Epigaeic arthropods are a vital component of agricultural landscape biodiversity. Research on their large-scale spatial distribution is essential for biodiversity conservation and sustainable land management. However, accurately predicting the spatial distribution patterns of epigaeic arthropods at the county scale remains challenging due to sampling difficulties and uncertainties in model applicability. This study was conducted in Changtu County, Liaoning Province. Using data from 120 sampling points and eight environmental variables, we systematically compared the performance of three models—Co-Kriging, Geographically Weighted Regression (GWR), and Random Forest (RF)—in predicting the spatial distribution of epigaeic arthropods. The results indicated that the GWR model achieved the highest prediction accuracy (R 2 = 0.72, MAE = 10), significantly outperforming RF (R 2 = 0.54) and Co-Kriging (R 2 = 0.38). NDVI, Rainfall, and PR were identified as key factors influencing their distribution. In Changtu County, the number of epigaeic arthropods exhibited a spatial pattern of “high in the southeast and low in the northwest.” This study confirms that in agricultural-dominated regions, local regression models that consider spatial non-stationarity (e.g., GWR) are more suitable for predicting the large-scale distribution of epigaeic arthropods, providing a theoretical basis for targeted zoning management. The findings offer a new methodology and scientific basis for farmland biodiversity assessment and landscape optimization at the county scale.
Guo et al. (Tue,) studied this question.