An Outlier-Aware Hybrid Framework for Spatial Prediction and Risk Assessment of Agricultural Soil Heavy Metal Pollution
Abstract
Agricultural soils can contain heavy metals from natural materials and human activities, but their concentrations rarely follow a uniform pattern across a landscape. Direct sampling and laboratory analysis provide reliable measurements at selected locations; they do not, by themselves, describe the many unsampled places between them. Prediction is further complicated when a small number of observations differ sharply from their neighbours. Such observations may be measurement errors, but they may also mark genuine local pollution. This paper reviews spatial prediction and risk-assessment methods for agricultural soil heavy metals, with particular attention to the complementary ideas in two published hybrid studies: one separates normal and outlierinfluenced areas before prediction, while the other combines semi-supervised concentration prediction, pollution classification, and geographic information system analysis. Building on these studies and related reviews, the paper presents a conceptual outlier-aware workflow that links data checks, spatial outlier investigation, region appropriate prediction, uncertainty assessment, risk interpretation, and mapping. The workflow is a proposed synthesis, not an experimentally validated model. The review argues that reliable pollution maps should preserve meaningful hotspot signals, use spatially appropriate validation, and communicate uncertainty alongside predicted concentrations and risk classes.