Estimating the spatial distribution and exploring the factors influencing cultivated land quality through a hybrid random forest and Bayesian maximum entropy model.

Cultivated land is one of the most valuable agricultural resources; its quality is not only the foundation of national food security but also a crucial issue for global sustainable development. However, owing to data limitations and spatial heterogeneity, large-scale cultivated land quality (CLQ) estimation remains challenging, which further leads to insufficient exploration of the mechanisms of CLQ variation. In this research, a hybrid random forest (RF)-Bayesian maximum entropy (BME) method integrating a field-sampled dataset and environmental auxiliary information was proposed to predict the characteristics and main factors influencing CLQ efficiently. The results demonstrated that the hybrid RF-Bayesian maximum entropy model substantially outperformed the ordinary kriging (OK), BME, RF and RF-OK models, with the overall estimation accuracies increasing by 39.92 %, 29.64 %, 29.33 % and 14.72 %, respectively. The CLQ in the research area was generally moderate, with high values occurring mainly in the northern and eastern coastal plains, medium values occurring mostly in the central hilly basin area and eastern hilly islands, and low values occurring primarily in the western hilly and southern mountainous areas. Natural factors influenced CLQ the most, with a relative importance of 53.4 %, followed by soil and anthropogenic factors, with relative importance values of 25.8 % and 20.8 %, respectively. The hybrid model could efficiently account for the correlation between CLQ and environmental variables, reducing the need for expensive field sampling and mitigating prediction errors. Overall, the proposed research framework emphasizes the potential of multisource heterogeneous datasets and RF- and BME-based strategies in promoting sustainable CLQ management.