The study area is located in Yugan County, Shangrao City, Jiangxi Province ( Figure 1 ). It has a total area of 2,350.36 km² and spans from 28°21’36″ to 29°03’24″ N and 116°13’48″ to 116°54’24″ E ( Figure 1 ). It is under the jurisdiction of 9 towns and 11 townships, with a total of 372 village committees. The topography of the city is dominated by low hills and lakeside plains with high terrain in the southeast and low terrain in the northwest. It belongs to a subtropical humid monsoon climate, with an average annual precipitation of 1,548 to 1,692 mm, average annual temperature of 15.4°C to 19.5°C, and frost-free period of 240 to 300 days. Land use mainly consists of arable land and forest land, with arable land accounting for 37.44% of the total area and forest land constituting 19% ( Figure 2 ). Rice is the main grain crop, accounting for more than 90% of the total grain production. The soil types are dominated by rice and red soils and vegetation by broad-leaved and coniferous forests. The concentration of cultivated land and gentle slopes in the region make it ideal for studying cropland degradation.

Yugan County represents the typical ecological and agricultural conditions of the Poyang Lake region, characterized by the concentration of cultivated land, gentle slopes, and proximity to the lake’s dynamic water system. Its location within the Poyang Lake basin makes it a hotspot for understanding the interactions between land use, water resource dynamics, and soil degradation. The area is subject to significant land degradation pressures, including soil acidification, soil fertility decline, soil erosion, and soil physical structure degradation, driven by intensive agricultural practices, heavy fertilizer use, and terrain features conducive to runoff and nutrient loss. These issues mirror broader degradation trends across the Poyang Lake region, making Yugan County a critical case for studying the mechanisms and drivers of farmland degradation. Moreover, its role as a major agricultural producer in the region highlights the broader environmental and socio-economic implications of land degradation, such as the risks of reduced crop yields, water eutrophication, and downstream ecological impacts, underscoring its representative significance in regional and national agricultural studies.

The data sources covered a wide range of aspects and provided solid support for scientific analysis ( Table 1 ).

The topsoil physical and chemical data, including soil organic matter content, pH value, total nitrogen content, and bulk density, along with fertilizer application amount, and straw return amount data were obtained from the Soil Testing and Formula Fertilization Dataset of Yugan County for 2012 and 2019.

Land use data, obtained from the Third National Land Survey of China for 2019 were incorporated to map the spatial distribution of land use types.

Remote sensing data, including Landsat 7 images, were obtained from the Google Earth Engine platform and ASTER GDEM V2 data were provided by the Geospatial Data Cloud (http://www.gscloud.cn/).

Socioeconomic data were obtained from population data from ORNL LandScan Viewer and The China GDP Spatial Distribution Kilometer Grid Dataset - Data from the Registration and Publishing System for Resource and Environmental Science Data (Xu, 2017).

For climate data, 1-km-resolution precipitation and temperature data were provided by the National Tibetan Plateau/Third Pole Environment Data Center (Peng, 2024) and 1-km resolution month-by-month mean temperature dataset (1901–2023) by the National Tibetan Plateau Data Center (Peng, 2019) was used.

Based on the Cultivated Land Quality Grade Standard (GB/T33469-2016) and relevant documents issued by the Ministry of Agriculture and Rural Affairs, the grading standards for the physicochemical properties of the cultivated land in the study area were determined through comprehensive consideration ( Table 2 ). Following the approach proposed by Tang et al. (2023), which emphasizes the integration of natural condition assessments and quality factor thresholds, we adopted a refined method for evaluating cultivated land quality. Specifically, we incorporated their equation for determining quality grades to ensure consistency and accuracy in assessing the degradation thresholds of cultivated land. Cultivated land quality grade was assessed using the 2019 Third National Land Survey Cultivated Land Map as the evaluation unit. The five indicators, including slope, soil organic matter content, soil total nitrogen content, soil pH, and bulk weight, were categorized into four grades based on their respective thresholds. The cultivated land quality grade of each evaluation unit was determined by the cumulative scoring of these indicators in accordance with the method outlined in Equation 1, derived and adapted from the formula used by Tang et al. (2023). This approach allows for a comprehensive evaluation of the degradation and quality status of cultivated land while aligning with region-specific thresholds for sustainable land management.

Equation 1 and its detailed parameters are elaborated to align with the methodology proposed in Tang et al. (2023), ensuring a robust framework for cultivated land quality assessment.

where grade is the quality indicator for cultivated land quality evaluation, n is the number of indicators, i respect the ith evaluation unit, and j is the indicator grade. i is taken as 1,2,3…n and j is taken as 1,2,3,4.

The level of degradation of cultivated land quality was determined using the grade of the two-year cultivated land quality evaluation, calculated using Equation 2.

where Grade(d) is the degradation grade of arable land, Grade12 is the quality grade of arable land in 2012, and Grade19 is the quality grade of arable land in 2019; a positive number represents an increase in grade, that is, no degradation, and a negative number represents a decrease in grade, that is, degradation.

The type of degradation was categorized into four grades: no degradation (the evaluation grade remained unchanged or increased), slight degradation (the grade decreased by one level), moderate degradation (the grade decreased by two levels), and severe degradation (the grade decreased by three levels or more).

Random forest (RF) is a widely used integrated learning method for classification and regression tasks, proposed by Breiman and Cutler in 2001. The core idea is to improve the overall accuracy and robustness of the model by constructing multiple mutually independent decision trees and synthesizing the prediction results for each tree (Sheykhmousa et al., 2020; Smith and Wang, 2020). The construction of decision trees by randomly selecting features and data subsets greatly improves the resistance of the model to overfitting and performs well in dealing with high-dimensional data, noisy data, and nonlinear relationships (Jain and Jana, 2023). RF has been widely used in remote sensing image classification and for predicting forest degradation (Gong et al., 2018; Zhou et al., 2023). In this study, the RF model was implemented using the Scikit-learn library for Python 3.11.4. Environmental variables such as topography and climate were selected as inputs, and a spatial distribution map of soil attributes with spatial information was generated by training the model.

The optimal parameters of the model were configured as follows: number of estimation trees (n_estimators) = 300, maximum tree depth (max_depth) = 30, and maximum number of features (max_features) = 1.0. 10-fold cross-validation was used during the model training process and the random seeds of both the training and test datasets were set to 30 to ensure the stability and reproducibility of the results.

SHAP is a game theory-based analytical method for interpreting the output of complex machine learning models (Huang et al., 2023a). By calculating the contribution of each feature to the model predictions, SHAP provides an intuitive and quantifiable framework for model interpretation. Its core is based on the Shapley value, an assignment method derived from cooperative game theory that measures the impact of each feature on model predictions fairly, thus providing a solid theoretical foundation for the interpretation of complex models (Han et al., 2024). In recent years, SHAP has been widely used in biomedicine, environmental monitoring, and other fields to help decision-makers better understand the importance of features and their impacts on the final results by explaining the prediction mechanism of machine learning models (Zhang et al., 2023a; Du et al., 2024). The contribution rate of each characteristic variable to the target variable, represented by the Shapley value, can be calculated using the following formula.

Where φ i represents the Shapley value for feature i, S R i is the subset of features preceding i in a permutation R, ∣K∣ is the total number of features, and v ( S R i ) denotes the predictive contribution of subset S R i ​. The Shapley value quantifies the average marginal contribution of feature iii across all possible feature subsets.

Light Gradient Boosting Machine (LightGBM) is an efficient machine learning algorithm based on gradient boosting designed for processing large-scale data and high-dimensional features (Wei et al., 2021). Its core innovation lies in the leafwise growth strategy, which generates a deeper tree structure and improves the prediction accuracy of the model by prioritizing the leaf nodes with the largest error reduction during expansion. In addition, LightGBM discretizes features using a histogram-based algorithm, which significantly reduces memory occupation and computational cost and improves training efficiency (Wang et al., 2023a). Compared to traditional gradient boosting models, LightGBM can efficiently handle sparse data and large-scale datasets while maintaining good prediction performance and scalability (Lokker et al., 2024).

This study employs LightGBM and SHAP to identify the importance of multiple factors and explore the primary drivers of cultivated land quality degradation. Eleven factors from five aspects of topography, climate, location, socioeconomics, and agricultural management were selected that may affect cultivated land quality degradation, including Digital Elevation Model (DEM), Slope Direction (ASPECT), Slope Gradient (SLOPE), Mean Annual Precipitation (MAP), Mean Annual Temperature (MAT), Distance to the Rivers (Dist. Rivers), Distance to the Rodes(Dist. Roads), Population Density (POP), Gross Domestic Product (GDP), Straw Return Rate (SRR), Fertilizer Application Rate (FAR). This comprehensive selection ensures that various dimensions influencing land quality, such as physical environment, climatic conditions, accessibility, human activity, and agricultural practices, are represented in the model.

The integration of these factors into the LightGBM model, combined with the interpretative power of SHAP, enables a robust analysis of the underlying mechanisms and key contributors to farmland degradation in the study area.

Using the cultivated land quality degradation type identification model and degradation degree evaluation method, this study presents a comparative table of the degradation degree of each degradation type in the study area during the decade ( Table 3 ). The overall health of cultivated land quality degradation in the study area was good, with only approximately 11.86% showing signs of degradation, of which lightly degraded cultivated land accounted for 11.81% of the total cultivated land, with very little moderate degradation and no heavily degraded cultivated land. To visualize the spatial distribution of the degree of degradation of the overall quality of cultivated land in Yugan County, we measured the degree of degradation in each township and constructed a spatial distribution map ( Figure 3 ). Degradation was mainly concentrated in the central and western plains and along the rivers in the southern hilly areas, whereas the degradation degree of the northern lakeshore plain was the least significant.

Soil acidification accounted for 3.92% of the total cultivated land area, primarily at the slight degradation level (3.87%), followed by moderate degradation (less than 0.1%) ( Table 3 ). Fertility decline made up 27.55% of the total cultivated land quality degradation, predominantly light degradation, which accounted for 25.54% of the total cultivated land area. Moderate degradation represented 2% of the total cultivated land area, significantly higher than other types of degradation, while heavy degradation was observed in only 0.002% of the plots. Soil erosion contributed to 0.03% of total cultivated land quality degradation, mostly slight degradation, with almost no moderate or severe degradation. Lastly, physical structure degradation accounted for 0.13% of the total cultivated land quality degradation, with no cases of moderate or severe degradation. Taken together, most of the degradation types predominantly resulted in slight degradation, especially soil acidification and soil physical structure degradation. However, the proportion of moderate degradation of fertility decline and soil erosion was relatively high, which needs to be studied further. Heavy degradation is rare and occurs only during fertility declines and soil erosion.

To visualize the spatial distribution of different types of cultivated land quality degradation and their degree of degradation in Yugan County, we measured the degradation types and degree of degradation in each township and drew corresponding spatial distribution maps ( Figure 4 ). The results showed that the spatial differentiation pattern of cultivated land quality degradation presented obvious regional differences. Soil acidification was widely distributed throughout the county, with a lower degree of acidification in the northern lakeshore plain area, whereas acidification is more severe in the central and southern hilly areas, especially in the central area, where acidified cultivated land is more concentrated ( Figure 4A ). In contrast, fertility decline has the widest impact and is particularly problematic in the central, river, and lake littoral areas ( Figure 4B ). Soil erosion was mainly concentrated in the southern hilly areas and along rivers and lakes. Although the overall distribution was small, it had significant impacts in localized areas ( Figure 4C ). In contrast, the area of soil erosion was limited, and the degree of degradation was relatively mild. The physical structure of the soil was degraded to the least extent and was mainly distributed in the southern hilly area, with a relatively scattered and limited range ( Figure 4D ). Comprehensive analysis showed that fertility decline in the central part and soil acidification in the south were the most prominent problems, whereas the impacts of soil erosion and physical structure degradation were more localized.

In this study, 11 factors influencing the degradation of cultivated land quality were systematically selected, encompassing five dimensions: topography, climate, geographic location, socio-economic conditions, and agricultural management practices. After the VIF test (VIF < 5) for the 11 factors, it was confirmed that there was no multicollinearity problem for these independent variables. On this basis, the influence characteristics of each driving variable on the degradation of comprehensive quality of cultivated land and different cultivated land quality degradation types were analyzed by the SHAP value interpretive method, and the results are shown in Figure 5 and Figure 6 . The target variable for this study is the score for a particular type of degradation, where an increase in the score indicates mitigation of degradation and vice versa. For example, an increase in the soil acidification score mitigates acidification, whereas a decrease in the score intensifies acidification.

By analyzing the SHAP values, this study revealed the main factors influencing the degradation of each type of cultivated land. Overall, MAP and agricultural management factors had significant effects on the different types of degradation but the drivers of each degradation type differed. For soil acidification, MAP was the most important mitigating factor, and its increase significantly reduced the acidification risk, whereas the excessive application of fertilizers exacerbated the acidification process. The fertility decline was mainly driven by precipitation and straw return rate, both of which increased to help mitigate the fertility decline, whereas excessive fertilizer application exacerbated this problem. Soil erosion was mainly affected by slope, with larger slopes exacerbating erosion; however, appropriate increases in annual precipitation and rational fertilizer use helped to mitigate it. The degradation of the soil physical structure was more mitigated by the straw return rate; however, it tended to be exacerbated in areas with greater slopes and the degradation was more pronounced farther away from the river. Changes in the overall cultivated land quality were also positively correlated with precipitation and straw return rate but excessive fertilizer application negatively affected the overall quality.

This study indicates that cultivated land in the Poyang Lake area is facing a series of degradation risks, including soil acidification, soil fertility decline, soil erosion and soil physical structure degradation, a finding that is consistent with existing research. For example, Zeng et al. conducted a spatial assessment of farmland soil pollution across China, revealing similar degradation trends in other regions, such as a decline in soil fertility and the accumulation of heavy metals, which may lead to significant ecological and health risks (Zeng et al., 2019). In the Poyang Lake area, the long-term application of chemical fertilizers has led to a decline in soil organic matter content, looser and less stable soil structure, and fertility decline (Liu et al., 2023). Excessive use of nitrogen and phosphorus fertilizers breaks the acid-base balance of the soil, which affects the healthy growth of the crops and increases the risk of soil acidification (Huang et al., 2012). Intensive land use in areas with high precipitation and complex terrain has led to the destruction of vegetation and degradation of the soil physical structure, triggering serious soil erosion and nutrient loss (Yuan et al., 2016; Zhang et al., 2021b). These issues are further corroborated by Zeng et al., who highlighted the widespread challenges of soil fertility decline and contamination across China (Zeng et al., 2019). If these issues are not addressed, they could lead to a series of environmental and socio-economic problems, including water eutrophication (Cai et al., 2017), acid rain, and a decline in crop yields (Zhang et al., 2021a), which would severely impact both the environment and local livelihoods. This study further highlights that the most prominent risk in the region is the degradation of arable land fertility, which poses a significant threat to sustainable agriculture. Given the urgency of this issue, it is critical to implement nature-based solutions (NbS) and other adaptive management strategies to mitigate the negative impacts and restore soil health.

The mean annual precipitation, straw return rate, and fertilizer application rate exhibited important effects on various types of cultivated land quality degradation. High precipitation exacerbates soil erosion, especially on slopes or in areas where soil and water conservation measures have not been implemented. Precipitation in the Poyang Lake Basin is concentrated during the rainy season and its complex topography makes precipitation-induced erosion a serious problem (Chen et al., 2020). Under excessive precipitation, topsoil is easily washed away, leading to the loss of soil organic matter and nutrients, a gradual decline in soil fertility, and destruction of the physical structure of the soil, making it loose and unstable (Zhan et al., 2023). In addition, precipitation affects the acid-base balance of the soil and can exacerbate soil acidification, especially under the condition of long-term application of nitrogen fertilizer (Guo et al., 2018). Higher precipitation increases the risk of nutrient loss, especially soluble nutrients, such as nitrogen and phosphorus, which are more likely to be leached by rainfall into groundwater or discharged into river systems, weakening soil fertility (Siddique et al., 2020). In terms of soil structure, excessive precipitation leads to the dispersion of soil particles, weakening the water-holding and carrying capacity of the soil, making it more susceptible to erosion, and thus accelerating the degradation of the physical structure of the soil (Jin et al., 2021).

Straw returning can increase soil organic matter content, improve soil structure, and alleviate soil acidification (Hu et al., 2023b). Straw returning strengthens the carbon and nitrogen cycles and increases soil microbial activity, thus enhancing the water and fertilizer retention capacity of the soil and reducing the risk of soil erosion (Shi et al., 2022). In addition, straw returning can reduce the dependence on chemical fertilizers and the associated problems stemming from their overuse, including increased soil degradation (Wang et al., 2023b) and acidification and decreased soil fertility (Islam et al., 2023).

Runoff and leaching transport excess fertilizer nutrients into water bodies, aggravating eutrophication and pollution problems (Zhang et al., 2021a). To address overfertilization, fertilizer management strategies such as soil testing and formulation have been developed. Precise fertilization plans can reduce the waste of chemical fertilizers and alleviate soil acidification and water pollution (Smith and Siciliano, 2015). Appropriate farmland management practices, such as the combination of straw return and rotational cropping systems, can also significantly improve soil health and promote the sustainable use of arable land (Zhang et al., 2024b).

This study found that the effect of river distance on the degradation of soil physical structure in the Poyang Lake area was particularly significant, supporting the results of past studies (Yuan et al., 2016; Li et al., 2021a). Simultaneously, population growth had a significant effect on the fertility decline of arable land, which was observed in past studies (Bao et al., 2021; Liu et al., 2021).

Vegetation-based cultivated land restoration solutions offer ecologically friendly and economically feasible strategies for addressing cultivated land degradation by optimizing vegetation selection and management (Feng et al., 2022). These solutions leverage the ecological functions and regulatory capacities of vegetation to effectively mitigate various types of soil degradation (Seddon et al., 2020). In the Poyang Lake region, such approaches have shown significant potential and effectiveness in tackling key challenges, including soil acidification, soil fertility decline, soil erosion, and soil structural degradation (Zhang et al., 2023b).

Soil acidification, the prominent degradation issue in the Poyang Lake region, often results from prolonged fertilizer use and acid rain. In this context, Cerda et al. (Spain) found that planting acid-tolerant species and employing vegetative cover are essential for mitigating soil acidification. For example, in acidic soils, Lolium perenne and Trifolium repens improve soil microbial activity and enhance buffering capacity through root exudates (Cerdà et al., 2022). Furthermore, straw incorporation has proven to be a highly effective strategy for neutralizing acidic soils. Research shows that applying 6–10 tons of straw per hectare significantly raises soil pH and increases the availability of calcium and magnesium ions. This method chemically balances soil acidity through the release of organic carbon and carbonate ions during straw decomposition (Chen et al., 2023). Such integrated vegetation management strategies provide a scientific foundation for the sustainable development of agriculture in the region.

Cultivated land fertility decline, largely caused by long-term monocropping and nutrient overexploitation, remains a critical issue. A typical approach is to plant cover crops, such as legumes like Medicago sativa and Trifolium repens, which, by forming symbiotic relationships with rhizobia, significantly increase soil nitrogen content and improve microbial activity in the soil (Lal, 2022). Furthermore, implementing crop rotation and intercropping systems, such as alternating leguminous crops with cereal crops, effectively alleviates nutrient depletion caused by monoculture practices and promotes nutrient cycling (Zhang et al., 2016). Studies in the Poyang Lake region indicate that incorporating 6–10 tons of straw per hectare, combined with moderate nitrogen fertilization (30–50 kg/ha), can substantially increase soil organic matter and enzyme activity, thereby enhancing nutrient availability. This practice releases significant amounts of nutrients during straw decomposition while optimizing microbial community structure, effectively restoring soil fertility (Lin et al., 2017; Hu et al., 2023a).

Soil erosion presents a major challenge for sloped cultivated lands in the Poyang Lake region, particularly during the rainy season. Vegetative buffer strips, an essential vegetation-based solution, effectively reduce surface runoff and soil erosion. The research conducted in the Poyang Lake region showed that using vegetative buffer strips with species such as Salix spp. and Phragmites australis along sloped cultivated lands not only reduces sediment loss but also contributes to stabilizing soil structure (Huang et al., 2023b). Similarly, straw mulching has proven highly effective in preventing soil erosion. Studies reveal that applying 6–8 tons of straw per hectare can reduce soil erosion rates by 35%–50% (Xing et al., 2023). In southern China, deep-rooted plants such as Medicago sativa significantly improve soil infiltration and stability, reduce erosion and enhance water retention and utilization efficiency (Xie et al., 2022). Additionally, Robinia pseudoacacia, used as a biological barrier, provides effective runoff interception and enhances soil stability on sloped lands through its deep-root system and dense canopy (Pan et al., 2024).

Soil physical structural degradation, characterized by compaction, reduced porosity, and poor aeration, significantly impairs root growth and water circulation. In the Poyang Lake region, planting deep-rooted crops like Brassica napus and Beta vulgaris effectively addresses these issues. The deep roots of these crops penetrate compacted layers, improving aggregate stability and enhancing soil aeration and water retention (Zhang et al., 2014). Additionally, incorporating biochar derived from straw into the soil has emerged as an innovative solution. This practice not only increases soil porosity and water retention but also creates favorable habitats for microorganisms, further enhancing soil ecological functionality (Xu et al., 2021). These solutions provide a scientific pathway for restoring heavily degraded cultivated land in the Poyang Lake region while underscoring the critical role of vegetation in soil health management.

In conclusion, vegetation-based natural solutions fundamentally improve soil health and ecosystem service capacity through scientific vegetation management (Qiu et al., 2024). From employing cover crops to enhance fertility to utilizing deep-rooted plants and straw incorporation technologies to improve soil structure, these strategies highlight the central role of vegetation in mitigating cultivated land degradation. Additionally, straw incorporation, as a circular utilization of vegetative resources, demonstrates comprehensive benefits in enhancing soil fertility, alleviating acidification, and stabilizing soil structure. Moving forward, integrating digital agriculture technologies and region-specific management strategies will be crucial for optimizing these solutions and achieving greater ecological and economic benefits.

While this study advanced the understanding of cultivated land quality degradation risks and drivers in the Poyang Lake area, there were some limitations. First, although the use of LightGBM and SHAP improved the accuracy to a certain extent, these models have limitations in capturing complex nonlinear relationships and may fail to comprehensively reflect all potential influencing factors. In addition, key variables, such as socioeconomic factors, were not fully incorporated into the analytical framework, which may have led to the limited comprehensiveness of some of the findings. Furthermore, the data and indicators used in this study may have some measurement errors, and the accuracy of the soil’s physical and chemical properties may have been affected by differences in local monitoring methods. Therefore, future studies should consider expanding the scope of data collection, incorporating more variables, and exploring different analytical models to comprehensively reveal the complex mechanisms of cultivated land quality degradation.