In 2007, the Intergovernmental Panel on Climate Change released its fourth assessment report based on a new group of climate models, the value of which earned the IPCC the Nobel Peace Prize. The new models—collectively referred to as CMIP3—revealed information about climate change that sealed the deal on a number of lingering questions: Is global temperature increasing? Yes. Is it caused by humans? Almost certainly. But one major question—perhaps the most important one of all—remained. What should we do about it?
To answer that question, stakeholders would need more locally relevant information on much shorter time scales. So when the next generation of climate models was released in 2012, scientists had high hopes. These models included more physical, chemical, and biological processes, often considered at a much finer-grain than the previous models. Surely CMIP5, as the new ensemble of models is called, would provide credible projections at the scales relevant to stakeholders.
But a new study from the lab of Northeastern University associate professor of civil and environmental engineering Auroop Ganguly tells a different story. According to the team’s analysis, which will be published in the journal Climate Dynamics, the CMIP5 models don’t do a better job. In fact, some projections are even worse than the CMIP3 models that were released in 2007.
“Some of the models say an area will be wetter in the future, others say it’ll be drier,” said Devashish Kumar, a doctoral candidate in Ganguly’s Sustainability and Data Sciences laboratory and lead author on the paper. “So which is it and which should we plan for?”
A handful of earlier studies made similar claims prior to the release of the CMIP5 results. But unlike Northeastern’s study, most of that work was anticipatory, since the actual data weren’t yet available.
While the results are sobering for a field that has, in recent years, relied on the accuracy of its models to make the most profound impact statements, Ganguly’s team does not think climate modeling has necessarily reached its end. It’s just that “these models may not be able to keep pace with the urgency with which the stakeholders require the problem to be solved,” he explained.
That’s because of longstanding gaps in our understanding of the science in areas such as cloud physics and ocean-land-atmosphere interactions, Ganguly said. While our understanding in these areas may eventually get better, the timeline will be too slow for policy makers to wait around. Additionally, climate and earth system models have intrinsic variability, which may never be overcome, “but still need to be characterized,” said Ganguly.
Instead, he is advocating for the use of Big Data tools to advance the field. Between observed data from various types of sensors stationed around the globe and even in space, and the data coming from the simulation models themselves, climate scientists are treading a veritable flood of data.
But climate is a complex field. A change in one variable can cause ripple effects throughout the entire system, a phenomenon that poses challenges to the scientists studying it. For that reason, Ganguly said, the data deluge is only being used in pockets of the field.
For instance, researchers interested in rainfall extremes look at rainfall data itself as well as data they already know may influence these extremes, such as specific atmospheric and sea surface temperatures. However, a vast treasure trove of complex data—which could also yield important insights—lies unexplored. As Ganguly put it, “We need to look at it all as an integrated whole.”
In another upcoming paper, which will appear in the journal Nonlinear Processes in Geophysics, Ganguly and doctoral candidate Evan Kodra collaborated with a large team of climate scientists, hydrologists, computer scientists, and statisticians to further address this question. This work is part of a multi-institution five-year, $10 million grant from the National Science Foundation.
While the researchers believe models will remain invaluable to the field, they argue that an optimal blend of data-driven insights and physical understanding—beyond what may be easily captured within the current generation of climate models—may be a path forward. For this to happen, Big Data tools need to be customized for complex climate data, and optimized for extremes to characterize what may ultimately be small data, from possibly elusive indicators.
“Ultimately, future projections will have to be based on models,” said Ganguly. “And actionable predictive insights will need to be generated based on such projections. But physics-guided data mining may need to work in tandem to inform stakeholder decisions.”