Authors: Dane Liljestrand – University of Utah
Title: Advancing Proglacial Soil Moisture Prediction in the Arctic Using LSTM Networks
Presentation Type: Poster Presentation
Abstract: The rapid retreat of Arctic glaciers is exposing extensive proglacial landscapes where soil moisture dynamics play a crucial role in microbial development, carbon cycling, and geomorphic evolution. Despite its importance, modeling soil moisture in these environments remains challenging and underperformed due to complex terrain, sparse in-situ observations, and limitations of existing remote sensing products at high-latitudes. To address these research gaps, we have applied a high-resolution spatio-temporal application of Long Short-Term Memory (LSTM) neural networks to model near-surface soil moisture in a proglacial moraine field in Svalbard. This approach integrates high-frequency in-situ soil moisture measurements, and topographical data with nearby surface meteorological observations, leveraging the LSTM’s ability to capture complex spatio-temporal patterns across the heterogeneous proglacial landscape.
Early findings demonstrate strong predictive performance across different soil types and topographic positions, with validation Kling-Gupta Efficiency (KGE) scores exceeding 0.75 for seasonal predictions. The framework effectively resolves fine-scale patterns while maintaining temporal accuracy across both wet and dry periods, capturing key hydrological transitions in the proglacial environment. To assess broader applicability, we propose the model transferability between Arctic proglacial environments. By comparing predictions with ERA5 reanalysis data across both Svalbard and Western Greenland, we establish the framework’s potential for broader application in data-sparse glacial environments. This research advances our ability to monitor and predict soil moisture in rapidly evolving Arctic landscapes at high resolution, providing a valuable tool for understanding fine-scale hydrological processes in Arctic proglacial environments.