Glacier mass balance modeling using a Long Short-Term Memory network

Glacier mass balance is a key indicator of climate change and a central driver of glacier evolution, yet most glaciers worldwide lack long-term in-situ measurements. For estimating glacier mass balance, data-driven models offer a complementary pathway to traditional numerical approaches by learning empirical relationships between climate forcing, topography, and mass balance directly from observations. Here, we develop a recurrent neural network based on a Long Short-Term Memory (LSTM) architecture within the Mass Balance Machine (MBM) framework and evaluate its ability to predict seasonal and annual point surface mass balance across the Swiss Alps. MBM is trained on 30'000 point observations from 30 glaciers and tested on eight glaciers excluded from training to assess spatial generalization. MBM predicts both winter and annual mass balance with high accuracy on unseen glaciers, and its recurrent structure provides a clear advantage by allowing the model to learn temporal dependencies, which improves the representation of seasons with strong accumulation or ablation. Beyond point predictions, MBM produces spatially distributed mass balance maps that closely resemble reference products directly derived from in-situ data, capture elevation-dependent gradients, and yield glacier-wide mass changes consistent with the differencing of repeated terrain models. Monthly outputs further show that the model captures the seasonal transition from winter accumulation to summer ablation and its dependence on elevation with realistic timing and magnitude. These results indicate that a recurrent neural network approach can recover key characteristics of glacier mass balance dynamics from sparse in-situ observations and that the learned relationships are transferable across glaciers with distinct meteorological and topographic settings. The demonstrated generalization skill and suitability for transfer learning highlight the potential of MBM for predicting glacier mass balance in regions with limited or no direct measurements.