Downscaled precipitation

BCSD and BCCA downscaled precipitation over the upper Mississippi River basin

In support of hydrological modeling (using SWAT) and agronomic modeling we have looked at precipitation in the BCSD and BCCA downscaled output from MIROC 3.2 over the upper Mississippi River basin over the period 1981-1999. It appears that BCCA precipitation is unrealistically low over much of the region. The following plots (prepared by Daryl Herzmann) show what we have found.

We chose MIROC 3.2 because it performs well for ENSO compared with other CMIP3 models. We have checked output from downscaled GFDL 2.1 and find that its precipitation also is too low in the BCCA downscaled output.

Raymond Arritt, Iowa State University, email rwarritt@bruce.agron.iastate.edu

Left: gridded 2 degree observations (from downscaling site). Right: PRISM climatological precipitation for 1961-1990. As expected, these two agree well. This was simply a check that the 1981-1999 period was reasonably representative of climatology.

Left: MIROC 3.2 raw precipitation. Right: MIROC 3.2 bias-corrected precipitation on the 2 degree grid. The bias correction procedure is giving realistic results.

Left: Downscaled MIROC 3.2 BCCA on the 1/8 degree grid. Right: Downscaled MIROC 3.2 BCSD. Notice that BCCA is drier than BCSD (and the observed climatology) throughout the domain.

Taking the difference between the downscaled fields (BCSD minus BCCA) shows that BCCA is lower by about 6 inches over most of the upper Mississippi River basin. There are some interesting geographical variations but no obvious trend with respect to climatological mean amount, other than very dry locations in the western U.S. having smaller differences. For example the difference is nearly as large in Nebraska as in Ohio despite observed mean precipitation in Ohio being around 50-80 percent greater.

Additional results

Following are plots for the entire continental U.S. instead of just the Mississippi River basin. A sampling of models is shown from both the CMIP3 and CMIP5 ensembles. These models were chosen in part to include members from the same "family":

GFDL CM2.0 and CM2.1 from CMIP3 along with GFDL CM3 from CMIP5;
MIROC 3.2 medres from CMIP3 along with MIROC 5 from CMIP5;
ECHAM5 from CMIP3 along with MPI ESM-LR and ESM-MR from CMIP5. (Note ESM-LR and ESM-MR have the same T63 horizontal resolution in the atmosphere. The main differences are that ESM-MR has 95 atmospheric layers versus 47 in ESM-LR, and finer horizontal resolution in the ocean.)