Some weather models predicted the Texas floods. Why did the official forecast miss it?
Weather models that meteorologists use to predict thunderstorm activity and heavy precipitation suggested on Thursday the possibility of 10 inches of rain — and even as much as 20 inches — across some parts of Central Texas on the Fourth of July. And yet the National Weather Service’s forecasts that day suggested about half that much would fall in the wettest spots.
In the wake of a deluge that killed at least 100 people — and potentially many more, with more than 170 still missing, Texas Gov. Greg Abbott (R) said Tuesday — the weather model data shows that meteorologists had reason to warn of higher flood risks as the Texas Hill Country readied for Independence Day.
The National Weather Service issued a flood watch across Central Texas early Thursday afternoon, as the models continued to show heavy rain potential, and later expanded it to include more counties. That evening, even as data began to suggest imminent rainfall rates of more than 3 inches per hour, the Weather Service’s predictions held steady, warning at most of 5 to 7 inches — until torrential downpours began early Friday.
By then, some areas were seeing rainfall at rates of several inches per hour. In total, about 10 to 15 inches fell by Friday morning in some parts of Kerr County.
Meteorologists who spoke with The Washington Post said the Weather Service meteorologists appeared to do what they were supposed to do to develop an accurate forecast and relay it to the public. They said the agency made reasonable predictions based on the information it had.
The floods shocked a region that was already known as “Flash Flood Alley” and quickly raised questions about how it could have been caught so unprepared.
Weather Service officials did not immediately respond to questions about how they may have factored the weather model predictions into forecasts and warnings.
Some have suggested that key vacancies at the National Weather Service office that covers this part of Central Texas may have contributed to the tragedy, something Senate Minority Leader Charles E. Schumer (D-N.Y.) called on the Trump administration to investigate. An official known as a warning coordination meteorologist left the Weather Service’s Austin/San Antonio office in April, one of hundreds of employees across the agency to take the administration’s latest buyout offer.
At the same time, a recently issued Trump administration budget proposal would eliminate all of the university-based weather and climate research centers that the National Oceanic and Atmospheric Administration funds across the country, and shut down NOAA offices including its National Severe Storms Laboratory, where scientists study how to better model dangerous storms and downpours.
Others have raised concern that those most vulnerable to the floodwaters, including girls at summer camp and people vacationing for the holiday weekend, may not have received flood warnings in time to act, if at all. The direst warnings came as rain intensified between about 1 a.m. and 5 a.m. local time, and many along the banks of the Guadalupe River may have lacked cell service.
But so far, there has been little scrutiny outside meteorological circles of exactly why the forecasts failed to indicate even the possibility of catastrophe.
“The amount of rain that fell in this specific location was never in any of those forecasts,” W. Nim Kidd, director of the Texas Division of Emergency Management, said Saturday night as the state reeled from the devastation across the Hill Country.
In public forecast materials, the Weather Service did not convey the risks, however slim, of 10 or more inches of rain.
In a flood watch and forecast briefing materials that were shared with local authorities Thursday afternoon, the Weather Service wrote: “Rainfall Amounts of 1 to 3 inches with isolated 5 to 7 inches are possible within the Flood Watch area.” In a flood watch update posted around 8 p.m. local time Thursday, the agency said: “Can’t rule out isolated 5 to 7 inches.”
The Weather Service did not update its flood watch again Thursday night. It began issuing flood warnings — later elevating them into dire warnings of a “flash flood emergency” — after 1 a.m. local time Friday.
Weather Service officials and the National Weather Service Employees Organization, a staff union, along with several independent meteorologists, say those at the agency’s San Antonio/Austin forecast office worked to warn the region of the flood potential on Thursday and to quickly issue flood warnings as disaster struck on Friday.
Federal meteorologists are trained to consider the outputs of dozens of weather models — supercomputer-based programs that use information about current weather to predict what will happen in the future. Some have known biases, while others are better for predicting large-scale weather patterns.
A handful are best for identifying what areas might see the heaviest rain or snow, including two NOAA models known as the High-Resolution Rapid Refresh (HRRR) and High Resolution Ensemble Forecast (HREF).
Both of those models indicated Thursday morning that more than 10 inches of rain could fall in isolated areas of Central or South Texas. By late Thursday, the HRRR suggested as much as 20 inches was possible in the region.
It “upped the ante all day,” Matt Lanza, a Houston-based meteorologist, wrote in a post on Substack.
Another model commonly used by forecasters — the three-kilometer North American Mesoscale Model, which produces new outputs every six hours — also predicted more than a foot of rain in roughly the right area on Thursday afternoon. Its Thursday evening output maintained a severe rainfall forecast.
No meteorologist would take a single model as gospel.
The high-resolution models, while useful for short-term predictions of storms and precipitation, can be inconsistent and occasionally exaggerate extreme weather potential, for example. And even as different models reliably produce similar expectations of extreme rainfall, their predictions of where it will fall can vary, leaving meteorologists with a sense of the potential impacts across an area tens of miles wide.
Lanza said in an interview that forecasters’ challenge — especially in a place like Texas, where it can “rain buckets” — is knowing when forecasting models are “going ballistic for no reason, or if they’re actually hinting at something significant.”
“It is very, very easy in hindsight to go back and say, ‘Well, look at that,’” said Lanza, who operates a website called Space City Weather. “I can count on two hands the number of times something shows up like that on the models and it doesn’t actually happen, or it happens to a far lesser extent.”
This time, as Thursday night wore on, forecasting data continued to reinforce the extreme predictions of those few models.
One technical analysis issued at 7:10 p.m. Eastern time by the Weather Service’s Weather Prediction Center — a team of meteorologists that monitors severe-weather potential across the country — warned that moisture levels were higher than more than 99% of the data in the climatological record for this region at this time of year. It noted “concerning trends” that could lead to rainfall at rates of more than 3 inches per hour in the Texas Hill Country.
But the forecast was a tricky one for meteorologists, who aim to strike a balance between alerting the public to the most likely weather conditions and maintaining its trust in their forecasts.
A Weather Service official, who spoke with The Post on the condition of anonymity because they were not authorized to comment publicly, said there is concern in such situations against issuing a dire warning that could confuse people or make them mistrust meteorologists if it does not come to pass.
For example, a forecast for up to 10 inches of rain could raise expectations of extreme rainfall, even though meteorologists know that much will fall only in isolated areas, said the official, who is the meteorologist-in-charge at one of the agency’s offices outside Texas.
“Everyone latches onto, ‘I’m going to get 10 inches of rain,’” the official said. “We always worry about how that will be interpreted.”
— — -
Ben Noll and Jason Samenow contributed to this report.