For my spatial problem I will examine the role of spatial autocorrelation and seasonality in developing a land use regression (LUR) model. In particular I am interested in optimizing the incorporation of spatial autocorrelation and seasonality for prediction of air pollution in the City of Eugene.

For those unfamiliar with a LUR, it essentially combines GIS variables that are predictive of air pollution concentrations along with actual air pollution measurements in order to predict air pollution at unmonitored locations using ordinary least squares (OLS) regression. The problem with a typical LUR model is that they don’t account for spatial autocorrelation. The value of accounting for spatial autocorrelation is due to the fact that spatially based data, such as air pollution, is typically spatially correlated.

This past quarter in my GEO580 course I developed a LUR that did account for spatial autocorrelation by modeling the covariance of air pollutant concentrations of adjacent zip code boundaries, using a spatial CAR model. For this class I wish to develop this idea even further by using multiple techniques, namely geographically weighted regression (GWR), a spatial CAR model, and OLS to compare the model results to actual air pollution measurements. This work will require me to use both ArcGIS spatial analyst toolbox and the R statistical software.

As mentioned above, I am interested in including seasonal trends in air pollutant variation in order to see if inclusion of seasonal variation is capable of improving model estimates. To do this I propose to incorporate seasonal ratios to annual ratios of air pollutant concentrations.

To keep this work focused I will use data on just one air pollutant, as opposed to last quarter wherein I developed a LUR for seven different pollutants. By focusing on just one pollutant I hope to keep the work efficient and effective toward achieving my goals in this class. Ideally, this work will help to inform my dissertation proposal work.