This study highlights the advantage of functional data approach in assessing and comparing the PM10 pollutant behaviour as an alternative statistical approach during and between the two extreme haze years (1997 and 2005) that have been reported in Selangor, state of Malaysia. The aim of the study was to improvise the current conventional methods used in air quality assessment so that any unforeseen implicit information can be revealed and the previous research findings can be justified. An analysis based on the daily diurnal curves in place of discrete point values was performed. The
analysis results provided evidences of the influence of the change in the climate (due to the El-Nino event), the different levels of different emission sources and meteorological conditions on the severity of the PM10 problem. By means of the cummulative exceedence index and the functional depth method, most of the monitoring stations for the year 2005 experienced the worst day of critical exceedences on the 10th of August, while for the year 1997 it occurred between 13th and 26th September inclusively at different dates among the stations.
In most research including environmental research, missing recorded data often exists and has become a common problem for data quality. In this study, several imputation methods that have been designed based on the techniques for functional data analysis are introduced and the capability of the methods for estimating missing values is investigated. Single imputation methods and iterative imputation methods are conducted by means of curve estimation using regression and roughness penalty smoothing approaches. The performance of the methods is compared using a reference data set, the real PM10 data from an air quality monitoring station namely the Petaling Jaya station located at the western part of Peninsular Malaysia. A hundred of the missing data sets that have been generated from a reference data set with six different patterns of missing values are used to investigate the performance of the considered methods. The patterns are simulated according to three percentages (5, 10 and 15) of missing values with respect to two different sizes (3 and 7) of maximum gap lengths (consecutive missing points). By means of the mean absolute error, the index of agreement and the coefficient of determination as the performance indicators, the results have showed that the iterative imputation method using the roughness penalty approach is more flexible and superior to other methods.