Objective The aim was to explore the time series characteristics of normalized vegetation index could effectively improve the accuracy of digital soil mapping, in order to provide new ideas for digital soil mapping research.

Method Taking Jiyuan Nanshan Forest Farm as the study area, terrain factors, climate factors, vegetation factors, and time series characteristics of normalized difference vegetation index (NDVI) were simultaneously extracted as environmental covariates. Ordinary kriging, regression kriging, random forest, decision tree, support vector machine, and other methods were used to construct soil organic carbon (SOC) prediction models to test whether the long-term time series data of NDVI could effectively improve the mapping accuracy of SOC.

Results Among different models, elevation, and terrain wetness index played dominant roles in SOC mapping. The time series characteristics of NDVI were consistently more important than single-phase vegetation factors in all models. The introduction of time series features of NDVI resulted in a decrease in mean error (ME) and root mean square error (RMSE) and an increase in the coefficient of determination (R²) and concordance correlation coefficient (CCC) for models such as RK, RF, SVM, and CART. Among different models, the random forest model incorporating time series has the most significant improvement in mapping accuracy and model reliability, making it the optimal model for this study. (R² increased by 69.57%, ME decreased by 88.04%).

Conclusion The introduction of NDVI time series features as variables can effectively reduce the prediction errors of models to varying degrees, enhancing the reliability of the models. This result provides new environmental variables for SOC mapping and opens up new possibilities for digital soil mapping.