Biases in temperature and precipitation over the CONUS in ensembles of the historical model simulations of CMIP6

Abstract:

The characterization of model biases at regional scales in CMIP6 historical simulations of surface air temperature and precipitation forms the basis for developing future projections. Given the need to downscale models due to their coarse resolution relative to local projection needs and the computational expense of doing so, model bias determination also helps in prioritizing which models and/or model ensemble members to downscale.

We therefore evaluate these biases across the Conterminous United States (CONUS) for 51 models and 367 ensemble members relative to gridded hydrometeorological datasets developed from station observations. The most common feature across models is that they are biased low in precipitation across quantiles for the West Coast and the Gulf Coast. Taylor diagrams of model ensemble members developed from slightly different initial conditions show that differences relative to observations of model ensemble members are nearly indistinguishable from each other. These findings indicate that model structural error, rather than model realizations of internal climate variability, dominate biases over the CONUS, and also indicate that downscaling efforts should focus on different models rather than different ensemble members of the same model.