Assessment of areas of selective logging and burned forests in Mato Grosso State, Brazil, from satellite imagery

This paper presents a semi-automated procedure for mapping degraded forest areas in Mato Grosso state, Brazil, from medium (30m) resolution satellite imagery. In the Brazilian Amazon region forest degradation in the context of UNFCCC REDD+ (i.e. considered as long term reduction of Carbon stocks in forests) is mainly driven by logging activities and forest fires. Medium to fine spatial resolution satellite data allow assessing the extent and distribution of deforestation, selective logging and burned forest areas with high reliability. However, it is difficult to obtain the full coverage of a region with cloud free satellite imagery for mapping logging tracks and burned areas. Such dynamic landscape features are detectable on satellite imagery only during a limited period of time in humid tropical regions (a few weeks / months). The proposed method consists in the further development of the approach used in the TREES-3 pan-tropical forest area survey conducted by the European Commission’s Joint Research Centre. It is applied to Mato Grosso State, Brazilian Amazon, using Landsat satellite imagery (Thematic Mapper sensor), acquired close to the year 2010. The approach is using a sample of sites systematically distributed at confluence geographical points (77 sample sites in Mato Grosso). The semi-automated procedure for analysis of satellite imagery extracted over the sample sites consists into: a) image pre-processing, b) production of ‘fraction images’: vegetation, soil and shade, and c) creation of spatially and spectrally homogeneous areas (‘objects’). In each of the 77 sample sites, spatial objects are then labelled as burned areas or logged forest areas through visual image interpretation. The resulting maps for the 77 sample sites are then combined with the forest cover maps of year 2010 from the TREES-3 survey. Statistical estimates of forest degradation by fires and selective logging are finally generated. The results show that 18,452 km 2 and 9,435 km 2 of forests appear as burned or affected by selective logging in year 2010, respectively, representing c. 4% and 2% of the total forest area of this state. The future availability of time series of finer spatial resolution data from the Sentinel-2 satellite (10m resolution) is expected to allow improving the assessment of forest degradation processes using such approach.