In search of CI: Developing a strategy for detection and verification of thunderstorm initiation in convection-allowing models

Convective initiation (CI) was a focus of two recent experiments in the NOAA Hazardous Weather Testbed, but unlike previous work on the topic, these experiments focused less on the cloud-scale process of CI and more on developing a systematic proxy-based framework for detecting and verifying CI in convection-allowing models (CAMs). This framework involved multiple stages of interrogation of gridded CAM and observational datasets. Initial stages of testing focused on simply identifying convectively active (CA) grid points using different proxies for convection. Ultimately, reflectivity (exceeding 35 dBZ at the -10º C level) was chosen as a suitable proxy because it was readily available in both CAM and observational datasets. Plus, there was some precedent for its use in similar contexts. Next, a decision was made to focus on “new” convection – points not associated with previous ongoing convective activity. An algorithm designed specifically for this purpose grouped together CA points that were contiguous in space and time (using a 4-km spatial grid with 5-minute temporal updates), yielding time-domain CA objects. Within each object, CI points were found by identifying local time mimima (i.e., CA grid-points without adjacent CA points occurring at earlier times). Objects having a lifetime spanning less than 30 minutes were not considered, effectively eliminating isolated, short-lived convective cells and focusing attention on larger systems that were more likely to be disruptive to human activities.

This proxy-based approach allowed us to establish a baseline for the skill of WRF-based CAMs in predicting CI, with a focus on timing. An equally important set of results came from lessons learned in developing appropriate proxies for CA and algorithms that can automatically discriminate between new convective activity and that associated with pre-existing storms. A new framework for detecting and verifying CI will be presented, its sensitivities will be discussed, and preliminary estimates of WRF-based CI predictive skill will be provided. An oral presentation is preferred.