Releasing acidity and metals into watercourses, acid sulfate soils represent a critical environmental problem worldwide. Identifying the spatial distribution of these soils enables to target the strategic areas for risk management. In Denmark, the occurrence of acid sulfate soils was first studied during the 1980’s through conventional mapping (i.e. soil sampling and the subsequent determination of pH at the time of sampling and after incubation, the pyrite content and the acid-neutralizing capacity). Since acid sulfate soils mostly occur in wetlands, the survey specifically targeted these areas. Recently, a digital soil mapping approach was assessed to create a predictive map for potential acid sulfate soil occurrence in the wetlands of Jutland (c. 6500 km2; Beucher et al., 2016). An Artificial Neural Networks method was applied using 8000 soil observations and 16 environmental variables, including geology, landscape type and terrain parameters. Visible-Near-Infrared (Vis-NIR) spectroscopy constitutes a rapid and cheap alternative to soil analysis, and was successfully utilized for the prediction of soil chemical, physical and biological properties. In particular, the Vis-NIR spectra contain diagnostic features for hydroxides, clay minerals, iron oxides and iron sulfates which are typically present in acid sulfate soils (Shi et al., 2014). Soil spectroscopy may thus efficiently supplement the mapping of acid sulfate soil occurrence. The present study aims at predicting acid sulfate soil occurrence in the Skjern River catchment (c. 2500 km2). Different machine learning approaches will be assessed using soil and environmental data, together with laboratory Vis-NIR spectral data available for the study area. Absorbance values (400–2500 nm) were measured for 600 soil samples with a DS2500 instrument (Peng et al., 2015). The spectral data were summarized using principal component analysis (PCA). The first two principal components (PC) explained 99% of the variability in the spectra. Kriging was applied to upgrade PC scores information from point to image scale for further use within the acid sulfate soil occurrence modelling.
