IMAGe Brown Bag- Extremes in Regional Climate: What to do with 8,000 Histograms?

04/08/2016 - 12:00pm to 1:00pm
ML- Chapman Room

Doug Nychka


Friday, April 8, 2016
12:00-1:00 pm
Mesa Lab, Chapman Room
(Bring your lunch)

As attention shifts from broad global summaries of climate change to more specific regional impacts there is a need for the data sciences to quantify the uncertainty in regional predictions. A regional climate model (RCM) is a large and complex computer code based on physics that simulates the detailed flow of the atmosphere in a particular region from the large scale information of a global climate model. Part of the value of these simulations is to explore the potential extremes in weather that are due to natural variation and also to climate change. Here we present an application that combines logspline density estimates to discern tail behavior in a distribution with spatial methods for large data sets (LatticeKrig). This is applied to estimate return levels for daily precipitation from a subset of the North American Regional Climate Change and Assessment Program. Here the regional models comprise about 8000 grid locations over North America and so pose challenges for the statistical analysis of functional data. Besides efficient algorithms this application also explores using embarrassing parallel steps using the Rmpi package.