Abstract
We outline a framework for identifying the geographical sources of biases in climate models. By forcing the model with time-averaged short-term analysis increments [tendency bias corrections (TBCs)] over well-defined regions, we can quantify how the associated reduced tendency errors in these regions manifest themselves both locally and remotely through large-scale teleconnections. Companion experiments in which the model is fully corrected [constrained to remain close to the analysis at each time step, termed replay (RPL)] in the various regions provide an upper bound to the local and remote TBC impacts. An example is given based on MERRA-2 and the NASA/GMAO GEOS AGCM used to generate MERRA-2. The results highlight the ability of the approach to isolate the geographical sources of some of the long-standing boreal summer biases of the GEOS model, including a stunted North Pacific summer jet, a dry bias in the U.S. Great Plains, and a warm bias over most of the Northern Hemisphere land. In particular, we show that the TBC over a region that encompasses Tibet has by far the largest impact (compared with all other regions) on the NH summer jets and related variables, leading to significant improvements in the simulation of North American temperature and, to a lesser degree, precipitation. It is further shown that the results of the regional TBC experiments are for the most part linear in the summer hemisphere, allowing a robust interpretation of the results.
© 2019 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).
