WORLDSOILS-Monitoring Topsoil Organic Carbon at Continental Scale Using Earth Observation Data

Kurzfassung

The adoption of multidimensional and integrated approaches is crucial to support relevant economic operators with sustainable soil related policies and services. The combination of innovative technologies based on Earth Observation (EO), Artificial Intelligence (AI), and cloud computing has been identified as a game changer for operational and cost-effective soil organic carbon (SOC) monitoring, reporting and verification approaches. In close collaboration with users and stakeholders, the European Space Agency funded WORLDSOILS project aims to develop an EO-driven soil monitoring system on a suitable cloud environment, utilizing open-source EO data, additional variables, and indices derived from EO sources, along with reference data from the European LUCAS soil data archive. The system focuses on monitoring top SOC using AI techniques and is designed in a modular manner to allow future extension to additional soil indices. With a grid resolution of 50m over Europe, it enables the assessment of temporal changes in the topsoil layer at least once a year, even at the intra-field level. At the outset, the system prepares the EO database for SOC retrievals by utilizing per-pixel reflectance composites from satellite imagery, using a novel histogram separation thresholding approach. This allows the collection of exposed soil spectra expanding the analyzable area and eliminating the effect of ambient factors, such as soil moisture, for subsequent modeling steps. The composites are then segmented into exposed soil (croplands) and permanently vegetated pixels. For exposed soils, SOC estimation relies on multi-input 1-D convolutional neural networks using Sentinel-2 spectral bands, while digital soil mapping techniques based on quantile random forest are effectively employed to generate a SOC product for vegetated areas, utilizing environmental covariates from various sources and spectral composites. Three pilot regions in Belgium, Czech Republic and Greece have been identified for validation purposes, representing different bioclimatic European territories, vegetation types, land uses, and soil compositions.