Widespread loss of safe lake ice access in response to a warming climate

2. Methods 2.1 Modeled data We employed the lake simulation results in the Community Earth System Model Version 2 Large Ensemble (CESM2-LE) project to estimate phenological shifts in lake ice thickness across the Northern Hemisphere [19]. Lake ice thickness was derived from the Lake Ice, Snow, and Sediment Simulator (LISSS), which is a one-dimensional process-based model that incor- porates ice growth, with the additional complexities of aerosol deposition on snow albedo, freeze-thaw cycles, ice physics, and sediment heat flux throughout the water column [20]. The model is integrated at a 30-min time step and forced by meteorological variables, including air temperature, pressure, snow, humidity, wind speed, shortwave radiation, and downward long- wave radiation [21] from the atmospheric model in the CESM2. We used daily mean of mod- eled ice thickness to detect the day of the year when ice growth reached a specified thickness deemed safe. We then compared the temperature anomalies at 1˚C, 2˚C, and 4˚C to a histori- cal climate period of 1851–1880, representing a 30-year period of unperturbed climate by human activities. For a more thorough explanation of the lake ice variables derived from CESM2-LE and LISSS, see [16,20]. 2.2 Validation The spatial extent and resolution of the study include the Northern Hemisphere on a 0.9˚-by- 1.25˚ gridded scale. Within each grid, the LISSS coupled to the CESM2-LE uses a sample lake to model ice thickness, where the sample lake has a depth and surface area of the mean of individual

lakes within the grid, taken from the Global Lake and Wetland Database [22] and global gridded depth data at a 1 km resolution [23]. CESM2-LE offers data for lake ice-on and ice-off timing, using the daily lake ice thickness from the one-dimensional lake ice model. The LISSS model has shown accurate representation of ice thickness and other variables (e.g., water temperature and surface fluxes) in previous studies [16,20,24–27]. Ice thickness was validated with records from 15 Canadian lakes with an R2 = 0.86. While the correlation coefficient was high, the modeled data deviated from the 1:1 line, such that the model tended to underpredict ice thickness (slope = 2.1) [16]. This bias toward thinner ice cover values is likely due to the warm bias of the simulated land surface air temperature in the fully coupled model, which was warmer than the ERA5 reanalysis data between the study years 1981–2020 and 30˚N-70˚N latitude by 1.5˚C [16]. Additional valida- tion information can be found in [16] and the associated supplementary information. Owing to the small sample size in [16], in this study we sought to further validate the ice thickness component of LISSS. We aggregated 71 lakes from datasets from Finland [28], North America [29,30], and Russia [31]. Using 71 ice thickness records with observations between 1980-present, we found that a comparison of observed and modeled ice thickness yielded a root mean square error (RMSE) of 18.5 cm and an R2 = 0.74 (n = 71) (Fig 1). As with the validation procedure in [16], the modeled data tended to underpredict ice thickness, which

is evident from the slope of 0.49 (Fig 1A). When directly examining lake ice thickness time series, the mean ice thickness of the ensemble members tends to underestimate the peak ice thickness (Figs 1B and S1). The mean and 2σ may also underpredict the peak ice thickness; however, they tend to capture the ice growth period and ice loss period, which are most impor- tant for this study. The model also tends to predict lake ice every year (1980–2024); however, some lakes did not freeze in years during data collection. It should be noted that the lake ice thickness model (LISSS) uses a mean lake area and depth to model ice thickness [16,20]. Therefore, peak lake ice thickness is likely to vary between lakes that do not approximate that mean area and depth. PLOS ONE Changing ice conditions inhibit safe use PLOS ONE | https://doi.org/10.1371/journal.pone.0313994 December 11, 2024 3 / 19