Análises texturais de produtos derivados de LiDAR para discriminação de cangas lateríticas, Serra Sul de Carajás (PA)

Abstract

The characterization of geological features through products derived from remote sensing in humid tropical regions suffers severe restrictions due to the influence of vegetation on the radiometric measurements. This is the motivation to investigate new methodological approaches aimed at extracting remote sensing information data applied to geological mapping. This research aimed to develop a methodological approach to discriminate types of lateritic surfaces from textural parameters of the Haralick extracted shaded relief image generated Digital Model Land of high spatial resolution (1 m) derived from the last return signal (ground) of LiDAR data. The study area is located in body S11CD in Serra Sul of Carajás Mineral Province, including the city of Canãa dos Carajás (Pará State - Brazil). The body S11CD is characterized by a residual relief supported by ferruginous crust thick developed on Archean banded iron formations (Carajás Formation / Grão Pará Group) and is covered by savannah (campus rupestres) that contrast with the surroundings tropical rain forest. The lateritic crust in the study area is classified into (1) structural duricrust and (2) detrital duricrust. The difference between the types of lateritic crust was analyzed by microtopography relief metrics (Hrms), measured in the field and textural parameters of the Haralick extracted from the shaded relief image. Statistical tests of averages Hrms comparison (Student t test) showed that it is possible to identify the terrain microrugosidade kinds of lateritic duricrust present in S11CD. The object-oriented classification (GEOBIA) was used dissimilarity textural parameter of the Haralick to discriminate the types of lateritic duricrust. For this, it was used the standard deviation of the average dissimilarity threshold for separating the detrital duricrust (26.1 ↔ 33.234) and structure duricrust (20.573 ↔ 28.515). The result of this classification overlaid around 89.35% the study area, remaining ~ 11% of the image not classified, possibly as a result of noise in the LiDAR data. The validation of this classification attested that the overall accuracy of field data and the classification was 78.8%. Note also, that the structural duricrust occurs in the area as small "windows" through the detrital duricrust, which made it difficult to carry out field sampling at this site, so the errors of commission and of omission become high.