Intracloud and Cloud-to-Ground Lightning Prediction using a Dynamic Lightning Scheme in WRF

Recent advances in computing technology make it feasible to run convection-allowing forecast models that better depict cloud and precipitation processes relative to coarser model forecasts with parameterized convection. Yet, these forecast models do not resolve clouds, and the grid-resolution sufficient in most is not fine enough to explicitly predict electrical charge generation and charge dissipation processes within clouds. Moreover, explicit calculation of lightning in real time forecasts would require a significant increase in computer resources. For this reason, these processes (and hence the prediction of lightning flashes) must be parameterized in convection allowing forecast models. Towards this end, a new prognostic spatial and time dependent variable was added to the Weather Research and Forecasting Model (WRF). This variable is referred to as the Potential Electrical Energy (Ep), and it was used to predict the dynamic contribution of the grid-scale resolved microphysical and vertical velocity fields to the production of cloud-to-ground (positive and negative) and intra-cloud lightning. The source of Ep is assumed to be the non-inductive charge separation process involving collisions of graupel and ice particles in the presence of super-cooled liquid water. The dissipation of Ep occurs when it exceeds pre-assigned threshold values and lightning is generated. Analysis of 4 case studies shows the capability of this scheme in predicting both cloud-to-ground and intra-cloud lightning. A single cloud-to-ground lightning event was forecast with about equal values of Probability of Detection and False Alarm Ratio on the 4 km simulation grid. However, when lighting was integrated on 12 km and then 36 km grid overlays, there was a large improvement in the forecast skill, and even as many as 10 cloud-to-ground lighting events were well forecast on the 36 km grid. The impact of initial conditions on forecast accuracy is briefly discussed, including an evaluation of the scheme in wintertime when there were relatively few severe weather reports. The dynamic algorithm forecasts are also contrasted with statistical lightning forecasts and differences are noted between them. The scheme is being used operationally with Rapid Refresh (13 km) data; the skill scores in these operational runs were very good in clearly defined convective situations.