Mapeamento da área do lago da UHE Sinop com imagens de Radar Sentinel-1

Abstract

Understanding the dynamics of rivers is crucial for the conservation and maintenance of the water system, flood prevention and energy production, especially during rainy periods. Radar systems, such as Sentinel-1, enable mapping even in the presence of clouds. In this context, the aim of this work is to use Sentinel-1 radar images to map the reservoir area of the Sinop Hydroelectric Power Plant (UHE Sinop) by means of supervised classification using machine learning techniques. The database used was the Google Earth Engine (GEE)data catalog, from the Sentinel-1 C-band satellite (5.405 GHz), with a temporal resolution of 12 days and a spatial resolution of 10 m. The Maximum Likelihood (MaxVer) and Random Forest (RF) algorithms were used for classification. The classification was evaluated based on the accuracy, F-score and Kappa indices. The two classification methods proved to be efficient, with no significant difference between them. For the RF algorithm, the F-score ranged from 0.9994 to 0.9997 and the Kappa from 0.9788 to 0.9897. For MaxVer, the F-score ranged from 0.9995 to 0.9996 and the Kappa from 0.9866 to 0.9872. The Sentinel-1 satellite has the potential to map water mass dynamics, contributing to environmental flood management and energy productionplanning, especially during times of maximum cloudiness.