Online Program

Fitting a GAM to Estimate Hourly Ozone Levels in the Air from Climate Variables (302782)

Keywords: Linear models, Smoothing, Regression, Nonparametrics, Splines

We use climate variables to construct a generalized additive model to estimate hourly ozone levels in the air. Instead of fitting a linear model using the original variables, or fitting a linear model using functions of the original variables picked by experts, we fit an additive model using data-driven functions obtained from one-dimensional smoothing techniques. For illustrative purposes, we fit a model using the ozone levels as the response. Then, we fit another model using as a response a dichotomous variable indicating whether a particular hourly ozone level has been reached. It highlights the fact that GAM models are useful for predicting a quantitative response, but also a non-numerical one. On both models, we used as predictors the covariates solar radiation, relative humidity, temperature, and wind speed. Smooth functions were based on spline smoothing. We show the procedure in such a way that potential users may be able to reproduce it step-by-step on their own projects. Computations and figures were done using The R Project for Statistical Computing. Data come from a monitoring station of the Air Quality Surveillance System at Cali, Colombia.