UTK Daily Beacon, Robbie Hargett, August 31, 2010,
Researchers from UT and ORNL are participating in a new multi-institutional project dedicated to predicting climate change more accurately.
The project uses data mining, the process of discovering patterns in data, to " ... discover hidden patterns among model-simulated variables that are relatively better predicted and establish their relations with those that are not, with the goal of improving predictions of the more crucial variables," Auroop Ganguly, senior staff member in ORNL's Computational Sciences and Engineering Division and adjunct professor at UT, said.
While some climate variables, such as temperatures over atmospheres and oceans, are relatively easy to predict with models, more extreme variables, such as hurricanes, are not.
"Turns out that some of the less well predicted model variables may be more important for climate change impacts and policy," Ganguly said. "Climate change policies and decisions could impact the nation's and the world's critical infrastructures and key resources."
Ganguly also noted that many climate models are able to reliably predict change from continental to global levels but become less accurate at local and regional levels.