Sayang Mohd Deni

Anomaly detection and assessment of PM10 functional data at several locations in the Klang Valley, Malaysia

Abstract In environmental data sets, the occurrence of a high concentration of an unusual pollutant, more formally known as an anomaly, may indicate air quality problems. Thus, a critical understanding of the behavior of anomalies is increasingly becoming very important for air pollution investigations. This study was conducted to detect anomalies in daily PM 10 functional data, to investigate the patterns of behavior as well as to identify possible factors that determine PM 10 anomalies at three selected air quality monitoring stations (Klang, Kuala Selangor and Petaling Jaya) in the Klang Valley, Malaysia. The statistical method employed to detect these anomalies consisted of a combination of the robust projection pursuit and the robust Mahalanobis distance methods using air quality data recorded from 2005 to 2010. Analysis of obtained anomalous PM 10 profiles showed that data recorded during El Nino years (2005, 2006 and 2009) contained the highest frequency of anomalies. More frequent anomalies appeared during the southwest (SW) monsoon which occurs in the months of July and August as well as during the northeast (NE) monsoon in February. A lesser number of anomalies were also observed during weekends compared to weekdays. The weekend and monsoonal effect phenomena were shown to be significantly existent at all stations while wind speed was positively associated with extreme PM 10 anomalies at the Klang and Petaling Jaya stations. In conclusion, anomalies detection was found useful for air pollution investigation in this study. The findings of this study imply that the location and background of a station, as well as wind speed, seasonal (monsoon) and weekdays-weekend variations play important role in influencing PM 10 anomalies.

Improved estimation of radar rainfall bias over Klang River Basin using a Kalman Filtering approach

Bad weather, consisting of thunderstorms, normally causes the presence of strong winds and heavy rain that may develop into a storm over a certain area. Radar has been the most potential and powerful instrument used to detect and monitor the development of thunderstorms over a large area; however, it also has certain weaknesses. Weather radar can be affected by different sources of errors, which have to be well considered and quantified for a proper interpretation of the collected data. We design a method that combines the Kalman Filter with a multivariate analysis technique. The implementation of this technique is for the purpose of developing a formulation that may help to reduce error. These studies involved parameters such as temperature, humidity, point of gauge rainfall, and weather radar reflectivity. The approach of using the Kalman Filter combined with multivariate analysis is still a new way to improve radar rainfall estimates by prediction (time update) and correction (measurement update). This particular research was developed purposefully to reduce radar rainfall bias due to the uncertain sources of error seen in the weather radar, and many studies have been developed, but still did not achieve suitable values between radar readings with rain gauge returns.

Geometric median for missing rainfall data imputation

Missing data is a common problem faced by researchers in environmental studies. Environmental data, particularly, rainfall data are highly vulnerable to be missed, which is due to several reasons, such as malfunction instrument, incorrect measurements, and relocation of stations. Rainfall data are also affected by the presence of outliers due to the temporal and spatial variability of rainfall measurements. These problems may harm the quality of rainfall data and subsequently, produce inaccuracy in the results of analysis. Thus, this study is aimed to propose an imputation method that is robust towards the presence of outliers for treating the missing rainfall data. Geometric median was applied to estimate the missing values based on the available rainfall data from neighbouring stations. The method was compared with several conventional methods, such as normal ratio and inverse distance weighting methods, in order to evaluate its performance. Thirteen rainfall stations in Peninsular Malaysia were selecte...

The construction of control chart for PM10 functional data

In this paper, a statistical procedure to construct a control chart for monitoring air quality (PM10) using functional data is proposed. A set of daily indices that represent the daily PM10 curves were obtained using Functional Principal Component Analysis (FPCA). By means of an iterative charting procedure, a reference data set that represented a stable PM10 process was obtained. The data were then used as a reference for monitoring future data. The application of the procedure was conducted using seven-year (2004–2010) period of recorded data from the Klang air quality monitoring station located in the Klang Valley region of Peninsular Malaysia. The study showed that the control chart provided a useful visualization tool for monitoring air quality and was capable in detecting abnormality in the process system. As in the case of Klang station, the results showed that with reference to 2004–2008, the air quality (PM10) in 2010 was better than that in 2009.

Data preparation for functional data analysis of PM10 in Peninsular Malaysia

The use of curves or functional data in the study analysis is increasingly gaining momentum in the various fields of research. The statistical method to analyze such data is known as functional data analysis (FDA). The first step in FDA is to convert the observed data points which are repeatedly recorded over a period of time or space into either a rough (raw) or smooth curve. In the case of the smooth curve, basis functions expansion is one of the methods used for the data conversion. The data can be converted into a smooth curve either by using the regression smoothing or roughness penalty smoothing approach. By using the regression smoothing approach, the degree of curve’s smoothness is very dependent on k number of basis functions; meanwhile for the roughness penalty approach, the smoothness is dependent on a roughness coefficient given by parameter λ Based on previous studies, researchers often used the rather time-consuming trial and error or cross validation method to estimate the appropriate numbe...

Generalized linear model for estimation of missing daily rainfall data

The analysis of rainfall data with no missingness is vital in various applications including climatological, hydrological and meteorological study. The issue of missing data is a serious concern since it could introduce bias and lead to misleading conclusions. In this study, five imputation methods including simple arithmetic average, normal ratio method, inverse distance weighting method, correlation coefficient weighting method and geographical coordinate were used to estimate the missing data. However, these imputation methods ignored the seasonality in rainfall dataset which could give more reliable estimation. Thus this study is aimed to estimate the missingness in daily rainfall data by using generalized linear model with gamma and Fourier series as the link function and smoothing technique, respectively. Forty years daily rainfall data for the period from 1975 until 2014 which consists of seven stations at Kelantan region were selected for the analysis. The findings indicated that the imputation me...

Predicting the “graduate on time (GOT)” of PhD students using binary logistics regression model

Malaysian government has recently set a new goal to produce 60,000 Malaysian PhD holders by the year 2023. As a Malaysia’s largest institution of higher learning in terms of size and population which offers more than 500 academic programmes in a conducive and vibrant environment, UiTM has taken several initiatives to fill up the gap. Strategies to increase the numbers of graduates with PhD are a process that is challenging. In many occasions, many have already identified that the struggle to get into the target set is even more daunting, and that implementation is far too ideal. This has further being progressing slowly as the attrition rate increases. This study aims to apply the proposed models that incorporates several factors in predicting the number PhD students that will complete their PhD studies on time. Binary Logistic Regression model is proposed and used on the set of data to determine the number. The results show that only 6.8% of the 2014 PhD students are predicted to graduate on time and the results are compared wih the actual number for validation purpose.

Identification of Seasonal Rainfall Peaks at Kelantan Using Fourier Series

This study aimed to identify the wet period, peaks of mean rainfall amount per rainy day at Kelantan River basin. Daily rainfall data from 12 selected rainfall stations for the period of 1975–2014 which comprises of three main areas such as inland, river, and coastal are analyzed. The estimated wet period, date, and value at peak of mean rainfall amount per rainy day will be identified using Fourier series and compared with the actual flood events in 2013/2014 and 2014/2015. The findings indicate that duration of wet periods obtained from the results of best fitting justifies the flood event occurred most recently. Moreover, the coastal area of Kelantan also shows the highest probability of rainfall amounts exceeding 60 mm.

Investigating the capability of a control chart based on functional data for daily PM10 monitoring

A control chart based on functional data (FD) has been proposed to be used as a tool for detecting anomalies and for assessing the trend of daily PM10. In this study, the capability of the FD control chart is investigated. The performance of the FD control chart is compared to the control charts based on the average (AV) and multivariate (MV) data. Daily PM10 indices for AV control chart are the average values while the indices for the MV and FD control charts are computed based on the Principal Component Analysis (PCA) model. The experimentation is conducted using real PM10 data from the Shah Alam air quality monitoring station located at the west of Peninsular Malaysia to investigate the performance of the control charts. The results of the first stage analysis have shown that the FD control chart outperforms the AV and MV control charts in detecting PM10 anomalies of extreme levels. Using a similar number of principal components, it is also found that the PCA model based on FD is able to capture more i...