Objective Developing efficient and accurate methods for mapping the spatial distribution of Albic bleached soil in cultivated layer is crucial for managing soil conditions in the Sanjiang Plain.

Method This study focused on cropland in a small watershed at Shuguang Farm, Huanan County, Heilongjiang Province. The dual temporal Sentinel-2 remote sensing imagery and terrain data were used, and the Random Forest algorithm was combined with, to construct a model for extracting albic bleached soil at the plot scale. By comparing the mapping accuracy of different temporal images and terrain parameters, the impact of soil moisture content on mapping precision was analyzed and the algorithm was applied across the entire Shuguang Farm.

Result The findings were as follows: ① During the bare soil period, the spectral reflectance of albic bleached soil was significantly higher than that of non-albic bleached soil. ② The sensitive spectral bands for distinguishing albic bleached soil from non-albic bleached soil generally range from the blue band to the narrow near-infrared band (458 ~ 875 nm), with the strongest separability in the blue and green (518 ~ 578 nm) bands. ③ the optimal time window was from late April to early May for mapping Albic bleached soil using remote sensing imagery. The lower soil moisture improved mapping accuracy. ④ Dual temporal remote sensing imagery had higher mapping accuracy than single temporal imagery, with the highest accuracy achieved when incorporating the valley depth parameter, resulting in an overall accuracy of 96.55% and a Kappa coefficient of 0.93.

Conclusion The Random Forest model, based on dual temporal remote sensing imagery and valley depth, could accurately extract albic bleached soil in the cultivated layer. This method provided a basis for soil survey and offered scientific guidance for the precise improvement of albic bleached soil.