Objectives To develop a timely, efficient, and accurate method for dynamically monitoring soil organic carbon (SOC) and soil total nitrogen (STN) in Area 1 of Plot 9 of the Millennium Forest in Xiong’an New Area, to characterize their spatiotemporal distribution patterns, support sustainable soil management, and ensure the long-term stable development of the Millennium Forest.

Methods A total of 95 soil samples were collected in Area 1 of Plot 9 within the Millennium Forest, Xiong’an New Area, and their SOC and STN contents were determined. Six Landsat 8 OLI images acquired during September or October from 2018 to 2023 were collected for the study area. Spectral reflectance values and remote sensing indices were extracted and calculated, and their relationships with measured SOC and STN contents were analyzed. Feature selection methods were employed to identify optimal predictors and reduce model complexity. Based on the Random Forest (RF) algorithm, three types of prediction models were developed for SOC and STN, including band-based models, band + single remote sensing index models, and band + multiple remote sensing index models. For SOC prediction, Band 4 and Band 6 were selected to construct the band-based model. The Redness Index (RI) was incorporated into the band + single-index model, whereas RI, Soil Salinity Index 2 (SI2), and Soil Salinity Index 3 (SI3) were included in the band + multiple-index model. For STN prediction, Band 4 and Band 3 were selected for the band-based model. SI3 was used in the band + single-index model, while RI, Soil Salinity Index 1 (SI1), and SI3 were incorporated into the band + multiple-index model. Model performance was comprehensively evaluated using the coefficient of determination (R²), relative percent deviation (RPD), mean absolute error (MAE), and root mean square error (RMSE). The optimal models were subsequently applied to predict the spatiotemporal distributions of SOC and STN.

Results The SOC-Band 4+Band 6+RI+SI2+SI3 model achieved the highest prediction accuracy for SOC, with R²=0.778, RPD=1.933, MAE=0.719, RMSE=0.829. For STN, the STN-Band 4+Band 3+SI3 model performed best, with R²=0.679, RPD=1.742, MAE=0.047, RMSE=0.054. From 2018 to 2023, the SOC content in the study area showed an overall fluctuating upward trend, with a slight decrease in 2022. The annual average SOC content ranged from 7.274 to 8.334 g/kg. Spatially, SOC content remained relatively stable in peripheral areas but showed considerable variation within the forest interior. The ecological basic forest and the western part of the near-natural forest consistently exhibited higher SOC contents than the other forest types. The overall changing trend of STN content was basically consistent with that of SOC content, and its annual average changing range is 0.676−0.772 g/kg.

Conclusions Incorporating remote sensing indices into RF band-based models improved the prediction accuracy of both SOC and STN. The predicted results indicated that SOC and STN contents generally increased over time with interannual fluctuations, although both decreased in 2022. Among the five landscape forest types, the ecological basic forest and the western part of the near-natural forest consistently maintained higher SOC and STN contents than the other forest types.