Using Canadian GEM output for forecasts of thunderstorm initiation on the Canadian Prairies

Predicting the location and timing of thunderstorm initiation (TI) is critical for forecasters to issue timely and accurate severe weather watches and warnings. Observational and modeling studies highlight the importance of processes near and within the atmospheric boundary layer (ABL) for promoting or inhibiting TI. These include convergence and moisture availability (at and above the surface) and vertical wind shear within and above the ABL. In Canada, real-time observations of ABL characteristics are extremely limited with no operational profiling network, widely-spaced upper air stations, AMDAR profiles without humidity information, and only partial radar coverage (especially for clear-air echoes).

Operational forecasters routinely utilize NWP output to complement a thorough analysis of observational data, calculate derived stability and other parameters, and fill in data-void areas. Currently, operational NWP models do not consistently provide accurate forecasts of the timing and location of convective precipitation. While the details may not be resolved, in many cases the general pre-storm environment is reasonably simulated, often including mesoscale boundaries associated with TI. A variety of model-derived parameters, fields, and indices are available to characterize the evolution of the troposphere and environments conducive to the development of severe storms. Few of these focus on ABL processes, and fewer still on the problem of TI specifically. Given current trends towards automation of forecasts, an improved first-guess of TI areas would be beneficial to forecasters and provide guidance for nowcasting applications and watch/warning decisions.

As part of Environment Canada's Research Support Desk (RSD) initiative, the Hydrometeorology and Arctic Lab (HAL) is investigating the utility of model-simulated ABL characteristics for hourly TI forecasts. Our methodology uses an ingredients-based approach considering conditions thought to be important for promoting or inhibiting TI. Forecast fields are produced using full-resolution (58 eta levels, 15-km horizontal spacing) hourly output from the Canadian Meteorological Centre's Global Environmental Multi-scale (GEM) model. These focus on model-simulated low-level water vapour, convergence, and ABL or near-ABL wind shear.

In 2007, a suite of hourly model fields targeting ABL processes was developed and tested in an operational setting via the RSD in Edmonton, Alberta. Fields were compared with observed lightning from the Canadian Lightning Detection Network (CLDN) to assess their utility, and develop preliminary thresholds, for TI forecasts. A subset of composite fields was then developed and evaluated to combine elements of selected individual fields and forecast instability. During the 2009 convective season, the composite fields were subjectively verified in real time on the RSD. An objective verification of the fields was conducted in winter 2009-10. This included comparisons with readily available objective TI forecasts (Kain-Fritsch rain rate, a modified SCRIBE forecast) and the cloud physics thunder parameter (CPTP, with a minimum threshold of 25) in use at the SPC.

In addition to acting as a visualization tool, preliminary verification results suggest the composite forecasts and CPTP may show improved skill over existing TI forecasts currently available to Environment Canada forecasters. A consistent weakness of these fields is incorrect model placement of areas or axes of forecast convection relative to observed lightning despite often reasonable spatial coverage. It is suggested however, that incorrect placement could be adjusted by the forecaster based on real-time observations. In this way, forecasters could use the automated forecasts to highlight potential TI areas early in the forecast period and adjust position/timing using observational data as TI became more imminent.

Work is underway to update selected TI fields in real time using hourly surface observations. The updated fields will be verified against the raw model fields to assess if there are improvements. Complementary work includes the use of classification and regression tree analysis to define TI forecast thresholds and modify existing lightning forecast models. The goal is to implement a first-guess, area-based TI field (or fields) to be modified by forecasters in future forecast and warning production software. A prototype version of the nowcasting component of this software called iCAST (interactive Convective Analysis and Storm Tracking) is undergoing development and testing within the Nowcasting and Remote Sensing Lab of Environment Canada.