10 Questions: The Meteorologist Who Saves Lives
Leveraging worldwide weather data, Tech professor Peter Webster helps to predict potentially deadly events across the globe.
Summer 2016 | by Tony Rehagen
Peter Webster is not your typical meteorologist. While your local TV weatherman is doing the green-screen dance and hedging bets against this afternoon’s cloudburst, Webster, a professor in the School of Earth and Atmospheric Sciences, is using Big Data and advanced algorithms he personally developed to predict monsoons in Bangladesh and heat waves in India—almost two weeks ahead of time.
Webster’s foresight is more than a matter of mere convenience to those affected by his forecasts. With advance notice, farmers and residents in these severe-weather-prone areas can get themselves and their possessions to safety. Indeed, Webster and his team are using Big Data to save lives—and livelihoods—on a large scale.
How did you begin to apply Big Data to predicting monsoons in Asia?
I was speaking at a United Nations conference in 1998, showing some new results in the oscillation of the Indian Ocean Dipole, and there had just been enormous flooding in Bangladesh—tremendous loss of life and property. As I was leaving the podium, I rather cockily said, “With our new knowledge, we could have forecast this.” Then I started getting phone calls from the U.S. Agency for International Development: “Can you really do that?” I thought I could.
A southeast Asian monsoon affects 30 to 40 percent of the world’s population. And most of those nations are not very wealthy. So the variations that occur when you have a mini-drought or a big flood affect millions. We just wondered if we could do something about that in a predictive sense.
And by “predictive,” you mean more than tomorrow’s forecast?
Our mantra is that the minimum length of a forecast has to be the timescale for the slowest person in the village to be able to get himself and his cow to higher ground, which means at least seven days. The reason I include the cow is because if you lose your cow, that’s five years you have to labor to get a new cow.
So you’re not just saving lives, you’re saving livelihoods?
If you can enable someone to save their personal effects, save their cattle, do some early cropping, you’ve increased the resiliency of these societies.
How does Big Data enable you to do that?
To do a one- to two-day forecast for a country like India, you just need to develop a forecasting model that covers India and its immediate neighbors. But as the forecast horizon increases, the influence of weather events thousands of miles away also become important. So for a 10-to 15-day forecast, we have to use a global model. We use the rainfall forecasts from the European Center for Medium Range Weather Forecast model in the United Kingdom. The model has grid points every 25 km over the globe and 130 levels in the vertical. Each day, the model is run twice and integrated out to 15 and 30 days. At these two points, the model is run 51 times with slightly different initial data to simulate the uncertainty in what we know about the state of the atmosphere and the physics of how the atmosphere works. Terabytes of data are generated each day and streamed to Georgia Tech where we stream it to obtain a regional forecast over Bangladesh, where we eventually stream the data via cellphone.
Do you have trouble convincing people who might not understand these complex models?
These people are living on the edge. They realize how vulnerable they are, so they’re accepting these forecasts.
It also helps that you’ve been correct?
In our first year there, 2007-08, we forecast all three major floods. There were no false positives. More importantly, there were no false negatives—a flood never came when we didn’t predict one.
Forecasting weather in Bangladesh from Atlanta using data from Europe—is that problematic?
Ideally, every nation should be able to do its own hazard forecasting. People in Indonesia better understand Indonesian risks. But it’s difficult for these places to ingest these terabytes of data every day. They’re not ready for Big Data. They don’t have the infrastructure, the bandwidth. It’s like trying to drink from a firehose—especially when you only want one small stream coming out of that firehose.
You can’t talk about Big Data without talking about privacy, can you?
There are issues with privacy when it comes to meteorological data. Some nations sell their data. Some nations believe that data belongs to them. We wanted to give India a flood forecasting scheme for the entire Ganges, and all we needed was their river data at a few points, and they wouldn’t give it to us. They wouldn’t give us sea-level data. That makes it very difficult.
Yet you’ve been able to help India, right?
Ahmedebad, a city of 4 million people, is impacted adversely by extreme heat waves that occur in the months before the monsoon rains. They wanted to be able to forecast these waves in advance so that they could allocate their scant resources optimally and develop a heat action plan.
Do you find Americans to be more open to Big Data?
I think people here are oblivious to Big Data. It’s in every aspect of their lives now. I saw a TED Talk that said we’re in the age of the algorithm—for everything we do, there is an algorithm somewhere. In sales, it’s anticipating what you’ll buy. Maybe that’s the way it should be. Our society is so complex that if you had to think through every decision that is made by you or for you, you’d have a nation of insanity.