9 May 2016

9 May 2016 turned out to be an extremely active day for severe convection, with tornado, wind, and hail reports stretching from Iowa and Nebraska south to the Texas/Oklahoma border. As of this writing, 25 tornadoes were reported. This count may fluctuate, as damage assessment is ongoing. The system that spawned this outbreak continues to cross the United States today, with 16 tornado reports occurring as of this writing. This system will likely warrant further study down the line, but I'd like to give a brief overview of our forecast considerations, review the experimental forecasts we issued yesterday in light of today's verification, and discuss our evaluation of the forecasts.

A view looking east from the roof of the National Weather Center in Norman, Oklahoma at 5:29PM CST. This supercell produced a brief tornado just east of Norman at 4:15PM CST.

Each week in the SFE begins with an orientation, but the first activity participants undertake is a hand-analysis of upper-air and surface charts, to get a feel for the current state of the atmosphere. All participants contour a chart, and then the entire group discusses the features relevant to the day's convection. This ensures that our understanding of the numerical models is grounded in the observations. On the morning of 9 May, we noted a very strong jet ejecting across the region of interest, with a 110 kt wind at 250 mb over Oklahoma City.

We also noticed that there still was less moisture at 850 mb and the surface than is typical in May, but that a plume of moisture was starting to return from over the Gulf of Mexico, aided by a 50 kt low level jet at 850 mb positioned over Dallas. The dryline was in the Panhandles of Texas and Oklahoma, and as we suspected that it could provide a focus for convective initiation.The models' handling of this feature was a large forecast consideration. We were also concerned about the instability over our area of interest, which stretched from central-eastern Texas northward to the Kansas-Nebraska border. (Each day we have to choose a centerpoint, and forecasts are subjectively verified only for a certain area around each centerpoint). 

The convection-allowing guidance gave us a good idea that the dryline would set up near or even west of the I-35 corridor, whereas the operational models were suggesting a dryline east of I-35 by the time convection initiated. The UH tracks from the ensembles were very strong, reaching values of over 400-500 m2/s2 in some members. The signal of a more western dryline combined with the winds and surface convergence patterns at the time of our forecast led us to believe the solutions that had a dryline slightly west of I-35. The CAMs also gave us an idea that convection would locally occur around 3-4 PM. The initial probabilistic forecast from the total severe desk is below on the left, overlaid with the local storm reports (LSRs):

The panel on the right is the practically perfect forecast of any severe report. Blue hatched areas in both figures indicate a 10% chance or greater of a significant weather report occurring within a 40 km radius of a point. The red dots are tornado LSRs, light green dots are hail LSRs, and light blue dots are wind LSRs. Significant wind LSRs are indicated by a dark blue square, such as the one near the Arkansas/Oklahoma/Texas border. Significant hail reports are indicated by a dark green triangle, such as the cluster of reports over northern Kansas.

The individual hazard probabilities and corresponding practically perfect forecasts are as follows:






Objective skill metrics are also listed in these plots - the meaning of these metrics will be discussed in a subsequent blog post. Overall, these forecasts captured the severe reports, though each has some faults. The total severe forecast is a bit too far east and slightly too low, and the individual hazard probabilities are generally a bit too low compared to the practically perfect probabilities. The uncertainty surrounding how much activity these storms would produce caused us to hedge our probabilities lower than they would have been had we had more confidence in the quality of available moisture and subsequent instability.

Participants on average rated the wind forecasts the highest out of the individual hazards, with a mean score of 7.67/10. The tornado forecasts had the next-highest mean of 6.5/10, and the hail was rated lowest, with a mean of 6/10. Six ratings went into each of these means. Participants noted the need for extending the significant hail probabilities further north, and docked the tornado probabilities for the area of 10% probability being too far east relative to the reports.

The total forecasts garnered a mean rating of 7.64/10, with eleven participants rating the forecast. Participants found the total severe probabilities to be quite skillful over Oklahoma, and considered the higher probabilities issued in southeast Oklahoma to accurately portray the risk posed by those storms, despite the practically perfect forecast encircling the area with a 30% contour rather than a 45% that the desk forecast had. However, they did note the need for higher probabilities to the north across Kansas, as well as the need for significant probabilities further north. These higher ratings overall reflect both that forecasting all severe weather is often easier than determining individual hazard threats, and that the conceptual model of the forecast that the total severe desk was working off of did quite well. While the forecast did not have the reports in the center of the outlined area, the desk correctly anticipated a sharp dryline off of which storms initiated.

The outlooks discussed above are merely the full period outlooks. More experimental forecasts for four-hour periods of total severe, tornadoes, hail, and wind were also issued, and they are well worth looking at. Week 2 started off quickly, and it looks like the weather will keep us busy for the next two days at least.