Pavan Racherla

Can regional climate models provide realistic precipitation and surface temperature 
information for continental-sized regions over the course of long simulations?

Despite several outstanding issues, regional climate models continue to be deployed 
in increasing numbers for a variety of applications including climate forecasts. We 
ask here if these models are capable of providing realistic climate information for 
the continental U.S., and address this question by performing two 10 year 
simulations (corresponding to the 1980s and 2000s, respectively) with the Weather 
Research & Forecasting model (at a 20X20 km2 resolution) and constrained at the 
lateral boundaries by the North American Regional Reanalysis. Both the ability to 
reproduce the climatology for a given time-slice (2000s) and trends across two 
time-slices (1980s to 2000s) are assessed. The degree of realism of the simulated 
precipitation and surface temperature data is shown to vary widely depending on 
the region and season. Although there are some large biases at an individual model 
grid-cell level, the high quality of the spatial correlation between the simulated and 
observed surface temperature (0.86 to 0.98), as opposed to the generally lower 
quality for precipitation (0.48 to 0.67), suggest that, with some simple statistical 
filtering, the simulated surface temperature information in such models is much 
more realistic from an end-use standpoint than the precipitation. The simulation 
does a generally poor job of capturing observed daily precipitation extremes. 
Interestingly, these biases appear not to matter all that much when it comes to 
capturing observed trends in precipitation and surface temperature, particularly the 
former, which the model does quite well.