Statistical Methods in Climatology
Speaker: Michael Stein, University of Chicago
Location: Warren Weaver Hall 1302
Date: Monday, February 29, 2016, 3:45 p.m.
Climate data (observational or model output) often exhibit space-time structures that are effectively impossible to capture in statistical models. For example, even a basic quantity like surface temperature can exhibit complex non-stationarities in space and time due to geography, seasonality and diurnal cycles, as well as clear non-Gaussian features due to passing weather fronts and anomalies that occur near the freezing point of water. Statistical modeling of space-time precipitation fields is even more challenging because of the preponderonce of zeroes when looking at precipitation at fine time scales. These difficulties raise some fundamental questions, including:
- What kinds of problems can statistical approaches address that do not require complete modeling of the space-time structures of processes?
- How can we evaluate how well statistical procedures work when we are unable to specify even roughly realistic models for capturing all the relevant modes of space-time variation?
I will discuss these general issues through a series of applications in climatology, including simulation of future temperature and precipitation processes at daily and finer time scales and inferences about climatological extremes.