Research Progress and Prospects of the Geographical Detector

Abstract

Geographical detectors have become a widely used statistical tool in soil science for identifying dominant factors and interactions driving spatial differentiation of soil properties. This paper systematically reviews the basic principles, core functions (factor detection, interaction detection, risk area detection, and ecological detection), and application progress of the geographical detector in soil heavy metal pollution identification, soil erosion risk assessment, and digital soil mapping. By analyzing existing case studies, we summarize current limitations including sensitivity to discretization of continuous variables, scale dependence, lack of causal inference, insufficient characterization of interaction mechanisms, static single-time-snapshot analysis, and absence of systematic uncertainty evaluation. To address these challenges, future directions are proposed: (1) developing adaptive discretization methods (e.g., MIC, genetic algorithms) and continuous geographical detectors; (2) conducting multi-scale modeling across watershed, county, and national levels; (3) coupling with causal inference frameworks (SEM, Bayesian networks, causal forests); (4) integrating machine learning (random forest, XGBoost, SHAP) and nonlinear models (GAM) to quantify interaction forms; (5) constructing spatiotemporal geographical detectors using multi-temporal remote sensing data; (6) establishing a standardized uncertainty assessment framework based on NUSAP and Monte Carlo simulation; and (7) fusing multi-source big data (Sentinel-2, GEDI, MODIS) with deep learning. These advancements will significantly enhance the scientific value of geographical detectors in soil pollution source tracing, erosion risk early warning, and ecological restoration decision-making, with important implications for soil protection in key regions such as the black soil region of Northeast China and the karst region of South China.