Articles | Volume 22, issue 1
Nonlin. Processes Geophys., 22, 33–46, 2015
https://doi.org/10.5194/npg-22-33-2015

Special issue: Physics-driven data mining in climate change and weather...

Nonlin. Processes Geophys., 22, 33–46, 2015
https://doi.org/10.5194/npg-22-33-2015

Research article 13 Jan 2015

Research article | 13 Jan 2015

On the data-driven inference of modulatory networks in climate science: an application to West African rainfall

D. L. González II1,2, M. P. Angus1, I. K. Tetteh1, G. A. Bello1,2, K. Padmanabhan1,2, S. V. Pendse1,2, S. Srinivas1,2, J. Yu1, F. Semazzi1, V. Kumar3, and N. F. Samatova1,2 D. L. González II et al.
  • 1North Carolina State University, Raleigh, NC 27695-8206, USA
  • 2Oak Ridge National Laboratory, P.O. Box 2008, Oak Ridge, TN 37831, USA
  • 3University of Minnesota, Minneapolis, MN 55455, USA

Abstract. Decades of hypothesis-driven and/or first-principles research have been applied towards the discovery and explanation of the mechanisms that drive climate phenomena, such as western African Sahel summer rainfall~variability. Although connections between various climate factors have been theorized, not all of the key relationships are fully understood. We propose a data-driven approach to identify candidate players in this climate system, which can help explain underlying mechanisms and/or even suggest new relationships, to facilitate building a more comprehensive and predictive model of the modulatory relationships influencing a climate phenomenon of interest. We applied coupled heterogeneous association rule mining (CHARM), Lasso multivariate regression, and dynamic Bayesian networks to find relationships within a complex system, and explored means with which to obtain a consensus result from the application of such varied methodologies. Using this fusion of approaches, we identified relationships among climate factors that modulate Sahel rainfall. These relationships fall into two categories: well-known associations from prior climate knowledge, such as the relationship with the El Niño–Southern Oscillation (ENSO) and putative links, such as North Atlantic Oscillation, that invite further research.

Download
Short summary
We applied coupled heterogeneous association rule mining (CHARM), Lasso multivariate regression, and dynamic Bayesian networks to find relationships within a complex system, and explored means with which to obtain a consensus result from the application of such varied methodologies. Using this fusion of approaches, we identified relationships among climate factors that fall into two categories: well-known associations from prior knowledge, and putative links that invite further research.