Spatiotemporal modelling of cropland soil pH dynamics in Southern China using a process-guided machine learning

Intensified fertilization and continuous cultivation have substantially accelerated cropland soil acidification in China, posing increasing risks to agricultural sustainability and ecological security. However, existing approaches are often limited in their capacity to simultaneously reconstruct long-term soil pH trajectories and resolve the mechanisms underlying soil pH dynamics at high spatial resolutions. To address this gap, we developed a process-guided machine learning framework (meta model) that couples long term soil pH series simulated by the VSD+ model with Random Forest spatial extrapolation. We applied this meta model to assess spatiotemporal pH dynamics in Chengmai (dominated by Ferralsols) and Yujiang (dominated by Acrisols) from 1980 to 2023. The proposed meta model demonstrated high performance to capture major spatial heterogeneity (R 2 > 0.6). Our results reveal a persistent acidification trend in the two regions, with a cumulative pH decline of 0.48 units and 0.42 units in the last 40 years. Temporally, Chengmai followed a staged pattern that rapid early acidification followed by deceleration and partial recovery, and Yujiang showed a more persistent declining trend, although at a relatively moderate rate after 2010. Spatially, upland soils generally showed more monotonic declines than paddy soils, with areas of pH < 5.5 becoming dominant across the two regions. Mechanistic analysis indicates that acidification is governed by climatic and geological sensitivities, with nitrogen inputs regulating acid load intensity and sustained biomass harvest and HCO 3 − leaching, driving base cation depletion. Crucially, we identified a progressive shift in soil buffering mechanisms from base cation exchange to aluminum-iron reactions. By integrating mechanistic consistency with high-resolution mapping capabilities, this framework provides a robust basis for diagnosing long-term soil acidification and informing targeted management strategies in intensively cultivated regions.