Proposta metodológica de mapeamento geomorfológico semiautomático na bacia hidro-gráfica do Rio Iguaçu-Sarapuí, RJ

Abstract

Due to the territorial extension and the diversity of the geomorphological landscape in Brazil, there are several approaches for geomorphological mapping. In this sense, geomorpho-logical maps present several problems such as: a) the absence of interoperability between relief classification systems; b) subjectivity in geomorphological maps and c) thematic assessment. In this context, we opted for the semiautomatic approach, in order to reduce these problems, thus, the objective of the present study is to propose the use of semiautomatic methods for the elaboration of geomorphological mappings. Thus, to verify the discrepancies between the proposed semiautomatic geomorphological map and the other geomorphological maps referring to the study area and the scale adopted in the present study was 1 / 100,000. The methodological proposal of the study is based on the assessment of altimetric precision; pre-processing of the DEMs; extraction of geomorphometric variables and the application of the OBIA, KDD method, using two algorithms: a) J48 (C4.5) and b) Random Forest and evalua-tion of thematic precision, applying the index: a) kappa and b) global accuracy. The MDE NASADEM presented the best results of the evaluation of altimetric precision and in the pre-processing of the MDEs, the Fill Sinks tool (Planchon and Darboux, 2001) and the median filter, window (7x7) were the most satisfactory. The geomorphological maps of the first taxon presented the most satisfactory result, from the J48 algorithm with the kappa indexes and the global accuracy as the values: a) 0.86 and b) 0.91, in the second taxon, the best performance algorithm was the Random Forest, having as the kappa index and the global accuracy respec-tively: a) 0.88 and b) 0.92 and in the fourth taxon, the Random Forest algorithm prevailing in the values regarding the kappa index and the global accuracy: a) 0.85 and b) 0.87. The Ran-dom Forest algorithm presented the most satisfactory result, however not so different from the J48 algorithm. The semi-automatic geomorphological map proposed in the present study was satisfactory, enabling the differentiation between technogenic landforms: a) technogenic plains and b) technogenic deposit. The main difficulty was in the differentiation between the classes of hills, showing that these classes vary in terms of altimetric criteria and topographic roughness. The semiautomatic method allows the integration of several techniques that pro-vide gains in the production of geomorphological maps, as it offers the possibility to under-stand criteria in the classification of landforms. Thus, it is necessary to commit efforts to the imprecision of concepts, so that it can be facilitated in the production of geomorphological maps.