Using formulas (1) and (2), the single dynamic degree and comprehensive dynamic degree of the 6 land use types can be obtained (Fig. 2).
From 2000 to 2023, the primary characteristics were the single dynamic degree changes of cultivated land and construction land, which were - 2.953% and 1.897%, respectively. This indicates that the area of cultivated land continuously decreased, while the area of construction land continuously increased. These two types had the highest negative and positive growth among the six land use types. During the 2000-2005 and 2005-2010 periods, the single dynamic degree of cultivated land was - 0.479% and - 0.664%, with the largest absolute values, indicating a rapid and extensive reduction phase. Although it increased to -0.071% after 2010, the growth was small, and in subsequent years, the dynamic degree remained less than 0, showing that cultivated land faces significant stress. The single dynamic degree of construction land was the highest from 2005 to 2010, with little difference in other periods, indicating that during the study period, the demand for construction land was rigid.
From 2000 to 2023, the average comprehensive land dynamic degree was 0.152%. Among them, the maximum comprehensive land dynamic degree was 0.323% between 2005 and 2010, indicating that land changes were frequent and large in this period, leading to significant disturbances to the ecosystem.
Using formula (3), the land use transfer matrix Sankey diagram (Fig. 3) was calculated. Between 2000 and 2023, the land use type conversions mainly occurred between cultivated land, forest, and grassland. Among them, the largest area of land conversion was from cultivated land, totaling 626,78.998 km², and the largest area of land conversion to was grassland, totaling 59,010.457 km².During the study period, land use type changes in the middle reaches of the Yellow River were large, with high intensity of human activity in the development of construction land and unused land.
From the spatial transfer map of land use in the middle reaches of the Yellow River, it can be observed that cultivated land is the largest land use type in this area (Fig. 4), occupying 36.514% of the total basin area, followed by grassland and forest, which occupy 34.181% and 23.796% of the area, respectively. These three land use types are distributed across most of the basin. The conversion of cultivated land is more pronounced, with its transformation into forest and grassland concentrated in the central part of Shaanxi Province, including counties such as Yanchuan, Yanchang, and Anse, as well as parts of Gansu Province's southern region and northern Shanxi Province. The interconversion of cultivated land to construction land is mainly distributed in the Jinzhong urban agglomeration, Guanzhong urban agglomeration, and Central Plains urban agglomeration. The conversion of forest into grassland is concentrated in the Ziwuling-Huanglongshan ecological protection area and water conservation area in Shaanxi Province, as well as areas in the Lüliang Mountains, Taihang Mountains, and the Yellow River and its tributaries in Shanxi Province. From 2010 to 2023, compared to 2000-2010, the conversion of cultivated land and forest to construction land became more pronounced, concentrated in counties such as Xi'an in Shaanxi Province, Luoyang in Henan Province, and the central urban area of Taiyuan in Shanxi Province, with a tendency to spread to surrounding areas. The conversion of unused and other land is concentrated at the junction of the Ordos Plateau and the Loess Plateau of Shanxi and Shaanxi.
Data from six key years (2000, 2005, 2010, 2015, 2020, and 2023) in the Middle Reaches of Yellow River were selected to estimate the ESV, and the results are shown in Table 2. From 2000 to 2023, due to the inclusion of the change in ESV for construction land, the overall ESV in the middle reaches of the Yellow River showed a gradual decline, with the greatest decline occurring between 2005 and 2010, totaling a reduction of 747.189 × 10 yuan. Specifically, over the 24 years, the ESV of forest, grassland, and water area showed some fluctuation and increase, with increases of 89.947 × 10 yuan, 20.351 × 10 yuan, and 11.649 × 10 yuan, respectively. The ESV of unused land remained almost unchanged, while the ESV of cultivated land and construction land gradually declined. Among these, the change in ESV for construction land was particularly significant, with a change rate of 80.930% (-1351.974 × 10 yuan). The main reason is that with the development of China's economy, infrastructure construction gradually improved, the living standards of people increased, and the population grew accordingly, leading to an increase in construction land area and a decrease in ESV. Especially between 2005 and 2010, China's industrialization and urbanization processes advanced continuously, and after the outbreak of the global financial crisis in 2008, the government implemented infrastructure investment-led plans to offset external economic shocks, leading to significant changes in land development patterns.
From Table 3, it can be seen that in the primary service types of the Yellow River Basin's middle reaches, the ESV of provisioning services shows a declining trend year by year, with a total decrease of 1337.946 × 10 yuan. The service values of food production, raw material production, and water supply have been consistently decreasing, indicating the ongoing encroachment on cultivated land during urban development. Water supply occupies a central position in the materials needed for human life, and industrial development consumes large amounts of water resources. The increase in construction land area also requires significant water resource input, which has led to a year-by-year decline in the ESV of water resource supply in type of secondary services, with a decrease of 1319.424 × 10 yuan over the 24 years. The ESV of regulating and supporting services has gradually increased. The value of regulating services accounts for over 80%, indicating their dominant role in the ecosystem of the middle reaches of the Yellow River. Among them, hydrological regulation and climate regulation play important roles in regulating the ecological environment and increasing ESV, with increases of 24.261 × 10 yuan and 25.116 × 10 yuan, respectively. In supporting services, soil conservation and biodiversity account for 54% and over 41%, respectively, playing a key role in supporting the ecological environment. The ESV of cultural services shows a fluctuating increase, but it remains relatively stable. Overall, it can be concluded that in the middle reaches of the Yellow River, the ESV of hydrological regulation accounts for the largest proportion, having a significant driving effect on the improvement of the ecological environment.
A 5 km × 5 km grid unit was used to divide the study area into 16,368 grid points. The ESV was categorized into five intervals (low-value area, lower-value area, middle-value area, higher-value area, and high-value area) using the natural breaks method. ESV for the middle reaches of the Yellow River was calculated, and visualizations of ESV for the years 2000, 2005, 2010, 2015, 2020, and 2023 were produced using ArcGIS 10.8.
As shown in Figs. 5 and 6, during the study period, the ESV in the middle reaches of the Yellow River primarily exhibited a reduction in the lower-value area and middle-value area, while the other areas expanded. Spatially, a gradual decrease in ESV from northwest to southeast is evident. Specifically, the areas of the lower-value area and middle-value area decreased by 3.671% and 1.690%, respectively, while the higher-value area expanded by 4.068%. The expansion of the low-value and high-value areas was not significant. The low-value area occupies a small area, approximately 1.076%, and is scattered spatially, specifically in the southern part of Inner Mongolia, northeastern Gansu, southern Shaanxi, and central and southwestern Shanxi. These areas overlap with the spatial location of the lower-value area. The lower-value area is mainly concentrated in the Inner Mongolia Plateau and the Guanzhong Plain urban agglomeration. The land types are mainly construction land and cultivated land, covering approximately 12.497% of the area. The middle-value area covers the largest area, approximately 56.512%, and is distributed across most of the Loess Plateau, with land use types dominated by forests and grasslands. The higher-value area and high-value area cover approximately 29.307% and 0.603%, respectively, and are primarily distributed in the ecological function areas of the Loess Plateau's hilly and ravine regions and the Qinling Mountains, with land use mainly consisting of forests. The higher-value area and high-value area generally have abundant precipitation, mild climates, lush grasslands and forests, high vegetation coverage, and rich biodiversity. The lower-middle-value area and low-value area are distributed in major agricultural production areas and key development zones, with higher land-use intensity and less vegetation. The Inner Mongolia Plateau and Loess Plateau receive less precipitation, with large diurnal temperature variation and low tree survival rates. Additionally, the Loess Plateau has deep soil layers and loose soil, making it suitable for farming. This region contains many major agricultural production areas and abundant energy resources, which, during energy development and urban expansion, have adverse effects on the ecological environment. The Guanzhong Plain, with its flat terrain and population concentration, is home to key development zones and major agricultural production areas. While driving economic development, it also faces relatively severe environmental pollution.
The secondary ecosystem services in the middle reaches of the Yellow River were analyzed for trade-offs and synergies, examining the degree and direction of interactions among individual ecosystem services during the periods 2000-2005, 2005-2010, 2010-2015, 2015-2020, and 2020-2023. The final results are presented in Fig. 7.
During 2000-2005, synergies accounted for 45.455% of the trade-off-synergy relationships. Food Production, Raw Material Production, Water Supply, and Gas Regulation exhibited trade-offs, reflecting the characteristics of resource competition. In contrast, a significant synergy (8.920) was maintained between Hydrological Regulation and Soil Conservation, partly due to soil and water conservation projects in the upper and middle reaches of the Yellow River. In particular, the implementation of the Grain for Green program and the Soil and Water Conservation Law effectively reduced soil erosion, thereby strengthening the synergy between Soil Conservation and Hydrological Regulation. Nevertheless, services such as Food Production and Raw Material Production continued to exhibit strong trade-offs with other ecosystem services, primarily due to excessive cultivation and agricultural expansion placing great pressure on land resources.
The changes from 2005 to 2010 were particularly striking, as the relationship between Water Supply and Soil Conservation shifted from a trade-off to a high level of synergy (63.947). This transformation was closely linked to the implementation of multiple ecological restoration projects starting in 2005. In particular, the Grain for Green and Grassland Program and the Natural Forest Protection Program significantly altered land use structures, substantially enhancing Water Supply and Soil Conservation capacities. These policies not only curbed excessive agricultural expansion but also increased vegetation cover, restored natural hydrological cycles, reduced soil erosion, improved soil infiltration, and stabilized groundwater fluctuations, thereby fostering stronger synergies between Water Supply and Soil Conservation.
Between 2010 and 2015, the proportion of synergistic relationships among ecosystem services further increased to 67.273%, indicating more pronounced interactions among ecosystem functions. Gas Regulation demonstrated strong synergies with Biodiversity and Aesthetic Landscape, while trade-offs with Hydrological Regulation and other services persisted. This period marked an important policy transition in the Yellow River Basin, including the launch of the Ecological Civilization Construction strategy and the strengthening of ecological protection policies, such as wetland conservation and grassland restoration. These policies enhanced Biodiversity protection and Aesthetic Landscape restoration, while simultaneously reinforcing support for Water Supply and Climate Regulation. However, despite the overall increase in synergies, trade-offs between Hydrological Regulation and Food Production persisted, reflecting ongoing tensions between agricultural production pressure and ecological protection.
From 2015 to 2020, synergies between Water Supply and other ecosystem services were significantly enhanced, particularly with Food Production, Raw Material Production, Soil Conservation, and Nutrient Cycling. With the deepening of ecological restoration projects and the implementation of ecological compensation mechanisms, soil and water conservation and ecological restoration efforts in the Yellow River Basin were further strengthened. For example, ecological compensation policies encouraged local farmers to adopt more environmentally friendly production practices, reducing excessive land exploitation and enhancing the overall benefits of ecosystem services. However, despite the strengthened synergies, certain trade-offs among services remained unresolved. In particular, during years of water scarcity, conflicts between Water Supply and irrigation demands for Food Production became more pronounced.
During 2020-2023, synergies accounted for 67% of the relationships; however, except for Gas Regulation, Water Supply largely exhibited trade-offs with other services. This reflected the dual impacts of post-pandemic economic recovery and land use changes. Although ecological protection policies were gradually implemented during this stage, some regions remained heavily dependent on traditional industries for economic recovery, leading to new conflicts between Water Supply and services such as Soil Conservation and Nutrient Cycling. This suggests that, despite improvements in ecological conditions, tensions between economic development and environmental protection persist, particularly as the restart and expansion of resource-intensive industries continued to challenge the coordination between land use and ecosystem functions.
Driven by human activities and natural factors, land use types in the study area underwent substantial changes, and shifts in ecosystem structure significantly influenced the transformation of ecosystem service functions, thereby further affecting trade-offs and synergies among ecosystem services. Therefore, it is essential to formulate and implement appropriate management policies to improve land use practices and optimize economic development structures. Such measures can mitigate the negative impacts of current unsustainable development patterns on trade-offs and synergies among ecosystem services, and help to alleviate potential conflicts to some extent.
ESV changes are influenced by various factors. Based on the actual situation of the study area and data availability, natural factors such as average elevation (x1), average slope (x2), annual average temperature (x3), annual average precipitation (x4), normalized vegetation index (NDVI) (x5), and forest proportion (x10) were selected, along with human factors such as nighttime light brightness (x6), GDP per capita (x7), population density (x8), and cultivated land proportion (x9). The ESV in the middle reaches of the Yellow River was set as the dependent variable, and the potential influencing factors were set as independent variables. After standardizing the data, machine learning algorithms were used to identify the impact degree and direction of each factor on ESV.
Based on the data from 226 counties in the middle reaches of the Yellow River between 2000 and 2023, the training and testing sets were divided at a ratio of 70% to 30%. Machine learning algorithms such as decision tree (LGB), random forest (RF), and XGBoost were used for fitting and training. The XGBoost algorithm achieved the lowest RMSE (0.117) and MAE (0.072), and the highest R2 (0.914), with the predicted values closest to the actual values. Therefore, this study uses the XGBoost algorithm to explore the influencing factors of ESV changes in the middle reaches of the Yellow River.
The SHAP mean reflects the contribution of each influencing factor to coupling coordination, which is an important method for identifying the main driving factors. Nighttime light brightness objectively reflects industrial production, commercial activities, and energy consumption in human society, and it is the most significant factor affecting ESV changes, accounting for 23.027%. The cultivated land proportion and population density also significantly influence the change in ESV in the middle reaches of the Yellow River, accounting for 18.593% and 17.979%, respectively. This suggests that the degree of human activity plays a significant role in the ecological environmental changes in the middle reaches of the Yellow River. In addition to nighttime light brightness, cultivated land proportion, and population density, other influencing factors in order of importance are average slope, average elevation, forest proportion, GDP per capita, annual average temperature, NDVI, and annual average precipitation (Fig. 8).
Although the SHAP mean can measure the relative importance of each variable, it cannot intuitively reflect the specific direction of their effects. Therefore, this study further constructs a SHAP visualization summary diagram to reveal the direction of influence of each driving factor on ESV. Where each point represents a sample value, the larger the value the redder the colour and vice versa the bluer. The greater the feature importance of a variable, the higher its ranking, and the positive and negative SHAP values represent the positive and negative impacts on the predicted value, respectively.
As shown in Fig. 9, increases in average slope, forest proportion, and annual average temperature have a positive impact on ESV. Since the 21st century, the government has implemented the policy of returning farmland to forest and grassland, explicitly prohibiting the cultivation of slopes greater than 25°. This has significantly reduced soil erosion in these areas and restored vegetation conditions. Dense vegetation roots and leaf litter layers can effectively intercept rainwater, slow down runoff, reduce soil erosion, protect water sources, and enhance biodiversity. At the same time, an increase in forest area can improve the local climate, mitigate soil erosion and land desertification, which plays a crucial role in increasing seedling survival rates and promoting the cultivation of superior species. Therefore, an increase in the forest proportion area promotes ESV. During 2000-2023, the frequency of extreme temperatures in the middle reaches of the Yellow River was relatively low, and the overall temperature showed a moderate increase, which played an important role in extending the growing season of plants and increasing vegetation coverage.
The factors that have a negative impact on ESV are primarily nighttime light brightness, cultivated land proportion, and population density, all of which can to some extent reflect the intensity of human activities. According to the land use change map and land use transfer matrix, during the 24 years, the land use types in the middle reaches of the Yellow River have changed significantly, with a large area of developed construction land. The expansion of construction land is a key indicator of urbanization, but its negative impact on ESV is significant and continuous. High-intensity development and construction drive changes in land use, especially the conversion of cultivated land, forest, and grassland to construction land, which may lead to a substantial reduction in ESV. Additionally, average elevation and NDVI also have a suppressive effect on ESV. When the average elevation is lower, it still has a positive effect on ESV, but as the elevation increases, ESV gradually decreases. This is mainly because the middle reaches of the Yellow River flow through the core region of the Loess Plateau, where the elevation is high, leading to a decrease in temperature, changes in water-heat conditions, and vegetation coverage.
According to the degree of influence of each factor, six important influences were selected to analyse the impact of key features on model predictions, as shown in Fig. 10. Among them, the effects of nighttime light brightness, cultivated land proportion and population density on ESV show a more significant negative linear relationship, nighttime light brightness and population density are more concentrated when they are less than 0, and the values of cultivated land proportion are clustered between - 1 and 1, but the whole is more dispersed, which is related to the spatial distribution of cultivated land, construction land and other land types; the effects of average slope on ESV show a positive linear relationship and is more clustered between slope values of 0.051 and 1.612; the factors of average elevation and forest proportion show a non-linear relationship. With the increase of elevation, the ESV shows that it first rises and then decreases at -1.248, because the middle reaches of the Yellow River passes through the hilly and gully areas of the Loess Plateau, and most of the counties and districts have higher elevations, so the average elevation is concentrated at higher values; forest proportion has a greater effect on the ESV in the beginning, and then tends to be stable gradually.