NSF Awards Samorodnitsky a Grant to Improve Climate Models

Extreme weather events such as heat waves, droughts, and heavy rains are always in the news. They have a very significant impact on society in general and agriculture in particular. However there is evidence that large scale mathematical models currently used to study current and future climate do not incorporate sufficient variability to accurately predict the frequency of such extreme events. Climate scientists have now proposed methods to improve the models.

ORIE Professor Gennady Samorodnitsky teamed up with Cornell colleagues from Civil and Environmental Engineering (CEE) and Earth and Atmospheric Sciences (EAS), as well as from the National Center for Atmospheric Research (NCAR), to propose an approach to evaluate one such improvement. The National Science Foundation recently awarded them $510K for a three year project to pursue this approach. Samorodnitsky is the Principal Investigator on the project, with Dr. Judith Berner (NCAR) and Professors Mircea Grigoriu (CEE) and Natalie Mahowald (EAS). Success could help predict the frequency and hopefully mitigate the effects of extreme weather events.

Felipe Tagle, a Ph.D. student in Operations Research, is working full time on the project. He points out that NCAR's widely used Community Climate System Model (CCSM) incorporates the complex dynamics of the atmosphere, land, ocean, and sea ice - and their interactions - in simulations of weather spanning the entire globe up to the resolution of a 1 degree by 1 degree grid. By changing the input conditions,CCSM can simulate both past and current weather patterns and more importantly, can examine possible future climate scenarios. However available computational power constrains the size of the grid.

Tagle says that "extreme weather events can already be seen in output from the Community Atmosphere Model (CAM), which is the atmospheric component of CCSM, But the frequency and magnitude of such events may be misestimated since the CAM does not capture processes that occur at so-called 'sub-grid' level. These are processes that originate within the grid squares and are essential to accurately forecast extreme events."

Tagle, who has a BS degree in Industrial Engineering from Pontificia Universidad Catolica de Chile, first came to Cornell as a Master of Engineering student in ORIE. After completing that degree in 2008 he took the unusual step of undertaking a Ph.D. in ORIE.

Some climate scientists propose to get around computational limitations and obtain better estimates of the frequency of extreme events by introducing randomness within the CAM in a physically consistent way, feeding the model with historic data, and tuning the randomness to replicate the historic frequency of extreme events. Thus enhanced, the model can be run over multiple time periods and with various input assumptions to generate histograms that plausibly predict the distribution of extreme events in the future. One such proposal to improve the estimation is to use a technique called "stochastic kinetic energy backscattering" as a randomizing mechanism that is consistent and controllable within the model.

Central to Samorodnitsky's project is the use of a statistical approach, called extreme value theory (EVT), that Samorodnitsky and ORIE professor Sidney Resnick have studied extensively. EVT will be used to determine whether techniques such as backscattering improve the estimation of the frequency distribution of extreme weather events occuring in reality and in the model. Since extreme events (such as 100 year floods or so-called 'black swans' occurring in finance and elsewhere) are extremely rare, EVT analysis is a daunting statistical challenge. Even more difficult is the prediction, not only of individual extreme events, but of connections between different extreme events, according to the grant proposal.

Samorodnitsky and his colleagues on the NSF proposal have also teamed up with Professor Peter Hess of Biological and Environmental Engineering as Principal Investigator to propose analyzing "extreme event impacts on air quality and water quality with a changing global climate" for the Environmental Protection Agency. This project will look at models of ozone and aerosols over the United States, in order to provide insights concerning the probability, frequency, duration and severity of high pollution episodes in relation to changes in emissions and climate.  It is expected to be funded at a higher level than the NSF-supported project.