A novel spatial stratification and performance evaluation framework for optimizing soil environmental quality monitoring networks

This study develops and evaluates a spatially stratified sampling framework for optimizing soil environmental quality monitoring networks. The framework aims to identify stratification strategies that maintain reliable monitoring accuracy while improving sampling efficiency under different precision requirements.

This study used 1,416 sampling sites from Jiangxi Province, China. Four stratification methods were compared: administrative division stratification, land use type stratification, K-means clustering stratification, and minimum spanning tree (MST)-based spatial clustering stratification. The optimal number of retained monitoring sites was determined by analyzing the relationship between monitoring sites and prediction error. Moran’s I was used to quantify intra-stratum spatial autocorrelation and to evaluate how spatial dependence affected monitoring network optimization.

Administrative division stratification and MST-based spatial clustering showed stronger overall performance than the other strategies, but their advantages depended on the precision requirement. Under the 20% relative error rate threshold, administrative division stratification required the fewest retained monitoring sites (n = 628). For routine monitoring with lower precision requirements, approximately 400 retained sites may be sufficient based on the stabilization of ASE and NRMSE, with MST-based spatial clustering showing comparatively stable uncertainty estimation. A significant negative correlation (r = -0.42, p < 0.01) was observed between Moran’s I and ASE, indicating that strata with stronger spatial autocorrelation generally supported more efficient network optimization.

The proposed spatially stratified sampling framework provides a quantitative basis for optimizing regional soil environmental quality monitoring networks. The results show that no single stratification strategy is universally optimal; instead, strategy selection should depend on the target precision, spatial coherence, and practical implementation requirements of the monitoring program.