GA-Optimized Sampling for Soil Type Mapping in Plain Areas: Integrating Legacy Maps and Multisource Covariates

Abstract

Soil mapping plays a crucial role in optimizing agricultural production by providing spatially explicit information on soil types and properties, which supports decision-making in precision fertilization, irrigation, and crop selection. Traditional soil mapping methods, which rely on field surveys and laboratory analyses, face challenges related to efficiency and scalability. Although combining legacy soil maps with environmental covariates can reveal soil–environment relationships and improve sampling layouts, low soil spatial variability and significant human activity in plain areas often hinder the effectiveness of existing algorithms, making them sensitive to sample density and environmental variability. This study proposes a genetic algorithm (GA)-based sampling optimization framework tailored to plain areas with low soil spatial variability. By integrating legacy soil maps and environmental covariates, the GA dynamically balances spatial dispersion and environmental representativeness, addressing the limitations of traditional methods in homogeneous landscapes. In a case study conducted in Tongzhou District, Beijing, China, the GA sampling method combined with random forest modeling, applied to soil type mapping, achieved the highest kappa coefficient of 70.25% with 5000 sampling points—an average improvement of 10% over fuzzy C-means clustering and K-nearest neighbor methods. Additionally, field-validated accuracy reached 89.69%, representing a 13% improvement over the other methods. This study demonstrates that the GA-based sampling approach significantly enhances sample representativeness and efficiency, thereby improving the accuracy of digital soil mapping. The proposed method offers an efficient and reliable solution for soil mapping in plain areas, contributing to optimized land use and more informed precision agriculture decisions.