Evaluating Approaches to Lake Representation in Operational Forecasting

Lake representation in operational weather and climate models can have important impacts on forecast accuracy. Recent work has demonstrated that predictions of lake-effect precipitation, cloud formation, and air temperature can be sensitive to lake surface conditions (Fujisaki-Manome et al., 2022; Benjamin et al., 2022; Fujisaki-Manome et al., 2020). However, most weather and climate models used for operational forecasting or climate projection do not adequately resolve lake physics (Briley et al., 2021; Minallah and Steiner, 2021). The only NOAA model to include either a lake model or integrate operational lake conditions into the forecast cycle is the High-Resolution Rapid Refresh (HRRR; Benjamin et al., 2016; Dowell et al. 2022). For large lakes, specifically the Great Lakes, the current version of the HRRR integrates surface conditions from the NOAA Great Lakes Coastal Forecast System (GLCFS; Anderson et al., 2018), which is built on a three-dimensional hydrodynamic model of each lake. For all other lakes that are represented in the 3km domain, cycling of lake conditions is modeled using the Community Land Model (CLM) lake model (Subin et al., 2012; Benjamin et al., 2022). While lake cycling under this approach is computationally inexpensive as part of the land surface scheme, lakes are represented by one-dimensional 3-km pixels (in the case of the HRRR domain) that do not interact except through indirect atmospheric means. As a result, physical limitations are introduced through lack of horizontal advection, fetch limitation, and other factors. In particular, ice cover, surface temperature, and surface heat fluxes can become unrealistic. Efforts to improve small lake representation have shown that optimized bathymetry can in some cases have important impacts on surface conditions but data is limited for model validation (Kessler et al., 2022; Sorensen et al., 2024). In this study, we use a comparative analysis of three-dimensional and one-dimensional approaches to investigate optimized approaches to lake representation in numerical weather prediction. A systematic approach to lake representation, based on a variety of parameters, may help determine an optimal configuration. However, computational expense and validation data continue to serve as limiting factors.