Investigating the Spatial Distribution and Influencing Factors of Non-Grain Production of Farmland in South China Based on MaxEnt Modeling and Multisource Earth Observation Data.

Excessive non-grain production of farmland (NGPF) seriously affects food security and hinders progress toward Sustainable Development Goal 2 (Zero Hunger). Understanding the spatial distribution and influencing factors of NGPF is essential for food and agricultural management. However, previous studies on NGPF identification have mainly relied on high-cost methods (e.g., visual interpretation). Furthermore, common machine learning techniques have difficulty in accurately identifying NGPF based solely on spectral information, as NGPF is not merely a natural phenomenon. Accurately identifying the distribution of NGPF at a grid scale and elucidating its influencing factors have emerged as critical scientific challenges in current literature. Therefore, the aims of this study are to develop a grid-scale method that integrates multisource remote sensing data and spatial factors to enhance the precision of NGPF identification and provide a more comprehensive understanding of its influencing factors. To overcome these challenges, we combined multisource remote sensing images, natural/anthropogenic spatial factors, and the maximum entropy model to reveal the spatial distribution of NGPF and its influencing factors at the grid scale. This combination can reveal more detailed spatial information on NGPF and quantify the integrated influences of multiple spatial factors from a microscale perspective. In this case study of Foshan, China, the area under the receiver operating characteristic curve is 0.786, with results differing by only 1.74% from the statistical yearbook results, demonstrating the reliability of the method. Additionally, the total error of our NGPF identification result is lower than that of using only natural/anthropogenic information. Our method enhances the spatial resolution of NGPF identification and effectively detects small and fragmented farmlands. We identified elevation, farming radius, and population density as dominant factors affecting the spatial distribution of NGPF. These results offer targeted strategies to mitigate excessive NGPF. The advantage of our method lies in its independence from negative samples. This feature enhances its applicability to other cases, particularly in regions lacking high-resolution grain crop-related data.