Meeting Report : Mid-latitude convective storms and their initiation

 

Convective storms produce some of the most damaging weather experienced in the UK, but predicting the precise location and timing of such small features remains a difficult forecasting problem. The development of a new generation of numerical weather prediction (NWP) models, with grid-spacings of the order of one kilometer, has however led to a resurgence of interest in this area. In particular, the Convective Storm Initiation Project (CSIP, a consortium of several UK universities, the Met Office and the Institut für Meteorologie und Klimaforschung, IMK) aims to understand and quantify the mechanisms controlling the location and timing of convective storms in the UK. This should allow improvements in the NWP of such events. Professor Alan Blyth (University of Leeds) gave an introduction to the CSIP project, and in particular to the CSIP field campaigns, which took place in the summers of 2004 and 2005 in southern England.

 

Convective storms can only form where there is potential instability (also known as

convective instability). The synoptic scale regions of instability could be reasonably predicted by the previous generation of hydrostatic NWP models, with grid-spacings of approximately 10 km. However, as Professor Keith Browning (University of Reading) explained, in general there is a “lid” of Convective Inhibition (CIN) which must be overcome for a storm to form. This process, of overcoming CIN, was generally not resolved by the previous generation of NWP models, but normally controls if, where, and when storms occur. A wide variety of initiation mechanisms were observed during CSIP, including: surface heating allowing thermals to reach their level of free convection; areas of deeper convection over higher ground in low winds; convergence lines generated by coastlines; upper level circulations (potential vorticity anomalies) reducing CIN (these were usually visible in the water vapour channel of Meteosat); longitudinal cloud organization (“cloud streets”); frontal upglide leading to mid-level convection and secondary initiation at the edges of cold-pool outflows from previous storms. In addition, the CSIP observations suggested that on one day storms were initiated within surface “hot-spots”, generated by gaps in cirrus cloud cover and on another day deep convection was initiated by gravity waves. Dr John Marsham (University of Leeds) showed how idealised modelling could be used to test these hypotheses. For the cirrus case, the magnitude of the surface flux variations induced by the cloud were evaluated from surface and Meteosat observations; applying these fluxes in Large Eddy Model (LEM) simulations showed that the “hot-spots” were sufficient to control convective initiation, despite the clouds moving at approximately 15 m/s. For the gravity wave case, LEM modelling showed that the positions and timings of the observed arcs were consistent with an earlier storm generating waves, which initiated the subsequent arcs of deep convection. Interestingly, the Unified Model (UM) forecast the generating storm, but did not forecast the convective initiation by the waves, perhaps due to the time-stepping used in the model. This is being investigated at the Joint Centre for Mesoscale Meteorology (JCMM) at the University of Reading. Dr John Marsham also discussed the effects of thin stable layers at mid-levels. The model results showed that the UM’s limited resolution of such features could affect the timing of modelled precipitation by approximately one hour, and the rain-rate by a factor of two.

 

The CSIP field campaigns, were centered around the Chilbolton radar facility and Dr John Nicol (University of Reading) showed cases from CSIP which highlighted the unique capabilities of this instrument. In particular: the dual-wavelengths allow the observer to distinguish between Rayleigh scattering from particles and Bragg scattering from turbulence; polarisation data allows insects, liquid water and ice to be distinguished (e.g. Figure 1) and gives information on raindrop size distributions and the radar’s Doppler facility also allows velocities to be retrieved.  The CSIP field campaigns have also provided an opportunity to develop the Fabry technique; this method allows mapping of the near-surface refractive index, by measuring the backscatter from stationary “ground-clutter”.  The refractive index field depends on the air temperature and humidity, but tends to be dominated by variations in humidity, and data from automatic weather stations deployed during CSIP will be used to evaluate this technique. Dr Ulrich Corsmeier (IMK) described how data from the numerous instruments deployed during the CSIP campaign could be combined to give a clear picture of one day; in this case the development of a sea-breeze over the south coast. Data from radiosondes, automatic weather stations, aircraft and lidar were combined to show the movement of the cooler wetter air inland during the day. In particular, IMK’s Doppler lidar gave very clear cross-sections through the head of the sea-breeze (Figures 2 and 3). Overall, the observations showed the sea-breeze suppressing convection near the coast, but enhancing convection, where uplift occurred at the sea-breeze head. Although turbulence and convection was detected along the several hundred km long convergence line generated by the sea-breeze, significant deep convection leading to subsequent showers was only observed in that area where sea-breeze and synoptic scale instability in the lower troposphere coincided. This leads to the conclusion that a sea-breeze alone does not cause severe weather if there is no other driving force enhancing the destabilization processes.  

 

 Peter Clark (Met Office) showed results from the new non-hydrostatic UM, which is now run operationally with a grid-spacing of 4 km, and was run for CSIP with a grid-spacing of 1 km. At these resolutions, the model can often resolve the mesoscale triggers, which allow precipitating storms to form. As a result, the model tends to perform well, where storms are initiated by well resolved convergence lines. For example, the 2004 Boscastle storm formed at the intersection of two thermally driven convergence lines and is predicted by the new UM, but not by the old mesoscale model. However, the model does not resolve individual cumulus cells, and so requires a spontaneous upscale organisation, with resolved storms forming from unresolved convection, and in general the model results are sensitive to the sub-grid turbulence scheme used. The new UM also frequently represents secondary initiation by cold pool outflows remarkably well, although the results can be very sensitive to cloud microphysics, particularly snow and ice fallspeeeds. Again, such cases of secondary initiation do not tend to be well represented by the old 12 km mesoscale model. The poorest UM forecast during CSIP was probably for the 24th June 2005, when mid-level storms led to floods at the Glastonbury festival; for such mid-level convection it appears that current data assimilation techniques do not tend to significantly improve the model’s representation of the initiation process.

 

Established data assimilation schemes have been designed for coarser resolution hydrostatic models and Dr Mark Dixon (Met Office) highlighted the three main problems with applying these techniques at high-resolutions (i) the failure of the established balance constraints (ii) positional adjustment and (iii) the multiple scales involved. Latent heat nudging techniques, which are currently used to assimilate cloud and precipitation observations, are also based on assumptions that break down at high-resolutions, but despite this, such schemes have enjoyed some success. Developing 4DVariational assimilation (4DVar) for high-resolutions should provide a more complex and computationally expensive, but hopefully more accurate method of assimilation. It is however not yet clear how to solve these three main challenges of high-resolution data assimilation in the context of 4DVar.

(i)                  Balance constraints: The balance constraints on the model error covariance matrix can be switched off and the model trajectory should establish balance between the variables, but it may be possible to improve on this.

(ii)                Positional errors: a forecast can give an incorrect position for a feature, such as a  storm, but otherwise represent it well. A flow dependent model error covariance matrix is one solution, but this is difficult; an ensemble of runs is another approach, although this would be very computationally expensive.

(iii)               The Multiscale problem: A high-resolution model can give errors in the small scale features, due to errors in the large scale fields inherited from the coarser models used to provide boundary conditions. For example, errors in upper level PV anomalies can lead to positional errors in forecast convection. In this situation observations from outside the domain of the high-resolution model must be assimilated into the model. This can be done by applying corrections applied to a coarser scale model to the high-resolution model, but the background states of those models are different, so it may be better to use these corrections as a weak constraint on the high-resolution model. An alternative solution is to use a variable grid, rather than the nested grids currently used in the UM, but this makes ensembles very expensive.

These approaches will all need several years of development to exploit their potential, but the Met Office aims to have an operational assimilation system for high-resolution models by 2010. This is comparable with other leading NWP centres worldwide.

 

In conclusion, CSIP has provided an exceptional dataset for investigating the initiation of convective storms in the UK. Together with modelling studies, these data can be used to evaluate and quantify the processes that control the convective initiation. Although this analysis is in its early stages, it has already shown the variety of initiation processes in the UK and some of the strengths and weaknesses of our forecasts. In particular, the new non-hydrostatic UM has proved to be much more successful at forecasting storms initiated by some mechanisms than others, and CSIP has already highlighted some aspects of the UM that could be improved. As the new generation of NWP models becomes established, further research should significantly improve both our forecasts of convective storms and our understanding of their limitations.

 

John Marsham

 

Correspondence to: Dr J. H. Marsham, Institute for Atmospheric science, Environment, School of Earth and Environment, University of Leeds, LS2 9JT, UK. Email: jmarsham@env.leeds.ac.uk