Super-resolution for terrain modeling using deep learning in high mountain Asia

High Mountain Asia (HMA) is characterized by some of the most complex and rugged terrain conditions in world. However, high resolution data are not easy to quickly acquire from area due difficulties accessing region. In this study, we trained a modified super-resolution residual network (MSRResNet) develop (SR) digital elevation models (DEMs) HMA areas using freely available DEM region limited (HR) DEMs other train model. network, new loss function was constructed that considered parameters slope curvature constrain learning convergence. The proposed method applied validated Hengduan Mountains southeastern part HMA, which world-famous longitudinal belt mountains canyons. A comparative analysis between current existing methods (i.e., SRGAN Bicubic interpolation) conducted assess effectiveness approach. experimental results were also investigated evaluated visual inspection parameters. demonstrate MSRResNet process can achieve highly accurate downscaling HMA. This SR outperforms comparable methods. Compared interpolation method, RMSE MAE accuracy improved 32.17% 33.97%, compared 39.15% 32.47%. HR generated more conducive improving extracted features, such as stream networks. It promising apply model on Earth or even planets with similar