyzes the key roles of specific climate variables, such as precipitation and temperature fluctuations on ESV. We found that the precipitation of the wettest month (Bio13) and annual precipitation (Bio12) had a significant positive impact on ESV profits and losses during the period from 1990 to 2000, while temperature seasonality (Bio4) showed a negative impact. These findings are consistent with global trends observed in previous studies but provide more specific regional data. Additionally, we discovered that in future scenario simulations, ESV significantly increases under the ecological protection scenario (EPS), whereas it decreases under the urban development scenario (UDS). These findings align with those of Wilkes et al. (2024) in Northern China but further detail the spatiotemporal variations under different scenarios [19]. The innovation of this study lies in its detailed analysis of the specific impacts of climate change on ESV profits and losses across different regions and scenarios within China, filling a gap in previous research that lacked detailed regional analyses. By employing various scenario simulation methods, our study not only validates the conclusions of previous research but also provides more localized scientific evidence. This evidence offers more targeted guidance for future ecosystem restoration and management strategies. The results of our study will provide policymakers with scientific evidence to help them formulate more effective climate change adaptation and mitigation strategies, promoting ecological sustainability.

While studies based on historical data can provide a theoretical foundation for current land planning and ecosystem management, predicting future ESV changes based on existing land use and analyzing the impact of climate factors on ESV changes can offer early warnings. This can prompt us to preemptively mitigate potential future risks and help us take more robust actions in the future. Our study employs several advanced methods, including the PLUS (Patch-Generating Land Use Simulation) model, LSTM (Long Short-Term Memory) model, and SHAP (Shapley Additive Explanations) method, to dynamically analyze the impact of climate change on historical and future ESV profits and losses. The PLUS model combines the random forest algorithm and Markov Chain model, enabling the precise capture of the spatial heterogeneity and temporal dynamics of LUCs. Lu et al. (2022) utilized the PLUS model to simulate future land spatial patterns under different development scenarios in the lower Yellow River, revealing the spatiotemporal differentiation of ESV and providing references for optimizing land use and ecological protection [25]. Zhu et al. (2024) used the PLUS model to analyze the driving factors of ecosystem function changes, identifying gross domestic product (GDP) and population density as the main social factors influencing wetland area changes [26]. Zhang et al. (2024) integrated the InVEST-PLUS models to evaluate and predict changes in ecosystem carbon storage under different land use scenarios in the Chengdu metropolitan area of Central and Western China by 2050 [27]. Evidently, applying the PLUS model not only simulates LUCs under different future scenarios but also provides data support for formulating rational land use planning. However, these studies primarily focus on simulating LUCs without deeply exploring the specific impacts of climate variables such as precipitation and temperature fluctuations on ESV. Our study not only employs the PLUS model but also integrates the LSTM model and SHAP method to offer a more comprehensive analysis of the dynamic impact of climate change on ESV, thereby filling this research gap.

In handling time series data, the LSTM model excels. LSTM is a special recurrent neural network that effectively addresses the gradient vanishing and exploding problems in traditional RNNs for long time-series data by introducing forget gates, input gates, and output gates [28]. Our study utilizes the LSTM model to analyze the time series data of climate variables, providing more accurate dynamic analysis by predicting ESV changes under future climate scenarios. Deng et al. (2024) constructed four LSTM-based models, LSTM, Stack-LSTM, Bi-LSTM, and CNN-LSTM, to assess hydrological changes under future climate change [29]. Yang et al. (2023) explored the feasibility of improving flow simulation performance in the tropical Kelantan River Basin (KRB) of Peninsular Malaysia by combining the hydrological model SWAT (Soil and Water Assessment Tool) with the deep learning model Bi-LSTM (Bidirectional Long Short-Term Memory) [30]. To interpret the outputs of the LSTM model [31], this study employs the SHAP method. SHAP can decompose the model’s output into contributions from each input feature, revealing the independent impact and interactions of each feature on the prediction results [32]. In this study, the SHAP method is used to interpret the outputs of the LSTM model, helping to identify the independent impacts and interactions of climate variables on ESV changes. Huang et al. (2024) transformed four meteorological features—precipitation, temperature, relative humidity, and wind speed—into six hydrological features and used an LSTM-SHAP composite model to analyze the potential relationships between feature variables and the target variable (streamflow), predicting daily streamflow in the Sandu River Basin (SRB) [33]. Sun et al. (2024) established a regionally applicable multi-output LSTM prediction model to determine the performance of semi-rigid asphalt pavement and conducted a stochastic analysis of the model based on SHAP, providing references for preventative maintenance decision making for China’s highway network [34]. By comprehensively employing these methods, we not only verify the conclusions of previous research but also provide more localized data support, revealing regional differences and the specific impacts of climate factors on ESV, which are rarely seen in the existing literature [35]. The application of the PLUS model [36], LSTM model [37], and SHAP method [38] allows for a comprehensive and systematic simulation and interpretation of the impacts of climate change on ESV profits and losses. The PLUS model is used to simulate LUCs, the LSTM model predicts the dynamic impacts of climate variables on ESV, and the SHAP method interprets the model’s prediction results. The combined application of these methods not only validates prior research conclusions but also provides localized data support, offering more targeted guidance for future ecosystem restoration and management strategies. The uniqueness of this study lies in its detailed analysis of the specific impacts of climate change on ESV profits and losses across different regions and scenarios within China, filling a gap in regional detailed analysis. Using multi-scenario simulation, this study provides scientific evidence for policymakers to help formulate more effective climate change adaptation and mitigation strategies, promoting ecological sustainability.

By simulating ESV profits and losses under different climate scenarios from 1990 to 2030, this study aims to better predict the potential impacts of future climate change on ESV profits and losses. The research will address the following scientific questions: (1) What is the spatiotemporal heterogeneity of ESV profits and losses in China under different times and scenarios? (2) Which climate factors have the greatest impact on ESV under different climate scenarios? (3) How can multi-scenario climate change simulations provide scientific evidence for future ecosystem restoration and management? By gaining an in-depth understanding of the impact of climate change on ESV profits and losses, this study can provide more precise guidance for ecosystem restoration and management. It will provide policymakers with scientific evidence to facilitate the formulation of more effective climate change adaptation and mitigation strategies. The research results not only reveal the dynamic impacts of climate change on ESV profits and losses but also offer scientific evidence for future ecosystem management and policymaking. The ultimate goal is to provide pathways for achieving ecological sustainability.

2. Materials and Methods

2.1. Study Area

China is a vast country (Figure 1) with a land area of 9.6 million square kilometers. The country spans several climatic zones from north to south, including subrigid, temperate, warm temperate, subtropical, and tropical zones. China’s climate types vary widely, ranging from the subrigid monsoon climate in the northeast to the tropical monsoon climate in the south. These regions exhibit significant climatic differences, with uneven distributions of precipitation and temperature. China’s ecosystems include forests, grasslands, wetlands, deserts, and farmlands that serve essential functions like water conservation, soil retention, biodiversity protection, and climate regulation [39]. Over recent years, climate change’s impact on China’s ecosystems has become ever more obvious. Alterations to precipitation patterns and more frequent extreme weather events threaten their stability and sustainability, endangering their very existence as ES. Human activities, including urbanization, agricultural expansion, and industrial development, have had profound effects on ecosystems, leading to habitat fragmentation, biodiversity loss and degradation in ecosystem functions. This research seeks to reveal the dynamic impacts of climate change on ESV profits and losses and provide scientific evidence for ecosystem restoration and management.

2.2. Data Sources

We carefully selected high-quality data from reputable sources (

Table 1. Details of the data and its pre-processing steps.

Data Type	Data Source	Format	Resolution	Time Range	Pre-Processing Steps
Land use data	Chinese Academy of Sciences Resource and Environment Science Data Center (http://www.resdc.cn/ (accessed on 20 January 2024))	Raster	1 km	1990, 2000, 2010, 2020	Reclassified into six categories: farmland, forest, grassland, water, built-up land, and unused land using ArcGIS 10.8.
Agricultural statistics data	China Rural Statistical Yearbook (https://www.stats.gov.cn/ (accessed on 20 January 2024))	Table	-	1990, 2000, 2010, 2020	Standardized data processing, including grain production, prices, and planting areas.
Digital elevation model	http://www.gscloud.cn/ (accessed on 20 January 2024)	Raster	1 km	-	Generated slope data, unified resolution to 1 km.
Annual precipitation	http://www.geodata.cn/ (accessed on 20 January 2024)	Raster	1 km	-	Resampled data, unified resolution to 1 km.
Annual mean temperature	http://www.geodata.cn/ (accessed on 20 January 2024)	Raster	1 km	-	Resampled data, unified resolution to 1 km.
Population density	http://www.resdc.cn/ (accessed on 20 January 2024)	Raster	1 km	-	Resampled data, unified resolution to 1 km.
Regional average gross domestic product	http://www.resdc.cn/ (accessed on 20 January 2024)	Raster	1 km	-	Resampled data, unified resolution to 1 km.
River data	http://www.resdc.cn/ (accessed on 20 January 2024)	Vector	-	-	-
Major road data	http://www.resdc.cn/ (accessed on 20 January 2024)	Vector	-	-	-
Railway data	http://www.resdc.cn/ (accessed on 20 January 2024)	Vector	-	-	-
Distance to rivers, major roads, and railways	-	Raster	1 km	-	Calculated Euclidean distance using ArcGIS 10.8 based on 1 km × 1 km grid to determine the distance from each grid centroid to the nearest river, road, and railway.

). The land use data were reclassified into 6 categories (farmland, forest, grassland, water, built-up land, and unused land) using ArcGIS Desktop 10.8. This ensures consistency and comparability across different periods (1990, 2000, 2010, and 2020). The agricultural statistics data are standardized to ensure consistency and accuracy, including grain production, prices, and planting areas for the years 1990, 2000, 2010, and 2020. Digital elevation model (DEM) data were processed to generate slope data, unified to 1 km resolution to match other datasets. Annual precipitation and the average temperature data were resampled to 1 km resolution to ensure consistency across different climate variables. Population density and regional average GDP data were also resampled to 1 km resolution to ensure consistency. Vector data for rivers, roads, and railways were used to calculate the Euclidean distance from each 1 km × 1 km grid center to the nearest river, road, and railway using ArcGIS 10.8. This method provides a robust representation of distances, which is crucial for accurately analyzing spatial relationships.

This study analyzed the impact of climate variables on ESV profits and losses and applied the SHAP method to reveal the nonlinear effects of climate variables on ESV profits and losses changes. To ensure the stability and reliability of the analysis, we utilized high-resolution climate data from WorldClim (https://worldclim.org/data/worldclim21.html (accessed on 20 January 2024)), provided in raster format with 1 km resolution (Supplementary Material S1). To avoid the influence of multicollinearity, we calculated the correlation coefficients between variables and set a threshold to exclude highly correlated variables. Additionally, we removed variables with low contribution rankings to simplify the model and enhance its interpretability and predictive capability. We ultimately retained the following climate variables: Bio2 (mean diurnal range), Bio4 (temperature seasonality), Bio12 (annual precipitation), Bio13 (the precipitation of the wettest month), Bio15 (precipitation seasonality), Bio16 (the precipitation of the wettest quarter), Bio18 (the precipitation of the warmest quarter), and Bio19 (the precipitation of the coldest quarter).

2.3. Calculation of the Ecological Contribution of LUC to ESV Profits and Losses

We adopt the standard equivalent factor value calculation method proposed by Xie Gaodi et al. [40,41], using the net profit of grain production per unit area of farmland as the ESV of one standard equivalent factor. Rice, wheat, corn, and soybeans are the four main grains in China [42]. Considering the fluctuations in the value and planting area of these four grains across different years, we use the average net profit of these four grains from 1990 to 2020 as the standard equivalent factor value in this study. The calculation formula is as follows [43]:

$$V_a={\displaystyle \frac{1}{7}}\sum _{i=1}^n{\displaystyle \frac{a_i{p}_i{q}_i}{A}}(i=\mathrm{1,2},\cdots ,n)$$

(1)

$${VE}_{ij}=C_{ij}{V}_a (i, j=\mathrm{1,2},\cdots ,n)$$

(2)

$$ESV=\sum A_k{E}_k$$

(3)

where V_a represents the economic value of crops per unit area in China; i represents the type of crop, p_i represents the current year’s price for the ith crop; q_i represents the yield per unit area of the jth crop; a_i represents the total planting area of the ith crop; and A represents the total planting area of the four types of crops. VE_ij represents the ESV coefficient of the jth ecosystem service function contained in the ith ecosystem; C_ij represents the economic value of the jth service in the ith ecosystem relative to per unit area of farmland; V_a represents the economic value generated by crops per unit area; ESV represents the total ESV; A_k represents the area of the kth land type; and E_k represents the ESV corresponding to per unit area of the kth land type. The revised ESVs are shown in

Table 2. Revised ecosystem service value of different land use types (USD/hm²).

Function	Farmland	Forest	Grassland	Water	Built-Up Land	Unused Land
FP	175.78	40.17	37.12	69.47	0.00	0.80
RM	38.97	92.27	54.62	38.71	0.00	2.39
WS	−207.60	47.72	30.23	691.47	0.00	1.59
GR	141.58	303.45	191.96	151.13	0.00	10.34
CR	73.97	907.95	507.47	340.96	0.00	7.95
ED	21.48	266.06	167.57	493.68	0.00	32.61
HA	237.83	594.17	371.72	7084.40	0.00	19.09
SC	82.72	369.47	233.85	171.81	0.00	11.93
NC	24.66	28.24	18.03	13.26	0.00	0.80
BD	27.04	336.46	212.64	553.07	0.00	11.14
AL	11.93	147.55	93.86	355.81	0.00	4.77
Aggregate	628.37	3133.49	1919.04	9963.76	0.00	103.40

Note: FP: food production; RM: raw material production; WS: water supply; GR: gas regulation; CR: climate regulation; ED: environment depuration; HA: hydrological adjusting; SC: soil conservation; NC: nutrient cycle maintenance; BD: biodiversity; AL: esthetic landscape.

.

Furthermore, the following formulas are used to calculate the profits and losses of ESV and the ecological contribution of LUC to these profits and losses. The specific formulas are as follows [44,45]:

$${LP}_{ik}=\sum A_{ij}\times ({VC}_{ik}-{VC}_{ik}^{\prime })$$

(4)

$${EL}_{ij}={\displaystyle \frac{\left({VC}_{ik}-{VC}_{ik}^{\prime }\right)\times A_{ij}}{\sum _{i=1}^n\sum _{j=1}^n\left[\left.\left({VC}_{ik}-{VC}_{ik}^{\prime }\right)\times A_{ij}\right]\right.}}\times 100\%$$

(5)

where LP_ik is the value profit and loss of ESV after Class i land is converted to Class k land. A_ij represents the area of Class i land converted into Class j land. ${VC}_{ik}$ − ${ VC}_{ik}^{\prime }$ is the ESV equivalent of the kth ESV of the ith land use type before and after the land use transition, respectively.

2.4. Multi-Scenario Land Use Projections for 2030

The PLUS model is an LUC simulation tool that combines the random forest algorithm and Markov Chain, capable of accurately capturing the spatial heterogeneity of land use changes [46,47]. This method is not only suitable for predicting future LUCs but also provides scientific evidence for land use planning under different scenarios [36]. Specifically (Figure 2), the PLUS model first extracts land expansion information from two phases of land use data and generates development probabilities for various land use types using the random forest algorithm combined with driving factors (such as distance to major rivers, railways, and main roads, elevation, slope, per capita GDP, population density, potential evapotranspiration, and annual average rainfall). To select the driving factors reasonably, this study, based on existing research and prior knowledge, chose the 7 most significant driving factors from over 20 potential factors. These driving factors and their processing and application in the PLUS model are shown in

Table 3. Rationality of driver selection and its application in the Patch-Generating Land Use Simulation model.

Driving Factor	Rationale	Processing and Application in PLUS Model
Population density	Population density is a key driver of urbanization and land use change; high-density areas experience significant land changes [49,50].	Input to random forest algorithm to generate development probabilities for different land use types.
Regional average gross domestic product	Economic development level directly influences land use; areas with high GDP are more likely to undergo land development [51,52].	Input to random forest algorithm to generate development probabilities for different land use types.
Digital elevation model	Elevation affects the distribution of land types; low elevation areas are more suitable for agriculture and construction [53,54].	Input to random forest algorithm to generate development probabilities for different land use types.
Slope	Slope is a crucial determinant of hydrological characteristics, affecting land usability [55,56].	Data input to random forest algorithm to generate development probabilities for different land use types.
Annual precipitation	Precipitation affects ecosystem functions and biodiversity; high precipitation areas are usually forests and grasslands [57].	Data input to random forest algorithm to generate development probabilities for different land use types.
Distance to major rivers	Proximity to rivers influences water resource availability; areas close to rivers are suitable for agriculture and construction [58].	Used in random forest algorithm to generate land development probabilities, simulating dependence on rivers.
Distance to major roads	Proximity to roads affects accessibility and economic activity; areas with high traffic accessibility are easier to develop [59].	Used in random forest algorithm to generate land development probabilities, simulating the impact of transportation accessibility.

. Then, the Markov Chain is used to predict the total number of pixels for future land use, serving as a constraint for land use layout. Finally, in the CARS (Conversion of Land Use and its Effects at Small Regional Scales) module [48], we set the total number of future pixels, cost matrix, neighborhood weight, and related parameters to simulate and predict the spatial distribution of various land use types.

To validate the accuracy of the model, we used the 2020 land use data as a basis and employed the PLUS model to simulate the land use situation in 2030. Additionally, we simulated the 2020 land use situation using the 2010 land use data and compared the simulation results with the actual data from 2020. The results showed a Kappa value of 0.9734 and an FOM value of 0.9218, indicating that the model has very high simulation accuracy. This provides a solid foundation for future land use predictions in this study. In this study, we set up three scenarios (natural development scenario [NDS], ecological protection scenario [EPS], and urban development scenario [UDS]), each representing different land use strategies and environmental planning. The detailed settings of each scenario are provided in Supplementary Table S2. In the conversion cost matrix, a value of 1 indicates permitted conversion, while a value of 0 indicates restricted conversion (

Table 4. Conversion cost matrix for multi-scenario land projections.

Type	Natural Development Scenario						Ecological Protection Scenario						Urban Development Scenario
	I	II	III	IV	V	VI	I	II	III	IV	V	VI	I	II	III	IV	V	VI
I	1	1	1	1	1	1	1	0	0	0	0	0	1	1	1	1	1	1
II	1	1	1	1	1	1	0	1	0	0	0	0	1	1	1	1	1	1
III	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1
IV	0	0	0	1	0	0	0	0	0	1	0	0	0	0	0	1	0	0
V	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0	0	1	0
VI	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1

Note: I, II, III, IV, V, and VI represent farmland, forest, grassland, water, built-up land and unused land respectively.

). All scenario predictions use the same driving factors to ensure the comparability and scientific validity of the results.

2.5. Deep Learning Simulation of the Impact of Climate Variables on ESV Profits and Losses

In this study, we used the LSTM (Long Short-Term Memory) model to simulate the impact of climate variables on ESV profits and losses. LSTM combines three gating mechanisms, forget gate, input gate, and output gate, which effectively regulate the flow of information and prevent the issues of gradient explosion and vanishing, thereby modeling time series features more accurately [60]. We applied the LSTM model to the climate variables in China, using 19 climate variables from the Shared Socioeconomic Pathway (SSP) 126 scenario for the period 2021–2040, to capture their impact on ESV profits and losses. The SSP126 scenario represents a low greenhouse gas emission pathway aligned with sustainable development goals. These 19 climate variables, including temperature and precipitation, were downloaded from the WorldClim database, and previously described variables were utilized in this analysis. In the implementation process, we used Python 3.8 along with its deep learning libraries TensorFlow and Keras to build and train the LSTM model [61]. The input for the LSTM model consists of climate variable data, and the output is the ESV profits and losses. The model is trained using past data through supervised learning, optimizing the model parameters by minimizing the error between the predicted values and the actual values. To help readers better understand the architecture of the LSTM model, Figure 3 illustrates the schematic of the LSTM neural network we used, including the specific operations of the forget gate, input gate, and output gate.

To better interpret the outputs of our LSTM model, we utilized the SHAP method [62]. It decomposes feature importance into individual effects and interactions with other features, thus providing explanations for black-box models [63]. SHAP is locally accurate and consistent, making it suitable for interpreting various machine learning models, including LSTM. In this study, we combined the LSTM model with the SHAP method (LSTM-SHAP) to reveal the nonlinear relationships between climate drivers and ESV profits and losses. Through the LSTM-SHAP method [34], we were able to identify the independent effects of individual climate variables on ESV changes and analyze their interactions.

In terms of implementation, we used Python 3.8 and the SHAP library [64], combined with the LSTM model, to predict the contribution proportions of climate-driving factors (Figure 4). The specific steps are as follows: First, the time series data of climate variables are input into the LSTM model. The LSTM model performs temporal prediction on these climate variables and outputs the contribution proportions of each climate-driving factor to ESV profits and losses. Then, we use the SHAP toolkit to calculate the Shapley values for each input variable, a process based on the SHAP formula:

$$SHAP\left(x_i\right)=\sum _{S\subseteq N\backslash \left\{i\right\}}{\displaystyle \frac{\left|S\right|!(n-\left|S\right|-1)!}{n!}}\left[f\left(S\bigcup \left\{i\right\}\right)-f\left(S\right)\right]$$

(6)

Here, x_i represents each climate variable, S is a subset of the remaining variables in the model, and f(S) is the predicted value of the model output for the subset of variables. n indicates the size of the current subset or the number of specific variables. N denotes the total number of all variables. This formula calculates the marginal contribution of each climate variable to the changes in ESV.

3. Results

3.1. Analysis of Driving Factors in Scenario Forecasting of LUC in China

Based on the existing research results and prior knowledge, we selected 7 driving factors from more than 20 possible factors. The importance of these seven driving factors is shown in Figure 5. By comparing the contribution of driving factors to various LUCs, we found differences in the main factors affecting various LUCs (Figure 6). For farmland and built-up land, population density is the most important factor driving LUC. The high population density means more frequent human activities, which promote the development of urbanization, the utilization of farmland, and the conversion of farmland into built-up land. For grassland and forests, rainfall is the most important transformation driver. Natural resource-rich areas are often accompanied by adequate rainfall, which is crucial in maintaining grassland and forest biodiversity, soil health, and landscape ecological value. The slope is the main driving factor of water transformation, which is more conducive to forming lakes and wetlands in low-lying areas. GDP mainly drives the transformation of unused land. Economic development promotes the development of unused land and turns it into built-up land.

3.2. Multi-Scenario Simulation of LUC in China from 1990 to 2030

According to the spatial–temporal pattern evolution of land use (Figure 7), farmland is mainly distributed in the eastern and northeastern regions of China. The forest is mainly distributed in the south and southeast of China, and a large amount of unused land, including sandy and bare land, is widely distributed in the northwest of China. Grassland, built-up land, and water areas are dispersed. From 1990 to 2020, unused land significantly expanded in Central and Eastern China, and built-up land showed an aggregation trend in Eastern China. In contrast, the changing trend of other types of land was not obvious.

Compared with the information shown in Figure 7d, clear temporal and spatial changes occur in all types of land under different scenarios in 2030 (Figure 7e). Under the NDS, the density of unused land increased, and grassland decreased in Southwest China. The amount of built-up land increased in the eastern region, but there was no obvious spatiotemporal variation in other directions. This suggests that under the NDS, China’s urban construction is likely to achieve further development in 2030. In the EPS (Figure 7f), the forest density in Southwest China increased, the built-up land density in Eastern China was lower than that in the NDS, and there was no significant variation in other directions. In the UDS (Figure 7g), the density of unused land increased, and the density of grassland decreased in Southwest China. The density of built-up land in Eastern China increased further than in other scenarios. There was no significant variation in other directions.

In the different scenarios, while different types of land use showed spatial differentiation, the amount of land use also significantly changed (Figure 7h). Under the NDS, except for the relatively high increased proportion of built-up land (8.58%), the changes in other types of land were relatively balanced. Under the EPS, the rates of increase in farmland (9.48%) and forest (6.84%) were the highest, followed by the decreased rates of built-up land (−22.05%) and grassland (−11.4%), whereas the water area did not significantly change. The ecological protection policy mainly protects farmland, forest, and water, which indicates that the scenario simulation is consistent with the actual situation. Under the UDS, built-up land (23.11%) increased by a large proportion, and the rest of the land changed similarly to the natural development scenario.

3.3. Multi-Scenario Simulation of ESV in China from 1990 to 2030

China’s ESV generally shows a trend of “high in the south and low in the north”. The ESVs ranged from USD 21.9 to USD 37.1 million for the southern region and from USD 0 to USD 21.9 million for most of the northern region (Figure 8). In most regions of the northwest, ESV < USD 10.9 million, and in the southwest, ESV > USD 10.9 million. There is a significant difference in the total ESV between the east and the northeast. From 1990 to 2020, the areas with ESV > USD 37.1 million gradually increased in Southwest China, and the areas with ESV < USD 10.9 million also increased in Southwest China, resulting in the obvious polarization of ESV in Southwest China. This spatial variation can be partly attributed to terrain and geological factors, as regions with higher elevations and more complex geological structures often provide diverse habitats and richer biodiversity, leading to higher ESV. Conversely, flatter and more homogeneous terrains with less geological variation tend to have lower ESV. The changing trend of ESV in other regions is not clear.

Based on the land use data for 2030 under different scenarios after the forecast, the corresponding ESVs were calculated. Compared to the 2020 ESV (Figure 8d), the 2030 NDS shows an increase in the area of the ESV distribution of USD 0–10.9 million in the Southwest and a denser distribution of USD 37.1–64.9 million. The ESVs in the Southwest showed a high–low value differentiation pattern, whereas there was no significant variation in other directions. In the EPS, ESVs of USD 21.9 to USD 37.1 million show an expanding trend in South, Southwest, Northwest, Central, and Eastern China. Compared with the EPS, the ESV (USD 10.9–21.9 million) under the UDS increases in South, Southwest, Northwest, Central, East, and Northeast China, and the ESV (USD 21.9–37.1 million) decreases significantly in the distribution area. This reduction refers to the decrease in ESV compared to 2020, which is typically associated with changes in land use, such as conversion to non-vegetated areas, thereby diminishing the ecosystem’s ability to provide services.

From 2020 to 2030 (Figure 8h), the ESV changes in all land use types under the NDS were relatively balanced. The ESVs of farmland (0.17%), built-up land (4.71%), and unused land (3.29%) slightly increased, and the ESVs of water area (−6.21%), grassland (−6.21%), and forest (−1.55%) moderately decreased. Under the EPS, the ESV of the forest (5.77%) and farmland (8.35%) significantly increased, while the ESV of built-up land (−24.83%) and water area (−12.31%) significantly decreased. Under the UDS, the ESV of built-up land (18.72%) significantly increased, and the ESV of other land showed a small change in range.

3.4. Multi-Scenario Simulation of ESV Profit and Loss

From 1990 to 2000, the ESV profits and losses in Northwest China were relatively concentrated in the range of USD 0 to USD 0.079 billion, while in other regions, the distribution of ESV profits and losses was more uniform, primarily fluctuating within the range of USD −0.053 to USD 0 billion and USD 0 to USD 0.079 billion (Figure 9). During the period from 2000 to 2010, there were significant ESV profits and losses in the northwest and eastern regions of China, with a decrease from the USD 0 to USD 0.079 billion range to the USD −0.053 to USD 0 billion range and a slight increase within the USD 0 to USD 0.079 billion range in the southwest region. From 2010 to 2020, some areas in Southwest and Northwest China experienced more severe fluctuations in ESV profits and losses. This reduction refers to the decrease in ESV compared to the year 2020, which typically corresponds to changes in land use, such as conversion to non-vegetated areas, thereby diminishing the capacity of ecosystems to provide services. The decrease in ESV can be attributed to increased developmental pressures such as urbanization and industrialization, leading to habitat loss and degradation and consequently affecting the provision of ecosystem services. Conversely, areas that saw an increase in ESV may have benefited from conservation efforts and ecological restoration. Other regions did not show significant changes. From 2020 to 2030, under the NDS and UDS, the trends in ESV profits and losses continued from the 2010 to 2020 period. Under the EPS, certain areas in Southwest China experienced a more noticeable increase in ESV, suggesting that the EPS is more conducive to promoting the growth of ESV.

3.5. Multi-Scenario Simulation of the Effect of LUC on ESV Profit and Loss

According to Formula (5), the contribution degree of each LUC to ESV was calculated (Figure 10). This study only analyzed the LUC for which the contribution degree was greater than 0.1%. From 1990 to 2020, the conversion of farmland to forest (11.94%), unused land to grassland (7.20%), and farmland to water area (6.23%) contributed the most to ESV profit. The conversion of forest to farmland (−12.46%), grassland to unused land (−7.18%), and forest to grassland (−5.74%) had the highest contribution to ESV loss. From 2000 to 2010, the conversion of farmland to waters (15.37%) and unused land to waters (9.04%) showed an ESV profit contribution that was the highest. The conversion from water area to unused land (−9.00%), from water area to farmland (−6.27%), and from grassland to farmland (−6.06%) had the highest contribution to ESV loss. From 2010 to 2020, the contribution of farmland to forest (9.02%), unused land to grassland (8.39%), and grassland to water area (6.14%) to ESV profit was the highest. The grassland conversion to unused land (−13.42%), forest conversion to farmland (−9.37%), and water conversion to unused land (−6.14%) contributed the most to ESV loss.

Under the NDS, grassland shifted to unused land (105,435 km²), with the highest contribution to ESV loss (−47.55%) and ESV reduction at USD 1815.60. In addition, converting grassland to farmland contributed more to ESV loss (−15.53%), reducing the ESV by USD 1290.70. The transformation of grassland to water area improved the ESV (17.66%), whereas the transformation contribution of other land groups was lower. Under the EPS, the transitions that contributed the most to ESV profit were grassland to forest (32.59%), grassland to water area (16.61%), and built-up land to farmland (5.56%). At the same time, the conversion of grassland to farmland of 119,020 km² (−28.27%) and grassland to unused land of 307,333.3 km² (−10.27%) resulted in ESV losses of USD 1290.70 and USD 1815.60, respectively. The contribution of other transformations to ESV change was low. In the UDS, the contribution of grassland to unused land (−38.37%), forest to grassland (−9.54%), and grassland to built-up land (−7.95%) to ESV loss was the highest. Grassland to water (14.27%), grassland to forest (4.74%), and unused land to grassland (3.90%) had the highest contribution to ESV profit. The above-mentioned analysis indicates that grassland transformation is crucial in regulating ESV.

3.6. Impact of Climate Variables on ESV Profits and Losses

The impact of climate variables on ESV profits and losses exhibits significant differences across various periods and scenarios (Figure 11). During the 1990–2000 period, Bio13 (the precipitation of the wettest month) and Bio12 (annual precipitation) had a significant positive impact on ESV profits and losses. However, the negative impact of Bio4 (temperature seasonality) indicated that temperature fluctuations weakened ecosystem stability, with notable ESV losses during years with large temperature variations. In the 2000–2010 period, the negative impact of Bio15 (precipitation seasonality) gradually became apparent. At the same time, the negative effect of Bio2 (the mean diurnal range) also became significant. Moving into the 2010–2020 period, the negative influence of Bio4 (temperature seasonality) continued to intensify, while precipitation-related variables such as Bio12 (annual precipitation) and Bio13 (the precipitation of the wettest month) continued to have a positive impact on ESV. The variable Bio13 (the precipitation of the wettest month) may indeed depend on which specific month is the wettest and the amount of precipitation received during that month, which influences whether there is a gain or decrease in ESV. For the 2020–2030 forecast under different scenarios, Bio13 and Bio12 maintained a strong positive impact under the NDS. However, under EPS, while the positive impact of precipitation on ESV remained, temperature-related variables, especially Bio4 and Bio16 (the precipitation of the wettest quarter), exhibited stronger negative and positive impacts, respectively. In the UDS, the negative impact of Bio4 (temperature seasonality) was the most pronounced. All these climate variables (Bio1 to Bio19) are derived from the WorldClim database, which provides high-resolution global climate data (Table S1). Specifically, Bio15 refers to precipitation seasonality, which is particularly important in regions with monsoonal climates.

4. Discussion

4.1. Ecological Impacts of LUC on ESV Under Multiple Scenarios

Similar to the results of prior studies at different scales, this study confirms that LUC causes ESV changes at the national scale; where LUC may be related to human activities and changes in natural resources, different land types can provide a certain amount of ESV, and a specific ESV can be affected by human activities and changes in natural resources [10]. The human activity factor is mainly exhibited by the increased development of land with accelerated urbanization, and more land will be transformed into land for construction [1]. In addition, the over-exploitation of land by humans causes land degradation, resulting in the desertification and sanding of land, eventually forming unused land; overgrazing by herders also leads to grassland degradation. Unlike prior studies [13], this study found that grasslands are critical in regulating ESV changes. In particular, the conversion of grasslands to unused land from 1990 to 2000 and 2010 to 2020 caused a significant decrease in ESV in Southwest China, which indicates that grasslands are more advantageous for maintaining soil health and ensuring a series of other ecological functions, such as biodiversity, which also indicates that the reasons for ESV changes are complex [11]. In contrast, other studies indicate that the ecosystem services of forests and wetlands also play a crucial role in maintaining ecological stability [65]. This study complements the role of grasslands in ecosystem maintenance, particularly their critical importance under specific regional environmental pressures. This finding highlights the significance of grassland conservation and management, especially in the context of rapid urbanization.

From 1990 to 2020, the total ESV in most regions of China was in the range of USD 0–37.1 million, and the improvement in ESV during the study period was most sensitive to the conversion of farmland to forests, grasslands, and watersheds and the conversion of grasslands to watersheds. However, the deterioration of ESV was most sensitive to the transformation of forests to grasslands and grasslands and watersheds to unused lands, which shows that forests, grasslands, and watersheds play an integral role in environmental purification. This is consistent with some studies [66,67,68], which emphasize the necessity of the rational use and protection of natural resources. Thus, forests, grasslands, and watersheds impact environmental purification, soil health maintenance, biodiversity enhancement, and hydrological regulation [45]. Therefore, in future ecological planning, certain amounts of forests, grasslands, and watersheds need to be included in the scope of protection to ensure the stability of the total ESV.

Grasslands play an instrumental role in ecological restoration, and in this context, the government needs to strengthen the management of grassland ecology and improve ecological protection policies. Government departments can try to link farmers’ income, grassland ecological protection, and ecological industry development so that social, ecological, and economic benefits are linked, and the three can develop synergistically; the government can also go deep into the grassroots to help farmers establish new concepts and learn new methods of ecological restoration and ecological management, adopt production methods that match environmental capacity and ecological carrying capacity, and realize the harmonious coexistence of herders and ecology. The government should improve relevant laws, define the scope of use of grasslands, prohibit herders from using the grasslands in the governance area, and impose economic or criminal penalties on violators.

4.2. Spatiotemporal Characteristics of ESV Profit and Loss

Based on the land use assessment and prediction of the spatial and temporal dynamics of ESV loss and profit in China, there is a scientific basis for the optimal allocation and effective use of future national land resources to guide ecological conservation and regional economic development [69]. From 1990 to 2000, economic development activities in Northwest China were relatively limited. The natural environment in this region is quite fragile, but due to the relatively small impact of human activities, the change in ESV was not significant. Between 2000 and 2010, economic development in Northwest China and the eastern regions accelerated, with urban expansion and industrialization leading to the degradation of the natural environment. Particularly in the eastern region, as the economic center of China, urbanization and industrialization exerted significant pressure on the ecological environment, with ESV losses decreasing from USD 0–0.079 billion to USD −0.053–0 billion. The slight increase in ESV within the USD 0–0.079 billion range in the Southwest may be attributed to the implementation of ecological protection policies and the impact of ecological restoration projects. From an economic perspective, it is projected that by 2030, under the NDS, the ESV in the southwest region could generate economic benefits of approximately USD 37.1 million to USD 10.9 million, while under the EPS, with the protection of forests and grasslands, the ESV is expected to increase by about USD 10 million. This increase in economic benefits not only reflects the direct contributions of ESV but also helps to indirectly promote local economic development and green industries such as ecotourism. In contrast to the findings of Ai et al. (2024), who emphasized the impact of population density on ESV [42], this study reveals the direct threat that rapid economic development poses to ESVs, providing a new perspective for understanding the dynamic relationship between urbanization and ecology. Moreover, in areas experiencing rapid urbanization, increased environmental pressure may lead to a loss of ESV activities indeed exerting pressure on ESVs. In this study, the fluctuations in ESV in the southwest region align with ecological protection policies, further validating the timeliness of these policies in influencing environmental changes. From 2010 to 2020, parts of Southwest and Northwest China experienced more intense fluctuations in ESV, which may be related to China’s increased development efforts in the western regions, such as the western development strategy, and the increased vulnerability of the ecological environment in this region under the impact of climate change. This study reveals that under high pressure, especially from urbanization and agricultural expansion, ESV is expected to fluctuate significantly. The fluctuations in ESV in the southwest region during this period indicate that potential economic losses in the future could increase further, with projected losses reaching USD 21 million. Consistent with the findings of Carbone et al. (2022) [70], our results emphasize the interaction between economic policies and environmental policies, highlighting the long-term impacts of climate change on ecological services, which provides practical cases for local governance. The trends in ESV changes under the three different predicted scenarios (NDS, EPS, and UDS) for 2020–2030 are similar to those from 2010 to 2020, which may indicate that the development trends and environmental policies of the past decade continue to have an influence. Under the EPS, certain areas in Southwest China exhibited a more noticeable increase in ESV; this result is consistent with the conclusions of Boonman et al. (2023) [71], emphasizing the effectiveness of ecological protection measures. It is anticipated that by 2030, the economic benefits under the EPS will further increase, with the growth in ESV potentially bringing an additional economic value of approximately USD 15 million to the region.

From 1990 to 2020, the conversion of farmland to forest in Northern, Central, and Southwestern China provided the maximum contribution to ESV profit, while the conversion of forest to farmland in the same region provided the greatest contribution to ESV loss, indicating that the policy of returning farmland to forest in China plays an important role in ESV improvement. Thus, the implementation of this policy needs to be maintained in the future. Consistent with the views of Li et al. (2019) [72], this study further confirms the effectiveness of the Grain for Green policy and indicates that this policy needs to be continued in the future to ensure the stability of ecosystem services. The conversion of unused land to grassland and grassland to unused land in Northwest and Southwest China was second in terms of contribution to both ESV profit and loss, indicating that the role of grassland in maintaining ESV is second only to that of forest; thus, the protection of grassland needs to be strengthened. In 2020–2030, under the NDS, EPS, and UDS, grassland conversion to cropland in Northwest and Southwest China and grassland conversion to unused land in East and Central China all contributed larger ESV losses, while grassland conversion to water and forest in Southwest China both contribute to larger ESV profits, indicating that the development of farmland and unused land depletes ESV. However, forest, grassland, and water all improve ESV. In the future, China must insist on reforestation, grass conversion, and the rational use of water resources to achieve greater ESV profits [73].

In this study, the PLUS model was used for multi-scenario prediction, which can intuitively reflect the LUC and ESV under different scenarios, and the research results were found to be reliable after testing. However, in selecting driving factors, government decisions, human factors, and social development were not considered. In contrast to the study by Li et al. (2022) [74], our research highlights the impact of more complex social and economic changes on ecosystem services. Therefore, the effect of LUC on ESV in the context of multiple complex relationships should be further analyzed in subsequent studies, and the land use data can be further subdivided to improve the accuracy of the data and analysis. In studying the spatiotemporal dynamic changes in ESV profit and loss in China, the quantitative accuracy of these figures is still insufficient. Although ESV evaluation based on a remote sensing model has strong computational power and a scientific nature, it cannot accurately reflect the spatiotemporal characteristics of ESV profit and loss in China only via numerical and formula simulation. In the future, it is essential to combine the measured data from the study area with continuous debugging and simulations to improve the accuracy of determining the actual levels of ESV profit and loss.

4.3. Impact of Climate Change on ESV Profits and Losses

The impact of climate variables on ESV profits and losses shows significant differences across different time periods and scenarios. We found that precipitation and temperature fluctuations are key climate factors affecting ESV profits and losses. Specifically, during the 1990–2000 period, Bio13 (the precipitation of the wettest month) and Bio12 (annual precipitation) had a significant positive impact on ESV profits and losses, indicating that increased precipitation helps enhance ES functions. However, the negative impact of Bio4 (temperature seasonality) suggests that temperature fluctuations weaken ecosystem stability, particularly in years with large temperature variations, resulting in notable ESV losses. These results reflect the different mechanisms through which precipitation and temperature fluctuations influence ESVs. Increased precipitation, especially annual precipitation and precipitation during the wettest month, can provide sufficient water resources, promoting plant growth and biodiversity and thereby enhancing ES functions. Conversely, intensified temperature fluctuations may lead to stress responses in ecosystems, disrupt ecological balance, and reduce ESV. Consistent with the findings of Austhof et al. (2024) [75], this study further reveals the nuanced impact of temperature seasonality on specific ecosystem service functions, which has not been thoroughly explored in previous research. Compared to the study by Swenson et al. (2023) [76], these findings emphasize the fundamental role of precipitation in sustaining ecosystem services, while also highlighting the adverse effects of temperature fluctuations on service functions.

From 2000 to 2010, Bio15 (precipitation seasonality) gradually manifested negative impacts, signaling that its uneven distribution put undue stress on ESV. Changes in precipitation seasonality can cause water shortages during certain times of the year that fail to meet biological community needs and disrupt the service functions of ecosystems, thereby diminishing their service functions. Simultaneously, the negative impact of Bio2 (the mean diurnal range) also became significant, suggesting that under extreme climate conditions, increased day–night temperature differences are similarly detrimental to the maintenance of ES functions. Large diurnal temperature variations can affect plants’ physiological processes, such as photosynthesis and respiration. Although temperatures remain above freezing and day length is unchanged, significant temperature fluctuations can cause thermal stress on plants. This increased physiological stress on plants may impact their growth cycles and overall health, reducing productivity and stability within the ecosystem. Compared to the study by Yao et al. (2022) [77], this research provides more detailed insights into the specific impacts of diurnal temperature variation on the profits and losses of ESVs, revealing the high sensitivity of ecosystem services to changes in temperature fluctuations. Entering the 2010–2020 period, the negative effect of Bio4 (temperature seasonality) continued to intensify, while precipitation-related variables such as Bio12 (annual precipitation) and Bio13 (the precipitation of the wettest month) continued to have a positive impact on ESV. Bio4, which represents temperature seasonality, is calculated as the standard deviation of monthly temperature averages multiplied by 100. This metric indicates the degree of temperature variation throughout the year. Increased temperature seasonality means greater fluctuations in temperature between seasons, which can lead to environmental stress on ecosystems. Precipitation-related variables such as Bio12 (annual precipitation) and Bio13 (the precipitation of the wettest month) continued to have a positive impact on ESV. During this period, the challenges posed by climate fluctuations to ecosystems intensified, and precipitation stability became a crucial factor in maintaining ESV. The worsening negative impact of temperature seasonality on ecosystems may be related to the increased frequency of extreme weather events caused by climate change. These events put significant pressure on ecosystems, leading to decreased service functions. Through the analysis of long-term data series, this study confirms the universal importance of precipitation stability on ESV across different climatic contexts, providing a new reference perspective for research on similar ecosystems globally.

Future scenario projections indicate that from 2020 to 2030, under the NDS, Bio13 and Bio12 will continue to have a strong positive impact, especially in years with abundant precipitation, significantly enhancing ESs. However, under the EPS, although the positive effect of precipitation on ESV remains, temperature-related variables, particularly Bio4 and Bio16 (the precipitation of the wettest quarter), show stronger negative and positive impacts, respectively. Lastly, under the UDS, the negative impact of Bio4 (temperature seasonality) is the most pronounced. Urbanization exacerbates temperature fluctuations, further weakening ESV, while the influence of precipitation-related variables decreases. Compared to traditional climate model predictions, this study incorporates human activity factors across multiple scenarios, such as different urbanization pathways, providing a more comprehensive analysis of the response mechanisms of ESV. Overall, precipitation and temperature fluctuations are key climatic factors affecting ESV profits and losses. Over time, the intensification of climate change and human activities has made the stability of precipitation and the variability of temperature crucial determinants of ESV changes. Future climate change projections suggest that the impact of climate variables on ESV under different scenarios carries a certain degree of uncertainty, necessitating further research and monitoring to formulate effective ecosystem protection and management strategies. The multi-period and multi-scenario analysis approach of this study not only validates existing theories but also reveals new impact pathways, providing innovative analytical perspectives for future evaluations of ESVs. Understanding how climate change impacts ESVs will enable us to more accurately anticipate and respond to future impacts, providing a scientific basis for protecting and managing ecosystems. This study highlights the significance of precipitation stability and temperature variability for maintaining ESVs, providing insights for climate adaptation strategies in the future.

Although this study deeply analyzed the impact of 19 climate variables on ESV profits and losses across different regions and scenarios in China using the LSTM-SHAP combined model, it still has certain limitations. Currently, the data we can access are limited to the 19 climate factors officially designated for the SSP126 scenario from 2021 to 2040, which may restrict our in-depth analysis of other potentially critical climate variables. Using these forecast-based data in analysis and modeling might introduce potential biases; therefore, decision makers should fully consider this when making relevant management and policy decisions. Future research will strive to overcome this limitation. We plan to introduce more climate factors, such as wind speed and humidity, which will help provide a more comprehensive understanding of the impact of climate change on ESV profits and losses. Additionally, we will integrate more scenario simulation data to further refine and enrich the research content. These efforts will contribute to providing more comprehensive and accurate scientific evidence for ecosystem restoration and management strategies. In this context, we also recognize that using the distance from the center of a 1 km grid to the nearest roads, rivers, and railways as an indicator in this study may introduce some uncertainties. While the 1 km grid scale offers a balance between environmental feature coverage and data processing operability in geographical and ecological research, it can still be inadequate in accurately representing certain regional characteristics. Specifically, although the distance from the grid cell center to the nearest roads, rivers, and railways simplifies complex spatial analyses and maintains consistency in overall analyses, the presence of multiple transportation and water networks within a 1 km unit and the relatively large area of the unit mean that the center point distance might not fully reflect the spatial characteristics of all features within the unit. For example, the actual roads or water bodies in a 1 km unit might be closer to the boundary of the area, and this spatial approximation might reduce the precision in depicting more complex relationships between features. In future research, considering higher resolution data could be a potential solution to reduce this uncertainty. Specifically, for each 1 km grid cell, we could use more detailed spatial analysis methods (such as area-based or multipoint approaches) to assess the actual impact of roads and water bodies. These methods, by integrating proximity from multiple locations or utilizing diverse distance calculations, can enhance the understanding of ecosystem services within specific regions. This improvement not only helps enhance data representativeness but might also improve the responsiveness and accuracy of analyses concerning ecosystem changes and impacts.

5. Conclusions

This study employed multi-scenario simulations to examine the profits and losses of ESV in China from 1990 to 2030, with particular attention paid to climate change’s influence. The primary findings are as follows: Over time, China’s ESV displays an overall high trend in the south and low trend in the north, with ESV levels significantly higher in southern regions than northern ones from 1990 to 2030. From 1990 to 2020, however, southwest regions show distinct polarization in ESV performance between different areas, with significant increases and decreases over this timeframe. Under the NDS, ESV changes in various land uses are relatively balanced. In the EPS, the ESV of forests and farmland increases significantly, while the ESV of built-up land and water decreases significantly. Under the UDS, the ESV of built-up land increases significantly, with smaller changes in other land use types. The observed variations in climate variables impacted ESV profits and losses. During the 1990–2000 period, precipitation had a significant positive impact on ESV, while temperature seasonality had a negative impact. While these findings suggest potential implications related to climate variation, the study’s scope was limited to analyzing observed data and did not comprehensively assess climate change impacts. Further research is needed to draw more definitive conclusions regarding the long-term effects of climate change on ESV. From 2000 to 2010, the negative impacts of precipitation seasonality and the mean diurnal range on ESV gradually became apparent. From 2010 to 2020, the negative effect of temperature seasonality continued to intensify, while the positive impacts of precipitation-related variables persisted. Future scenario projections indicate that under the NDS, the positive impact of precipitation on ESV remains significant; in the EPS, the impacts of precipitation and temperature variables are more complex; and in the UDS, the negative effect of temperature seasonality is most pronounced. Converting farmland to forest, converting unused land to grassland, and converting farmland to water contribute the most to ESV profits, while converting forest to farmland, grassland to unused land, and forest to grassland contribute the most to ESV losses. Under future scenarios, grassland transformation plays a crucial role in regulating ESV. This paper provides new insights into the dynamic changes in ESV and their driving factors by comprehensively considering the dual impacts of climate and land use. This not only enriches existing research but also provides important theoretical support for practical policymaking. In future research, the further exploration of the interactions between different regions and long-term trend changes will be of significant importance for achieving sustainable development goals nationwide.

Based on the above research conclusions, the following policy recommendations are proposed to enhance the ESV and promote sustainable development in China: (1) Consider long-term climate impacts in regional planning: when formulating regional development plans, fully consider the long-term impacts of climate change and implement comprehensive measures to address extreme climate events and mitigate the adverse effects of temperature seasonality and precipitation seasonality changes on ecosystem. (2) Strengthen water resource management: in areas with significant precipitation seasonality changes, enhance water resource management by building reservoirs, improving rainwater collection, and increasing water use efficiency. (3) Integrate ESV into economic policies: When formulating economic development policies, use ESV as an important indicator. Encourage green GDP accounting and consider environmental benefits when evaluating economic performance. Actively promote green industries such as ecological agriculture and ecotourism, encourage businesses to utilize more eco-friendly production methods, reduce environmental damage through regional ESV, and enhance environmental education. (4) Raise public awareness on ESV/sustainable development issues. Encourage and assist public participation in ecological protection activities. (5) Consider climate change in land use and ecological policies: Integrating climate change factors into land use planning and ecological protection policies to enhance ecosystem resilience and adaptability while mitigating the negative effects of extreme climate events on ecosystems can help us better navigate climate change’s complexities and drive the sustainable development of ecosystems in China. By adhering to this recommendation, we can navigate its complexity for sustainable ecosystem development in China.