Forecasting carbon monoxide concentrations near a sheltered intersection using video traffic surveillance and neural networks.

In this study, the authors investigated the relationships between traffic and carbon monoxide (CO) concentrations measured near an intersection that is sheltered from the wind by multistory buildings. Detailed information on traffic parameters was obtained by video camera technology during a single, 4-hour period (3PM to 7PM) of calm winds (<1 m/s). A neural network was trained with both lane specific traffic information as well as on-site wind parameters and used with an independent test set to predict one-minute average CO concentrations with reasonable accuracy (0.68 < R squared < 0.8). Standard linear regression models as well as an EPA dispersion model (CAL3QHC) could not reliably predict CO levels from the same data set (0 < R squared < 0.28). (A) This paper was first published by the Civil Engineering Department of Washington University.