Quantitative assessment of uncertainties for a model of tropospheric ethene oxidation

 

 

Judit Zádor1, Tamás Turányi1, Michael J. Pilling2

 

1 Department of Physical Chemistry, Eötvös Loránd University, H-1518,

Budapest 112, P.O. Box 32, Hungary

2 Department of Chemistry, University of Leeds, Leeds, LS2 9JT, England

 

 

Detailed and comparative uncertainty analysis was carried out on the photochemical degradation model of ethene currently implemented in the Master Chemical Mechanism version 3. The methods of uncertainty analysis were developed for comparison with measurements made in the European Photoreactor (EUPHORE) at Valencia, Spain. There are three sources of error when chemical models are tested against smog chamber measurements: (i) measurement errors, (ii) errors introduced by chamber specific effects and (iii) errors and uncertainties in the model itself. This work aimed to reveal the significance of the different error sources. First, uncertainties for individual rate coefficients and other similar parameters were carefully established. Following the calculation of first-order sensitivity coefficients local uncertainty estimates were evaluated. Globally the importance of the reactions was established by Morris's one-at-a-time design and finally quantitative uncertainties were estimated by means of Monte Carlo simulations.

The concentrations of ozone and formaldehyde are significantly overpredicted. According to the local uncertainty analysis the most important contributor to uncertainty are the HOCH2CH2O2 + NO ® HOCH2CH2O + NO2 and the OH + NO2 ® HNO3 reactions. The Morris’s method revealed the considerable interplay between the rate coefficients as they affect model outputs and that influential reactions tend to influence the system in a nonlinear way. Monte Carlo simulations estimate the 2s uncertainties for ozone as approximately 20 % at the end of the experiment and also around 20 % for the peak formaldehyde concentration. The results suggest systematic disagreement between measurements and model calculations, although the reasons are not entirely understood at the current level of analysis.