Predictability of extreme Mediterranean cyclones in past and current ECMWF models
Open access
Author
Date
2023-04-03Type
- Master Thesis
ETH Bibliography
yes
Altmetrics
Abstract
Cyclones are one of the main factors that drive the variability of Mediterranean weather. If they reach high intensities, they pose a major environmental risk to the densely populated coasts along the sea. Although progress has been made in modelling the dynamics and physical processes relevant for the formation of Mediterranean cyclones, the prediction of extreme Mediterranean cyclones still remains a challenge.
While forecast performance is often verified in a climatological context based on years of forecast data, this thesis takes a novel approach and aims to understand how model developments actually affect forecast performance for individual extreme weather events. To this end, we compare three different forecast types from the ECMWF: deterministic forecasts (higher resolution), ensemble fore- casts (lower resolution), and ensemble hindcasts (more modern model versions). On the one hand, this allows us to study the relation between forecast accuracy and model version, and on the other hand, this shows how well a historic event would have been predicted with more recent model ver- sions. In addition, we compare different initialisation times to analyse how forecast accuracy depends on forecast lead time. Our evaluation is based on mean sea level pressure, horizontal wind speed, and precipitation. We focus on both the spatial distribution of forecast errors, and the accuracy of spatially and/or temporally averaged quantities.
We find three main errors for forecasts at long lead times: an uncertain position of the cyclone, a timing error of the mature stage (at least 0.5 d), and an underestimated cyclone intensity (winds too weak, precipitation too scarce). These uncertainties become smaller with decreasing lead time, and we notice a distinct jump in forecast accuracy at lead times between 4.5 d and 5 d before the mature stage of the cyclone. However, even the short lead time forecasts do not predict strong winds and heavy precipitation with sufficient accuracy to infer the societal impact of the Mediterranean cyclones.
Overall, the ensemble hindcasts are the most accurate forecasts, followed by the deterministic forecasts, and the ensemble forecasts. The advantages of the hindcasts are smaller positional uncer- tainties, and better accuracy at longer lead times (extended skill horizon). Since the hindcasts are based on more recent model versions, this highlights the benefits of the model improvements over time. Still, the forecast accuracy is very variable and each event shows unique error patterns. This reflects the large case-to-case variability of extreme Mediterranean cyclones. Our case study approach indicates potential sources of forecast errors, but specifically designed experiments would be needed in future research, to obtain a detailed analysis of forecast uncertainties. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000608732Publication status
publishedPublisher
ETH ZurichSubject
Meteorology; Weather prediction; extratropical cyclonesOrganisational unit
03854 - Wernli, Johann Heinrich / Wernli, Johann Heinrich
More
Show all metadata
ETH Bibliography
yes
Altmetrics