Spatial prediction of soil organic matter content using multiyear synthetic images and partitioning algorithms

Accurate assessment of the spatial distribution of soil organic matter (SOM) is of great significance for regional sustainable development. However, due to the strong spatial variability in soil, there is still a lack of robust SOM mapping methods. The use of remote sensing satellite technology to map soil parameters has been effectively applied in many areas. Soil mapping in Northeast China with less cloud cover, higher planting intensity and longer bare soil period of cultivated land is expected. In this study, multiyear Landsat 8 and Sentinel-2 images of bare soil periods were used to generate median synthetic images according to different time intervals in Google Earth Engine (GEE). Then, the spectral index and image band were used as input variables, and the prediction accuracies of different combinations were evaluated by the random forest (RF) algorithm. Finally, the best combination of two partitioning methods (based on different soil types and cascade simple K-means clustering) was used for SOM prediction of local regression and mapping. The results show that 1) the best time window for SOM prediction in the Songnen Plain is in May, but precipitation will affect the prediction accuracy of SOM; 2) Sentinel-2 synthetic images are not superior to Landsat 8 synthetic images for SOM mapping, although Sentinel-2 has better temporal and spatial resolution; and 3) compared with the global regression model, the local regression method based on two partitioning methods can improve the accuracy of SOM mapping, but the actual mapping effect based on the soil type partitioning algorithms is affected by the distribution of soil samples. This study extends the application of GEE to soil mapping and improve the accuracy of large-scale regional SOM mapping by partitioning images.

Automated summary

Accurate assessment of the spatial distribution of soil organic matter is crucial for regional sustainable development, but there is a lack of robust soil organic matter mapping methods. This study utilizes remote sensing satellite technology and image partitioning to improve the accuracy of large-scale regional soil organic matter mapping.