On 18 October 2010 Typhoon Megi made landfall in the Philippines. The central pressure of 885hPa was the lowest seen for 20 years and one of the lowest ever recorded at landfall.
Accurate forecasting of tropical cyclones is vital in providing information to decision makers and emergency services so that lives can be saved and property protected. While current numerical models are able to provide useful guidance (e.g. as displayed in StormTracker and used by the National Hurricane Centre) our scientists continue to assess and improve these models. Recent work at the Met Office has used typhoon Megi as a case study to assess how well our current models perform and to explore the possibilities of how higher resolution models might improve predictions.
Higher resolution simulations are better able to resolve small scale features of the atmospheric flow and so rely less on approximations of the subgrid physics. However, with greater accuracy comes a greater computational cost; these simulations cannot yet be run in real time and so remain a research tool until made affordable by the next generation of supercomputers. Among operational configurations of the Unified Model(UM) there are currently models running global forecasts using a 25km grid, models covering Europe and the North Atlantic with a 12km grid and models running over the UK using 4km and 1.5km grids (see current Numerical Weather Prediction model configurations). These fine resolution models get close to resolving some of the convective scale motions; a very important aspect in tropical cyclone development and evolution.
With these grid resolutions in mind, a series of hindcasts were performed. A nesting procedure was used such that a 12km model over a limited area of 4800 x 2400km was forced at its boundaries by the global 25km forecast, with the 1.5km and 4km models both driven at their boundaries by the 12km model. Each of the finer resolution models spanned an area of 3000 x 1500km. The forecast was initialised approximately 5 days before landfall and each model domain positioned to capture the expected path of the typhoon.
The image above shows a snapshot of the typhoon after it has developed and travels westward to make landfall. This image is from the 1.5 km model and displays the simulated outgoing longwave radiation (OLR). By clicking on the enlarge button, an animation can be seen comparing the simulated OLR with infra-red satellite imagery from the MTSAT-2 satellite. The comparison is quite extraordinary and to the untrained eye it is difficult to tell which of the panels contains the simulated data and which the observed. (The top image shows the model data) 10m wind speed for simulations of Typhoon Megi Central pressure for simulations of Typhoon Megi
After an initial period during which the model simulation "spins-up" smaller scale features, the cloud fields look very realistic. As the typhoon increases in strength an "eye" can be seen to form at the centre of the storm and, just as in reality, it fills in again once the typhoon makes landfall.
The effects of resolution on the accuracy of the simulation can be clearly seen by looking at statistics of the central pressure and peak 10m wind speeds of the storm. The images on the right show how these two metrics vary as the model resolution changes. The black line in these plots shows the estimated values derived from satellite imagery. Despite coarser models providing useful guidance to the track of the storm, there appears to be very little skill in the prediction of these measures. Although some of the stronger winds start to be resolved in the 4km model, it isn't until we get right down to 1.5km that the model values really start to look similar to those from the observations.
Work continues with high resolution modelling as a research tool; simulations of Megi can be run at even higher resolution (100m) to capture even more detail and investigate the eye-wall structure and detailed processes within the storm.
Despite the computational expense of the high resolution runs, we are not far away from seeing these sorts of model simulation coming into operational forecasts. For example, the Australian Bureau of Meteorology (a UM collaboration partner) will soon be looking to run operational forecasts of Tropical Cyclone formation and development using a 4km grid.