Indicative mapping of peat and other hydromorphic organic soils across Austria with legacy data

Understanding the spatial distribution of peat soils and other hydromorphic organic soils is essential for terrestrial carbon storage, biodiversity, and sustainable land management. In addition to these overarching purposes, there is now an urgent need to better localize these soils in the wake of the recently adopted European Nature Restoration Law (NRL). The NRL requires EU member states to formulate plans for the restoration of organic soils under agriculture (and other land-uses such as forest) constituting drained peatlands from 2026 onwards. A comprehensive, high-resolution map of peat and other hydromorphic organic soils covering the entire Austrian territory (i.e. land-uses other than agriculture) to support this objective is currently lacking.

To model the distribution of these soils across Austria, we propose a strategy based on regression and machine learning (ML), driven by legacy occurrence data and spatial covariates. Owing to the nature of the legacy data, we define six operational classes of peat and other hydromorphic organic soils, derived from the Austrian Soil Taxonomy in conjunction with the IPCC 2013 Wetlands Supplement. Response variables (i.e., expected soil classes) are constructed by re-classifying 14 legacy datasets containing land-use specific information on soils and/or vegetation. These include the Agricultural Soil Map of Austria, the Austrian Soil Taxation Survey, the Austrian Soil Information System BORIS, various forest site and vegetation mapping projects and the currently updated Austrian moorland protection database. In the process, real point data is supplemented by synthetic points generated by sampling from input polygon maps.

Separate occurrence probability models are built for each response class using (ensembles of) Environmental Niche Models (ENMs, also known as Species Distribution Models). These models predict the likelihood of occurrence for each peat soil type based on spatially explicit, high-resolution predictor variables (100m and higher). Covariates include soil properties, climate, relief features from Digital Elevation Models (DEMs), vegetation indices from remote sensing, and parent material, following the well-established SCROPAN approach of Digital Soil Mapping. The final indicative map is created by combining predictions from individual models. We employ cross-validation to tune ML hyperparameters and assess predictive model performance. Additionally, external evaluation will be carried out through a field campaign in spring 2025. A total of 90 transects will be established to collect data on soils and vegetation which will serve map validation and iterative model refinement.

The resulting indicative map will depict peat and hydromorphic organic soil distribution across Austria at high spatial resolution. Amongst others, it will serve as valuable guidance for field campaigns that likely will precede any targeted restoration measures under the NRL.