COMPARING PER-PIXEL AND REGION-BASED CLASSIFICATION METHODS USING CBERS-4/MUX IMAGES TO ANALYSE LAND COVER CHANGE CAUSED BY THE MARIANA DISASTER

In November 5th, 2015, the Fundão dam’s rupture in Mariana, Minas Gerais, Brazil, dumped millions of cubic meters of tailing into the river, causing abrupt changes in
the land cover (LC). Remote Sensing (RS) techniques and image analyses allow monitoring LC changes, that can contribute for decision making. In this paper we show results of LC change detection caused by the disaster applying per-pixel and region-based classifiers. For this purpose, three CBERS-4/MUX images were independently classified to assess LC in different periods: prior the incident, right after and its current situation. The per-pixel classification distinguished rivers from other classes, better than the region-based classification. In addition, the changes detected in the LC helped to highlight vegetation areas affected by the incident and also to evaluate its effects. Furthermore, the analysis was able to identify regenerated vegetation areas.