The operational High Resolution Window WRF model runs at NCEP: Advantages of multiple model runs for severe convective weather forecasting

In April 2004 the National Centers for Environmental Prediction (NCEP) Environmental Modeling Center (EMC) began running a once daily experimental high resolution version of the WRF model with the Non-hydrostatic Mesoscale Model dynamic core. The WRF-NMM was integrated out to 36 hr over a large domain (three-fourths CONUS) with 4.5 km grid length and with no parameterized convection for testing and evaluation in the NOAA Hazardous Weather Testbed (HWT) Spring Experiment. The Spring Experiment is highly collaborative activity organized annually by the Storm Prediction Center (SPC) and National Severe Storms Laboratory to bring together numerical model developers, research scientists, operational forecasters, and academia to accelerate the transfer of cutting edge research to National Weather Service operations. A primary goal of the HWT interactions is to improve forecasts and warnings of hazardous weather such as severe convective storms and flood producing rainfall.

Feedback from the 2005 Spring Experiment was very positive, indicating that convection-allowing WRF models provided unique guidance to forecasters on important details of thunderstorm structure such as the convective mode, even at forecast times as long as 36 hr. Consequently, EMC continued running an experimental version of the model year round to provide support for SPC forecasters. Periodic improvements to the WRF-NMM have been made since then, including an increase in resolution to a 4 km grid length, upgrades to newer WRF versions, modifications to model physics, and the creation of additional specialized convective products.

In the fall of 2007, an operational version of the 4 km WRF-NMM was implemented in the NCEP High Resolution Window (HiResW) operational run slot, along with a version of the Advanced Research WRF (WRF-ARW) run at 5.1 km grid length. Over the Continental U.S. (CONUS), these two WRF configurations are run twice daily (at 00 and 12 UTC) over an eastern two-thirds CONUS domain, and once daily at 06 UTC over a western two-thirds CONUS domain, with all runs producing forecasts out to 48 hr.

A key challenge in severe thunderstorm forecasting is to predict correctly the evolution of nocturnal convection that occurs prior to the onset of the next day's diurnal heating cycle. Errors in the prediction of overnight convection can result in modification of the model predicted pre-convective environment well into the next day. This can impact the location of low-level boundaries, amount of cloud cover, and air mass thermodynamic characteristics, strongly impacting model forecasts of deep convection during the subsequent convective cycles. For example, erroneous persistence of model generated deep convection after the time observed convection dissipates has been found to result in excessively strong and persistent cold pools and convective outflow boundaries in the model forecasts that are not present in the atmosphere. These features may improperly focus subsequent model development of thunderstorms along the spurious boundary and inhibit destabilization within the model generated cold pool.

The updated 12 UTC WRF model runs in the HiResW have been found to provide improved convective scale thunderstorm guidance to severe weather forecasters for the afternoon and evening time period on some occasions during the spring and early summer of 2008. In addition, when overnight convection diminishes or dissipates before 12 UTC and/or the 00 UTC run predicted the overnight convection reasonably well, the forecast guidance from the 12 UTC run is more likely to be similar to the 00z guidance and this can increase forecaster confidence in the model solutions.

Several cases focusing on WRF-NMM forecasts are presented to illustrate this finding. These results highlight the importance of incorporating improved initial conditions at the start of the diurnal heating cycle into updated convection-allowing WRF models, especially during periods of active overnight thunderstorms. They also suggest high resolution models that are capable of providing guidance for smaller scale, high impact weather events may need to be run more frequently to take advantage of later observational data that will be incorporated into the initial conditions.