Breaks for Additive Seasonal and Trend (BFAST) algorithm applied to support the detection of changes in Portuguese forests

#BFAST #wildfires #GPU

In this seminar the application of the Breaks for Additive Seasonal and Trend (BFAST) algorithm is described to support the detection of changes in Portuguese forests, observed by the Sentinel-2 satellite. In particular, the considered implementation of the algorithm relies on the use of PyOpenCL, which supports co-processing between CPU and GPU, from Python wrapping OpenCL, and allows the efficient multi-core implementation of the BFAST to reduce processing time. A total of 204 Sentinel-2 images from a central region of Portugal were used to detect and evaluate the changes that occurred there from 2017 to 2020, and to test the use of BFAST for known fires. Also, a set of 5 tests was carried out to assess the accuracy of the application of BFAST, and to inspect the impact of different time series on results quality, and the impact of the size of the region on the computation time.