Identificação do leito regular de cursos d’água com técnica geobia aplicada a imagens de ultra alta resolução espacial

Abstract

The processes of occupation of the territory by man, in a disorderly way, cause changes in land use that impact the environment and the life of the population. The consequence of these appropriations interferes not only with the landscape, but also with the hydrological cycle of the rivers, due to the permeability of the soil and deforestation around the water courses, directly affecting the availability and quality of this resource. In this sense, laws and regulations were implemented to improve the monitoring of the territory and water resources. In the Brazilian context, the current Forest Code – Law No. 12,651/2012 stands out, which provides for Permanent Preservation Areas (APPs), whose environmental function includes the protection of rivers and also the fauna and flora. According to the aforementioned Law, APPs are delimited based on the width of the edge of the regular river bed. However, the correct delimitation of the edge of the aforementioned gutter is not well defined in the current Forest Code, presenting subjectivity, which compromises the preservation and conservation of these resources. In this sense, there is a global trend in the use of Remote Sensing (RS) for environmental and urban monitoring, highlighting the acquisition by sensors on board Remotely Piloted Aircraft (RPA). Therefore, the classification techniques of these images have also evolved, with emphasis on the Image Analysis Based on Geographical Objects (GEOBIA). Considering these aspects, the objective of this work was to delimit the edge of the regular bed of a stretch of watercourse located in the Monjolinho sub-basin, in the municipality of São Carlos - SP. For that, images obtained with the use of RPA and subsequent application of the GEOBIA technique were used. The research methodology began with a previous analysis of the rainfall seasonality in the municipality, since according to the current Forestry Code, the channel of the regular bed of a river is defined by “the channel through which the waters of the watercourse run regularly during the year”, therefore, it is necessary to visualize the minimum flow of the river, that is, the periods of little rain. Then, the planning of field activities, aerial survey and image processing were carried out, in order to obtain the orthomosaics of the periods of high and low rainfall and later apply GEOBIA, followed by the validation of the results, through the field truth and application of the Kappa Index (IK). As a result of this project, first the orthomosaics were obtained, referring to the aerial surveys carried out in the months of March and September, high and low pluviometry respectively, followed by the elaboration of the maps by manual vectorization, and later application of GEOBIA, this step consisted in the execution of the segmentation, feature extraction, sample collection and finally the classification through the decision tree algorithm. The validation of the results demonstrated a satisfactory classification through GEOBIA with Kappa Index values of 0.6682 and 0.7102 for the periods of high and low rainfall, a fact corroborated by the Accuracy values of 0.8382 and 0.8727 for both periods respectively. The values found indicate that for the period of low rainfall the riverbed was better delimited. Therefore, the results of this research can serve as a basis for public and private agencies, where there is a need to monitor water resources, for the purposes of preserving the APP.