AI and the World's Hard Problems · Earth & climate science · arXiv · 2026-10-20
Permafrost augmented simulation and forecast using hybrid modeling through spectral neural networks
Physics-based permafrost simulations are slow and energy-intensive, which limits long-term forecasting. The authors built a fast surrogate model using a spectral neural network trained on high-performance numerical simulations of water and heat transfer at a forested continuous permafrost site, and enhanced it with measured data in a hybrid framework. The hybrid model showed markedly better accuracy on test measurements and was used to forecast permafrost behavior over the coming decade under CMIP6 climate scenarios.