Extending the air2water Model to Simulate Lake Ice Phenology, Thickness, and Composition Using Minimal Inputs

Ice cover in seasonally frozen lakes plays a crucial role in ecosystem dynamics and human activities, while also serving as a sensitive indicator of climate change. Accurate yet efficient modeling of lake ice timing and thickness is therefore increasingly important. This study presents a new ice module integrated into the existing air2water model, a hybrid physics-based/statistical model originally developed to predict lake surface water temperature (LSWT) using air temperature as its sole input. The extended model simulates ice cover dynamics, distinguishing between black ice (formed by direct lake water freezing) and white ice (from accumulated snow submerged and frozen over black ice). The extension preserves the original framework's key advantages: minimal input requirements (air temperature and precipitation), low parametrization (8–11 parameters), and physically based governing equations. Calibration was performed using an automatic optimization algorithm with a multi-objective performance metric combining LSWT and ice thickness Nash-Sutcliffe efficiency indices (NSEs). Evaluation employed long time series (1960–2023) of LSWT and ice thickness data from three Finnish lakes spanning different climatic conditions. When calibrated with equal weighting of LSWT and ice thickness NSEs, the model demonstrated robust performance across all lakes, with daily LSWT root mean square errors (RMSEs) near 1°C and ice thickness RMSEs of about 10 cm. Notably, consistent ice thickness predictions were achieved even when calibration relied solely on LSWT data. LSWT simulations also improved relative to the original model. These results show that the extended air2water model offers a competitive and data-efficient alternative to more complex lake ice models.