Evaluation of the Integrated Use of Nanosatellite Images and Classifiers based on Machine Learning for Studies of Hydrological Dynamics in the Nhecolândia Region (Pantanal)

Abstract

The Lower Nhecolândia region is one of the most iconic landscapes in the Pantanal Basin. Its unique morphology comprises more than 10,000 lakes with saline-alkaline water and fresh water that coexist in an area of approximately 12,000 km2. This region is subject to seasonal flooding that acts on runoff; however, little is known about its flooding dynamics. Recent advances in the area of geoprocessing have helped expand our knowledge about lacustrine environments. This work evaluates the performance of two supervised classifiers based on machine learning (Support Vector Machine and Random Forest), for characterizing the hydrological dynamics of the Nhecolândia region. The classifiers were applied to nanosatellite images (PlanetScope) using the Google Earth Engine cloud computing platform. The results showed satisfactory and similar performance of these two classifiers.