Climate Model Biases in Rain-on-Snow Days Across the Central United States

Regional climate simulations have explored how rain-on-snow (ROS) events may change in the future; however, the confidence in these projections depends on how ROS is represented in models during the historical period. Here, we use observations from 1982-2011 to characterize ROS days over North America, then use the same methodology to detect ROS days in Coupled Model Intercomparison Project Phase 6 (CMIP6) simulations. The CMIP6 simulations produce fewer ROS days and less variability from year to year in ROS days than the observational data for the region east of the Rocky Mountains, which also shows different trends in the number of ROS days. To understand where these biases arise from, we consider simulations with fixed sea surface temperatures (SSTs), higher resolutions, and regional domains and find that these biases are not sensitive to model resolution or SST biases. We then explore the relationship between ROS days and precipitation and snowpack characteristics in the models. Models that produce the fewest number of ROS days tend to have the lowest fraction of days where snow melts after precipitation; however, when a ROS day occurs, these models also tend to produce a larger decrease in snow water equivalent and the smallest amount of surface latent heat flux. This suggests that the representation of land-atmosphere processes in these models is central to understanding the biases in ROS days. To further explore the relationship between surface latent heat flux and ROS representation in models, WRF simulations are run with different surface coupling schemes, which are found to change the characteristics of ROS events.