Near-real time deforestation detection in the Brazilian Amazon with Sentinel-1 and neural networks

Optical-based near-real time deforestation alert systems in the Brazilian Amazon are ineffective rainy season. This study identify clear-cut deforested areas through Neural Network (NN) algorithm based on C-band, VV- and VH-polarized, Sentinel-1 images. Statistical parameters of backscatter coefficients (mean, standard deviation, difference between maximum minimum values – MMD) were computed from 30 images, 2019, used as input NN classifier. The samples manually selected, including forested areas. After deforestation, mean signals decreased average 2 dB for VV 2.3 VH May to September–October. A Multi-Layer Perceptron (MLP) network was detecting forest disturbances larger than ha. Case studies performed both polarizations considered following sets MLP: mean; deviation; MMD; mean, MMD. For 2019 dataset, latter showed best performance with accuracy F1 score 99%. Automatic extraction using 2018 images reached 89% MapBiomas reference data 81% 79% PRODES data.