Evaluating decadal predictions of northern hemispheric cyclone frequencies

Mid-latitudinal cyclones are a key factor for understanding regional anomalies in primary meteorological parameters such as temperature or precipitation. Consequently, it is vital for a decadal prediction system to adequately represent extra-tropical cyclones. Analyses of prediction skill with respect to cyclone frequency will thus contribute to the understanding of the model's performance regarding the related quantities. Additionally, extreme cyclones can cause significant impacts on human society and economy, due to, for example, enormous economic losses through wind damage. Hence, prediction skill on decadal time scales is of great socio-economic value.

Within MiKlip, 41 retrospective forecasts, initialised annually on 1st January for the period 1961-2001 were analysed within the project EnsDiVal, This study evaluates the ability of the first two development stages of the MiKlip decadal forecast system to provide skillful forecasts for winter (October to March) extra-tropical cyclone frequencies over the Northern Hemisphere with lead times from 1yr up to a decade. Each forecast is set up as an ensemble forecast that is consisting of several model simulations (ensemble members). The members start from slightly different initial conditions to account for uncertainty. EnsDiVal assesses the probabilities of three discrete categories for cyclone frequency (below normal, normal, above normal) which are derived from relative frequencies of ensemble members forecasting the respective category.

It is shown that these probabilistic predictions are skillful, that is significantly better than forecasting climatological probabilities for each category and forecast, mainly for lead times of 2-5yr, especially over the North Atlantic and Pacific. Additionally, these initialised forecasts are also significantly better than the results of uninitialised transient model simulations, containing potential changes (and hence skill) arising from long-term trends of greenhouse-gas and aerosol concentrations. Skill for intense cyclones (especially relevant for impact assessment) is generally higher than the skill for all detected systems.

A comparison of the two development stages, characterised by different initialisation techniques, indicates that initialising directly from reanalysis data yields slightly better results for the first forecast winter; initialisation based on an assimilation experiment provides instead better skill for lead times between 2 and 5yrs.

The mechanisms behind this predictive skill are subject to future work. Preliminary analyses suggest a strong relationship between the model’s skill over the North Atlantic and the ability to predict upper ocean temperatures. The latter modulate temperature differences in the lower troposphere for the respective area and time scales. These temperature differences between subtropical and polar latitudes are one main ingredient for enhanced or decreased occurrence of extra-tropical cyclones.

Back to List