Author(s): Shahin Nourinezhad; Jiahui Qiu; Heini Postila; Nasim Fazel; Ali Torabi Haghighi
Linked Author(s): Nasim Fazel
Keywords: Lake ice cover; Ice thickness; Climate change; XGBoost; Winter instability; Ice phenology; Finland lakes
Abstract: Lake ice is a defining component of cold climate regions, influencing thermal stratification, winter mixing, oxygen availability, light penetration, ecosystem structure, and greenhouse gas dynamics. This highlights the importance of accurate and continuous records of ice thickness and duration in cold-climate regions. Lake ice thickness observations in different lakes globally vary widely in duration and sampling frequency. Some lakes have long-term records, while others have major gaps, making comparisons and trend analysis difficult. Recent advances in artificial intelligence have demonstrated strong potential for accurately simulating hydrological variables under data-scarce conditions. In this study, we employ a gradient boosting framework (XGBoost) to reconstruct long-term lake ice thickness for three Finnish lakes, namely Tuusulanjärvi, Pieksänjärvi, and Vuokkijärvi over the period 1961–2024. The root mean square error (RMSE) ranged between 3.99 and 6.24 cm, and the R-squared (R²) between 0.79 and 0.91 in the validation phase, indicating that ice thickness can be predicted with substantial accuracy. Spatiotemporal analysis of the reconstructed ice dynamics reveals that the decline in ice thickness and duration was more pronounced in Tuusulanjärvi than in Pieksänjärvi, and in Pieksänjärvi more than in Vuokkijärvi. This pattern indicates that the magnitude of ice thickness and duration decline has been greater at lower latitudes.
Year: 2026