Optimizing SAR data processing and thresholding for forest change detection: an application for early deforestation warnings on eastern Amazonia

Abstract: The present work proposes a prototype for an operational method for early deforestation detection of cloudy tropical rainforests. The proposed methodology makes use of Sentinel-1 SAR data processed into the Google Earth Engine platform for flag the areas where the probability of recent deforestation is high. The evaluation of the results over a region on the Eastern Amazon basin showed that copolarized data (VV band) offers the best results in terms of producer's accuracy (95,4% for a 5% significance, 88,9% for 1% significance), while crosspolarized data (VH band) offered excellent results in terms of user's accuracy (86% for a 5% significance, 100% for 1% significance).