HIGH-RESOLUTION SNOW DEPTH DISTRIBUTION ESTIMATION BY MEANS OF OUT-OF-SAMPLE LSTM

Abstract

This study conducted snow depth distribution estimation by means of out-of-sample LSTM (Long Short-Term Memory Network). The training dataset was divided based on the snow-depth observation points. The target data was the snow depth at the observation points scattered in the target area. As for the input data, an atmospheric reanalysis and the grid data of precipitation were used. Because the resolution of the air temperature data obtained from the atmospheric reanalysis is lower than that of the precipitation data, the air temperature data were refined by means of a high-resolution digital elevation model (DEM) and the temperature lapse rate. As the result, it was shown that the model constructed by out-of-sample LSTM has the potential to estimate the snow depth distribution that reflects the influence of elevation. This study shows that the usefulness of out-of-sample LSTM in estimating the snow depth distribution.