Improving Accuracy of Gridded Livestock Mapping by Combining Trend Modeling and Residual Assignment

ABSTRACT The accurate distribution of livestock is a fundamental data set upon which sustainable land development in agricultural and pastoral areas can be promoted. Nevertheless, the current gridded livestock mapping methods exhibit considerable scope for improvement in accuracy, given that the issue of residual assignment has not been addressed effectively. We attempt to propose a combinatorial methodological framework that decomposes gridded livestock mapping into two stages, trend modeling, and residual assignment, through the perspective of uncertainty. Investigating the relative efficacy of different combinations of methods within the framework aims to identify the optimal total mapping solution. Ridge regression (RR), geographically weighted ridge regression (GWRR), and random forest (RF) methods with varying characteristics were used to represent the trend modeling component and combined with area‐to‐point kriging interpolation based on spatial autocorrelation for residual assignment, respectively. The Inner Mongolia Autonomous Region from China provides a multi‐scale case validation of the framework. The experimental results show that GWRR has high robustness and prediction accuracy for predicting livestock distribution trends across scales. By using residual assignment optimization as a post‐processing tool, the total accuracies of the three trend models were improved in different degrees, with the greatest accuracy gains obtained by RF (RMSE reduced by 13.731 and MAE reduced by 12.247, in units/km 2 ). Therefore, the combined two‐stage methodological framework proposed in this study can effectively improve the accuracy of gridded mapping. This provides an effective complement to current gridded livestock mapping, thereby facilitating more sustainable land use, ecological environmental protection, and livestock development.