Flood hazard and water resources management efforts across the planet would be well served by focusing on regional issues such as urbanization and land use change -- in addition to global warming and greenhouse gas emissions -- for predicting rainfall extremes and developing effective adaptation or mitigation strategies, according to a team of researchers funded by the National Science Foundation's Expeditions in Computing program.

The Expeditions project team, led by principal investigator Vipin Kumar of the University of Minnesota's Department of Computer Science and Engineering, and co-principal investigators including Auroop Ganguly of the civil and environmental engineering department at Northeastern University in Boston, has been developing new data-driven methods or novel adaptations for understanding climate change.

A new study led by Ganguly focused on rainfall extremes in India and identified a steady and significant increase in geographical variability in that country over the past half-century. The data-driven methods used, say the researchers, can be generalized not just to other regions beyond India, but to both observed and model-simulated climate data as well. Traditional physics-based computational models cannot even begin to provide such spatially-explicit insights, they say.

"Rainfall extremes are rather difficult to characterize over space and time, particularly at regional or local scales. However, our current understanding of the geographical patterns of heavy rainfall and their changes over time guides water resources and flood hazards management as well as policy negotiations related to urbanization or emissions control," the researchers note. "Thus, in vulnerable regions of the world where floods may claim many lives and water drives the economy, or in emerging nations which may contribute significantly to the atmospheric inventory of greenhouse gases, major science advances are needed.

"If we were to use India as a case study, we find that top scientists and peer-reviewed publications do not agree on the nature of observed trends in heavy rainfall over the country," they add. "This has led to scientific controversies and uncertainties about adaptation and mitigation strategies in a vulnerable yet rapidly growing region of the world."

Study insights are reported in the journal Nature Climate Change, authored by Ganguly, Subimal Ghosh of the Indian Institute of Technology Bombay, Shih-Chieh Kao of the Oak Ridge National Laboratory (ORNL), and Temple University graduate student Debasish Das. Ghosh was funded by India's Department of Science and Technology; Ganguly, Kao and Das were partially funded by ORNL. The paper appeared as an advance online publication on December 18, 2011, and will appear in the print version in the February 2012 issue.

The study exemplifies how an NSF "Expedition-scale" project can bring together a team of interdisciplinary experts, develop new methods or adaptations which are both statistically rigorous and generalize to large data, and lead to new scientific insights with deep societal relevance.

The suite of data-driven methods required for any specific research may differ in their implementations and need for computational power. Thus, in the Nature Climate Change article, the extreme value theoretic methods by themselves were relatively easy to implement and faster to run on simple desktops, while the corresponding uncertainty characterization methods were more involved and required larger computational capabilities even though the datasets analyzed were relatively small and could fit into a spreadsheet on a desktop. The Expedition project's focus on remotely-sensed observations and climate model simulations is expected to fuel major innovations in computational and data-intensive sciences, such as the research thrust Vipin Kumar is leading, in collaboration with Ganguly and others, in the area of complex networks and graph-based methods for mining climate data.

Earlier research led by Ganguly, formerly of ORNL, also investigated precipitation extremes, but from both models and observations and for predictive insights into the 21st-century. He and Kao authored a Journal of Geophysical Research article earlier this year which showed that while models provide relatively credible predictive insights of precipitation extremes at aggregate spatial scales and over the extra-tropics, the uncertainty begins to grow significantly at localized spatial scales especially over tropical regions. This is where data-driven methods for understanding climate change can help address large science gaps with major societal relevance, they say.

Credits: Peter Kent, College of Engineering, Northeastern University
Correspondence for the Nature Climate Change paper: Auroop R Ganguly, Civil and Environmental Engineering, Northeastern University
Expedition PI and Contact: Vipin Kumar, Department of Computer Science, University of Minnesota