[Use of Machine Learning Methods to Identify Soil Parent Materials in a High-cadmium Geological Background Area]

Recently, the characteristics of high Cd content and low Cd mobility in karstic soil of a high geological background area in south China have received extensive attention. Parent material type is crucial for understanding soil Cd geochemical behavior and identifying soil ecological risk. However, the southern tropical climate leads to fewer rock outcrops, and it is difficult to obtain accurate parent material information. The aim of this study was to identify the main soil parameters that control the spatial distribution of lithology and affect soil Cd activity and ultimately uses these characteristics and machine learning methods to predict different soil parent materials in the high geological background area. In total, 5 096, 5 602, and 1 653 surface soil samples were collected from the carbonate rock, clasolite, and quaternary sediment regions, respectively. Hot spot analysis and the sequential extraction test showed that the spatial distribution patterns of soil properties and Cd were controlled by the underlying bedrock, and the ecological risk of soil Cd in the non-karst region was significantly higher than that in the karst region. Correlation analysis and importance analysis indicated that the content and mobility of Cd in the high geological background were mainly controlled by Fe/Mn oxides, total organic carbon （TOC）, CaO, and pH. Based on the big data of surface soil samples, the soil parent materials were then predicted using artificial neural network （ANN）, random forest （RF）, and support vector machine （SVM） models. The RF model had higher Kappa coefficients and overall accuracies than those of the ANN and SVM models, suggesting that RF has the potential to predict soil parent materials from big data, which provides a new idea and method for mapping lithology distribution and identifying soil Cd ecological risk in high background areas.