Evaluating collective significance of climatic trends: A comparison of methods on synthetic data

0474173 - ÚFA 2018 DE eng A - Abstrakt
Huth, Radan - Dubrovský, Martin
Evaluating collective significance of climatic trends: A comparison of methods on synthetic data.
Geophysical Research Abstracts. Göttingen: European Geosciences Union, 2017. EGU2017-4993. ISSN 1607-7962.
[EGU General Assembly 2017. 23.04.2017-28.04.2017, Vienna]
Institucionální podpora: RVO:68378289
Klíčová slova: climatic trends * multi-site stochastic generator
Kód oboru RIV: DG - Vědy o atmosféře, meteorologie
http://meetingorganizer.copernicus.org/EGU2017/EGU2017-4993.pdf

The common approach to determine whether climatic trends are significantly different from zero is to conduct
individual (local) tests at each single site (station or gridpoint). Whether the number of sites where the trends are
significantly non-zero can or cannot occur by random, is almost never evaluated in trend studies. That is, collective
(global) significance of trends is ignored.
We compare three approaches to evaluating collective statistical significance of trends at a network of sites, using
the following statistics: (i) the number of successful local tests (a successful test means here a test in which the
null hypothesis of no trend is rejected); this is a standard way of assessing collective significance in various
applications in atmospheric sciences; (ii) the smallest p-value among the local tests (Walker test); and (iii) the
counts of positive and negative trends regardless of their magnitudes and local significance. The third approach is
a new procedure that we propose; the rationale behind it is that it is reasonable to assume that the prevalence of
one sign of trends at individual sites is indicative of a high confidence in the trend not being zero, regardless of the
(in)significance of individual local trends. A potentially large amount of information contained in trends that are
not locally significant, which are typically deemed irrelevant and neglected, is thus not lost and is retained in the
analysis.
In this contribution we examine the feasibility of the proposed way of significance testing on synthetic data,
produced by a multi-site stochastic generator, and compare it with the two other ways of assessing collective
significance, which are well established now.
Trvalý link: http://hdl.handle.net/11104/0271282