From a science vision to a new probabilistic post processing system at the Met Office

A fully probabilistic post processing system called IMPROVER has been developed at the UK Met Office. IMPROVER provides frequently updated probabilistic gridded forecasts, as well as forecasts for point locations, for input into automated forecast generation and for users such as operational meteorologists. Although the outputs are probabilistic, a deterministic interpretation can be extracted if required.

The scientific rationale behind this endeavor is the need to make more optimal use of the current and future generations of convection-allowing Numerical Weather Prediction (NWP) models and ensembles. The aim is to provide seamless, calibrated, probabilistic forecasts that are a blend of NWP models/ensembles from nowcasting to medium range. Today’s NWP systems offer not only many ensemble members but also frequent updates making it very difficult for users to manage the data and exploit latest information, so a key capability of IMPROVER is the frequent cycling, providing a continuously updated forecast blending the most recent available data.

Several key scientific benefits arise from the probabilistic approach on top of the capability to provide probabilistic outputs. Probabilities allow much simpler and effective blending with older forecasts or between different models/ensembles. We have introduced a variety of probabilistic neighbourhood methods to account for the inherent limited predictability at small scales. Some of these can incorporate topographic variation which is particularly important for variables such as rain, sleet and snow or fog. The ensemble-probabilistic approach has also enabled the use of ensemble calibration methods, which can not only improve skill and spread, but create a much more seamless transition between models/ensembles at different resolutions.

The system is built with a modular software framework that allows flexibility for future development and includes verification at every stage of the processing. IMPROVER is now routinely running with operational support and is expected to become fully operational in 2022. This presentation will briefly describe the initial scientific vision and current IMPROVER capability and discuss where any compromise or re-evaluation had to be made along the way. Finally, thoughts about the future and lessons learnt will be shared.