Deep learning to simulate and predict glacier change in the high mountains of Asia

Abstract

Globally, climate change causes glaciers to retreat. Driving mechanisms at local scales are poorly understood. This study aims to discover which climatical and morphological variables most contribute to explaining the Specific Mass Balance (SMB) variability of 9098 individual glaciers in High Mountain Asia (HMA). We separate the data into 15 regions assumed climatically homogeneous. We calculated correla- tions between the selected variables and the SMB. Additionally, we calculated a multivariate linear model and assessed a linear combination of the variables that best explain the SMB variability. We exclude all predictor-predictor correlations > |0.7|. We concluded that in the heterogeneous climate setting (HMA), the selected variables explain 30.7% of the glaciers’ SMB variability, with the most important predictors being the presence of a lake, the slope and the mean temperature and precipitation in 2000-2020. All predictors, except for the slope, are found to be associated with negative SMB. In the subregions, we conclude that the selected variables explain 18.1% to 50.0% of the SMB variability, with generally the most important predictors being the morphological variables: the presence of a Lake, the slope, and the median elevation. The slope and the median elevation are found to be associated with positive SMB. In all analyses, we observed a large influence of glacial lakes on the SMB. We expect this partly results from unobserved correlations between the variables lake and slope. Only the subregion West Kun Lun was associated with positive SMB. Here, we expect this contrasting behaviour to result from the dry and cold climate settings. A major limitation of this study is the use of a linear model for non-linear data. This resulted in relatively low model performance in the climatically heterogeneous region HMA.