What is it about?
A spatial extreme-value statistical model, now called the propinquity model, is introduced that employs the Heffernan and Tawn conditional extreme-value (HT) model. A single time-series, say q(t), is derived from the space-time series that measures the amount of energy in the spatial field at each time point. The example in the paper uses the upper quartile over space, but other measures such as the mean could be used instead. Then, conditioned on q(t) > u, for large u, the HT model is fit at each grid point and the joint distribution across space is obtained.
Featured Image
Photo by NOAA on Unsplash
Why is it important?
This modeling approach differs from the usual spatial modeling approach where the joint distribution between pairs of locations is sought. Currently, in the extreme-value modeling context, such models are computationally intensive for even small spatial domains. Moreover, this propinquity model allows for observing large-scale patterns and making inferences about weather phenomena, such as hurricanes, torandic activity, etc. in terms of changes not just in time, but also in space.
Perspectives
I struggled to find an appropriate spatial extreme-value model that could handle the 884 grid points and result in a physically plausible product. In playing with the R package texmex, I was delighted with how quickly the results came out, and that they made physical sense. It is a sort of weird spatial model, but one that makes sense for a plethora of potential applications. I had wanted a different word to describe the method, and only arrived at "propinquity" model much later.
Eric Gilleland
National Center for Atmospheric Research
Read the Original
This page is a summary of: Spatial extreme value analysis to project extremes of large-scale indicators for severe weather, Environmetrics, September 2013, Wiley,
DOI: 10.1002/env.2234.
You can read the full text:
Contributors
The following have contributed to this page
