Emerging unprecedented lake ice loss in climate change projections

Seasonal lake ice plays a crucial role in both local communities and lake ecosystems. The observed widespread loss of lake ice in recent decades, characterized by delayed freezing and earlier melting, has raised concerns about the ecological consequences of climate change. Previous studies using offline lake models and statistical models have linked this observed loss to greenhouse warming and projected further reductions in the future. However, these studies often lacked the full coupling between the lake, lake ice, and the atmosphere, and did not resolve the diurnal cycle of thermodynamics—limitations addressed in this study. The fully coupled Earth system model used here explicitly represents thermodynamic processes including ice growth and melt, vertical mixing, and exchanges of heat and momentum between the lake and the atmosphere. This is particularly important for large lake systems which significantly influence local climate. The study aims to provide a mechanistic understanding of lake system sensitivity to greenhouse warming and to determine the emergence timescales of anthropogenic signals beyond natural variability. This is crucial for assessing risks to and planning adaptation for lake ecosystems facing unprecedented ice conditions.

Prior research has documented a significant decline in lake ice cover across the Northern Hemisphere. Studies utilizing offline lake models forced by multi-model climate simulations have established a clear link between greenhouse warming and observed lake ice loss, projecting further intensification of this trend. Other research employed statistical models based on empirical relationships between ice phenology and meteorological variables to forecast future changes, but these lacked projections of changes in ice thickness and duration. While recent studies have used process-based lake-ice models, these were typically offline models forced by daily meteorological data, lacking the crucial three-way interaction between lake, lake ice, and atmosphere, and often not resolving diurnal cycles. The lack of resolution on these dynamic interactions especially affects large lake systems, which have substantial impacts on local climate. This research directly addresses these limitations by incorporating interactive lake simulators within a fully coupled Earth System Model.

This study leverages a 90-member subset of the Community Earth System Model version 2 Large Ensemble (CESM2-LE) simulations. These simulations, spanning 1850–2100 under a historical/SSP3-7.0 emission pathway, were initialized with perturbed initial conditions, providing a physically consistent estimate of both anthropogenic signals (ensemble mean) and natural variability (ensemble spread). This approach is advantageous for determining the emergence of anthropogenic signals and no-analog conditions, as it avoids the suppression of natural variability and mixing of sensitivities from different climate models, common in multi-model ensemble forcing approaches. The model used here, CLM5’s Lake, Ice, Snow, and Sediment Simulator (LISSS), is a one-dimensional thermodynamic model that resolves the diurnal cycle and explicitly represents thermodynamic processes such as ice growth and melt, vertical mixing, and surface fluxes between the lake surface and the atmosphere. The model’s performance is validated against observational records from the Global Lake and River Ice Phenology dataset, focusing on records with durations of at least 20 years, and against ice thickness data from the Canadian Ice Thickness Program. Analysis includes assessment of changes in ice phenology (freeze and breakup dates, duration), ice thickness, and emergence of no-analog conditions by comparing projected ice duration against the natural variability range (estimated from the 1850–1950 period). The emergence of no-analog conditions is defined as the point when the projected ice duration consistently falls below the lower 2σ limit of the natural variability range, signifying a 2.5% probability of returning to the natural range. Regional acceleration of lake ice loss is investigated by examining relationships between ice duration and air temperature anomalies.

The CESM2-LE simulations accurately reproduce observed climatological lake ice phenology, with high correlation coefficients between simulated and observed mean ice duration, freeze date, and breakup date (r values of 0.94, 0.78, and 0.94 respectively). However, the model shows a slight underestimation of ice duration, partly attributable to a warm bias in simulated surface air temperature. The simulations project a substantial decrease in global mean lake ice cover duration by 38 ± 11 days by 2100. Maximum ice thickness is also projected to decline by 0.23 ± 0.07 m. The strongest reductions in ice duration are projected over the Tibetan Plateau and the Canadian Arctic, exceeding −0.45 days per year. A strong negative correlation (r = −0.8, P < 0.001) is found between annual mean air temperature and ice duration anomalies in the simulations, consistent with observations (r = −0.47). The model shows a sensitivity of 9.9 days reduction in ice duration per 1 °C warming, similar to the observed sensitivity of 8.9 days. Analysis of the emergence of no-analog lake ice conditions reveals that in many lakes, ice duration started deviating from its natural range between 1980–1990. No-analog conditions, where there is less than a 2.5% chance of the ice duration returning to the natural range, are projected to emerge in some regions within the next 4–5 decades under the SSP3-7.0 emission scenario. This corresponds to a global warming level of approximately 1.9 °C. The study identifies the Canadian Arctic and the Tibetan Plateau as hotspots for rapid lake ice loss. In the Canadian Arctic, this is driven by Arctic and Hudson Bay sea-ice loss and cold-season polar amplification, while on the Tibetan Plateau, it's amplified by strong ice-albedo feedback under high solar radiation.

The findings highlight the significant and rapidly emerging impacts of climate change on lake ice. The close agreement between simulated and observed lake ice phenology provides confidence in the model projections. The identification of the Canadian Arctic and Tibetan Plateau as hotspots of accelerated ice loss underscores the regional variations in climate change impacts on lake ecosystems. The emergence of no-analog conditions within the next few decades emphasizes the urgency of climate action. The projected changes in lake ice will have far-reaching ecological consequences, affecting biodiversity, nutrient cycling, dissolved oxygen levels, and the timing of algal blooms. The study's results are crucial for informing conservation efforts and adaptation strategies for lake ecosystems.

This study demonstrates that unprecedented lake ice loss is already underway and will dramatically accelerate in the coming decades, particularly in the Canadian Arctic and Tibetan Plateau. The emergence of no-analog conditions within the next 4-5 decades under a high emission scenario highlights the urgent need for climate mitigation to avoid severe ecological consequences. Further research should investigate the impacts of specific lake characteristics on ice loss and explore the development of regional adaptation strategies. Future work with single-forcing ensemble simulations could further refine our understanding of the relative roles of greenhouse gases and aerosols in driving these changes.

The study uses a gridded simulation method with one representative lake per grid cell, potentially overlooking the influence of lake-specific factors like morphology and discharge on lake ice dynamics. The one-dimensional lake model does not capture horizontal heterogeneity in water temperature and ice cover, which can impact ice phenology, particularly in large lakes. Uncertainty in snowfall from the atmospheric model could affect the accuracy of lake ice simulations. While this is the most advanced modeling practice available, incorporating improvements in lake modeling and high resolution will be essential to refining these projections further.