In June 2013, Keith Musselman was living in the Canadian Rockies when the nearby Bow River flooded. “We were in a valley, so we were stuck for about five days,” Musselman told Ars. “The community was devastated.”
The flood was one of the costliest and most devastating natural disasters in Canada’s history, with five people killed, more than 100,000 evacuated, and extreme property damage.
“Forecasters have a good handle on what happens when rain falls,” he says. “But when that rain falls in mountains where there’s deep snow, we don’t have a good handle on what the flood volume will be.”
Musselman and a group of colleagues used historical data to sketch out what the rain-on-snow flood risk looks like across western North America—and how that risk is likely to change in a warming world. Their paper, published in this week, finds that some areas will see reduced risk thanks to declining snowfall. But at higher elevations, there’ll still be snow, and the increasing amount of rain falling on that snow will have implications for flood risks.
The perfect storm
At drier and higher elevations, rainfall on snow isn’t so much of a problem, either because there’s not enough rain or not enough snow for it to matter. But when the conditions are just right, heavy rainfall can combine with snowmelt to produce floods that are more catastrophic than the rainfall would cause.
When Musselman moved to work at the National Center for Atmospheric Research (NCAR), his colleagues there had also identified this as something they wanted to study—and “they had a great dataset,” Musselman says. “We used that to understand how rare these large flood events are.” Work has been done on this before, but this team was able to use more fine-grained data to build up a detailed picture.
They built a simulated world of weather systems that would produce rain-on-snow floods in the same places and with the same frequency as the 13 years of historical data they used. Once they were sure that the simulation matched the historical patterns of snowpack, snowmelt, and other factors, they used it to build up a picture of how those kinds of events might change over the next 13 years.
The model showed shifting patterns: areas that currently have a high risk of flooding will see a decrease, as precipitation will occur more frequently as rain than snow. But at higher elevations, where once there would have been snow falling on a snowpack, rising temperatures will now cause rain to fall on that snow. The Sierra Nevada, Canadian Rockies, and Colorado River headwaters saw the highest increases in risk, with floods twice as likely in some cases.
Philip Mote, who studies climate-related snowpack change and wasn’t involved in this work, points out that the starting rate of floods might not be particularly high, so a doubling of the risk might not necessarily mean a region suddenly awash with floods. The shifting landscape will nonetheless demand a rethink of flood management—but for that, more data will be needed.
“Climate scientists like to look at [periods of] 30 years,” Musselman notes. But his team had to limit the time window because they were running into the limits of computational power available to them. “We’re up against this wall of computational demand to simulate all the physics that are involved,” he explains.
The two 13-year simulations took up a year of NCAR supercomputer time, so they were also only able to look at one climate change scenario—the average of all 19 different climate models in a “business-as-usual” emissions scenario. Ideally, it would also be good to know what the projections look like under additional climate scenarios.
The 13-year simulation is “OK if it’s treated carefully,” says Prof. Mote. “That’s a constraint if you’re running a 4km resolution over the whole of the US—that’s a whole heck of a lot of computing and data.” It does make things tricky for regions that have rare events that might be represented only a few times across 13 years, he explains: “The exact way those events change is probably less robust than if they had a 30- or 50-year simulation.”
Opening the floodgates
In anticipation of winter and spring rainfall, reservoir managers in many regions release large amounts of water in the fall, creating space to buffer the incoming deluge. “Reservoir managers follow tight rules,” Musselman explains. “They have guidelines based on historical data. The challenge is, as the climate changes, our rules and regulations aren’t adjusting accordingly.”
The simulation suggests that the flood risk isn’t the only thing set to change—seasonal timing will change, too. In some regions, the highest risk is likely to shift from spring to winter; in others, from winter to spring. This means new rules both for providing flood buffers and for ensuring enough refill during dry periods.
The current rules are “heavily focused on dates,” says Charles Luce, a hydrologist who studies water management. Changing the rules is complicated because of all the different people involved, but there’s already work underway to figure out how to update the rules in the face of climate change, he says. That work will likely be “looking to this paper for guidance.”
The flood protection infrastructure itself also needs a look, says civil engineer Farshid Vahedifard: “Levees and dams need to be re-analyzed to see if they can safely withstand and handle this new pattern of flooding.” This is particularly crucial, he adds, since “most of these dams and levees are old, and a good percentage of them are currently operating under marginal conditions.”
There’s still work to be done to turn this risk projection into useful information, though. “It’s the first step in a series of studies that aim to better understand the hydrology,” says Musselman. Future work will need to look at how this translates to a very local scale, including how factors like soil saturation and vegetation affect the risk of runoff becoming a flood. And ongoing observation of the real situation on the ground will also be essential.
Hydrologist Wouter Berghuijs agrees with Mote that the simulations match what intuition would predict, but it’s “good to see it confirmed in harder numbers,” says Berghuijs. “These models are not perfect, and there are uncertainties associated with them,” he adds “That does not mean they don’t provide useful insights.”