1 July 2010
Our new IBM supercomputer has enabled our scientists to implement a number of major advances in forecasting science. However, as Professor Brian Golding, Head of Forecasting Research & Development explains, it has also presented a major challenge to maintain the quality and integrity of our existing capabilities.
Preparations for the migration of existing code were set in motion as soon as the contract was awarded to IBM. The software developers in IBM and our own code owners kept in very close contact to ensure that the migrated code had the necessary reliability, accuracy and speed to take over the operational tasks. At the same time, scientists and code owners have been working together to bring many years of scientific research to fruition in new versions of our forecasting systems so that the expected performance gains could be achieved at the earliest opportunity.
Our new flagship capability is a 1.5 km mesh size model for the UK, which for the first time provides routine forecasts of typical sized thunderstorms. The science behind this capability has been under development for almost a decade by a dedicated team of scientists under Professor Peter Clark at the Met Office Joint Centre for Mesoscale Meteorology based at the University of Reading. It is a world-leading capability and has already
provided impressive demonstrations of its capacity to give realistic rainfall forecasts. For instance, it forecast a peak of 269 mm of rain over the Cumbrian mountains during the November floods (see page 5), substantially closer to observations than the operational guidance available from existing models.
The second major advance was to increase the number of atmospheric layers in several of our forecasting models from 38 to 70. It is difficult to make changes to the representation of the vertical structure of the atmosphere because they impact on many aspects of model formulation that interact in complex ways, and have to be adjusted for consistency. The benefits of this advance have been most dramatic in our short-range regional models
- both for the UK and elsewhere. This has shown some spectacular improvements in the accuracy of cloud, and therefore of surface-temperature, forecasts during some of the cold, dry spells of the past winter. The source of these improvements comes from better resolution of the shallow layer of the atmosphere about 1 kilometre above the ground, which marks the normal upper limit of direct mixing of air from the surface.
The last of the major upgrades was to refine the spatial mesh size of the global model from about 40x40 km to about 25x25 km at UK latitudes. This enables the model to represent the influence of small weather systems on future weather anywhere in the world; particularly energetic weather systems like hurricanes. This has lead to significant improvements in the accuracy of UK forecasts up to several days ahead, continuing the steady rate of increase in predictability of the weather, and enabling the Met Office to hold its leading position amongst global forecasting centres.
- See a profile on Professor Brian Golding, Head of Forecasting R&D.