Premium
Parametric, non‐parametric and mixed approaches to prediction of sparsely distributed pollution incidents: a case study
Author(s) -
PradaSánchez José Manuel,
FebreroBande Manuel
Publication year - 1997
Publication title -
journal of chemometrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.47
H-Index - 92
eISSN - 1099-128X
pISSN - 0886-9383
DOI - 10.1002/(sici)1099-128x(199701)11:1<13::aid-cem430>3.0.co;2-k
Subject(s) - parametric statistics , semiparametric model , smoothing , nonparametric statistics , computer science , parametric model , series (stratigraphy) , statistics , algorithm , mathematics , econometrics , geology , paleontology
This article discusses the prediction of ground‐level atmospheric sulphur dioxide concentration at multiple points from time series with a 5 min sampling period. For the computational resources at the authors’ disposal (a VAX 6000), parametric Box‐Jenkins methods are unsatisfactory because, given the extreme irregularity of the data, with long sequences of zeros, excessively long sample sequences would have to be used. Non‐parametric methods are satisfactory if one uses two predictor variables, a historical matrix summarizing information on past pollution episodes, and locally defined smoothing parameters (bandwidth), although a slight sacrifice of accuracy is necessary in order to avoid the computational burden of recalculating local smoothing parameters after every fresh sample acquisition. A mixed (‘semiparametric’) method in which the non‐parametric estimation of a general time series is assumed to generate residues conforming to a parametric model was found to afford best all‐round performance. The combination of a semiparametric approach with the use of a historical matrix for the non‐parametric component may be generalizable to the prediction of abnormally high or low values in other time series. © 1997 John Wiley & Sons, Ltd.