Blanc SIMI 1 - Sciences de l'information, de la matière et de l'ingénierie : Mathématiques et interactions

Multisupport conditional simulation of max-stable processes. Applications to the local prediction of extreme climatic events. – Mc Sim

Submission summary

Extreme events are a key manifestation of complex systems, in both the natural and human world. By definition, they are rare and unexpected. Extreme modeling for stationary time series is well-established, and a major effort is now underway to develop models and methods for dealing with more complex data that appear in modern applications, where spatio-temporal problems and non-stationarity are the rule rather than the exception. To predict extreme events corresponding to different scenarios, as well as to investigate the significance of a possible climatic change, numerical tools such as climate models are already available at both continental and regional scales, but are insufficient. The spatial resolution of global climate models (GCMs) is too coarse to provide a good local description of the processes that are of interest for human activities and the environment.
The question addressed here is how to predict extreme events at a local scale using information from different scales? Typical information includes the GCM data that can be treated as a trend, and physical measurements (e.g. data from rain-gauges) that just act as local calibration. Assuming a stochastic spatio-temporal modeling for the underlying process, a possible way is to generate conditional simulations that are realizations of the model that honor the data. This approach presents a number of advantages, namely

(i) The upscaling from physical measurements or the downscaling from GCM data are performed in compliance with the selected model. Accordingly, the compatibility relationships between distributions at different scales are automatically satisfied, whatever the scaling operation, average or maximum, considered.

(ii) Conditional simulations can be combined to derive a predictor of any quantity of interest (quantiles, probabilities of exceedance, return levels). Moreover, each predictor may be assigned a precision, for instance in terms of confidence bounds.

The “Mc Sim” project deals with the prediction of extreme climatic events starting from multisupport data, using conditional simulations. It implicitly has to cope with a number of challenges regarding the selection of a model, its statistical inference, the design of a conditional simulation algorithm, the prediction of extremes from simulated values.
This project proposes a new and original methodology for predicting spatio-temporal extremes. Special attention will be paid to the family of max-stable models (e.g. storm process, M4 process, Brown–Resnick model).The developed methodology will be applied to the prediction of extreme precipitations in the French region of Languedoc. Despite this narrow field of application (convective precipitations), it is clear that the proposed methodology has a much broader scope, and can be applied to any field of applications where spatio-temporal extreme phenomena may be encountered at various scales.
To make it available to a large audience, it is intended that an R-package will be written to summarize the work achieved on max-stable processes in the domains of the conditional simulation and the upscaling-downscaling.

Project coordination

BACRO Jean-Noel (CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE - DELEGATION REGIONALE LANGUEDOC-ROUSSILLON) – bacro@math.univ-montp2.fr

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.

Partner

LSCE CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE - DELEGATION REGIONALE ILE-DE-FRANCE SECTEUR SUD
ARMINES ASSOCIATION POUR LA RECHERCHE ET LE DEVELOPPEMENT DES METHODES ET PROCESSUS INDUSTRIELS (ARMINES)
UM2 CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE - DELEGATION REGIONALE LANGUEDOC-ROUSSILLON
INRA INSTITUT NATIONAL DE LA RECHERCHE AGRONOMIQUE - CENTRE VERSAILLES GRIGNON

Help of the ANR 140,000 euros
Beginning and duration of the scientific project: - 48 Months

Useful links

Explorez notre base de projets financés

 

 

ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter