The influences of different land uses associated with human activities on soil quality and the redistribution of heavy metal in soil have been widely concerned. Surface soil samples were obtained to assess comprehensive soil quality in a typical red soil region of southern China, combining the heavy metal pollution evaluation with fertility evaluation. It can be learned from the results that the overall level of soil fertility was at medium and lower level, and soil heavy metal pollution risk in the study area in a few regions had reached warning line and slight pollution line, and there was a risk of potential pollution. TOPSIS evaluation results showed that the comprehensive soil quality was mainly good quality and moderate quality, accounting for 31.7% and 29.0% of the total land area, respectively. Positive matrix factorization (PMF) model results showed that transportation source contributes a lot in terms of Cd and Pb. As for Cr, natural source contributes 53.8%. In terms of Cu and Zn, agriculture source contributes 50.7% and 38.7%, respectively. In a word, the comprehensive soil quality assessment in red soil region of southern China provides an important basis for the scientific management and sustainable utilization of soil resources.

Cow dung reduces the bioavailability of heavy metals (HMs) in wheat soils. However, interactive influence of soil chemical properties, microbial communities and functional genes in HM immobilization need to be further investigated. Therefore, the effects of cow dung on heavy metals content in soil and wheat, soil microbial community structure, and heavy metal resistance genes (MRGs) were tested with pot experiment. Cow dung reduced the bioavailable forms of Cd by 8% to 21%, thereby reducing the Cd content of wheat grains by 30.5% to 46.1%. Bioavailable forms of Cd in soil were significantly and negatively correlated with soil chemical properties, relative abundance of microbial genera (such as Bradyrhizobium spp. and Mycolicibacterium spp.), and relative abundance of MRGs (such as actA and fpvA). Concurrently, the bioavailable forms of Cu and Zn exhibited significant correlations with Shannon index and MRGs (such as copC). These results indicated that changes in soil environmental factors (soil chemical properties, microbial communities and MRGs) are conducive to reducing the bioavailability of Cd. However, the high levels of Cu and Zn in cow dung may also disrupt soil ecology, resulting in a decline in microbial diversity and impacting the abundance of MRGs. Therefore, the rates of application of cow dung should be carefully selected and should not exceed 2.5% (w/w). These results may provide scientific evidence for the safe use of cow dung and remediation of heavy metals in wheat soils.

Cadmium (Cd) pollution poses a severe threat to rice safety and human health, while traditional linear models exhibit significant limitations in predicting rice Cd accumulation due to environmental complexities. This study systematically evaluated the predictive performance of Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Residual Neural Networks (ResNet), using a multi-source soil–rice dataset comprising 57,200 samples from Hunan Province. The results showed that the RF model performed best on the test set (R² = 0.62), with the dominant features being soil’s available Cd (contributing 9.74%) and precipitation during the rice-filling stage (joint contribution of 15.96%). However, the model’s predictive performance experienced a sharp decline on the independent 2023 validation set comprising 393 samples from Yizhang County and Lengshuitan District, with R² values ranging from −0.12 to −0.31. This highlighted the fundamental limitations of static data-driven paradigms. Agronomic management measures, simplified by heterogeneous data and binary encoding, failed to effectively represent the actual intervention intensity. The study demonstrated that while machine learning models captured nonlinear relationships in laboratory environments, they struggled to adapt to the dynamic interactions and spatiotemporal heterogeneity of farmland systems. Future efforts should focus on developing hybrid models guided by mechanistic insights, integrating dynamic environmental processes and real-time data, and promoting localized “one model per region” strategies to enhance predictive robustness. This study provides methodological insights for the technological transformation of agricultural artificial intelligence, emphasizing that the deep integration of data-driven approaches and mechanistic understanding is crucial for overcoming the “last mile” challenge.