Farmland Soil pH in Guizhou, China: Spatial Patterns and Associated Factors

The spatial structure and multivariable associations of farmland soil pH remain poorly resolved in mountainous landscapes. We analyzed 487 topsoil samples (0–30 cm) from Guizhou, China, using spatial autocorrelation, geostatistics, random forest (RF), and structural equation modeling (SEM). Soil pH ranged from 4.37 to 8.45, with 17.9% of samples below the pH 5.5 screening threshold. Because sampling was not area-weighted, this percentage does not estimate province-wide areal prevalence. Coordinate-level means of pH, soil organic matter (SOM), and cation exchange capacity (CEC) exhibited modest positive global spatial autocorrelation. Local pH evidence after multiplicity correction was limited and neighborhood-dependent. The pH semivariogram indicated short-range dependence and substantial unresolved variation. RF showed modest performance under random and spatially blocked nested cross-validation, with higher errors under spatial validation. CEC ranked first in importance, followed by climate- and terrain-related variables. SEM explained approximately one-fifth of pH variance. Within the model, CEC had the largest standardized direct coefficient in the pH equation, SOM showed a positive indirect association with pH through CEC, and an ordered landform-relief score showed a negative total conditional association. Climate-related pathways yielded small, nonsignificant total associations. These findings inform regional screening and field-level diagnosis of soil acidity, nutrient availability, and exchange status.