Quantifying and revealing cultivated land ecological compensation performance: A machine learning-based approach

Abstract

Cultivated land ecological compensation enables the rational allocation of ecological resources and promotes the integrated development of the cultivated land system. Evaluating the effectiveness of ecological compensation policies helps reveal the extent of benefits, while identifying influencing factors uncovers critical bottlenecks hindering policy implementation. However, current research primarily focuses on compensation amounts and mechanisms, with few studies on evaluating performance and influencing factors. To fill this gap, this paper introduces an indicator called cultivated land ecological compensation performance, which assesses the specific impacts of cultivated land ecological policies on the ecological, economic, and social systems. The spatiotemporal variations of this indicator across mainland China and its provinces were analyzed, showing an increasing trend on the temporal scale and higher values clustering in the eastern part on the spatial scale. To objectively identify the influencing factors, the XGBoost model, combined with SHAP and PDP methods, was used to explore how univariate and multivariate factors influence this indicator. Through a series of robustness tests, the results confirm the effectiveness of the XGBoost model in identifying the influencing factors. At both the national and regional levels, environmental protection expenditure, GDP, and total machinery power all have a positive impact, and the collaboration between central fiscal support and local economic development is particularly notable. The findings validate the effectiveness of policies aimed at ecological compensation and furnish an empirical foundation for policy optimization.