Contributers: AAU, AU and I•GIS
By using the latest advances in Deep Neural Networks together with large amounts of multi-source data, we expect to achieve a much higher prediction accuracy than classical machine learning techniques. The project will not only produce predictive maps for the target variables, but also uncertainty maps and model interpretation, which constitute significant gaps within current Digital Soil Mapping studies. Since the predictive maps and their associated uncertainties will be used by decision-makers, it is crucial to get some insight into the developed models drawing on recent work in explainability and how to understand the models produced.
Work package lead: Mark Philip Philipsen, AAU.