Digital mapping of organic carbon content and stocks in soils of Cis-Salair drained plain using the Google Earth Engine online platform and the random forest algorithm

For a key site on the Cis-Salair drained plain, digital mapping of the content of soil organic carbon (SOC) in the topsoil (0–30 cm) was conducted using the random forest algorithm implemented on the Google Earth Engine cloud platform. The following were used as predictors in the random forest model: 1) 19 bioclimatic variables from WorldClim; 2) 5 climatic variables calculated based on WorldClim data and soil-climate atlas data; 3) 8 vegetation indices calculated based on Landsat 8 OLI images; 4) 26 morphometric characteristics of the terrain calculated based on the ALOS DEM; 5) 2 variables describing the spatial location. The correlation coefficients (R) between the content of SOC and the values of the predictors were taken into account when forming sets of predictors: 1) BIO11+RVI; 2) Longitude+CNBL; 3) SAT10+CC+Texture; 4) 60 predictors; 5) 42 (without relief curvatures, vegetation indices and predictors with zero values); 6) 37 (all with R > ±0.5); 7) 32 (all with R > ±0.3 without vegetation indices); 8) 27 (all with R > ±0.5 without vegetation indices); 9) 23 (without BIO1–19, relief curvatures, vegetation indices and predictors with zero values). The result of modeling the content of SOC based on 32 predictors and a training dataset (n=42) with a lower RMSE (0.72) was chosen as the best. Based on this model, a soil bulk density map was compiled using a pedotransfer function. This data, together with a map of the SOC content, was used to create a map of SOC stock. The SOC content in the arable layer (0–30 cm) varied from 1.3 to 6.1%, according to the actual data. The SOC stocks ranged from 84 t/ha to 203 t/ha. The highest levels of SOC content and stocks were found in the soils in the upper part of the slope. A gradual decrease in these values was noted as one moved downhill. The soil bulk density ranged from 1.20 g/cm³ to 1.36 g/cm³ and increased as one moved downhill, indicating a reverse trend compared to the SOC content and stocks. The total SOC stock in the arable layer (0–30 cm) of the soils of the studied territory with an area of 225 hectares amounted to 28.7 kt.