Product1 publisher2 min readPublished
TACLS puts satellite data and machine learning behind the forecaster who issues flash flood warnings
A team inside the National Weather Service built software that reads satellites with machine learning to flag flood-prone ground sooner, and the warning it feeds still ends with a forecaster deciding to press a button.
The Product Desk · Product desk
What happened
- Laura Lin saw pieces of her wood floating near the barn at her home in Lanesville, southern Indiana, closed her laptop, got her children up and sheltered with them at a neighbor's house.
- The town took more than 8 inches of rain within a few hours that day, far above the threshold for heavy rainfall, and the water had drained a few hours later.
- New software called the Transient Artifact and Continuous Learning System uses satellites and machine learning to spot areas that could flood sooner and help the National Weather Service decide when to issue flash flood alerts.
- Severe weather alerts are issued by 122 weather forecast offices across the United States and its territories, according to NWS senior service hydrologist Jayme Laber.
- Lin said the alerts telling people to get on their roofs went out after everyone in town had already been flooded.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint An emergency manager cannot rank this tool against a gauge network or an alert-delivery upgrade in minutes of warning gained, so it has to be judged on a forecaster's account of its usefulness.
- decision An office that adopts TACLS changes the inputs to a human judgement call. The call, and the person on shift at four in the morning, stay where they were.
- capability Flagging flood-prone ground earlier could let the watch window open nearer its top end, when there is still time for sandbags, road barriers and an evacuation plan review.
- exposure Until delivery changes, the risk sits with residents who have to act on what they can see in the yard before any warning reaches them.
The alert ends with a person pressing a button. Jayme Laber, senior service hydrologist at the National Weather Service office in Oxnard, California, said forecasters weigh rain and stream gauges, satellites and radar, plus computers that flag when rainfall passes what a specific place can take [13]. "Based on the type of soils we have, the steepness of slopes, we have developed flash flood guidance so that we know that if we get this amount of rain in this amount of time in that area, it could result in flash flooding," Laber said [14]. Years of expertise inside those buildings feeds the same decision, he said [15].
TACLS goes into that stack of inputs [6]. The strongest claim for it on the record comes from Ivory Small, science and operations officer at the NWS San Diego Weather Forecast Office, who worked on the project with Laber and a broader team of scientists and researchers [12]. "It will help you save lives," Small said [9]. Without TACLS, he said, a storm moving into your area "could kill some folks"; with it, "you can put out the warning and save some folks" [10]. The Verge's account does not report a lead-time figure for TACLS [19].
"We didn't get the 'get on your roof' warnings until I was already at that person's house," Lin said [7]. She had already seen wood from her barn floating across the yard, closed her laptop, woken her children and walked them to a neighbor's house [2]. More than 8 inches of rain fell on the town in a few hours, and the water was gone a few hours after that [3]. Six inches of fast-flowing water is enough to knock an adult off their feet [5].
A flood watch is the least serious alert and can be issued anywhere from 12 to 48 hours ahead, Laber said [16], a top end four times the bottom [18]. A county that hears a watch is in effect learns nothing from the word about whether it has two days of daylight or one evening, and Laber said the watch is the cue for local teams to begin "flood protection measures" [17].
This software is built for the forecaster at one of 122 desks [11]. For anyone deciding whether a new detection input earns a place in the workflow, the question about the last bad incident is whether the office knew late or the town learned late. TACLS works on the knowing. Lin said the alerts came "way too late", after everyone in town was already flooded [8].
What to watch
- A published pilot result or lead-time figure from TACLS at any of the 122 forecast offices.
- Whether TACLS becomes part of the standard toolset at forecast offices or stays a research project.
- Any change to how roof-level warnings reach residents in rural counties, which is the gap Lin describes.