ARB Research Seminar
This page updated June 19, 2013
Non-Linearity in Atmospheric Response: A Sensitivity Analysis Approach
Amir Hakami, Ph.D. Candidate, Department of Civil and Environmental Engineering, Georgia Institute of Technology
April 12, 2002
Cal EPA Headquarters, 1001 "I" Street, Sacramento, CA
Atmospheric models (most importantly photochemical models) may exhibit a non-linear response, mostly due to non-linear chemistry. This behavior cannot be captured by first-order (linear) sensitivity analysis. A direct sensitivity analysis technique [Decoupled Direct Method (DDM)] is extended to calculate the local second, and higher order sensitivity coefficients in three dimensional air quality models. The time evolution of sensitivity coefficients of different order is followed alongside that of the concentrations. Calculation of higher order sensitivity coefficients requires few modifications to the original (first order) sensitivity modules, and is carried with minimal computational overhead. By using higher order sensitivity coefficients, the non-linear responses are better understood and described.
The modeling results from the Multi-scale Air Quality Simulation Platform (MAQSIP), and for the SARMAP domain and August 1990 episode, are shown and discussed. Second order sensitivity coefficients of ozone concentration with respect to domain-wide NO emissions show reasonable agreement with brute force results and exhibit less noisy behavior. For a Taylor series projection from the base case, including the higher order terms (up to fourth order sensitivity coefficients) improves the accuracy. Most of the improvement is achieved by only incorporating the second order sensitivity terms. The same conclusion can be made about higher order cross sensitivity coefficients, i.e. sensitivity coefficients with respect to more than one independent variable. In general, higher order sensitivity analysis shows a noticeable improvement in terms of accuracy over the conventional first order analysis. Of particular interest, higher order sensitivity analysis is better equipped to address the non-linear behavior around the peak ozone in NOX-rich plumes, where first order analysis is least accurate. The method is also useful in three dimensional uncertainty analysis of organic reactivities.
Amir Hakami is a last year Ph.D. student at the department of civil and environmental engineering, Georgia Institute of Technology. His research interests lie in different aspects of air quality modeling and sensitivity analysis techniques. Mr. Hakami has a bachelor's degree in chemical engineering from Polytechnic of Tehran, and a master's degree in environmental engineering from Georgia Tech. He has been involved in two international projects for air quality improvement in Tehran. As a graduate student, he has investigated the uncertainty associated with episode selection in regional air quality modeling. His doctoral research focuses on formal sensitivity analysis of air quality models in general and development of high order sensitivity analysis techniques in particular. Applying a direct sensitivity analysis method, Mr. Hakami has worked on three dimensional reactivity assessment projects for central California and the eastern United States.