【Objective】Digital soil mapping of soil types commonly adopts a multi-type one-time modeling strategy, but this strategy struggles to fully consider the complex relationship between each soil type and its soil-forming environment. To address this gap, this study proposes a type-by-type random forest modeling strategy. 【Method】Taking Yizheng City, Jiangsu Province, as the study area, 21 environmental covariates and 860 typical virtual sample points extracted from soil species polygons in historical soil maps were used to construct random forest classification models for each soil species individually. The spatial distribution of predicted probabilities for each soil species was output, and the soil species with the highest probability was taken as the predicted result for each pixel. This method was compared with the conventional multi-type one-time random forest modeling strategy.【Result】Results showed that the overall accuracy and Kappa coefficient of the type-by-type multiple modeling strategy were 0.62 and 0.58, respectively, while those of the conventional modeling strategy were 0.61 and 0.57. The difference in overall accuracy between the two strategies was very small in the study area. However, the uncertainty of the former decreased significantly. The spatial mean of the maximum probability under the type-by-type multiple modeling strategy was 0.54, which was 0.08 higher than the spatial mean of the maximum probability under the conventional modeling strategy. Moreover, the type-by-type multiple modeling strategy could output, for each pixel, the probability of the second-most-probable soil species and the confusion index, which helps identify confusion-prone areas such as the northern uplands, where the probability of the second-most-probable soil species is relatively high. The type-by-type multiple modeling strategy had higher prediction accuracy for Horse-liver Soil, Yellow Rigid Soil, Yellow Variegated Soil, Oily Silty Soil,etc , than the conventional modeling strategy, but it did not achieve consistent improvement for all soil species. In addition, the type-by-type multiple modeling strategy could output the importance of environmental variables for individual soil species, facilitating analysis of the relationships between soil species and environmental factors. The results showed that elevation, topographic wetness index, and distance to the Yangtze River could indicate the spatial variation of most soil species.【Conclusion】Therefore, compared with the conventional one-time modeling strategy, the type-by-type multiple random forest model can significantly reduce the uncertainty of soil type prediction, improve the reliability of prediction results, and serve as a useful tool for understanding the relationship between each soil species and its soil-forming environment.