UW Information Technology

Climate scientists use supercomputer to predict flood risk in the Pacific Northwest

Severe flooding in downtown Snohomish in 2009 is a harbinger of the potential flood risks ahead.

UW Bothell Associate Professor of Physical Sciences Eric Salathé is developing the equivalent of a crystal ball to understand the impacts of climate change at a regional level.

He has teamed up with Guillaume Mauger, a research scientist with the University of Washington Climate Impacts Group, to develop a model that can help predict flood risk in the Pacific Northwest in the next century.

What they’re discovering could help regional agencies like King County understand how to better prepare — influencing a range of changes, from regulating development on floodplains to building stronger storm water systems.

To perform this data-intensive work, they rely on Hyak, the UW’s on-site shared cluster supercomputer managed by UW Information Technology (UW-IT).

The two scientists are developing methods to transform climate simulations on a global level into predictions that are region-specific and then using Hyak to spin the computationally intense calculations needed to predict regional climate scenarios.

For the two scientists, Hyak has been a game changer.

Developing regional climate models

Eric Salathé
Associate Professor of Physical Sciences, UW Bothell

Salathé’s aha moment came several years ago, when he discovered a new method for modeling regional climate change.

Up until then, Salathé had — like most climate scientists — relied on traditional statistical analyses that he ran on his office computer to downscale climate models developed by international organizations and other research institutions. These global climate models are the most advanced tools available. They simulate the effect of increasing greenhouse gases on the global climate into the future by representing processes in the atmosphere, ocean and on land.

In these models, the world is divided up into a grid, Salathé explained, which is a how a computer reads the information. The smaller the box in the grid, the more detail overall. Salathé downscales the global climate models into regional versions by essentially cutting up those big boxes into smaller and smaller ones. The detail is needed to account for differences in rainfall between Sequim on the Olympic Peninsula and Seattle and in temperatures on the top of Mount Rainier and in downtown Tacoma.

The problem was that the traditional method didn’t — and couldn’t — account for regional climate dynamics. The traditional analyses showed that in Seattle, it would continue to drizzle and rain in the future, with little change from today.

But when Salathé used a new approach that modelled a warming atmosphere with the thermodynamics of a severe storm — the cloud formation, the rain, and the direction of the wind — he got a totally different result. He discovered that Seattle would increasingly suffer from major flooding during severe storms, with potentially catastrophic outcomes.

Here’s why. A region can’t be divided into discrete locations because of the dynamic nature of the climate, Salathé explained. Clouds form and because of the Puget Sound convergence zone, air is moved around.

“If it’s raining in Seattle pretty hard, then it’s raining like crazy in Everett,” he said.

Using Salathé’s dynamic regional model to run simulations of precipitation and incorporate uncertainty — a range of numbers based on what-ifs — required much more processing power than his desktop computer could handle. Scientists can’t know the exact amount of greenhouse gas emissions from now and into the future, so a range of futures have to be imagined. That’s where Hyak came in.

Hyak: The right tool for the job

This new model caught the attention of King County, which wanted to understand the risk of flooding in the region over the next 150 years. The county asked the two scientists to run simulations of precipitation extremes and plug in the ensuing hydrological information — the flow of water on surfaces and to watersheds. The county wanted to use the projections to prepare for storm water surges in the area during heavy rain. This would allow them to identify how to protect homes and property from damage.

“In the past, because of computational constraints, we ran these really coarse-scale hydrological models with the same ballpark resolution as the regional climate model,” explained Mauger. These coarse-scale models lacked the detail needed to predict what would happen in the smaller areas.

Guillaume Mauger
Research Scientist, UW Climate Impacts Group

Hyak has made this possible. We have this better climate information from the regional climate models via Hyak.
Guillaume Mauger

“Now we’re running a hydrologic model that’s 150 meters in resolution, a football field and a half. Hyak has made this possible. We have this better climate information from the regional climate models via Hyak,” Mauger said.

Hyak’s super-fast processes can handle this massive amount of data.

“It runs in gigabytes per second and the scale of it is petabytes in size,” said Chance Reschke, director of Research Computing in UW-IT. “You plunk your data down in there and you can use it for analysis, you can use it for simulation, you can use it for anything, from anywhere.”

So whether Salathé is using Hyak from UW Bothell or Mauger is using Hyak on the Seattle campus, the experience is the same.

Chance Reschke
Director of Research Computing, UW-IT

The two scientists can also take advantage of “elastic supercomputing,” a term coined by Reschke, to describe the flexibility that Hyak provides. With Hyak, resources are shared among a group of researchers, so they only need to buy enough cycles to support their typical daily work and then tap into the pooled resources to satisfy peak demand.

Under this model, a researcher may use 20 percent one month and 300 percent the next month, without incurring additional costs, Reschke said. Overall, elastic supercomputing ensures that the whole system is used efficiently.

Now Salathé and Mauger are running two simulations for King County. Eventually, they want to use these models to help the county answer questions like how high levees should be built and where levees are no longer high enough to protect homes in floodplains. They want to be able to run models that can predict when in the summer salmon will no longer be able to swim in a particular stream.

“There’s a lot of room for growth in how we use Hyak,” said Mauger.

Planning for our future may depend on it.

Learn more:
Hyak, the UW’s supercomputer, is always ready to crunch massive amounts of data.