Abstract ID: 034
Domino: A framework for improving extreme event predictability using flow precursors
Lead Author: Joshua Dorrington
Karlsruhe Institute of Technology, Germany
Keywords: Extreme events, Mid-latitude dynamics, statistical-dynamical prediction
Abstract:
Current numerical weather prediction models show very little practical skill for predicting extreme precipitation events in the S2S range. This is, in part, due to the small spatial scales of extreme precipitation, influenced by many sub-grid scale parameterised processes. However, extreme precipitation events are by no means decoupled from the large-scale flow: the prevailing winds and wave-structures ‘set the scene’ for amplified or suppressed risk of an extreme, by controlling moisture availability and vertical stability. This is especially true in the mid-latitudes where frontal rainfall is dominant. The prospect of usefully predicting these large-scale flow configurations in the S2S range is more promising. A number of studies have investigated the flow patterns pre-cursing extreme precipitation in different geographical areas, using both process-oriented and regime-based approaches. These potentially predictable large-scale precursors may be used to statistically infer the probability of smaller-scale extremes; essentially mapping out a probabilistic ‘domino chain’ of processes which may cause a local extreme event. However, such precursor-based insights are not yet easily extensible to other regions, or available to the operational meteorologist.
In this talk we will present a framework for identifying the precursor patterns that tend to precede extreme events, how to reduce these patterns to skilful scalar indices that capture the activity of the precursors, and how to combine these indices with direct numerical weather model predictions to produce a more skilful hybrid forecast. Implementation and extension of this framework is supported by a new open-access Python package, Domino, which reduces such analyses to only a few lines of code.
Specifically we will consider the predictability of European regional daily rainfall extremes at lead times of several weeks. We will discuss how using information about large-scale precursors, in combination with direct model output, allows extra skill to be extracted from our existing models, and allows us to make the most of the high dimensionality and high data volumes produced by modern forecast systems. We will discuss the potential of this flexible framework to be applied to other geographical regions, different kinds of extreme events and to different time-scales, and plans for a semi-operational implementation of this approach using the IFS forecast model.
Co-authors:
Federico Grazzini (ARPAE- Emilia-Romagna)
Christian Grams (Karlsruhe Institute of Technology)
Linus Magnusson (ECMWF)
Frederic Vitart (ECMWF)
Laura Ferranti (ECMWF)
