Met Office Global and Regional Numerical Weather Prediction: Status and Plans

The Met Office is the UK's National Weather Service, providing global/regional weather, oceanic/wave/surge, air quality, atmospheric dispersion and space weather prediction services to a wide range of customers. This talk will focus on the Met Office's underpinning numerical weather prediction (NWP) scientific capabilities that provide the basis for our global and regional weather services from hours out to 1-2 weeks ahead.

For over 25 years, Met Office NWP has been based on the 'Unified Model', used for both traditional weather and climate predictions, and now also used by a growing number of institutes around the world to provide their own weather services via the 'UM Partnership'. This talk will briefly describe the fundamental design of the UM system, including the prediction model, observation pre-processing and data assimilation (DA) systems. The current main operational configurations will be described: global deterministic/ensemble (~10/20km lat-long grid, 70 vertical levels, 80km model top, hybrid 4D variational DA - 4DVar) and UK deterministic/ensemble (1.5/2.2km, 70 vertical levels, 40km top, 4DVar DA). A summary of the increasing number of km-scale NWP regional applications away from the UK will be given. Trends in performance improvements over the past 5-10 years will be described, including the impact of several recent major operational upgrades.

The next few years will see a number of additional upgrades to our NWP capabilities. For global NWP, latest results will be presented from candidate upgrades including coupled ocean-atmosphere-land NWP, the assimilation of all-sky radiances, and a new ensemble prediction system based on the 4DEnVar DA technique. Within our km-scale models, major current foci to be briefly summarised include a) The improvement of day 1 rainfall via the assimilation of radar reflectivity observations in 4DVar, b) Extending the km-scale 4DVar technique to include flow-dependent ensemble-based forecast errors (hybrid 4DVar), c) Modelling improvements to improve key forecast systematic biases e.g. max/min surface temperature, representation of showers, improved land characteristics, and d) Improvements to small-scale ensemble spread to improve the reliability of probabilistic NWP predictions.

As we consider the next-generation of applications, ranging from city-scale environmental to fully-coupled global Earth System prediction, the limitation of current modelling systems has become increasingly apparent. The need for such models to run efficiently on an increasingly diverse range of supercomputers over the next 10-20 years is also a major challenge for the next few years. The talk will provide a brief summary of plans to develop NWP capabilities beyond the UM.