Cadmium adsorption-desorption synergy drives spatial heterogeneity of grain cadmium enrichment in a subtropical limestone paddy region.

Traditional risk assessment methods frequently overlook the dynamic adsorption-desorption processes of cadmium (Cd) at solid-liquid interfaces and underutilize the integrative predictive potential of easily measurable soil indicators. This study consolidated data from literature, field experiments, and regional surveys to build a database on Cd adsorption-desorption in subtropical paddy soils, and introduced a machine learning-based approach to infer soil Cd adsorption-desorption capacity - a composite indicator of Cd retention-release potential predicted from readily available soil properties serving as proxy indicators - and combined this with total soil Cd to assess the spatial heterogeneity of regional Cd enrichment in rice grains. Then the application of adsorption-desorption capacity was focused on a typical limestone-dominated area characterized by complex soil-rice Cd relationships. Analysis of 336 soil-rice pairs indicated that rice grains Cd content remained low (median 0.04 mg kg-1), while total soil Cd showed significant accumulation (median 0.82 mg kg-1), highlighting a pronounced spatial dissociation between them. Soil Cd adsorption capacity was negatively correlated with rice Cd, whereas desorption capacity exhibited a positive correlation. Multiscale geographically weighted regression (MGWR) was applied, yielding a high goodness-of-fit (R2 = 0.73), confirming that the adsorption-desorption capacity combined with total soil Cd, effectively predicts the spatial heterogeneity of rice Cd under the MGWR framework. MGWR analysis further revealed multi-scale influence patterns: the regulatory effect of desorption capacity was approximately twice that of adsorption capacity. Rice Cd enrichment in 74.7 % of the study area was controlled by the synergistic effect of adsorption-desorption, operating at a township/small watershed scale. In contrast, total soil Cd was significant only at a semi-global scale (bandwidth 146, covering 43.5 % of the study area) and exerted a comparatively weak influence. This study presents a spatially explicit approach for assessing rice Cd risk by integrating inverted adsorption-desorption attributes with multi-scale modeling.