Given very large volumes of remote sensing data and climate model output,
one would like to be able to compare them in order to understand where,
when and why model predictions do not agree with observations. Due to
the large volumes, and to incongruities between instrument observation
techniques and models, the traditional approach is to reduce
both data sources by averaging important parameters up to coarse,
common resolution. This destroys information about high-resolution
dependencies among parameters, which are often important sources
of model-data discrepancies. Instead, we replace the means with
nonparametric multivariate distribution estimates of multiple quantities
of interest. We then perform statistical hypothesis tests to determine
whether distributions produced from model output agree with those for the
same coarse grid cell obtained from observations. If differences exist,
we can isolate them with another suite of hypothesis tests that identify
the distributional characteristics causing the problems. In this talk,
we report on work to assess and diagnose the Geophysical Fluid Dynamics
Laboratory's AM2 atmospheric model.
Meet the speaker in Room 212 Cockins Hall at 4:30
p.m. Refreshments will be served.