Modeling Temperature Data: An Illustration of the Use of Biplots and Bimodels in Non-Linear Modeling

摘要

The paper describes a strategy for exploration and modeling of multivariate data. The data illustrated are temperatures at 50 North, Central and South American stations; the techniques rely heavily on lower rank approximation, biplot and binomial display, least squares fitting, and examination of residuals. The biplot/bimodel display is shown to reveal the relation of temperatures to various geographical features, (No new meteorological findings are claimed--the example is used mainly to illustrate the technique.) More strikingly, the bimodel shows an elliptical configuration of the markers for the different months. The paper shows how this ellipse implies a harmonic model for the temperatures themselves. This model makes good physical sense and its parameters are readily understood in meteorological terms. Some summary comments are made about the usefulness of biplots and bimodels for diagnosing models that may fit the data. (Author)