Prediction of lead accumulation risk and safe planting zone delineation of rice in Guizhou Province using machine learning.

The uneven distribution of lead (Pb) in rice and soil across the primary rice-growing regions of southern China has led to challenges in assessing rice quality and associated health risks. Therefore, it is crucial to develop a fast and precise method for forecasting the accumulation of Pb in soils and rice to evaluate the environmental risks of heavy metals. We utilized eight machine learning models to fit the training data and find the optimal model based on 1,396 pairs of soil-rice samples collected during field surveys in Guizhou Province. Among them, the random forest model achieved higher prediction accuracy (rice: R² = 0.486; soil: R² = 0.518) and was further optimized using a Bayesian optimizer to enhance its performance (rice: R² = 0.662; soil: R² = 0.718). The importance of characteristics showed that annual precipitation and soil effective state were the main factors affecting rice Pb accumulation; distance to the nearest mine and annual rainfall were the main factors affecting total soil Pb. The area with higher risk of Pb accumulation in soil was located in the western part of Bijie, while the area with higher risk of Pb accumulation in rice was located in the southern part of Tongren. There were some differences between the two. About 88% of the areas in Guizhou Province are classified as priority protected areas regarding safe planting zoning, with safe utilization areas accounting for about 10%. However, areas in the eastern part of Qiandongnan, the southeastern part of Tongren, and the western part of Bijie require strict control. Our study attach great importance to the prevention of high Pb accumulation in rice as well as in soils in major rice growing areas.