Utilizing wind direction and wind speed as independent variables in multilinear receptor modeling studies

The problem of identifying sources of airborne pollutants and providing quantitative estimates of the contributions of each of those sources is important for airborne particulate matter. Various forms of factor analysis have been applied to this problem. However, in factor analysis, there is the fundamental problem of rotational ambiguity that makes the problem ill-posed. Thus, the incorporation of additional information can be useful in improving the solutions. Especially for identifying local sources, wind data (direction and speed) could be valuable additional information in such receptor modeling. However, wind data cannot be used directly as dependent variables in factor analytic modeling because the dependence of observed concentrations on wind variables is far from linear. An expanded multilinear model has been developed in which the wind direction, speed and other variables are included as independent variables. For each source, the analysis computes a directional profile that indicates how much of the concentrations are explained by the factors depending on wind direction, speed, and other values. This model has been tested using simulated data developed by the U.S. Environmental Protection Agency as part of a workshop to test advanced factor analysis methods. For most of the local sources, well-defined directional profiles were obtained.