Dangerous floods struck communities across the United States this July. 

Heavy rains in New York region killed at least two people after swamping roads and rails this week. In New Mexico, a flash flood hit a mountain town, killing three people. Torrential rain from a tropical storm destroyed homes and claimed at least four lives in North Carolina.

In Texas, over the July 4 holiday weekend, catastrophic floods swept through the central part of the state, claiming the lives of at least 120 people, including more than two dozen children.

Even with advances in technology, forecasting how much rain will fall, where it will land and the impact of flooding all remain difficult, according to Zhi Li, a researcher at CU Boulder’s Institute of Arctic and Alpine Research.

Zhi Li

“In Texas, different weather forecasting models were giving out different predictions,” said Li, an assistant professor in the Department of Civil, Environmental and Architectural Engineering, who uses computer models to simulate floods and forecast risk. “It wasn’t until just a day before heavy rain began to inundate the area that the models finally agreed there would be extreme precipitation. That left very little time for people to respond.”

CU Boulder Today sat down with Li to discuss where the current flood prediction system falls short and what’s being done to meet the urgent need for real-time flood forecasting.

How does the official current flood prediction system work?

The National Weather Service (NWS) is the primary government agency responsible for issuing weather forecasts and warnings in the United States. Currently, flash flood forecasting is mainly based on the amount of rainfall expected in a given region. If the predicted rainfall is likely to exceed a certain threshold and overflow riverbanks, streams or dams in the area, the NWS will issue a flash flood warning.

Where does this system fall short?

The current system can only predict where the floods might happen, but it doesn’t provide detailed information about how much water could overflow onto land, how far it will spread or how deep the flooding might get.

Without the data, it’s hard to know where exactly the danger is and who should evacuate. As a result, it offers little urgency or actionable information for residents.

Is there a better way to predict floods?

At my newly established Flood Lab at CU Boulder, we’re working on developing accurate, high-resolution flood models, which are computer programs that predict how water will behave during flood events. Our latest model can predict flooding impacts at one-meter resolution, meaning we can pinpoint exactly where water will go and how deep it will get. Running these models is very computationally expensive, which is why the National Weather Service doesn’t have real-time flood models currently in use.

We’re developing a flood prediction model powered by artificial intelligence that also follows the physical laws governing how water moves. This will significantly speed up forecasting compared to traditional flood models.

Can we prevent damage from these increasingly common, record-breaking weather events?

We need to be aware that climate change is making extreme weather events more frequent, and that’s why deploying a real-time flood model is very necessary. It would be able to tell us with more confidence where the vulnerable regions are.

I also want to point out that our buildings and flood infrastructure were built 50 or even 100 years ago. They are aging, and they’re also not designed for the kind of severe weather we’re seeing now.

Flood mitigation is a very complex problem that needs collective effort from research communities, government agencies and industry.

Are drier places like Colorado at risk of floods?

Floods can happen anywhere, especially as climate change drives more weather extremes.  Colorado has a drier climate, but it can still get very heavy thunderstorms. Because of the state’s mountainous terrain, it’s easy for water to rush down hillsides and trigger flash flooding.

And we have seen that happen. In 2013, heavy rains inundated Boulder and many towns along the Front Range, killing nine people and destroying nearly 2,000 homes.

What else can we do in addition to deploying better flood models?

Improving flood risk communication with the public is very important. Instead of just telling people there will be, say one meter or three feet per second of water flow in their area, we want to translate that into how much damage specific buildings might sustain and what actions people should take.

We’re also exploring customized alert systems. For example, we could provide tailored warnings for people with disabilities, parents with young children or others with specific needs. Current alerts are issued at the county level, but with better prediction tools, we could tell people exactly who’s at risk, where to go and when to leave.

Will federal funding cuts impact our ability to predict future disasters?

Budget cuts are certainly not helpful, but I also see it as an opportunity for industry and the research communities to step up.

I’m very optimistic about the future. I’ve seen firsthand how much the technology has improved in just the past few years. I believe we’ll soon have better warning systems, better flood models, and fewer lives lost to natural disasters.