Advances in Environmental Biology

Advances in Environmental Biology, 8(24) December 2014, Pages: 244-256

AENSI Journals

Advances in Environmental Biology ISSN-1995-0756

EISSN-1998-1066

Journal home page: http://www.aensiweb.com/AEB/

Spatial Assessment of Air Pollution Index Using Environ Metric Modeling Techniques Hamza AhmadIsiyaka, Hafizan Juahir, Mohd Ekhwan Toriman, Barzani M. Gasim, Azman Azid, Mohd Khairul Amri, Aminu Ibrahim, Usman Nasiru Usman, Auwalu Rabiu Ali Rano, Musa A. Garba East Coast Environmental Research Institute, Faculty of Bioscience & Food Technology, Universiti Sultan Zainal Abidin Malaysia ARTICLE INFO Article history: Received 26 September 2014 Received in revised form 25 November 2014 Accepted 21 December 2014 Available online 23 January 2015 Keywords: Multivariate techniques; Artificial neural network; Multiple linear regression; Air pollution index.

ABSTRACT The quest for industrial and urban development over the years has changed the pattern and source of atmospheric air pollution in Malaysia. This study aims to investigate the spatial variation in the source of air pollution; identify the most significant parameterscontributing to the air pollution and to develop the best receptor model for predicting air pollution index (API). Data from five monitoring stations base on five year's observation (2007-2011) were used. Multivariate techniques such as cluster analysis (HACA), discriminate analysis (DA), principal component analysis (PCA), factor analysis (FA) and modeling techniques comprising of artificial neural network (ANN) and multiple linear regression (MLR) were used in this study. HACA was able to group the five monitoring stations into three clusters, indicating that one station in each cluster can provide a reasonable accurate spatial assessment of air quality within the study area. The result for standard mode, forward stepwise and backward stepwise DA gave a correct assignation of 82.37% (p˂ 0.05) which indicate that all the parameters significantly discriminate spatially. PCA and FA for the three clusters account for more than 62%, 56% and 58% of the total variance respectively indicating that the source of air pollution are from anthropogenic induced point source and nonpoint source. ANN gave a better prediction at R2 = 0.93 and RMSE = 4.87 compared with MLR AR2 = 0.70 and RMSE = 9.57. This indicates that ANN can be applied with more precision in modeling and understanding the dynamic nature of API compared with MLR. However, it can be concluded that the atmosphere exhibit a complex reaction which is made more complicated by man through his daily activities and multivariate analysis, modeling techniques will help to understand the dynamic structure of the source of pollution, reduce the cost of monitoring unnecessary sampling site, save time and accurately predict air pollution index. These findings can provide stakeholders a basic for understanding and developing air pollution control strategies in order to reduce their possible health and environmental effect.

© 2014 AENSI Publisher All rights reserved. To Cite This Article: Hamza AhmadIsiyaka, Hafizan Juahir, Mohd Ekhwan Toriman, Barzani M. Gasim, Azman Azid, Mohd Khairul Amri, Aminu Ibrahim, Usman Nasiru Usman, Auwalu Rabiu Ali Rano, Musa A. Garba., Spatial Assessment of Air Pollution Index Using Environ Metric Modeling Techniques. Adv. Environ. Biol., 8(24), 244-256, 2014

INTRODUCTION Clean air constitutes the fundamental requirement for human health and well-being. Although, the level of atmospheric air quality has deteriorated over time by series of developments that occur as countries become industrialized and urbanized [1,2,3], since the general characteristics of the troposphere has the ability to emit, transform, disperse as well as deposit pollutants [4]. An estimate of about 2.4 million people die each year from causes directly attributed to air pollution [5]. This is because atmospheric air pollutants have the ability to penetrate the gas exchange region of the human lung when inhaled, causing cancer, asthma, cardiovascular and respiratory diseases [6,7]. It also limit the amount of oxygen fed to the foetus through the mother thereby retarding its anthropometric development by reducing its head circumference [8]. Most urban cities within the Peninsular faces severe air pollution [9] and about 5.4 metric tons of CO2 is produced in Malaysia as at 2002 which exceeds the benchmark of 3.9 metric ton per capital as provided by the world resource institute [10]. Although, the overall air quality status in Malaysia fall within good and moderate

Corresponding Author: Hamza AhmadIsiyaka, East Coast Environmental Research Institute, Faculty of Bioscience & Food Technology, Universiti Sultan Zainal Abidin Malaysia

245

Hamza AhmadIsiyaka et al, 2014 Advances in Environmental Biology, 8(24) December 2014, Pages: 244-256

base on air pollution index (API), but still show a slight fluctuation in trend from 2008 to 2011 [11]. In 2008, the status of good air quality fall around 59%, 55.6% in 2009, 63% in 2010 and 55% in 2011 [11]. The major sources of air pollution in Malaysia are; mobile source, stationery, and trans-boundary source [12,13,14]. The mobile sources are emitted from a non-static point such as motor vehicle (82 % of pollution load), boat engine, aircraft, and motorcycles. The stationery pollutants originate from static points such as manufacturing activities (18 % of pollution load), industrial fuel burning, power stations and domestic [14]. The trans-boundary sources are transported from point source to non-point source, especially from Indonesia by the south-westerly wind into Malaysia [12]. The basic pollutants used to determine the level of atmospheric air quality in Malaysia are (CO, O 3, PM10, SO2, and NO2) [15,16,17]. They are used to establish air pollution index (API), which is a non-dimensional number calculated according to the urban daily average concentration of pollutants [15,16]. The highest subindex values of these pollutants are used as the API value for a specific period base on Recommended Malaysia Ambient Air Quality Guidelines (RMAQG) issued by DOE since 1989 [16]. However, modeling the dynamic nature of air pollution is very challenging since some pollutants are emitted from point source while others are from non-point sources. Furthermore, air quality monitoring and analysis involve the use of large complex time series data from monitoring stations which require the integration of modern and robust statistical techniques (multivariate statistics and receptor modeling techniques) for simplification, classification, modeling, prediction, avoid misinterpretation and pattern recognition [18,19,20]. As such, the use of multivariate statistical analysis such as hierarchical agglomerative cluster analysis (HACA), discriminate analysis (DA), principal component analysis (PCA) together with factor analysis (FA) and modeling such as artificial neural network (ANN), multiple linear regression model (MLR) will enable us reduce the dimensionality of the data as well as extract meaningful statistical findings [21,22,23,24,25,26]. The objectives of this study is to; identify the spatial variation in the pattern of air quality within selected monitoring stations in Peninsular Malaysia; to identify the major possible sources of air pollution; and to determine the best receptor modeling techniques for predicting air pollution index in Peninsular Malaysia. MATERIALS AND METHODS Study Area: The study area comprises of five continuous air quality monitoring stations managed by a private company AlamSakitar Malaysia Sdn. Bhd. (ASMA) on behalf of the Department of Environment Malaysia (DOE). Three stations (ST1, ST2 and ST3) are located in the east coast Peninsular Malaysia (Terengganu) and two in the west of Malaysia Peninsular. The Kemaman station (ST1) is a residential and petrochemical industrial area, followed by Paka-Kertih (ST2) an industrial area which houses the PETRONAS Petrochemical Integrated Complex (PPIC) linking the entire range of oil and gas value chain, [17,25,27], The third station Kuala Terengganu (ST3) is located in the city center of Terengganu state associated with high population density, commercial activities, airport and serious traffic congestion. Putra Jaya station (ST4) is located in the administrative center of Malaysia in the federal state of Putra Jaya and it is the third most populated state in Malaysia [28] with high traffic congestion. Kuala Lumpur station (ST5) is a city center considered as the most densely populated city in Malaysia [28]. It is characterized with airport, commercial and industrial activities with serious traffic round the clock. 2.1

Table 1: Location and coordinates of monitoring station. Station ID Location ST 1 Kemaman ST2 Paka-Kertih ST3 Kuala Terengganu ST4 Putra Jaya ST5 Kuala Lumpur Source: Department of Environment (DOE) Malaysia.

Coordinates Latitude N04° 16.260 N04° 35.880 N05° 18.455 N 02° 55.915 N 03° 06.376

Longitude E103° 25.826 E103° 26.096 E103° 07.213 E101° 40.909 E 101° 43.072

2.2

Data collection: The air quality and meteorological data used for this study were provided by the Air Quality Division, Department of Environment Malaysia covering a period five years (2007 - 2011). Seven parameters including carbon monoxide (CO), ozone (O3), particulate matter with diameter less than 10 µm (PM10), sulfur dioxide (SO2), nitrogen dioxide (NO2), ambient temperature and wind speed were used in this study. The average time varies from 1 - 24 hours which represents the period of time over which measurement are monitored and reported in order to assess the possible health impacts of specific air pollutants [16,29]. 2.3 Data treatment: The data used in this study were provided in the form of hourly reading and were converted to daily average reading in order to carry out a precise and accurate statistical analysis using XLSTAT add-in 2014 and JMP 11

246


software. A total of 64036 data sets (9148 observations × 7 parameters) were used for the analysis. Furthermore, to facilitate this analysis, the nearest neighbor method was applied using XLSTAT add-in software to estimate missing values [13,30]. Nearest neighbor is used to predict unknown values using the known values at neighboring locations where sample points lack [25]. The missing data recorded is 4% of the original data sets. [30]suggested that nearest neighbor method can provide a simple scheme, where the endpoint of the gaps is used as estimates of all missing values. This equation can be shown bellow; y = y1if x ≤ x1 + [ (x2 - x1) / 2] or y = y1 if x ≥ x1 + [ (x2 - x1) / 2] (1) Where y represents the interpolate, x is the time point of the interpolate, y1 and x1 are the coordinates of the starting point of the gap and y2 and x2 are the end points of the gaps.

Fig. 1: Location of air quality monitoring stations in Terengganu. 2.4

Hierarchical Agglomerative cluster analysis (HACA): HACA is a statistical method used to classify observations into groups, or clusters (spatial) that are similar to each other and different from the observations in other groups [31,32]. The spatial classification of air quality monitoring station can be illustrated using a dendrogram that measures the degree of the risk homogeneity through Ward's method and Euclidian distance measurement [33]. Euclidian distance is based on a single linkage which denotes the quotient between the linkage distance divided by the maximal distance [(Dlink/Dmax)], by multiplying the quotient by 100 in order to standardize the linkage distance represented by the y-axis [34,35,36]. 2.5 Discriminate analysis (DA): DA is used to determine the variables that best discriminate between groups developed by HACA and construct new discriminant functions (DFs) for each group in order to evaluate the spatial variation in atmospheric air quality [36,19,37]. DFs are calculated using Eq. F (Gi) = Ki + ∑jn=1wijPij (2) Where i = the number of group G; kj = constant inherent to each group; n = the number of parameters used to classify a set of data into a given group; wj = the weight coefficient assigned by discriminant function analysis (DFA) to a given parameter Pj. In this study, DA was applied to the raw data for spatial analysis base on the three clusters developed by HACA using standard mode, backward stepwise mode and forward stepwise mode to determine whether the groups differ with regards to the mean of the variable and to use that variable to predict group membership. To achieve this, cluster 1, 2 and 3 were selected as dependent variable, while all the measured parameters made up the independent variables. Using the forward stepwise mode, variables were included step by step from the most significant until no significant changes are observed while in the backward stepwise mode, variables were removed step by step beginning from the less significant variable until no significant changes were observed [19]. 2.6 Principal component analysis for source identification (PCA): PCA is used to establish the most significant parameter that best discriminate the whole set of data by eliminating the less significant parameter with minimal loss of the original variables [38,39]. This will enable identification of the source of pollution [40]. The equation is expressed as:

247


Zij = ai1x1j + ai2x2j+ ai3x3j +........... + aimxmj (3) Where z is the component score, a is the component loading, x is the measured value of variables, i is the component number, j is the sample number and m is the total number of variables. FA is usually carried out after PCA has been successfully applied in order to identify the different possible sources of pollution base on the activities in the air quality monitoring environment [14,40]. It also helps to reduce the least significant variables in the dataset in order to simplify the data structure in the PCA [41]. [19] argued that the PCs generated by PCA are sometimes not readily available for interpretation, therefore, it is advisable to rotate the PCs by varimax rotation with eigenvalues equal to or greater than 1.The varimax rotation is considered significant in order to obtain new groups of variables called varimax factors (VFs) [3,40]. Furthermore, the number of varimax factors(VFs) obtained by varimax rotations is usually equal to the number of variables in accordance with the common features which can include unobservable, hypothetical and latent variables [42,43]. In addition, the VFs coefficient with a correlation from 0.75 are considered as strong significant factor loading, those that range from 0.50 - 0.74 have moderate, while of 0.30 - 0.49 are classified as weak significant factor loading [44]. Zij = af1 x1i + af2 x2i + ........+ afmfmi + efi (4) Where Z is the measured value of a variables, a is the factor loading, f is the factor score, e is the residual term accounting for errors or other sources variation, i is the sample number, j is the variable number and m is the total number of factors. 2.7 Artificial neural network modeling (ANN): ANN is an information processing system that evolves from the way biological nervous system works (brain) in processing information [45]. It is an interconnected group of nodes linked to a large number of networks of neurons in the brain. Each node depicts an artificial neuron which represents a connection from the input of one neuron to the output of another [46]. A multilayered perceptron feed-forward artificial neural network (ML-FF-ANN) with a non-linear activation function is applied using JMP 11 software to predict the most significant parameters affecting the air pollution index. This software is very flexible and easy to apply [25]. This network consists of multiple neurons which are organized in layers. Information is presented to the network via an input layer (independent variable), which then send signal to the hidden layer via a system of weighted connection, where the actual processing is done and pass to the output layer (dependent variable) [46]. The network is trained severally by adjusting the weight value in order to minimize error and optimize the number of hidden neurons in each layer [24]. To achieve this, a Backpropagation algorithm is introduced to the network in order to correlate the coefficient between the expected and the calculated using a supervised learning [24,47]. The data set use to predict API was divided into training (60%), testing (20%) and validation (20%) [48]. The training data set is use to adjust the weight in order to estimate and learn the pattern of the parameters, testing is use to evaluate the generalization ability of the network, while validation subset is use to evaluate the performance of the trained network [15,25]. Furthermore, the coefficient of the determination (R2) and the root mean square error (RMSE) were used to evaluate the result gotten in the ANN model. From the result, the model with the greatest R 2 value (R2=1) and the lowest RMSE value (RMSE=0) is considered the best model [13,49]. The network structure is shown in Figure 2 below.

Fig. 2: Network structure of a multilayer feed-forward artificial neural network (ML-FF-ANN)

248


2.8 Multiple linear regression model (MLR): MLR is a statistical technique that is used to predict the variability that exists between the dependent and independent variable [50,51]. The model has a k independent variables and n observation. Thus, the regress model can be represented as [52]. Yi= βo + β1xIi + ..............+ βkxki + ԑiEq (6) where i = 1.......n, β0, β1 and βk are regression coefficient, x1 and xk are independent variables and ԑ is error associated with the regression. However, in order to evaluate the result of the MLR model, the coefficient of determination R 2, Adjusted R2 and Root Mean Square Error (RMSE) must be considered as the best fitting regression linear equation. The R2 represents the proportion of variability that exists in the dependent variable use to measure the goodness of fit of a linear model base on the regression equation [53]. Furthermore, the value with the greatest R 2 (=1) represent the best linear model [13,49]. The R2 is usually adjusted to reduce its tendency to overestimating the validity of the model base on real world taken into account all the variables and observations used. The descriptive statistics of the entire data sets comprising of minimum value, maximum, mean and standard deviation are shown in Table 2. Table 2: (a) Descriptive statistics of daily average air pollutants (b) meteorological parameters and API 2007- 2011. (a) Variables Statistics Stations Avg time ST1 ST2 ST3 ST4 ST5 CO, ppm Min 0.000 0.000 0.000 0.000 0.000 1-h Max 3.180 0.810 2.740 2.830 4.750 Mean 0.452 0.049 0.923 1.045 1.624 Stddev 0.250 0.117 0.426 0.423 0.722 O3, ppb Min 0.000 0.000 0.000 0.000 0.000 1-h Max 3.180 0.050 0.140 0.120 0.170 Mean 0.452 0.003 0.027 0.047 0.055 Stddev 0.250 0.008 0.012 0.022 0.031 PM1O, ug m-1 Min 0.000 0.000 0.000 13.583 0.000 24-h Max 264.000 470.000 357.000 299.000 282.000 Mean 55.922 57.950 84.297 59.903 71.137 Stddev 26.155 33.051 39.352 26.518 27.875 SO2, ppb Min 0.000 0.000 0.000 0.000 0.000 1-h Max 0.060 0.010 0.010 0.060 0.060 Mean 0.002 0.001 0.000 0.004 0.003 Stddev 0.005 0.002 0.001 0.006 0.006 NO2, ppb Min 0.000 0.000 0.000 0.000 0.000 1-h Max 0.030 0.060 0.030 0.070 0.110 Mean 0.007 0.007 0.012 0.025 0.036 Stddev 0.006 0.009 0.006 0.011 0.014 (b) Variables Statistics Stations ST1 ST2 ST3 ST4 Ambient Temp 0C Min 22.208 21.463 21.463 22.371 Max 30.763 31.267 31.267 29.679 Mean 27.623 27.333 26.333 26.927 Stddev 1.597 1.701 1.848 1.252 Wind Speed KM h-1 Min 0.313 1.000 1.508 1.508 Max 27.213 210.200 30.700 13.983 Mean 11.139 14.240 10.532 4.383 Stddev 3.571 14.143 3.530 0.986 API Min 0.000 0.000 19.458 14.208 Max 95.000 71.000 81.000 119.000 Mean 45.126 37.926 50.053 53.513 Stddev 13.186 10.016 10.206 16.986

RMAQG

30

100

150

130

170

ST5 23.913 29.679 27.218 1.124 1.508 7.546 4.274 0.891 20.125 169.000 64.298 22.820

RESULTS AND DISCUSSION 3.1 Spatial classification of air quality monitoring stations: In this study, HACA was performed on the raw data sets in order to evaluate the spatial variation among air quality monitoring stations. This resulted in a successful grouping of the stations into three clusters base on their level of homogeneity. Cluster 1 comprises of three air quality monitoring stations; Kemaman (ST1), Kertih (ST2) and Kuala Terengganu (ST3) located in the east coast Peninsular Malaysia. These stations are characterized with residential area, city center with traffic congestion, airport, commercial and industrial activities, and can be considered as a good healthy site (GHS) [12,54]. Cluster 2 is a residential area embedded with one station; Putra

249


Jaya (ST4) and can be classify as moderate healthy site (MHS). It is considered as the administrative center of Malaysia with commercial activities and third most densely populated area. Kuala Lumpur (ST5) makes up cluster 3 which is a city center characterize with high Traffic congestion around the clock, high population density (the most populated city in Malaysia), airport, commercial and industrial activities [12,28]. However, we can conclude that one station in each cluster can provide a reasonable accurate spatial representation of air quality within the study area and can reduce the need for unnecessary sampling sites especially in Terengganu. Figure 3 below represents the spatial classification of air monitoring stations using a dendrogram.

Fig. 3: Dendrogram of different clusters showing sampling sites in the study area. 3.2 Spatial variation in atmospheric air pollution using DA: DA was performed on the raw data sets to determine the most significant parameters associated with the spatial variation in air quality monitoring parameters base on the clusters produce by HACA. Cluster 1, 2 and 3 represent the dependent variables while the seven parameters (PM 10, CO, O3, NO2, SO2, Ambient temperature and Wind speed) were used as independent variables. Using the standard mode, forward stepwise and backward stepwise mode, the accuracy of the spatial classification in the confusion matrix for air quality and meteorological parameters yielded a correct assignation of 82.37% each for both cluster 1, 2 and 3 with a p ˂ 0.05 for each parameter. Base on these findings, we can conclude that the seven parameters (PM10, CO, O3, SO2, NO2, Ambient temperature and Wind speed) have a strong variation in terms of their spatial distribution. The spatial variation associated with these parameters base on cluster is shown in Table 3. Table 3: Classification matrix for DA base on spatial variation in Peninsular Malaysia. Sampling stations Regions assign by DA Cluster 1 Cluster 2 Cluster 3 Standard mode (7 parameters) Cluster 1 5365 90 10 Cluster 2 456 956 445 Cluster 3 189 423 1214 Total 6010 1469 1669 Forward stepwise (7 Parameter) Cluster 1 5365 90 10 Cluster 2 456 956 445 Cluster 3 189 423 1214 Total 6010 1469 1669 Backward stepwise(7 Parameter) Cluster 1 5365 90 10 Cluster 2 456 956 445 Cluster 3 189 423 1214 Total 6010 1469 1669

% correct 98.17% 51.48% 66.48% 82.37% 98.17% 51.48% 66.48% 82.37% 98.17% 51.48% 66.48% 82.37%

250


3.3 Spatial identification of major possible source of pollution: PCA together with FA was performed on the normalized data sets (PM10, CO, O3, SO2, NO2, Ambient temperature and Wind speed) separately for the three clusters in order to compare the compositional pattern between the examined parameters and identify the major possible source of variation in the study area. Three PCs were obtained for cluster 1 and two PCs for both cluster 2 and 3 with eigenvalues greater than 1(˃1), which account for more than 62%, 56% and 58% respectively of the total variance in the data sets as shown in the PCA loading plot in Figure 4. Three varifactors (VFs) were obtained for cluster 1, and two (VFs) for both cluster 2 and 3 through the FA performed on the PCs. Only factor loadings greater than 0.75 were considered for interpretation. The result of FA is presented in Table 4, comprising of factor loadings, eigenvalues, the total variance and cumulative percentages. Table 4: Factor loadings after varimax rotation by clusters. Variables Cluster 1 VF1 VF2 CO 0.776 0.27 O3 0.361 0.691 PM10 0.784 -0.089 SO2 -0.102 0.715 NO2 0.767 0.006 ABT TEMP -0.111 0.181 WIND SPEED 0.059 -0.432 Eigenvalue 2.131 1.211 Variability (%) 30.444 17.297 Cumulative % 30.444 47.741

Cluster 2 VF3 -0.318 -0.111 -0.073 0.293 0.211 0.781 0.542 1.06 15.139 62.88

VF1 0.826 0.756 0.663 0.219 0.81 0.421 -0.272 2.646 37.795 37.795

VF2 -0.033 0.057 0.061 0.126 -0.054 0.763 0.864 1.343 19.183 56.978

Cluster 3 VF1 0.789 0.763 0.768 0.262 0.838 0.033 -0.054 2.571 36.734 36.734

VF2 -0.028 0.013 0.034 0.01 -0.046 0.882 0.879 1.554 22.206 58.94

a) Cluster 1: The first varimax factor in cluster 1 (VF1) accounts for 30.4% of the total variance, which indicate a strong positive loading for CO (0.776), PM10 (0.784) and NO2 (0.767). The spatial variation in the source of CO can be attributed to the incomplete combustion of fuel in automobiles (motor vehicles, motorcycles, engine boats) [55,56], due to high traffic congestion in the city center of Kuala Terengganu, increase in individual ownership of vehicles and busy port. CO accounts for 82% of pollution load in Malaysia [13]. The variation in the source of PM10 are dust fall from industrial activities and construction site [57], gases and particles from transport exhaust emission, resuspension of soil dust and transboundary pollution [58,59,60]. Industrial activities in Kemaman and Paka-Kertih can be linked to the emission of NO2 base on the previous study conducted by [61]. Furthermore, a report by Intergovernmental Panel on Climate Change [62] and study carried out by [63,64] shows that aircraft is an important source of NOx which accounts for 1.5% of the total global pollution. These emissions mostly come from Sultan Mahmud airport in Terengganu. The third varifactor (VF3) explains 15.1% of the total variance which indicate a strong positive loading for ambient temperature (0.781). High temperature especially during northeast monsoon season in Terengganu [65] provide a favorable condition for photochemical reactions [52,66,67]. Further justification is provided in the findings of [68,69] which shows that, sulfate formation increases with a rise in temperature due to faster SO 2 oxidation, which act as an important precursor for the formation of PM 10[53]. NO2 is also produced when nitrogen in the fuel is burnt and when at a very high temperature nitrogen in the air reacts with oxygen [70]. b) Cluster 2: In this cluster, the first varimax factor (VF1) account for 37.8% of the total variance which shows a strong positive loadings for CO (0.826), O3 (0.756), NO2 (0.81). The spatial variation in the source of CO is from engine boats and motor vehicles, especially when the engines are not turned properly which causes incomplete combustion [55,56]. O3 is strongly related to secondary pollutants produced by photochemical oxidation and forms the main component of smog [71]. Its concentration is produced by O 3 precursors (NOx CO and VOCs ) from industrial activities and motor vehicle emission from traffics in Putra Jaya the third most populated state and administrative center in Malaysia [72,73,74,75,76]. NO 2 is largely due to industrial activities as well as emission due to heavy traffic congestion. In Malaysia, an analysis of NO 2 emission proves that about 69% of this pollutant is from power stations and industrial activities, 28% from motor vehicles and the remaining 3% makes up other sources [29]. The second varimax factor (VF2) records 19.2% of the total variance with a strong positive loading for ambient temperature (0.763) and Wind Speed (0.864). According to a report by [28] on Compendium of Environment Statistics, Malaysia is a tropical region near the equator (high temperature) which provide favorable condition for nitrogen oxides (NOx) and other VOCs to react, and CO slowly helps to oxidize nitrogen monoxide (NO) to nitrogen dioxide (NO2) which indirectly accelerate O3 formation [52,66,67]. Wind speed help to transport both primary and secondary pollutants from point source to non-point source [68,69].

251


c) Cluster 3: This cluster is made up of two varifactors (VF1,VF2) which account for 36.7% and 22.2% respectively for the total variance with a strong positive loading for CO (0.789), O 3 (0.763), PM10 (0.768), NO2 (0.838), ambient temperature (0.882) and wind speed (0.879). This station comprises the most densely populated city (Kuala Lumpur) in Malaysia with a high traffic jam [54] which leads to emission of CO due to incomplete combustion from motor vehicle. However, the busy port in the Klang Valley contributes to NO 2 emission. According to [28], Kuala Lumpur houses the commercial and industrial center of Malaysia, which in turn emits a large proportion of O3 and NO2. High temperature provides a favorable condition for photochemical oxidation of both O3 and PM10 precursors. Strong positive loading for wind speed (0,879) (Wang et al.2012) enable O 3 precursors (CO, NOx, and hydrocarbon) to be transported by multi- scale circulations which increases the concentration of O 3 during the transport period, while PM10 are transported from source point to non-source point [68,69]. The Sumatrabush burning in Indonesia are usually transported into Malaysia by strong wind which in turn pollute the air (DOE, 2010).

Fig. 4: Plot diagram for PCA loading after varimax rotation (a) cluster 1 (b) cluster 2 (c) cluster 3. 3.4

Predicting API using artificial neural network and multiple linear regression: In predicting air pollution index (API), the raw data comprising of seven parameters (CO, O3, PM10, NO2, SO2, Ambient temperature and Wind speed) were used to develop two receptor model; multilayer perceptron feed forward artificial neural network (MLP-FF-ANN) model (A) and multiple linear regression (MLR) model (B) in order to compare and identify the best receptor model for predicting API using JMP 11 software. The model with the highest R2 and lowest RMSE is considered the best. In developing MLP-FF-ANN model A, the data were divided into training, testing, and validation. Twenty hidden nodes were examined in order to avoid over-fitting and to approximate any non-linear function with a high level of accuracy. The best prediction model for MLP-FF-ANN was gotten in node eighteen with R2 = 0.93 and RMSE = 4.87 as shown in Table 5 For model B, MLR was used to predict API by developing an equation model which represents its prediction capability. The regression coefficient of determination (R 2), adjusted regression coefficient of determination (AR2) and the root mean square error (RMSE) were used to test the performance of the model. The result shows that R2 = 0.71, AR2 = 0.71 and RMSE = 9.58. This result is shown in Table 5. The final models develop by MLR for API is; API = -1.59 (CO) + 451.76 (O3) + 0.162 (PM10) + 13.7 (SO2) + 149.76 (NO2) -0.03 (ambient temperature) + 0.06 + 22.93 Base on the result gotten from the two receptor models (A and B), it is glaring that MLP-FF-ANN (A) has more prediction capability and accuracy in predicting API at R 2 = 0.93 and RMSE =0.487 compare to MLR with AR2= 0.71 and RMSE = 9.58 . This is because ANN has a high degree of flexibility, efficiency and can model the complexity and unusual non-linear relationship between meteorological variables and air quality parameters [24]. ANN learns when presented with new data sets by forming functional relationship among the data and does not degrade much with noisy data [77,78]. The scatter plot diagram for the two models is shown in Figure 5. Table 5: Prediction performance of ML-FF-ANN model A and MLR model B. Models Hidden nodes [7,1,1,1] ML-FF-ANN Model A [7,2,1,1] [7,3,1,1] [7,4,1,1]

R2 0.82 0.89 0.89 0.90

RMSE 7.83 6.1 5.87 5.8

252

Hamza AhmadIsiyaka et al, 2014 Advances in Environmental Biology, 8(24) December 2014, Pages: 244-256 [7,5,1,1] [7,6,1,1] [7,7,1,1] [7,8,1,1] [7,9,1,1] [7,10,1,1] [7,11,1,1] [7,12,1,1] [7,13,1,1] [7,14,1,1] [7,15,1,1] [7,16,1,1] [7,17,1,1] [7,18,1,1]

0.91 0.91 0.91 0.91 0.91 0.92 0.91 0.92 0.92 0.92 0.92 0.92 0.92 0.93 0.71

MLR Model B

5.43 5.37 5.25 5.24 5.32 5.17 5.28 5.05 5.02 4.93 4.96 4.93 4.97 4.87 9.57

Fig. 5: Scatter plots diagram for predicted API performance: ML-FF-ANN (A) and MLR (B). Conclusion: In this study, multivariate statistics and environmental modeling techniques were used to assess the spatial variation of atmospheric air quality in some selected monitoring stations in Peninsular Malaysia. HACA successfully group the stations into three clusters base on their degree of homogeneity. This classification helps to eliminate unnecessary sampling sites which in turn reduce cost and time of monitoring. The result for DA show that each parameter have a p-value (p ˂ 0.05) with a correct assignation of 82.37% for both standard, forward stepwise and backward stepwise. This means that all the monitored parameters discriminate spatially within the study area. PCA together with FA indicates that only six parameters (except SO 2) with strong positive loadings ˃ 0.75 have a strong variation in the three clusters. The major possible sources of these pollutants are from anthropogenically induced emission (automobiles, industries, power plants, transboundary sources and construction sites). Comparism between MLP-FF-ANN and MLR was made in order to test their predictive capabilities and to select the best receptor model for predicting API. MLP-FF-ANN gave the highest R2 = 0.93 with the lowest RMSE = 0.487 compared to MLR with and R2, AR2 = 0.71 and RMSE = 9.58. However, ANN stands the test of providing a better prediction model compare to MLR and can provide factual and reliable estimation of API. ACKNOWLEDGEMENT The authors would like to extend their profound gratitude to the Department of Environment (DOE) Malaysia for their kind gesture in providing the data used for this study. A special thanks to the East Coast Environmental Research Institute, Faculty of Bioscience and Food Technology, Universiti Sultan ZainalAbidin, Malaysia for the constructive academic support towards putting this findings into reality REFERENCES [1] [2]

Ramanathan, V. and Y. Feng, 2009. “Air pollution, greenhouse gases and climate change: global and regional perspectives”, Atmospheric Environment, 43(1): 37-50. Latif, M.T., S.Z. Azmi, A.D.M. Noor, A.S. Ismail, Z. Johny, S. Idrus, A.F. Mohamad, M. Mokhtar, 2011. “The impact of urban growth on regional air qualitysurrounding the Langat River Basin, Malaysia”, Environmentalist, 31: 315e-324.

253


[3]

[4] [5] [6]

[7] [8] [9]

[10] [11] [12] [13] [14]

[15] [16] [17] [18] [19]

[20]

[21]

[22]

[23] [24]

[25]

Oyeyiola, A.O., C.M. Davidson, K.O. Olayinka, T.O. Oluseyi, B.I. Alo, 2013. “Multivariate analysis of potentially toxic metalsin sediments of a tropical coastal lagoon”,Environ Monit Assess, 185: 21672177. DOI 0.1007/s10661-012-2697-7 Monks, P.S., C. Grainer, et al., 2009. “Atmospheric composition changes-global and regional airquality”, Atmospheric Environment, 43: 4268-5350. World Health Organization (WHO), 2002. “World Health Report: Reducing Risk, Promoting Healthy Life”,Geneva, Switzerland. Ilayas, S.Z., A.I. Khattak, S.M. Nasir, T. Qurashi, R. Durrani, 2010. “Air pollution assessment in urban area and its impact on human health in the city of Quetta Pakistan”. Clean Technologies and environmental Policy, 12(3): 291-299. MacNee, W. and K. Donaldson, 2003. “Mechanism of lung injury caused by PM10 and ultrafine particles with special reference to COPD”, Eur. Respir, 40-47. Ballester, F., M. Estarlich, I. Carmen, L. Sabrina, R. Rosa, E. Ana, L. Marina, M. Rabagliato, 2010. “Preterm birth and exposure to air pollutants during pregnancy”, Environ. Research, 110(8): 778-85. Azmi, S.Z., M.T. Latif, A.S. Ismail, L. Juneng, A.A. Jemain, 2010. “Trend and status of air quality at three different monitoring stations in the Klang Valley, Malaysia”. Air Quality, Atmosphere and Health”, 3: 5364. doi: 10.1007/ s11869-009-0051-1. Razak, M.I.M., A. Ismail, B. Imabrine, A.H. Talib, Z. Ibrahim, 2013. “Economics of air pollution in Malaysia”, International Journal of Humanities and Social Sciences, 3(13). Department of Environment Malaysia (DOE), 2012. “Malaysia Environmental Quality Report”, Ministry of Science, Technology and Environment, Kuala Lumpur. Department of Environment Malaysia (DOE), 2005. “Malaysia Environmental Quality Report”, Ministryof Science, Technology and Environment, Kuala Lumpur. Dominick, D., H. Juahir, M.T. Latif, S.M. Zain, A.Z. Aris, 2012. “Spatial assessment of air quality pattern in Malaysia using multivariate analysis”, Atmospheric Environment, 60: 172-181. Mutalib, S.N.S.A., H. Juahir, A. Azid, S.M. Sharif, M.T. Latif, A.Z. Aris, S.M. Zain, D. Dominick, 2013. “Spatial and temporal air quality pattern recognition using environmetric techniques: a case study in Malaysia”, Environmental Science, Processes & Impacts 1-12. Department of Environment Malaysia (DOE), 2011. “Malaysia Environmental Quality Report”, Ministry of Science, Technology and Environment, Kuala Lumpur. Department of Environment Malaysia (DOE), 1997. “A Guide to Air Pollution Index in Malaysia (API)”, Ministry of Science, Technology and Environment, Kuala Lumpur. Department of Environment Malaysia (DOE), 2007. “Malaysia Environmental Quality Report”, Ministry of Science, Technology and Environment, Kuala Lumpur. Simeonov, V., J.W. Einax, I. Stanimirova, J. Kraft, 2002. “Environmental modeling and interpretation of river water monitoring data”, Analytical and Bio Analytical Chemistry, 374: 898-905. Juahir, H., S.M. Zain, M.K. Yusoff, T.I. Hanidza, A.S. Armi, M.T. Ekhwan, M. Mokhtar, 2011. “Spatial water quality assessment of Langat River Basin (Malaysia) is using environmetric techniques”, Environ. Monit. Assess, 173: 625-641. Al-Odaini, A.N., P. Zakaria, M. Zali, A.M. Juahir, H. Yaziz, I.M. Surif, 2012.“Application of chemometrics in understanding the spatial distribution of human pharmaceuticals in surface water”, Environ Monit Assess, 184: 6735-6748. DOI 10.1007/s10661-011-2454-3. Satheeshkumar, P. and A.B. Khan, 2012. “Identification of mangrove water quality by multivariate statistical analysis methods in Pondcherry Coast, India”, Environ Monit Assess, 184: 3761-3774. DOI 10.1007/s10661-011-2222-4. Deepulal, P.M., T.R. Kumar, C.H. Sujatha, R. George, 2012. “Chemometric study on the trace metal accumulation in the sediments of the Cochin Estuary-Southwest coast of India”, Environ Monit Assess, 184: 6261-6279. DOI 10.1007/s10661-011-2418-7. Wahid, A.B.N., T.M. Latif, S. Suratman, 2013. “Composition and source apportionment of Surfactants in atmospheric aerosols of urban and semi-urban area in Malaysia”, Chemosphere, 91: 1508-1516. Arhami, M., Nima Kamali, Mohammad Mahdi Rajabi, 2013.“Predicting hourly air pollutant levels using artificial neural networks coupled with uncertainty analysis by Monte Carlo simulations”, Environ Sci. Pollut. Res., 20: 4777-4789. DOI 10.1007/s11356-012-1451-6. Azid, A., H. Juahir, M.T. Ikhwan, M.K. Amri, A.S. Mohd, C.N. Hasnam, N.A. Azlina, F. Azaman, M.T. Latif, S.F.M. Zainuddin, M.R. Osman, 2014. “Prediction of the level of air pollution using principal component analysis and artificial neural network techniques: a case study in Malaysia”, Water Air Soil Pollut, 225: 2063.

254


[26] Mustaffa, H.I..N., T.M. Latif, M.M. Ali, F.M. Khan, 2014. “Source apportionment of Surfactants in marine aerosols at different locations along the Malacca Straits”. Environ. Sci. Pollut. Res., 21: 65906602. [27] Petrolian Nasional Berhad (PETRONAS), 2013. “Annual report”. [28] Department of Statistic Malaysia, 2010. “Basic Population Characteristics by Administrative Districts”, 2009 Report. [29] Department of Environment Malaysia (DOE), 2009. “Malaysia Environmental Quality Report”, Ministryof Science, Technology and Environment, Kuala Lumpur. [30] Junninen, H., H. Niska, K. Tuppurainen, J. Ruuskanen, M. Kolehmainen, 2004. “Methodsfor imputation of missing values in air quality data set”, Atmospheric Environment, 38: 2895e-2907. [31] Al-Odaini, A.N., P. Zakaria, M. Zali, A.M. Juahir, H. Yaziz, I.M. Surif, 2012.“Application of chemometrics in understanding the spatial distribution of human pharmaceuticals in surface water”, Environ. Monit. Assess, 184: 6735-6748. DOI 10.1007/s10661-011-2454-3. [32] Farmaki, G.E., S.N. Thomaidis, V. Simeonov, E.C. Efstathiou, 2012. “A comparative chemometric study for water quality expertise of the Athenian water reservoirs”, Environ. Monit. Assess, 184: 7635-7652. DOI 10.1007/s10661-012-2524. [33] Lau, J., W.T. Hung, C.S. Cheung, 2009. “Interpretation of air quality in relation to monitoring station’s surrounding”, Atmospheric Environmetric, 43: 769-777. [34] Singh, K.P., A. Malik, D. Mohan, S. Sinha, 2004. “Multivariate statistical techniques for the evaluation of spatial and temporal variations in water quality of Gomti River (India) a case study”, Water Research, 38(18): 3980-3992. [35] Singh, K.P., A. Malik, S. Sinha, 2005. “Water quality assessment and apportionment of pollution sources of Gomti River (India) using multivariate statistical techniques e a case study”,Analytica Chimica Acta, 538: 355-374. [36] Shrestha, S., F. Kazama, 2007. “Assessment of surface water quality using multivariate statistical techniques: a case study of the Fuji River basin, Japan”, Environmental Modeling and Software, 22(4): 464-475. [37] Pati, S., K. Mihir, C.K. Dash, B. Mukherjee, S. Dash, Pokhrel, 2014. “Assessment of water quality using multivariate statistical techniques in the coastal region of Visakhapatnam, India”, Environ Monit Assess 186: 6385-6402. DOI 10.1007/s10661-014-3862-y. [38] Osman, R., N. Saim, H. Juahir, P.M. Abdullah, 2012. “Chemometric application in identifying sources of organic contaminants in Langat river basin”,Environ Monit Assess, 184: 1001-1014. DOI 10.1007/s10661011-2016-8. [39] Wang, Y., C. Liu, P. Liao, J. Lee, 2014. “Spatial pattern assessment of river water quality: implications of reducing the number of monitoring stations and chemical parameters”, Environ Monit Assess, 186: 17811792 DOI 10.1007/s10661-013-3492-9. [40] Zhang, X., H. Jiang, Y. Zhang, 2013. “Spatial distribution and source identification of persistent pollutants in marine sediments of Hong Kong”, Environ Monit Assess, 185: 4693-4704. [41] Deepulal, P.M., T.R. Kumar, C.H. Sujatha, R. George, 2012. “Chemometric study on the trace metal accumulation in the sediments of the Cochin Estuary-Southwest coast of India”, Environ Monit Assess 184: 6261-6279. DOI 10.1007/s10661-011-2418-7. [42] Vega, M., R. Pardo, E. Banato, L. Deban, 1998. "Assessment of seasonal and polluting effect on the quality of river water by explanatory data analysis'' Water Research, 32: 3581-3592. [43] Helena, B., R. Pardo, M. Vega, E. Barrado, J.M. Fernandez, L. Fernandez, 2000. “Temporal evaluation of groundwater composition in an alluvial aquifer (Pisuerga river, Spain) by principal component analysis”, Water Research, 34: 807-816. doi:10.1016/S0043- 354(99)00225-0. [44] Liu, C.W., K.H. Lin, Y.M. Kuo, 2003. “Application of factor analysis in the assessment of ground water quality in a black foot disease area in Taiwan”. Science of the Total Environment, 313: 77-89. [45] Moustris, K.P., I.C. Ziomas, A.G. Paliatsos, 2010. “3-day ahead forecasting of regional pollution index for the pollutants NO2, CO, SO2, and O3 using artificial neural network in Athens, Greece”, Water, Air, Soil pollution, 209(1-4): 29-43. [46] Dongare A.D., R.R. Kharde, A.D. Kachare, 2012. “Introduction to Artificial Neural Network”, International Journal of Engineering and Innovative Technology, 2(1) ISSN: 2277-3754. [47] Juahir, H., S. Zain, Md., A.Z. Aris, K.Y. Mazlin, B. Mokhtar, 2009.“Spatial assessment of Langat river Water quality using chemometrics”, J. Environ.Monit., 12, 287–295 | 287 DOI: 10.1039/b907306j. [48] Xie, H., F. Ma, Q. Bai, 2009. “Prediction of indoor air quality using artificial neural networks”, 2009 Fifth International Conference on Natural Computation, 412-418. [49] Norusis, M.J., 1990. “SPSS base system user’s guide”, Chicago, IL, USA: SPSS.

255


[50] Petrie, A., C. Sabin, 2000. “Medical Statistics at a Glance. Blackwell Science, Oxford. Querol, X., Alastuey, A., Lopez-Soler, A., Plana, F., Mantilla, E., Juan, R., Ruiz, C.R., La Orden, A., Characterisation of atmospheric particulates around a coalfired power station. Int. J. Coal Geol., 40: 175-188. [51] Shi, G.L., F. Zeng, X. Li, Y.C. Feng, Y.Q. Wang, G.X. Liu, T. Zhu, 2011. “Estimated contributions and uncertainties of PCA/MLR-CMB results: source apportionment for synthetic and ambient datasets”, Atmos. Environ., 45: 2811-2819. [52] Kovac-Andric, E., J. Brana, V. Gvozdic, 2009. “Impact of meteorological factors on ozone Concentrations modelled by time series and multivariate statistical methods”, Ecological Informatics, 4(2): 117-122. [53] Ilten Nadir Aand Tülay Selici, 2008. “Investigating the impact of some meteorological parameters on air pollution in Balikesir, Turkey”. Environ. Monit. Assess, 140: 267-277. [54] Department of Environment Malaysia (DOE), 2010. “Malaysia Environmental Quality Report”, Ministry of Science, Technology and Environment, Kuala Lumpur. [55] Janssen, N.A.H., V. Van, F. Aarts, H. Harssema, B. Brunekreef, 2001. “Assessment of exposure to trafficrelated air pollution of children attending school near motorways”. Atmospheric Environment, 35: 38753884. [56] Morawska, L., E.R. Jayaratne, K. Mengersen, M. Jamriska, S. Thomas, 2002. “Difference in airborne particle and gaseous concentration in urban air between weekdays and weekends”, Atmospheric environment, 36: 4375-4385. [57] Pandey, S., B. Tripathi, V. Mishra, 2008. “Dust deposition in a sub-tropical opencast coalmine area, India”, Journal of Environmental Management, 86: 132-138. [58] Brook, R.D., B. Franklin, W. Cascio, Y. Hong, G. Howard, M. Lipsett, R. Luepker, M. Mittleman, J. Samet, S.C. Smith, I. Tager, 2004. “Air Pollution and Cardiovascular Disease”, Journal of American Heart Association, ISSN: 1524-4539. [59] Arsene, C., R.I. Olariu, N. Mihalopoulos, 2007. “Chemical composition of rainwater in the Northeastern Romania, Iasi Region (2003–2006)”, Atmospheric Environment, 41: 9452-9467. [60] Yusof, Md. N.F.F., N.A. Ramli, A.S. Yahya, N. Sansuddin, A.A. Ghazali, W. Al-Madhoum, 2010. “Monsoonal differences and probability distribution of PM10 concentration”, Environmental Monitoring and Assessment, 163. [61] Ghazali, N.A., N.A. Ramli, A.S. Yahaya, 2009. “A study to investigate and model the transformation of nitrogen dioxide into ozone using time series plot”.European.Journal of Scientific Research, 37(2): 192e205. [62] Intergovernmental Panel on Climate Change (IIPC), 1999. “IIPC Special Report: Aviation and the Global Atmosphere”, Summary for Policy Makers. IIPC-XV/DOC 99. [63] Propecus, F., I. Lonel, L. Belegante, N. Lontis, V. Cebrucean, 2011. “Direct measurement and numerical simulation issues in airport air quality”, International Journal of Energy & Environment, 3(5): 410-417. [64] Scheelhaase, J.D., W.G. Grimme, 2007. “Emissions trading for international aviation, estimation of the economic impact on selected European airlines”. Journal of Air Transport Management, 13: 253-263. [65] Malaysia Meteorological Department (MMD), 2012. “General Climate of Malaysia”, Kuala Lumpur: Ministry of Science, Technology and Innovation. [66] Atkinson, R., 2000. “Atmospheric chemistry of VOCs and NOx”, Atmospheric Environment, 34: 20632101. [67] Bauer, S.E., B. Langmann, 2002. “Analysis of summer smog episode in the Berlin-Brandenburg regionwith a nested atmosphere chemistry model”, Atmospheric Chemistry & Physics, 2(4): 259-270. [68] Dawson, J.P., P.J. Adams, S.N. Pandis, 2007. “Sensitivity of PM2.5 to climate in the Eastern U.S: a modeling case study”, Atmos. Chem. Phys., 7: 4295-4309. [69] Kleeman, M.J., 2007. “A preliminary assessment of the sensitivity of air quality in California to global change”, Climatic Change, 87: S273-S292. [70] Brunekreef, B., S.T. Holgate, 2002. “Air pollution and health Lancet”, 360: 1233-1242. [71] Banan, N., M.T. Latif, L. Juneng, F. Ahamad, 2013. “Characteristics of surface ozone concentrations at stations with different backgrounds in the Malaysian Peninsula”, Aerosol and Air Quality Research, 13: 1090-1106. doi: 10. 4209/aaqr.2012.09.0259. [72] Mittal, M.L., P.G. Hess, S.L. Jain, B.C. Arya, C. Sharma, 2007. “Surface Ozone in the Indian region”, Atmospheric Environment, 41: 6572-6584. [73] Pham, T.B., K. Munomaiphiboon, C. Vongmahadlek, 2007. “Updated emission of ozone precursors from energy consumption by power plants and industrial facilities in the central and eastern region of Thailand”, Asia Journal on Energy Environment, 8(1): 483-489. [74] Minoura, H., A. Ito, 2010. “Observation of primary NO2 and NO oxidation near the trunk road in Tokyo”, Atmospheric Environment, 44: 22-29. [75] Department of Statistic Malaysia, 2010. “Basic Population Characteristics by Administrative Districts”, 2009 Report.

256


[76] Sadanaga, Y., M. Sengen, N. Takenaka, H. Bandow, 2012. “Analysis of the ozone weekend effect in Tokyo, Japan: regime of oxidant (O3 + O2) production”, Aerosol & Air Quality Research, 12: 161-168. [77] Jiang, D., Y. Zhang, X. Hu, Y. Zeng, J. Tan, D. Shao, 2004. “Progress indeveloping an ANN model for air pollution index forecast”, Atmos Environ., 38: 7055-7064. [78] Shahin, M.A., M.B. Jaksa, H.R. Maier, 2008. “State of the art of artificial neural networks in geotechnical engineering. Electro. J. Geotech. Eng., 8: 1-26.