Mapping the terrestrial ecoregions of the Purus-Madeira interfluve in the Amazon Forest using machine learning techniques

Abstract

An ecoregion is a region with similar environmental conditions. However, many ecoregions represent regional habitat heterogeneity, and areas with little fieldwork information can benefit from ecoregion mapping by providing information about their biodiversity distribution. This work presents the procedure adopted to map the terrestrial ecoregions of the Purus-Madeira interfluve, in the Brazilian Amazon using Machine Learning techniques. A methodological approach with Self-Organizing Map and K-means algorithms is proposed for the ecoregion mapping and the resulting model is discussed. The final ecoregion map was built up from a set of variables including altitude, slope, drainage density, percentage of tree cover and a vegetation map. Discriminant analysis identified the extent to which the variables are similar or different between the ecoregions, with a Kappa index of 0.86. This indicates that the methodological approach is reliable and thus can reproduce valid results over different areas. We produced a map with 14 ecoregions to account for the environmental diversity in the Purus-Madeira interfluve. This map can be used for the planning of biodiversity conservation.