Remote sensing statistical ROC assessment of effluent discrimination

In previous works generalized techniques were developed for quantifying the probability of detection and false alarm of atmospheric pollutants. These techniques consisted of the extrapolation of Neyman-Pearson theory for use in non-Gaussian formulation of receiver operator characteristic (ROC) analysis. However, these studies failed to address the probability associated with effluent identification and discrimination. The present work expands upon these earlier works by formulating an approach to quantify the probability of misidentification of effluent gases. It is based upon a Bayesian statistical assessment in the formulation of effluent confusion matrices. These matrices convey information of the probabilistic likelihood of misidentifying a given effluent as something else. The number of potential gases associated with the problem determines the size of these matrices. To better understand the system sensitivity of sensor noise on effluent discrimination performance, a Monte Carlo study has been performed. The technical approach is summarized in this paper along with specific a specific example