Project at a Glance
Title: Investigation of error propagation in the California Air Resources Board air quality model
Principal Investigator / Author(s): Talat Odman, M
Contractor: Carnegie Mellon University
Contract Number: A932-091
Research Program Area: Atmospheric Processes
An extensive series of tests have been performed to quantify the error and uncertainty propagation in the CALGRID model and its modules. Further tests have been performed on calculating the sensitivity of aspects of the calculations (e.g. chemical kinetics, horizontal transport). Also, coding checks have been conducted both computationally and manually. Recommendations of specific changes have been made, and implemented, to improve and understand model performance (e.g. the chemical ODE solver, the use of a filter in the transport algorithm, vertical transport, etc.). Many of the tests performed are unique to this project. For one, effort has been expended to test system modules within the CALGRID modeling framework when ever feasible. In addition, some new methods to test error and uncertainty propagation have been applied. Also, new tests have been conducted on specific modules as applied to an actual simulation.
It was found that the horizontal transport algorithm used is a source of significant error when concentration gradients are high. This was tested in three ways. First, the standard rotating puff test, with solid body rotation was used. This analysis, and looking at the basic formulation of one-dimensional operators applied to that test, suggested that a more severe test was desirable. An extension of that test was developed and applied to the CALGRID model. A final, very telling, test was to compare three different transport algorithms as applied to the Southern California Air Quality Study (SCAQS) data. Each of these tests suggested that the error arising from the horizontal transport solution could be of the order of 20 to 40%.
Similarly, an in situ test, along with more standard tests and formal sensitivity analysis techniques, was used to quantify the effect of the choice of chemical kinetics solvers was tested. In general, after modifications to the original scheme used in CALGRID, the solvers provided much less error than the transport algorithm chosen.
A unique contribution of this research is the introduction of some new computational procedures for assessing error propagation. These methods, making use of stochastic finite elements, show considerable promise as a way to quantitatively follow the effects of parameter errors. Using the vertical transport code used in CALGRID, it was found that vertical diffusivity errors and uncertainties were more significant than dry deposition and chemical conversion errors.
In summary, CALGRID can be an effective air quality model, with errors typical of most of the photochemical models currently in use. The horizontal transport algorithm is likely the largest source of error and uncertainty propagation, and the use of the non-linear tilter will likely over-diffuse the emissions from some point sources. On the other hand, the transport algorithm is less diffusive than others currently in use. The code itself is relatively portable, though some non-standard FORTRAN statements were found. CALGRID should provide acceptable performance on most computational platforms, in terms of algorithm accuracy and efficiency, for use in typical photochemical air quality model applications.
For questions regarding research reports, contact: Heather Choi at (916) 322-3893
Stay involved, sign up with ARB's Research Email Listserver