A New Method for the Characterization and Verification of Local Spatial Predictability for Convective Scale Ensembles

The use of kilometre-scale ensembles in operational forecasting provides new challenges in the areas of forecast interpretation and evaluation. These challenges will be discussed and an alternative, spatial, view of ensemble spread presented.

A new method is presented for the characterization and evaluation of the local spatial agreement between ensemble members. Spatial scales over which ensemble forecasts agree (member-member Ensemble Agreement Scales, mm-EAS) are calculated at each grid point, providing a map of the spatial agreement between forecasts. It will be demonstrated that these maps of the mm-EAS summarize the forecast spatial predictability in a concise and physically meaningful manner which is useful for model interpretation and as a forecasting tool. To provide useful guidance, the mm-EAS must be representative of the true uncertainty in the forecast. This true uncertainty is quantified by calculating the spatial scales, at each grid point, over which ensemble members agree with radar observations (member-radar Ensemble Agreement Scales, mr-EAS). Comparing the mm-EAS and mr-EAS allows the spatial spread-skill relationship of the ensemble to be assessed.

To demonstrate these techniques, and their practical applications, the mm-EAS and mr-EAS are presented for instantaneous rain rates from six convective cases. These cases sample a variety of convective situations and were selected from the period of the COnvective Precipitation Experiment (COPE). Forecasts are considered from the 12-member, 2.2 km grid -spacing, Met Office UK ensemble (MOGREPS-UK), operational since July 2012.

The mm-EAS highlight predictability differences between cases , which are linked to physical processes. Spatial predictability also varies across the model domain, reinforcing the need for a spatially varying approach where local predictability differences can be detected. A comparison of mm-EAS and mr-EAS demonstrates the case by case and temporally variability of the spatial spread skill; this is again linked to physical processes. Overall, for these six convective cases, the mm-EAS and mr-EAS were similar and therefore the ensemble was well spread spatially. However, at small scales there was also an indication that the model was under spread: when the model was confident about the location of precipitation, it was over confident. This objective information about spatial spread and skill is very valuable for the understanding and improvement of convective ensemble systems, and the forecasts associated with them.