92nd American Meteorological Society Annual Meeting (January 22-26, 2012)

Tuesday, 24 January 2012: 2:45 PM
Need for Caution in Interpreting Extreme Weather Statistics
Room 238 (New Orleans Convention Center )
Prashant D. Sardeshmukh, Univ. of Colorado/CIRES/CDC and NOAA/ESRL, Boulder, CO; and G. P. Compo and C. Penland

Given the substantial anthropogenic contribution to 20th century global warming, it is tempting to seek an anthropogenic component in any unusual recent weather event, or more generally in any observed change in the statistics of extreme weather. This study cautions that such detection and attribution efforts may, however, very likely lead to wrong conclusions if the non-Gaussian aspects of the probability distributions of observed daily atmospheric variations, especially their skewness and heavy tails, are not explicitly taken into account. Departures of three or more standard deviations from the mean, although rare, are far more common in such a non-Gaussian world than they are in a Gaussian world. This exacerbates the already difficult problem of establishing the significance of changes in extreme value probabilities from historical climate records of limited length using either raw histograms or Generalized Extreme Value (GEV) distributions fitted to the sample extreme values.

A possible solution is suggested by the fact that the non-Gaussian aspects of the observed distributions are well captured by a general class of “Stochastically Generated Skewed distributions” (SGS distributions) recently introduced in the meteorological literature by Sardeshmukh and Sura (J. Climate 2009). These distributions arise from simple modifications to a red noise process and reduce to Gaussian distributions under appropriate limits. As such, they represent perhaps the simplest physically based non-Gaussian prototypes of the distributions of daily atmospheric variations. Fitting such SGS distributions to all (not just the extreme) values in 25, 50, or 100-yr daily records also yields corresponding extreme value distributions that are much less prone to sampling uncertainty than GEV distributions. For both of the above reasons, SGS distributions provide an attractive alternative for assessing the significance of changes in extreme weather statistics (including changes in the statistics of extreme precipitation events) over the 20th century, and of the changes projected over the 21st century.

Supplementary URL: