Project: Homogeneous and non-homogeneous Poisson process models for extremes in climate change studies

The interdisciplinary project is motivated by the recent advances in the field of statistical modelling of extremes and their applications in the non-stationary context of "climate change". The previous studies involved many ad hoc settings and/or decisions concerning the methodology, which need to be justified, and have not been adapted for time series models. The aims of this project are (i) to develop statistical models of extremes based on both homogeneous (with constant intensity) and non-homogeneous (with intensity dependent on covariates) Poisson process approaches combined with the "peaks-over-threshold" method that utilizes regression quantiles; (ii) to develop methods based on marked point process for time series models; (iii) to apply these models to estimating probabilities of extremes in the observed data and climate model simulations; (iv) to compare the performance of the homogeneous and non-homogeneous Poisson process approach; and (v) to compare the maximum likelihood and Bayesian techniques for the estimation of the models" parameters. The project makes use of recently released transient simulations by climate models, in particular regional climate models over Europe.

Funded by: GA ČR. GAP209/10/2045

Duration: 2010-2013

Investigator: Kyselý, J.

Tým: Rulfová, Z.