Menu
Assessment of convective-scale ensembles
The Met Office has begun running an operational ensemble with 2.2km horizontal grid-length, and essentially as a downscaling of selected members from its MOGREPS ensemble. There are many issues associated with the interpretation and analysis of such an ensemble that are specific to the fact that it is at convective scales. The verification of even a single forecast at convective scales raises issues about representivity. A well-known issue is the so-called double-penalty problem, whereby a rainfall forecast slightly displaced from the observations scores worse than a forecast of no rainfall according to simple point-to-point comparisons. For such reasons, a range of fuzzy, scale-dependent indicators of performance have been developed in recent years, such as the fractions skill score (FSS) of Roberts and coworkers.
The aim of this project is to develop scale-dependent techniques that have been used for deterministic forecasts to produce suitable diagnostics to aid the understanding of convective-scale ensembles. For example, what are useful measures of spread-skill relationships for such ensembles? As part of this work, we aim to exploit tools that are more commonly thought about in the context of data assimilation studies, such as multivariate covariances and spatial and temporal correlation structures. We would like to relate correlation lengthscales, for example, to spatial verification measures such as FSS.
Links for this work
Papers
1. A paper assessing spatial uncertainties in precipitation forecasts (featured article in QJRMS).2. PhD thesis (by Seonaid Dey) on the spatial assessment of convective-scale ensembles.
3. A paper introducing the ensemble agreement scale.
4. A paper on the FSS approach as a method for assessing ensemble spread.
Talks
1. A talk on the assessment of spatially-varying spatial predictability (presented by Seonaid Dey): a recorded presentation at the 27th Conference On Weather Analysis And Forecasting/23rd Conference On Numerical Weather Prediction in Chicago, July 2015. (Awarded best student presentation in the NWP category.)2. A talk on analysis of the 02/08/13 and 03/08/13 cases (presented by Seonaid Dey), at a COPE project science meeting in Leeds, April 2014.
3. A talk describing a combined correlation/FSS analysis, (presented by Seonaid Dey), at the EMS conference, September 2013
4. A talk on introducing the project (presented by Seonaid Dey), at a DIAMET project science meeting in Manchester, March 2013
Posters
1. A poster on ensemble assessment based on agreement scales, presented at the ECMWF/WWRP workshop on Model Uncertainty, April 2016. Also presented at the RMetS NCAS conference in Manchester.2. A poster on spatial assessment of convective-scale forecasts, presented (by Nigel Roberts) at the ECMWF/WWRP workshop on Model Uncertainty, April 2016.
3. A poster on reliably estimating vertical correlations (presented by Seonaid Dey), at at the Met Office Academic Partnership Poster Conference in Exeter, February 2014.
4. A poster on measuring spatial predictability (presented by Seonaid Dey), at a SEPnet summer school in at NPL, June 2014.