Mining soil data of Switzerland: New maps for soil texture, soil organic carbon, nitrogen, and phosphorus

Abstract

Maps of soil property are essential for soil monitoring, management, and conservation. These are available from regional to global scales, with global maps being urgently needed for global modeling and management endeavors (from soil degradation to climate change modeling and assessments). Currently, global maps are provided by SoilGrids (SG) and OpenLandMap (OLM). However, the number of samples used and spatial resolution for these maps indicate potentially a high uncertainty and leave room for discussion. A validation of global maps could be regional and national assessments with higher resolution data availability. Therefore, the objective of this study was to produce a new Swiss Soil Property Map (SSPM) for soil organic carbon, soil texture, total nitrogen, and phosphorus for Switzerland using higher resolution data compared to the previous global studies. To establish swiss scale maps, we fitted the Quantile Random Forest (QRF) machine-learning model for spatial predictions linking environmental covariates such as topography, climate, and vegetation with soil property data. Each model was evaluated using five-fold spatial cross-validation. The results showed concordance correlation coefficients (CCC) between 0.24 and 0.68 for predicted soil property. The new Swiss Soil Property Map (SSPM) as well as the global soil property maps were validated using the LUCAS dataset. The CCC for OC and clay maps of SSPM are 52%, and 24% higher, respectively, than the SG map. Therefore, this study demonstrates the potential impact of using sufficient data points to represent the individual country. Simultaneously we show that not just any additional data set will improve model performance. A well-planned high-resolution national soil monitoring is needed to improve the global maps in the future.