Digital mapping of soil organic carbon in a plain area based on time-series features

• Using time-series features to improve SOC mapping accuracy in plains. • Comparison the timeliness of variable screening methods to reduce data redundancy. • Evaluating the performance of machine learning models for SOC mapping. Improving the accuracy of digital soil organic carbon (SOC) mapping in plain areas is important for meeting the needs of agricultural development and environmental protection. Utilizing time-series environmental factors is thought to be helpful in digital soil mapping (DSM) of SOC, which is a current research hotspot. This study focused on the DSM of SOC in Fengqiu County, China, using terrain, climate, single-time ecological factors, and time-series features of time-series ecological factors as environmental covariates to investigate whether time-series environmental covariates could improve the accuracy in a plain area. SOC prediction models were established using random forests (RF), backpropagation neural networks (BP), and support vector machines (SVM). The results showed that ecological factors such as normalized difference vegetation index (NDVI) normalized difference built-up index (NDBSI), drought, and humidity indices, along with distance from rivers, played a dominant role in digital SOC mapping. The relative importance of the time-series features of the ecological factors was higher than that of the single-time-point vegetation indices. Introducing the time-series features of ecological factors resulted in a decrease in the mean error (ME) and root mean square error (RMSE), whereas the coefficient of determination (R 2 ) and concordance correlation coefficient (CCC) showed increasing trends across the different models. Comparing the various environmental variable screening methods, the Boruta algorithm achieved the most significant improvement in model accuracy. The RFSTB (RF + Conventional variables + Time-series variables + Boruta algorithm) model was identified as the optimal model, with R 2 increasing by 65.45 % and RMSE decreasing by 47.12 %. This study introduces new environmental covariates for SOC mapping and provides new insights into digital mapping of SOC in plain areas.