Urban Matanuska Flood Prediction using Deep Learning with Sentinel-2 Images

Abstract In this paper, we produce a novel raster dataset depending upon the Sentinel-2 satellite. They envelop over thirteen spectral bands. Our novel data set consists of ten classes within a total of 27000 Geo-referenced and labelled images. Gradient Boosting Model (GBM) used to explore this novel dataset in which the overall prediction and accuracy of 97% is obtained from the support of Graphics Processing Unit (GPU) afforded from Google Colaboratory (Colab). The obtained classification result can provide a gateway for numerous earth observation applications. Here, in this paper, we also elaborate on how this classification model might be applied for a conspicuous change in land cover and how it plays an important role in improving the graphical maps.