DETECTING CHANGES OF RETROGRESSIVE THAW SLUMPS FROM SATELLITE IMAGES USING SIAMESE NEURAL NETWORK

Retrogressive thaw slumps (RTSs) are among the most dynamic landforms resulting from the thawing of ice-rich permafrost; but their distribution and evolution are poorly quantified. In this study, we propose a change detection algorithm based on the Siamese neural network, aiming to automatically detect RTS development from remote sensing images. Firstly, we derived training data based on multi-temporal Planet CubeSat images and polygons covering the RTS expanding areas. Secondly, we designed a simple network and trained it using a portion of the training data. Lastly, we predicted RTS expanding areas using the well-trained algorithm. Validating against ground truths shows that the precision, recall, and F1 score are 0.43, 0.96, and 0.59, respectively, indicating that the proposed change detection algorithm can identify RTS expanding areas correctly. Potentially, this method can be applied to larger areas and fill in a major knowledge gap about distribution and evolution of RTSs in permafrost areas.