New operational forecasts of tropical storm activity using the Met Office dynamical seasonal prediction model

In June 2007 the Met Office issued, for the first time, a seasonal prediction of July to November North Atlantic tropical storm activity, using output from its dynamical coupled ocean-atmosphere global seasonal prediction system (GloSea). Use of a dynamical prediction model distinguished this forecast from most other published/operational forecasts for the Atlantic sector, which are based on statistical-empirical methods. The forecast predicted between 7 and 13 tropical storms for the Atlantic sector in the period July to November, comparing with 12 observed to date.

Motivation in starting this new operational forecast comes from recent research with European partners showing that, for tropical storm frequency, prediction skill with dynamical systems is now challenging/overtaking that of some well-known statistical-empirical methods (Vitart et al. 2007). Advantages for the dynamical models are seen both in skill scores calculated over multi-year re-forecasts, and also in recent real-time forecasts. For example, the dynamical models correctly distinguished between the exceptionally active North Atlantic season of 2005 and the near-normal activity of the 2006 season. This marked difference between seasons was missed by a number of well-known statistical prediction methods.

The horizontal resolution of the GloSea model is not sufficient to simulate the intensity and detailed structure of tropical storms. However, the larger-scale features are sufficiently realistic to allow counting of predicted storms using a parametric approach (Vitart et al., 2003). For example, tropical storms in the GloSea model exhibit a warm temperature anomaly above the centre of the vortex – a characteristic of observed storms. A calibration procedure is applied, which adjusts the number of detected storms according to past model performance. The GloSea system is run in a forecast ensemble of 41 members, each initialised with slightly varying ocean analyses to represent uncertainty in the initial state. This allows generation of a probability distribution for the predicted number of storms that reflects uncertainty in the forecast process.

A large proportion of the inter-annual variability of North Atlantic tropical storms is associated with sea-surface temperature (SST) variability in the tropical Pacific Ocean (associated with the ENSO cycle) and the tropical North Atlantic. The success of the dynamical model forecasts is rooted in good prediction skill for SST in these regions, and also in an ability to correctly translate predicted SST variability into variability of tropical storm frequency - through representation of ENSO impacts on Atlantic vertical wind shear and through local SST impacts.

The GloSea prediction system and the methodology used to track tropical storms will be briefly introduced. The skill of dynamical prediction systems will be compared with that of statistical methods, focussing on predictions for the North Atlantic sector. Examples of more detailed probabilistic forecast products based on information from the GloSea forecast ensemble will be discussed.

Vitart, F., Anderson, D. and Stockdale, T. 2003: Seasonal forecasting of tropical cyclone landfall over Mozambique. J. Climate, 16, 3932-3945.

Vitart, F., Huddleston, M.R., Déqué, M., Peake, D., Palmer, T.N., Stockdale, T.N., Davey, M.K., Ineson, S. and Weisheimer, A. 2007: Dynamically-based seasonal forecasts of Atlantic tropical storm activity issued in June by EUROSIP. Geophys. Res. Lett. 34, L16815.