THE TREATMENT OF ATMOSPHERIC DISPERSION DATA IN THE PRESENCE OF NOISE AND BASE-LINE DRIFT

Recently large quantities of dam from many different field experiment have become available to facilitate the examination of various proposed models of atmospheric dispersion. How ever. these data sets are invariably corrupted by some form of random noise and, also, significant baseline drift is often recorded. Consequently, these problems require careful consideration and treatment before the data can be used in model testing. In many papers, the noise is simply treated by 'thresholding' but this is unacceptable since many valid readings are discarded. This paper examines the performance of two different noise removal methods;hat are more: soundly based. both physically and mathematically The first is a Wiener filter with certain modifications. and the second is a maximum entropy inversion technique. A comparison of the results of applying these methods is presented, with the emphasis on the practical treatment of the numerical and computational problems that arise. The problem of baseline drift is treated initially by applying a number of subjective tits, Subsequently the noise removal methods are applied. In general, it is found that the maximum entropy method performs better than the Wiener filter for the data sets examined here.