Mapping Cerrado physiognomies using Landsat time series based phenological profiles

Monitoring of the Brazilian Cerrado biome is crucial in order to fully understand its ecosystem functions, its response to changes, and to keep track of ongoing change processes in one of the world's biodiversity hotspots. The huge extent, heterogeneity and complex diversity of the Cerrado makes monitoring very challenging, but optical remote sensing based approaches have been shown to be valuable to tackle this task. We explored the potential of Landsat data to derive seasonal phenological profiles for the main Cerrado physiognomies, in order to use this information for classification purposes. Therefore we gathered Landsat TM, ETM+ and OLI data from the open Landsat data archive. To overcome data gaps in the time series, we applied a Gaussian kernel based convolution filter on Tasseled Cap transformed Landsat data. Thus, it was possible to derive comprehensible phenological profiles for distinct Cerrado physiognomies at Landsat spatial resolution. We used these derived seasonal profiles to train a Support Vector Machine classification model in order to map the main Cerrado physiognomies around Brasilia, DF. Although classification errors were observed between similar classes in terms of their vegetation structure and density, we were able produce an accurate map that captures the spatial patterns of the vegetation physiognomies.