Delineation of terrain features in Demini’s Watershed – Setentrional Amazonia using regional geomorphometry

Abstract

The mapping of terrain features relies on the necessity of an object-based approach, which can be related to the constitution of terrain units. The regionalization of land surface parameters enables the assessment of characteristics inside terrain unities and enhances heterogeneity outside these patches. This study presents a classification methodology for hierarchical geomorphometric delineation, using visual and Random Forest (RF) classification in a regionalized dataset of terrain variables, applied in Demini’s watershed, north of Amazonia. These variables, calculated by segments derived from multiresolution segmentation, were evaluated in order to identify which had major contributions to RF’s classifications. The characterization of features had great correspondence with Environmental Informations Database (BDIA) data of geomorphological mapping, used as a reference in this work. The Overall Accuracy for RF taxon 1 was 96%, while Taxon 2 Highland and Lowland RF models reached 84% and 87%, respectively. Identification of subdomain classes was possible mostly using the digital elevation model (DEM) and variables directly derived from the DEM. The delineation of floodplain presented significant differences between visual and RF results, including BDIA’s data.

Keywords

• Modelo Digital de Elevação
• Segmentação
• Regionalização
• Classificação do relevo
• Aprendizado de máquina