Programme  OS3f Regional and nationwide scenarios 2  abstract 122

Regional tests for trend detection in maximum pluviometric series in the French Mediterranean region.

Author(s): Nicolas Pujol(1), Luc Neppel(1), Robert Sabatier(2)
NICOLAS PUJOL (1), LUC NEPPEL (1), ROBERT SABATIER(2) (1) Hydrosciences, Maison des Sciences de l’Eau, cc 57 Université Montpellier 2 – Place Eugène Bataillon 34095 Montpellier cedex 5, France e-mail : ; Tel :

Keyword(s): climatic change, Mann -Kendall’s test, Generalized Extreme Value distribution, maximum likelihood method, regional test for trend detection

Article: abs122_article.pdf
Get Adobe Reader

Session: OS3f Regional and nationwide scenarios 2
During the last years, French Mediterranean regions have been affected by

extremely violent precipitations, in particular the ones of November 1999, located in Aude, of September 2002 in

Gard. With the current prospect of a climatic change, these recurrent inundations are regularly ascribed to an

evolution of the rainy regime. In this paper, we are going to focus on extreme precipitations of the Mediterranean


Our purpose is to test the stationary hypothesis of extreme daily precipitations in the

Mediterranean region. However, a local approach of trend detection is limited. In fact, if 5 non stationary posts

among 100 are detected by a local test, with a 5% risk, what can be concluded? Are they due to chance or to true

changes? To cope with this problem, we propose to estimate the regional significance of locally noticed change, and

then, to check if they are indicative of a real climatic change or not.

The Mann-Kendall non

parametric test (denoted by MK) is locally used for each post, and a regional statistics is built. Furthermore, maxima

are generally distributed as a Generalized Extreme Value distribution. The maximum likelihood method (denoted by

ML) is built on this hypothesis. A regional interpretation of this test is also built. Both MK and ML regional statistics

are built by bootstrap. This re-sampling method allows to respect the initial cross-correlation in the


It is interesting to notice that, whereas they are no common points, the MK and ML tests

provide very similar results, locally as well as regionally. Moreover, we have seen how liberal the regional MK and

ML tests become when special correlation is ignored. In Languedoc Roussillon, we mainly observe, in March, a

regionally significant fall of the maxima and the sums. Simultaneously, a regionally significant increase of the maxima

and the sums has been observed in April. Moreover, we have observed, in the same way as previously, a regionally

significant fall of the number of days exceeding the threshold 20mm in March, and a significant increase in April.

Besides, we have applied this test to the data constituted of the two monthly sums of March and April: no trend has

been identified. Consequently, we can think that the most severe rainfall events in March have shifted today in April.

This gives substance to the idea of a seasonal gap in Languedoc Roussillon. Finally, a regional increasing trend of

annual maxima has also been detected in Cevennes.

Firstly, the implementation of these

regional statistical procedures requires a climatic zoning of the studied domain, homogeneous with regard to the

variables of interest. Secondly, it is necessary to properly consider the effects of cross-correlation in the data when

building a regional test. Finally, the causes of the detected unstationarities should be analysed: are there any changes

in meteorological factors like the surface pressure fields or the NOAA index that might explain the changes in rainfall

regime? Investigations of this question could be a future development of this first work.

  Return up