THE EVALUATION OF ACID DEPOSITION MODELS USING PRINCIPAL COMPONENT SPACES

An analytical technique involving principal component analysis is proposed for use in the evaluation of acid deposition models. Relationships among model predictions are compared to those among measured data, rather than the more common one-to-one comparison of predictions to measurements. A multidimensional comparison of principal component spaces, using general methodology developed by Krzanowski (J. Am. statist. Ass. 74, 703-707), is the basis of the technique. The technique is illustrated using concentrations measured by aircraft over the eastern United States of America. Model predictions are generated by the Regional Acid Deposition Model (RADM). A simple example provides motivation for the technique. More detailed applications are shown for systems of concentrations involved in the creation of sulfur and nitrogen deposition, respectively. Orthogonal vectors describing the two-dimensional principal component spaces are calculated for measurements and predictions, such that the angles between measured vector i and predicted vector i (i = 1, 2) are minimized. Using one of the two approaches that are presented (common scaling factors), the agreement between measured and RADM-predicted spaces is generally good in the sulfur system, with angles of 4.5 and 11.4-degrees. Agreement is weaker in the nitrogen system (17.2 and 27.7-degrees). Issues of interpretation and potential extensions of the technique's applicability are discussed.