Modeling and forecasting daily maximum hourly ozone concentrations using the RegAR model with skewed and heavy-tailed innovations
This paper considers the modeling and forecasting of daily maximum hourly ozone concentrations in Laranjeiras, Serra, Brazil, through dynamic regression models. In order to take into account the natural skewness and heavy-tailness of the data, a linear regression model with autoregressive errors and innovations following a member of the family of scale mixture of skew-normal distributions was considered. Pollutants and meteorological variables were considered as predictors, along with some deterministic factors, namely week-days and seasons. The Oceanic Niño Index was also considered as a predictor. The estimated model was able to explain satisfactorily well the correlation structure of the ozone time series. An out-of-sample forecast study was also performed. The skew-normal and skew-t models displayed quite competitive point forecasts compared to the similar model with gaussian innovations. On the other hand, in terms of forecast intervals, the skewed models presented much better performance with more accurate prediction intervals. These findings were empirically corroborated by a forecast Monte Carlo experiment.
KeywordsAdaptative LASSO Air pollution Forecasting Ozone SMSN-RegAR model
The authors thank to Instituto Estadual de Meio Ambiente e Recursos Hídricos of Espírito Santo state for making the data sets used in this paper available. The authors also thank the extremely helpful comments of the associate editor and reviewers, which have improved substantially the quality of the paper.
- Babu GJ, Rao C (2004) Goodness-of-fit tests when parameters are estimated. Sankhya 66(1):63–74Google Scholar
- Johnson N, Kotz S, Balakrishnan N (1994) Continuous univariate probability distributions, vol 1. Wiley, New YorkGoogle Scholar
- Kalbarczyk R, Kalbarczyk E, Niedźwiecka-Filipiak I, Serafin L (2015) Ozone concentration at ground level depending on the content of NOx and meteorological conditions. Ecol Chem Eng S 22(4):527–541Google Scholar
- Pankratz A (2012) Forecasting with dynamic regression models. Wiley, New YorkGoogle Scholar
- Percy K, Legge A, Krupa S (2003) Tropospheric ozone: a continuing threat to global forests? In: Karnosky D, Percy K, Chappelka A, Simpson C, Pikkarainen J (eds) Developments in environmental science: air pollution, global change and forests in the new millennium, Chap. 4, vol 3. Elsevier, Amsterdam, pp 85–117CrossRefGoogle Scholar
- R Core Team (2017) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, https://www.R-project.org/. Accessed 2 Oct 2018
- Seinfeld JH, Pandis SN (2016) Atmospheric chemistry and physics: from air pollution to climate change. Wiley, New YorkGoogle Scholar
- Tibshirani R (1996) Regression shrinkage and selection via the LASSO. J R Stat Soc Ser B (Methodological) 58:267–288Google Scholar