Abstract
This paper improves an extreme-value-theory-based detection and attribution method and then applies it to four types of extreme temperatures, annual minimum daily minimum (TNn) and maximum (TXn) and annual maximum daily minimum (TNx) and maximum (TXx), using the HadEX2 observation and the CMIP5 multimodel simulation datasets of the period 1951–2010 at 17 subcontinent regions. The methodology is an analog of the fingerprinting method adapted to extremes using the generalized extreme value (GEV) distribution. The signals are estimated as the time-dependent location parameters of GEV distributions fitted to extremes simulated by multimodel ensembles under anthropogenic (ANT), natural (NAT), or combined anthropogenic and natural (ALL) external forcings. The observed extremes are modeled by GEV distributions whose location parameters incorporate the signals as covariates. A coordinate descent algorithm improves both computational efficiency and accuracy in comparison to the existing method, facilitating detection of multiple signals simultaneously. An overall goodness-of-fit test was performed at the regional level. The ANT signal was separated from the NAT signal in four to six regions. In these analyses, the waiting times of the 1951–55 20-yr return level in the 2006–10 climate for the temperature of the coldest night and day were found to have increased to over 20 yr; the corresponding waiting times for the warmest night and day were found to have dropped below 20 yr in a majority of the regions.
Denotes content that is immediately available upon publication as open access.
Supplemental information related to this paper is available at the Journals Online website: http://dx.doi.org/10.1175/JCLI-D-15-0835.s1.
© 2017 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).
