关键词: 熵权TOPSIS法, XGBoost算法, 耕地质量, 耕地整治, 可恢复耕地, 局部空间自相关

Abstract: Taking the current cultivated land plots as the evaluation units, an evaluation index system for cultivated land quality was constructed based on three influencing factors: natural endowment, location conditions, and cultivated land contiguity. The comprehensive scores of the current cultivated land plots were calculated using the entropy-weighted TOPSIS method. Subsequently, the XGBoost algorithm was employed to train the evaluation indicators and comprehensive score data of the current cultivated land, thereby building a predictive model. By applying the evaluation indicator data of restorable cultivated land plots and this predictive model, the comprehensive scores of restorable cultivated land plots could be quickly and accurately obtained. Finally, the local spatial autocorrelation method was used to quantitatively analyze the spatial clustering characteristics of the comprehensive evaluation results, and the matrix method was applied to determine the delineation of cultivated land consolidation zones. The results indicated that the most influential indicators in the comprehensive evaluation index system for cultivated land quality in Panyu District were soil pH level, cultivated land contiguity, and soil texture level. The predictive model constructed using the XGBoost algorithm achieved an R² of 99.94%, demonstrating excellent predictive performance and verifying the accuracy of the entropy-weighted TOPSIS algorithm in weight determination. The total area of restorable cultivated land plots in the priority consolidation zones of Panyu District was 1 935.56 hm². The current land use types in these priority consolidation zones were mainly garden land, covering an area of 1 139.44 hm², which was closely related to local agricultural structural adjustments. The method for delineating cultivated land consolidation zones, which integrated the entropy-weighted TOPSIS and XGBoost algorithms while considering the spatial aggregation of current cultivated land and restorable cultivated land resources, could objectively and accurately identify plots for cultivated land consolidation and restoration.

Key words: entropy weight TOPSIS method, XGBoost algorithms, quality of cultivated land, cultivated land remediation, restorable cultivated land, local spatial autocorrelation