High-resolution, reliable soil data is crucial for addressing climate change and sustainable land management. Integrating remote sensing data, such as from Copernicus Sentinel, is essential for improving accuracy and relevance.

This study presents an overview of our Digital Soil Mapping (DSM) approach and its innovations. We combine satellite imagery, environmental covariates (e.g., elevation, weather data), and ground truth observations (e.g., LUCAS and other European and national datasets) to create high-resolution soil property maps using statistical models. These maps encompass primary properties (e.g., organic carbon, pH, texture), derived properties, and soil health indicators.

We used the Soil Composite Mapping Processor (SCMaP) to derive soil reflectance composites from Sentinel-2 time series. These composites aid in identifying bare soil areas and estimating their frequency, serving as a proxy for land management. They represent spectral reflectance and dynamics. Random Forest models, iin particular Quantile Random Forests for uncertainty assessment, are employed to predict soil properties.

This study delves into the advantages and challenges of using high-resolution remote sensing data with limited ground truth data. We also provide insights into product uncertainty assessment at a continental scale, including accuracy, spatial patterns, and user evaluation. We focus in particular on the relevance of finer resolution and accuracy for continental products.