Improving the Landsat-8 determinations of soil organic carbon and total nitrogen stocks through reducing the effects of buildings, water, and vegetation

Abstract

Soil mapping using Landsat-8 provides potential for a real-time synoptic estimation of soil organic carbon (SOC) and soil total nitrogen (STN) stocks. However, the few relevant studies at global and European scales have shown relatively poor model accuracy because of the effects from buildings, water and vegetation. Here, we used independent variable fuzzy learning based on several building, water and vegetation indices to reduce the effects of these land covers on the SOC and STN models. We applied the models to predict the SOC and STN stocks at the pan-European scale. Measured SOC and STN stocks obtained from the Land Use/Cover Area frame statistical Survey 2018 dataset and spectral data from the corresponding Landsat-8 images were used in this study. The results indicated that the SOC and STN models after learning performed better than the models before learning (the SOC model: R² = 0.56, mean absolute error (MAE) = 10.38 t C ha⁻¹, root mean squared error (RMSE) = 13.72 t C ha⁻¹, ratio of performance to interquartile range (RPIQ) = 2.02; the STN model: R² = 0.52, MAE = 0.92 t N ha⁻¹, RMSE = 1.20 t N ha⁻¹, RPIQ = 1.45). The two models predicted the 0–20 cm SOC and STN stocks across Europe as 28.13 and 2.25 Gt, respectively. The uneven distributions were clearly reduced when compared with the maps before learning, especially our STN map and STN relative standard deviation (RSD) map. The results indicated that our study provided a valuable reference for reducing the effects of buildings, water, and vegetation during satellite SOC and STN mapping. Considering the importance of SOC and STN for future global policies, we will also pay attention to the effects of other soil properties, weathering and land covers in the future.