In an analysis of the best available Earth systems models, Northeastern researchers found that by the turn of the next century, 850 million people will feel the effects of declining runoff from the world’s major rivers.
That is more than three times the number estimated by previous analysis of Earth system models, says Puja Das, a post-doctoral research fellow at AI for Climate and Sustainability within the Institute for Experiential AI at Northeastern University.
“What (Das) found is that some of the more skillful models seem to project the worst conditions in terms of the impact of water scarcity,” says Northeastern professor Auroop Ganguly, who served as co-author for the paper Das published recently in the journal npj Climate and Atmospheric Science.
Population estimates are important because they give policymakers an idea of what to expect in terms of the availability of food, water and energy, since river runoff recharges water supplies, enriches agricultural soil and generates hydropower, Das says.
Earth systems models are complex computer simulations of Earth’s processes, such as the atmosphere, oceans and human activity.
Her research shows that the five most skillful models project 40% of the world’s 30 major rivers will exhibit decreased runoff by 2100, affecting a population 100 times the size of New York City, as opposed to earlier estimates of 260 million.
“We chose the 30 biggest river basins around the world, including the Amazon, Congo, Ganges, Brahmaputra and Nile rivers,” she says. “We were trying to see how the runoff in those river basins, or water availability in those river basins, are presented in climate models.”
“We know that climate models use different equations and parameterizations to estimate these variables. We are trying to see how good they are,” she says.
The researchers compared two generations of Coupled Modeling Intercomparison Projects, the CMIP5 and the more recent CMIP6, to see how they performed against historical projections of annual runoff from 1960 to 2005.
The latter modeling system, the CMIP6, was more skillful and accurate, Das says. Applying it to the future, “we found that the more skillful models are finding a worse future scenario in terms of water availability.”
For one thing, the latest generation of models has higher resolution — one data point for every 100 kilometers instead of every 500 kilometers, she says.
The CMIP6 also did a better job of incorporating comprehensive physics, such as physics of land, ocean and ice into climate model equations, Das says.
And it performed better when it came to making mathematical equations, known as parameterizations, out of events such as cloud formation and convection.
“There are some critical parameterizations that need to be correct. We saw that the models that use those parameterizations are performing well,” says Das, a recent Ph.D. graduate from the Sustainability and Data Sciences Laboratory within Northeastern’s department of civil and environmental engineering.
Higher resolutions, more intricate parameterizations and comprehensive physics are hypothesized to improve model projections, but this is not guaranteed until the models are thoroughly evaluated against observations (skills) and against each other (consensus), she says.
Furthermore, Das says model-based uncertainty bounds may actually increase in some cases, even as understanding and projection performance improve.
“We try to look at all these kinds of metrics based on skills and consensus and that’s what (Das) has done,” says Ganguly, Northeastern distinguished professor of civil and environmental engineering.
“She’s saying that higher resolution, better parameterization and more physics components do add value,” he says.
The researchers also ran the models against five different carbon emission scenarios.
“We saw that if there is a greener world, the water availability will be higher and fewer people will be impacted because of the decrease in water availability,” Das says.
With lower carbon emissions, she says, “We found that 500 million people (would be affected) instead of 900 million people, but water availability will still decrease in certain parts of the world.”
Das says the research is important for two communities: policymakers and water resources managers who use Earth systems model results for understanding impacts and informing adaptation, as well as natural scientists, data scientists and computational modelers who build the Earth systems models and analyze the results.