An ensemble learning-based approach for large-scale cultivated land quality monitoring

Cultivated land quality is the foundation of agricultural production, directly affecting food security and the stability of the ecological environment. Based on MODIS remote sensing data, the Harmonized World Soil Database (HWSD), terrain information, and other data, combined with GBDT and Bagging ensemble learning algorithms, this study investigates the mapping and validating of cultivated land quality in Northeast China, exploring the feasibility of assessing cultivated land quality in this region using remote sensing data. The results show that over 95 % of the cultivated land in Northeast China has high quality, and remote sensing technology demonstrates high accuracy in identifying high-quality land. However, notable limitations exist in assessing low-quality croplands, with remote sensing detecting substantially lower proportions of Grade 3 croplands (0.50–0.60 %) relative to GIS-derived results (1.24–4.56 %). the assessment of low-quality land has certain limitations, mainly due to the insufficient capability of remote sensing data in identifying complex soil types and low-quality cultivated land. This study provides a new technical approach and theoretical support for cultivated land quality monitoring.