Verification of image processing based visibility models

Abstract

Methods are presented for testing visibility models that use simulated photographs to display results of model calculations. An experimental protocol is developed and used to obtain input data including standard photographs of chosen scenes on a clear day and during a smog event at Pasadena, CA. With the clear day photograph as a substrate, pollutant properties measured on the smoggy day are introduced into the visibility model, and results of the model calculations are displayed as a synthetic photograph of the expected appearance of the smog event. Quantitative comparisons are made between the predicted and actual appearance of the smog event. Diagnostic techniques developed are applied to the visibility modeling procedure proposed by Malm et al. That model is shown to reproduce the contrast reduction characteristic of urban air pollution but produces synthetic photographs with sky elements that differ substantially from a real photograph of the actual smog event.