Toward Minimally Calibrated Physics-Based Glacier Mass Balance Modeling

Glacier mass loss has accelerated globally in recent decades, yet many regional and global glacier models still rely on temperature-index melt formulations that limit physical realism and process attribution. Physics-based approaches, such as surface energy balance (SEB) models, can overcome these limitations but are challenging to apply at regional scales because they require climate forcing resolved at local scales. Global climate models and reanalysis products are known to perform poorly in complex mountainous terrain without downscaling and bias correction. In addition, albedo parameterizations in SEB models typically require calibration against in-situ observations, which are sparse for most glacierized regions.

Here, we evaluate the skill of a physics-based glacier modeling framework designed to reconstruct glacier mass changes with minimal parameter calibration and without downscaling, and with only limited bias correction of climate input data. We first assess a relatively simple SEB model forced by climate variables from the European Centre for Medium-Range Weather Forecasts ERA5 reanalysis. Surface albedo, a key input to the SEB model, is prescribed using a standalone machine-learning model trained on Moderate Resolution Imaging Spectroradiometer (MODIS) satellite observations. We show that calibration of selected model parameters—most notably precipitation correction and albedo bias correction—is required for the model to perform well from local to regional scales across western Canada. We then evaluate a more complex SEB formulation using the open-source COupled Snowpack and Ice surface energy and mass balance model in PYthon (COSIPY), with the aim of assessing whether parameter calibration can be further reduced or avoided. We find that neural-network-based albedo estimates substantially improve model performance and that calibration-dependent albedo bias correction is no longer required when Landsat data are used instead of MODIS. In contrast, wind speeds from ERA5 require bias correction to obtain realistic turbulent heat fluxes, highlighting the importance of improving the representation of katabatic winds over glacier surfaces. With these adjustments, the non-calibrated physics-based model performs well in simulating summer mass balance at the glacier scale, provided that snow accumulation at the onset of the melt season is adequately captured.