How to improve global lake water temperature projections: findings from calibrating 4 lake temperature models to 73 lakes

Global warming is impacting lakes and reservoirs through change in the water temperature and thermal stratification which are affecting ecosystem processes like nutrient recycling that can fuel re-eutrophication or increase methane emission. To quantify these impacts and plan mitigation strategies, process-based projections of water temperature and stratification are needed. For projections on individual lakes, these models are usually calibrated using historic water temperature observations. However, sufficient observations are generally not available, so for global simulations it is common to apply the models without a lake or region specific calibration, which adds additional uncertainty to the projections. As part of the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) we calibrated four 1D lake temperature models (FLake, GLM, GOTM, Simstrat) using a standardized methodology to a global set of 73 lakes for which in-situ water temperature observations are available. We evaluated the performance of the lake models, estimated the sensitivity of the calibrated model parameters, and related the results to the different model structures and lake characteristics. We highlight how each model differed in their ability to replicate the water temperature dynamics of specific lake types, but also acknowledge that each of the models performed best for a particular subset of lakes. Even though we did not find general relationships between model parameters and lake characteristics, we underscore modeling takeaways to improve global simulations without the need for model-specific calibration. For most models, the most sensitive parameter was the scaling factor for wind speed. Further, our results indicate that accounting for internal seiches in the model can likely increase model performance. From our findings we want to discuss possible paths forward to further improve the quality of global simulations, i.e. improvements in forcing data, model process description, and using (multi-model) ensemble techniques.