CISL Brown Bag Seminar: "On the Sensitivity of Bias Corrected Climate Projections to the Choice of Observational Products and Statistical Methods"

08/08/2018 - 12:00pm
 "On the Sensitivity of Bias Corrected Climate Projections to the Choice of Observational Products and Statistical Methods"
 
Keith W. Dixon
NOAA/Geophysical Fluid Dynamics Laboratory
Empirical Statistical Downscaling Team
 
Multi-decadal climate projections produced by CMIP-class dynamical models often are deemed to be unsuitable for direct use in climate impacts applications, as they may contain unacceptably large biases and/or lack the desired spatial resolution. To address dynamical model shortcomings, bias correction and statistical downscaling techniques are often employed, serving as a “value-added” processing step. Yet, various statistical processing methods each yield somewhat different results, revealing uncertainties that exist in the bias correction process – a factor that can be underappreciated by and/or confusing to some users, thus raising the question, “How correct are the corrections applied to future climate projections?”
 
NOAA GFDL’s ESD Team and collaborators have designed, conducted, and analyzed sets of experiments aimed at evaluating the performance characteristics of some bias correction methods used to statistically downscale climate model projections. A perfect model design allows one to determine to what extent some techniques’ performances may degrade when applied to 21st century projections, relative to performance during a historical training period. We find several methods that generally perform well have notable weaknesses that are exposed under certain climate conditions for some variables and locations. Also, we illustrate how the choice of observation-based precipitation product used in training can affect some evaluation metrics and not others. Overall, results support the general advice that informed matching of statistically processed climate projections to climate impacts applications benefits from an understanding of an application’s data requirements and sensitivities as well as an understanding of strengths and weaknesses of candidate bias correction and statistical downscaling methods.
 
Wednesday, August 8, 2018
12:00 p.m. - 1:00 p.m.
Mesa Lab, Chapman Room
(Bring your lunch!)