Wednesday, 10 January 2018: 3:15 PM
Room 19AB (ACC) (Austin, Texas)
General circulation models (GCMs) are often judged by metrics that are tied to how well climatological means, spatial variations and seasonal cycles of various parameters are simulated. However, such evaluations are made against imperfect observational datasets. A large component of uncertainty stems from observing system limitations (e.g. radiation/wavelength spectra utilized), retrieval algorithm limitations in differing environments, and sampling limitations (e.g. sun- vs. non-sun synchronous satellites). A number of studies have found that uncertainties arising from these limitations may approach 50% in many regions for a number of widely-used hydrological parameter products. Pervasive biases of this magnitude, if formally incorporated into model performance cost functions, may substantially alter the perceived skill of a GCM, and further influence the model developer’s choice of GCM parameterization tuning and free parameter combinations.
Using the GISS GCM as a test-bed, preliminary results show that when large uncertainty estimates are accounted for in a performance metric, the GCM free parameter state space comprising “good” or admissible GCM configurations substantially changes. Wide variations in parameter combinations are found, and are further associated with widely varying regional biases in radiation, despite similar “goodness” metrics (e.g. Taylor Scores) and TOA radiative balance. Varying biases in radiation fields, suggesting variations in the extent to which the atmosphere is “conflicted”, may suggest further impacts on simulated climate sensitivity. We discuss our methodology for accounting for systematic biases in observational datasets within the context of “tuning” and evaluating a GCM, and discuss the impact this has on improving regional radiation biases in a GCM.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner