The second DEM was produced interpolating contour lines digitized from the same topographic maps used to produce the soil map of \citet{MiguelEtAl2012}, the geologic map of \citet{MacielFilho1990}, and the land use map of (Figure \ref{fig:contour-dem}). Interpolation of the raster surface with 5-m pixel size was performed using the same hydrologically correct interpolation method employed by \citet{JarvisEtAl2008}. The algorithm is implemented in ArcGIS software by ESRI and includes an interpolation method based on the ANUDEM program developed by \cite{Hutchinson1989}. Vector files of contour lines at 10 meters equidistance, drainage network, lakes, and peaks, were used to generate the DEM and define the boundary of the study area.
The STRM DEM was registered and geocoded with the prediction grid using cubic resampling. This method was used to guarantee the creation of a smooth raster surface, and because it is more efficient in minimizing the double-oblique stripping \citep{Samuel-RosaEtAl2013c}.
The source of less detailed satellite images is Landsat-5 Thematic Mapper. The image used was acquired on 26 December of 2010 and is available through the Division of Image Generation of the National Institute for Space Research (INPE) of Brazil. It has eight bits radiometric resolution and \(\approx\) 30 meters spatial resolution. All spectral bands (except the thermal band) were submitted to orthorectification (Geomatica OrthoEngine), radiometric correction (), atmospheric correction ( with the 6S atmospheric model \citep{VermoteEtAl1997}), and topographic correction (). The source of high resolution orbital images is the RapidEye constellation, available through the Brazilian Ministry of the Environment \citep{Brasil2012}. The image (tile number 2225403) was acquired on 16 November of 2012, has 16 bits radiometric resolution, 6.5 meters spatial resolution, and was orthorrectified to 5 meters spatial resolution \citep{RapidEye2013}. All bands were submitted to atmospheric correction (6S atmospheric model \citep{VermoteEtAl1997} adapted by \cite{AntunesEtAl2013}) and topographic correction (). Environmental covariates derived from these satellite images are the spectral bands (except the thermal band of Landsat-5 Thematic Mapper) and vegetation indexes calculated using the red, infrared and red-edge (RapidEye) spectral bands. Vegetation indexes are the normalized difference vegetation index () and the soil-adjusted vegetation index () \cite{Jensen2009}. Eight covariates were derived from the Landsat 5 TM image, and nine from the RapidEye image.