Data informatics for the Detection, Characterization, and Attribution of Climate Extremes

Abstract:

The potential for increasing frequency and intensity of extreme
phenomena including downpours, heat waves, and tropical cyclones
constitutes one of the primary risks of climate change for society and
the environment. The challenge of characterizing these risks is that
extremes represent the "tails" of distributions of atmospheric
phenomena and are, by definition, highly localized and typically
relatively transient. Therefore very large volumes of observational
data and projections of future climate are required to quantify their
properties in a robust manner. Massive data analytics are required in
order to detect individual extremes, accumulate statistics on their
properties, quantify how these statistics are changing with time, and
attribute the effects of anthropogenic global warming on these
statistics.

We describe examples of the suite of techniques the climate community
is developing to address these analytical challenges. The techniques
include massively parallel methods for detecting and tracking
atmospheric rivers and cyclones; data-intensive extensions to
generalized extreme value theory to summarize the properties of
extremes; and multi-model ensembles of hindcasts to quantify the
attributable risk of anthropogenic influence on individual extremes.
We conclude by highlighting examples of these methods developed by our
CASCADE (Calibrated and Systematic Characterization, Attribution, and
Detection of Extremes) project.