CO atmospheric concentration modeling using non-parametric regression with non-homogeneous variability bands

Abstract

Air contamination by carbon monoxide (CO) is one of the main factors affecting the air quality in big cities, since  it’s  directly  related  to  urban  activities. The  CO  concentrations  variability  average  behavior  changes constantly mainly due to the traffic in the place. The objective of this article is to propose a non-parametric smoothing model for the hourly CO concentration in the air, considering non-constant variance that allows the description of its behavior through the day. To this end, contamination records by CO in a downtown pollution monitoring station in Cali, Colombia were used. Curves were estimated by using local lineal regression and variance function through an estimator of variance function. The estimated curves allowed describing the CO behavior, showing bigger concentrations in rush hours and smaller concentrations in the early morning, besides the variance function estimation allowed to better model the data’s heteroscedastic behavior.