CROWDSOURCING FOR DEFORESTATION DETECTION IN THE AMAZON
Keywords: Crowdsourcing, High-Resolution Land Cover, Deforestation, Rainforest, Land Cover, OpenStreetMap
Abstract. Every year, deforestation results in the loss of wide stretches of forest which is worsening the state of air quality, biodiversity, indigenous cultures, climate, meteorological conditions, etc. According to the Monitoring of the Andean Amazon Project (MAAP), roughly 20 million hectares of land were lost due to deforestation in 2020. To address the issue of deforestation, this study proposes a derivation of the deforestation risk model to target the spread of deforestation, which is the first step towards its prevention. The region of interest - North West of Mato Grosso, Brazil - was selected based on two characteristics: it is a deforestation hotspot according to MAAP and it comprises 4 indigenous lands. The sequence for developing the risk model comprises reference information collection, information cleaning, classification, postprocessing, and change detection. In a crowdsourcing mapathon, reference data were gathered, and they were refined in an iterative process using existing land cover maps and photo interpretation. Google Earth Engine and the Random Forest algorithm were used to classify Sentinel-2 imagery for 2019 and 2020. The results obtained are land cover maps from 2019 and 2020 and land cover change, and the risk model. The results are not demonstrating intensive deforestation in the region of interest, however, the deforestation appears to be systematic in two subregions, indicating that it has the potential to spread. An additional concern in the case of these subregions is their proximity to the indigenous land.