Integrating Ice Analyses and Varying Floe Size Parameterization Into Great Lakes Ice Prediction

Abstract

Accurate prediction of Great Lakes ice cover is critical for regional weather, navigation support, and safety. Current operational models often exhibit biases, such as over‐prediction of ice concentration and delayed spring melt, potentially due to simplified parameterizations. This study addresses this issue by first characterizing ice floe size variability using 12 years (2010–2021) of satellite‐derived ice charts from the U.S. National Ice Center (NIC). We tested spatially‐ and temporally‐varying floe size parameterizations in the Finite Volume Community Ocean Model (FVCOM), comparing its performance against the standard operational configuration, which uses a constant 300 m floe size. The analysis reveals that floe sizes are highly variable and correlate with winter severity. Model simulations for a high‐ice (2019) and low‐ice (2020) winter show that variable floe size parameterizations significantly reduce model bias in ice concentration and improve categorical accuracy for ice thickness when compared to a constant floe size configuration. For example, exact categorical accuracy for Lake Superior ice thickness improved from 41% to 67% in 2019. These findings demonstrate that integrating variable ice floe size from satellite data resolves a key model deficiency and offers a practical path toward improving operational ice forecasting in the Great Lakes.