The Catastrophic Tennessee Flash Flood - Why The United States Needs a National Disaster Review Board

While it is important for meteorologists to make accurate forecasts in situations of “partly cloudy and 70° with light winds,” it is vital to society that weather science provides timely and accurate forecasts and warnings of situations with the potential of considerable loss of life and extreme property damage. 


The computer model and human forecasts in the hours leading up to the catastrophic flash flood west of Nashville Saturday morning, where more than 17 inches of rain fell in less than 10 hours, were inadequate. None of the “guidance” prior to the start of the rain signaled that a major event -- which would take at least 21 18 lives with three missing -- was about to occur. The 17 inches (and, up to four additional inches fell later in the day), if confirmed, will shatter the Tennessee's state rainfall record by four inches. So, this is exactly the type of situation where weather science needs to shine. 

Below is the actual rainfall amounts for the morning (pre-dawn) event.

This preliminary report outlines what occurred from a  meteorologists'  perspective and explains why the United States needs a National Disaster Review Board to examine all aspects of situations like these in detail. We start about ten hours before the heavy rain commenced. 


National Quantitative Precipitation Forecast 

The QPF issued at 5pm CDT Friday, August 20 – less than ten hours before the event began – gave no indication that a major event was just hours away. The forecast for the area about to flood was for one to two inches of rain to fall. 

Thanks to Dr. Ryan Maue and WeatherModels

Ensemble Max Rainfall

While I have collected the mesoscale models from 7pm Friday; in order to keep this presentation straightforward and an appropriate length for a blog post, I am presenting only the mesoscale maximum ensemble rainfall (below). It would have become available about four hours before the event began. 

It did a fine job of depicting a stripe of seven-inch rains in the correct location (star). Unfortunately, it forecast similar seven inch amounts in multiple other locations. This and the other mesoscale models typically have unacceptably high false alarm rates that can cause them to be disregarded in situations like these. Plus, the seven inches was less than half of what actually fell -- which is a poor performance for a model that purports to be forecasting the “maximum possible.” 


HRRR Model Output

The HRRR model's 12-hour forecast from 1am (data available to meteorologists about 2:20am) forecasted a 12-hour maximum rainfall of 4.4 inches in the general area where heavy rain later occurred. Of course, this was less than one-quarter of what actually fell. 

Again, the location was pretty good but the amount was far too little. 

The 2am HRRR 12-hour rainfall amount forecast was better as to amounts with a maximum of six inches but it was too far northwest. There were also two other 6-inch maxima forecast outside of the band of heavy rain which made this forecast of questionable utility for this flash flood event. 


The 3am HRRR, which would have been available to forecasters at about 4:20am was excellent with a correct forecast of an additional 6.5 inches in the correct location. However, the event was well underway at this point and it may have had limited usefulness given the inconsistency in the HRRR's performance up to that point Saturday morning. Had I been on the warning desk, I would have been reluctant to depend on it given its failure to lock-on to a single threat location in the previous hours' versions.

The, at best, inconsistent quality of the models is especially perplexing given that rawinsondes (weather balloons with instruments) were being launched every six hours due to Hurricane Henri, which is twice as often as usual. That extra data should have led to better forecasts. 


Mesoscale Precipitation Discussion

The NWS’ Weather Prediction Center issued a timely excessive rainfall discussion at 1:20am pointing out the heavy rain threat. It forecasted, 


There is some disagreement among the 00Z HREF suite of models with respect to the timing of the more organized convective threat and the resulting rainfall amounts going through 12Z, but generally the expectation is for some spotty 3 to 4+ inch totals for this time frame, and then a threat for additional rainfall beyond 12Z as additional energy/forcing arriving from upstream maintains the convective threat.


Most all of the 17+ inches (98%) fell before 12Z (7am). This forecast also anticipated too little rain (“spotty 3 to 4+ inch totals”). While other discussions were issued, they were too late to influence the outcome of this situation. 


Flash Flood Warnings and FLASH Data

The first flash flood warning for the area was issued at 3:29am. It was timely and accurate, but it forecast too little rainfall with “between 1.5 to 2.5 inches of rain have fallen. Additional rainfall amounts up to 1 inch are possible in the warned area.”

Given the poor quality of the guidance, the fact that a warning was issued at this point should be lauded. There was nothing that indicated an all-time (subject to confirmation) record for state rainfall was about to be set. 


The flash flood warning was updated at 4:49am, but it was not upgraded to a flash flood emergency. This is perplexing because the 6-hr FLASH data was beyond the “200-year” recurrence interval over a large area which made it an extremely rare event. The data at that time is below.

The deep purple up to 4am was a large area of rainfall that fell at a 
once in 200 years (or worse). Lower right was the flash flood warning 
issued at 4:49am. Upper right is a map of the area.

The flash flood warning was upgraded to a flash flood emergency at 7:47am, presumably due to ground truth, as explained in the warning. 

However, the FLASH data had been depicting a 200+ year rainfall event for four hours. Given the goal of the FLASH data is to enable earlier and more accurate flash flood warnings, it is perplexing as to why an “emergency” flood warning was not issued earlier. Adding to the oddity is the river gage on the Piney River near Vernon had risen more than 9 feet in the preceding hour. Between FLASH and that extraordinary rise, an emergency probably should have been issued sooner. 

Why does the emergency timing matter? In the words of a storm chaser whose home was swept away and “not even on the same street it is supposed to be,” the warning was described as inadequate.

There was absolutely nothing. I got an emergency alert about 10 minutes before the “O-F*****” moment, and everyone was asleep. But it was not an evacuation order or anything. It was a simple flash flood warning which we get weekly here.

At least 21 18 people died and others are still missing. While the first warning was excellent, would the followups -- more strongly worded and earlier -- have made a difference? I don't know. 


This type of event is exactly why the United States needs a National Disaster Review Board (NDRB) that is modeled after the highly successful National Transportation Safety Board. 


In the meantime, I call on the National Weather Service to appoint a “service assessment” team of outside experts rather than their usual practice of having internal NOAA employees to do the job. Past NWS service assessments have, too often, glossed over the most serious issues. Only outside experts should do the evaluation and writing. 

A review of this event should include, but not be limited to, answering the following questions:
  • Do we have solid evidence, via data starvation studies, that demonstrates additional rawinsondes improve the accuracy of synoptic-scale and mesoscale models?
  • Have local National Weather Service office meteorologists received sufficient training in radar and applied hydrology to properly evaluate and warn of high-end situations?
  • Are flood gages sufficient in number and properly located?
  • How did the new National Water Model perform?
Again: I want to be clear that I do not wish to be overly critical of the local National Weather Service office: at first, there was little solid guidance upon which to base a high-end flood warning.  While things did get clearer over time, this was not an obvious situation. 

© 2021, Mike Smith Enterprises, LLC


Popular posts from this blog

[1:10am Update] Tornado Forecast for Rest of the Night

First Tornado Watch of the Day Issued

Hilary's Forecast Path Shifts West; Updated 9:20am PDT