Ways forward for Machine Learning to make useful global environmental datasets from legacy observations and measurements.

Abstract

Advances in geospatial and Machine Learning techniques for large datasets of georeferenced observations have made it possible to produce model-based global maps of ecological and environmental variables. However, the implementation of existing scientific methods (especially Machine Learning models) to produce accurate global maps is often complex. <i>Tomislav Hengl</i> (co-founder of OpenGeoHub foundation), <i>Johan van den Hoogen</i> (researcher at ETH Zürich), and <i>Devin Routh</i> (Science IT Consultant at the University of Zürich) shared with <i>Nature Communications</i> their perspectives for creators and users of these maps, focusing on the key challenges in producing global environmental geospatial datasets to achieve significant impacts.