Medium‐range multimodel ensemble combination and calibration

As part of its contribution to The Observing System Research and Predictability Experiment (THORPEX), the Met Office has developed a global, 15 day multimodel ensemble. The multimodel ensemble combines ensembles from the European Centre for Medium‐Range Weather Forecasts (ECMWF), Met Office and National Centers for Environmental Prediction (NCEP) and is calibrated to give further improvements. The ensemble post‐processing includes bias correction, model‐dependent weights and variance adjustment, all of which are based on linear‐filter estimates using past forecast‐verification pairs, calculated separately for each grid point and forecast lead time. Verification shows that the multimodel ensemble gives an improvement in comparison with a calibrated single‐model ensemble, particularly for surface temperature. However, the benefits are smaller for mean‐sea‐level pressure (mslp) and 500 hPa height. This is attributed to the higher degree of forecast‐error similarity between the component ensembles for mslp and 500 hPa height than for temperature. The results also show only small improvements from the use of the model‐dependent weights and the variance adjustment. This is because the component ensembles have similar levels of skill, and the multimodel ensemble variance is already generally well calibrated. In conclusion, we demonstrate that the multimodel ensemble does give benefit over a single‐model ensemble. However, as expected, the benefits are small if the ensembles are similar to each other and further post‐processing gives only relatively small improvements. © Crown Copyright 2009. Reproduced with the permission of HMSO. Published by John Wiley & Sons Ltd.