Speaker: Paula Gonzalez, Laboratory for Climate and Environmental Studies (LSCE) of the Pierre-Simon Laplace Institute (IPSL)

Title: Forecasting yearly record probabilities from historical measurements

Abstract: In climatology, a record refers to the highest (or lowest) value observed for a climate variable, such as temperature, precipitation, at a specific location or within a specific region over a given period. While the probability of observing such records should mathematically decrease in a stationary climate, temperature records are frequently broken nowadays. In this context, one can naturally wonder if it is now possible to assess how quickly the strong global warming signal affect records at a given local scale, i.e. investigate the rate of local record changes from year to year. Practically, stakeholders like a state official may be interested in knowing what are the chances of breaking a record next year with respect to past observations from a specific region of interest. To estimate such probabilities from only historical weather recordings, we leverage advanced statistical techniques based on record theory and multivariate extreme value theory, which helps us to model dependencies among observed annual block maxima time series with some strong spatial correlation. Non-stationarity in time is handled by a kernel smoothing approach. We illustrate our approach by analyzing yearly maxima of daily maximum temperatures from 15 meteorological stations in France spanning the period 1960-2022. Three regions with different climatological features are investigated. In those three regions, probabilities of record decrease at a much slower rate than what is expected under stationary conditions with an independent time-series at each station. The estimated rate of changes brings relevant information for local probability risk assessments about unprecedented events.