Abstract: To reveal the spatial distribution patterns of soil organic matter (SOM) and soil total nitrogen (STN) contents in tobacco-planting areas, Duping Town of Wushan County in Chongqing City was taken as the study area, the soil parent materials and topographic factors were used as predictors, and three machine learning algorithms, namely Random Forest (RF), Gradient Boosted Decision Tree (GBDT) and Extreme Gradient Boosting (XGBoost) were used for model construction and evaluation, and then the optimal model for digital soil mapping was selected and the importance of environmental variables was analyzed. The results showed that: 1) Soils developed from limestone of Permian Liangshan Formation had higher SOM and STN contents than those developed from limestone of Triassic Daye Formation. 2) The GBDT model had the best prediction accuracy with coefficients of determination (R²) of 0.616 7 and 0.746 8, mean absolute error(MAE)of 4.81 g/kg and 0.25 g/kg and root mean square error (RMSE) of 5.94 g/kg and 0.34 g/kg for SOM and STN contents, respectively. 3) The effects of the main environmental factors on SOM contents were in the order of parent material > altitude > topographic wetness index > valley depth, and those on STN contents was in the order of parent material > slope height > altitude.