Premium
Analysis of PM 10 air pollution in Brno based on generalized linear model with strongly rank‐deficient design matrix
Author(s) -
Veselý Vítězslav,
Tonner Jaromír,
Hrdlivčková Zuzana,
Michálek Jaroslav,
Kolář Miroslav
Publication year - 2009
Publication title -
environmetrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.68
H-Index - 58
eISSN - 1099-095X
pISSN - 1180-4009
DOI - 10.1002/env.971
Subject(s) - rank (graph theory) , air pollution , matrix (chemical analysis) , statistics , generalized linear model , linear model , stability (learning theory) , mathematics , basis (linear algebra) , meteorology , econometrics , covariate , computer science , geography , combinatorics , chemistry , materials science , geometry , organic chemistry , machine learning , composite material
Abstract An analysis of air pollution by suspended particulate matter (PM 10 ) in Brno, the second largest urban agglomeration of the Czech Republic, based on generalized linear model (GLM) is presented. Average daily concentrations coming from PM 10 monitoring for the period 1998–2005 have been processed. The measured meteorological factors: air temperature and humidity, direction and wind speed were considered as covariates along with some additional seasonal factors. Three standard and six GLMs with strongly rank‐deficient design matrix have been applied. The rank deficiency is due to overparameterization which allows one more precise modeling involving, among others, identification of significant air pollution sources (PSs). From each of them the parameter estimates were obtained using both standard estimation procedure and a new sparse parameter estimation technique based on a four‐step modification of the basis pursuit algorithm originally suggested for time‐scale analysis of digital signals. As the standard estimation algorithms often fail due to numerical instability caused by strong overparameterization, we have applied this new computationally intensive approach allowing us to reliably identify nearly zero parameters in the model and thus to find numerically stable sparse solutions. The goal of the analysis was to identify the model and algorithm yielding most precise 1‐day forecasts of the level of pollution by PM 10 with regard to the meteorological and seasonal covariates. Copyright © 2009 John Wiley & Sons, Ltd.