COMPARISON BETWEEN SPECTRAL ANGLE MAPPER AND EUCLIDEAN DISTANCE IN LANDFORM MAPPING

The supervised classification from geomorphometric signatures (AG) is a proceeding that can help landform mapping using similarity or distance measures. This study aims to compare the supervised classifi cation of similarity and distance methods to the landform mapping. The comparison was made in Campo de Instrucao de Formosa (GO), divided into the following steps: acquiring HydroSHEDS data, generation of bending images, selection of geomorphometric signatures; landforms classifi cation using the spectral angle mapper (SAM) and Euclidean Distance (DE) methods; comparing the classifi cation using the cross-tabulation matrix, elevation model analysis in 3D, evaluation of the mean and standard deviation of the curvatures for each mapped class, and field observation. Selecting geomorphometric signatures considered the following steps: (a) reduction of geomorphometric attributes the transformation Minimum Noise Fraction (MNF); (B) spatial reduction by the Pixel Purity Index (PPI); and (c) the manual selection by the n-Dimensional Viewer. The classifi cation process took 14 AG describing two behaviors: Type 1 - when the longitudinal curvature has a value greater than the transverse curvature; and Type 2 - when the opposite occurs. The process has been simplified to six landforms classes: Convex/Convex (Cx/Cx); Concave/Convex (Cc/Cx); Concave/Concave (Cc/cc); Concave/Rectilinear (Cc/Rt); Convex/Rectilinear (Cx/Rt); Rectilinear/Rectilinear (Rt/Rt). In SAM mapping, the predominant landforms are Cc/Rt, Cx/Rt and Cc/Cx, indicating heterogeneity in many transition and concave areas. The classifi cation from DE showed prevalence of rectilinear features (Rt/Rt). However, this FT showed the smallest standard deviations and average values close to zero for all curvatures, indicating that was the most efficient method for mapping the rectilinear areas.