Three-year assessment of the space-time dynamics of burned forest in the Brazilian Amazon, State of Mato Grosso

The use of fire in tropics for land management and for forest clearing contributes to forest degradation and to carbon emissions in the atmosphere. Yet, a comprehensive understanding of the spatial and temporal dynamics of forest fires is needed to mitigate such impacts. The primary aim of this study is to map burned forest in Brazilian Amazon by means of an object-based classification applied on Landsat imagery. Our secondary aim is to understand the spatial and temporal dynamics of forest fires. For this, we compare the forest cover maps with the map of burned forest generated during the dry season. In our study we consider three consecutive bi-annual periods (2013-2014, 2014-2015, 2015-2016) over four sites, in Mato Grosso State. From Landsat-8 imagery acquired during May or June of each year we derived NDVI and fraction images (soil, vegetation and shade) in order to map the following land cover classes: intact forest, burned forest, forest regrowth and non-forest. In addition, we map the burned areas during the fire season for each year. Intact forest, initially covering 55% of the area has lost 3.4% of its area over the course of three years. The annual burning increased from 2.6% to 5.9%, from 2013 to 2015. Within the annually burned areas, non-forest areas showed a lower percentage of burning, compared to forest areas, varying from 3.8% to 7.9%, while amongst the forest classes, burned forest showed the highest percentage. Re-burning on a biannual period is more than double of re-burning on a thrice-year.