hosting by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction
Globally, an estimated 7 million premature deaths from airpollution exposure were reported in year 2012, 88% of these werefrom developing countries (Lim et al., 2013; WHO, 2014). In India,air pollution associated death rates have registered an increase of12% between 2005 and 2010 (UNEP, 2014). Investigations fromdifferent geographical locations associate ambient air pollutionexposure with Respiratory Disease Infections (RDI): Acute respira-tory Infection, acute lower respiratory infection, lung cancer,chronic obstructive pulmonary diseases, cardio vesicular diseasesand ischemic heart disease etc (Balakrishnan et al., 2013; Braueret al., 2012; Ezzati and Kammen, 2001; Mehta et al., 2013; Popeet al., 2011; Pope et al., 2009; Pope and Dockery, 2006; Smithet al., 2014). Ambient particulate matter (PM) concentrations,
ail.com (A.K. Attri).ational Committee for Air
ttee for Air Pollution Research andg/licenses/by-nc-nd/4.0/).
particularly PM size fraction having aerodynamic diameter<2.5 m (PM2.5), stand as a metric for air pollution exposure (Braueret al., 2012). Statistical modeling of health outcome time series datawith corresponding air pollution exposure data, also takes intoconsideration other factors (e.g., meteorological factors) whichmayalso play a role in causing the health outcome (Arundel et al., 1986;Lowen et al., 2007). The association between health outcome andexplanatory variables is not always straightforward; temporalprofiles of data-sets in question display not only non-linearity butalso oscillations in the data on account of the presence of multi-cyclic influences of varying periodicity. Consequently, in anexploratory investigation the detection of association between thehealth outcome and explanatory variables requires appropriateconsiderations of multi-cyclic variable periodic influences: short-term, seasonal, and trend (Peng and Dominici, 2008; Merrill,2010). It is common to evaluate association between the two byusing overdispersed generalized additive model (GAM) (Merrill,2010; Peng and Dominici, 2008). De-composition of respectivetime series data over different time scale enables the separation oftrend from cyclic variations to examine their association separately(Cleveland et al., 1990). In the present context, the estimation of the
Control. Production and hosting by Elsevier B.V. This is an open access article under
A. Kumar, A.K. Attri / Atmospheric Pollution Research 7 (2016) 858e864 859
trend in RDI cases and its correlationwith the corresponding trendsin PM, and meteorological variables was accomplished by usingEnsemble Empirical Mode Decomposition (EEMD) method; theestimation of the trends using this approach follows the stepwisede-composition of the respective time series data-sets (RDI, PM andmeteorology variables) to separate the embedded cyclic influenceswith minimum assumptions from the trend (Franzke, 2012; Huanget al., 1998). This approach was used in the present context toanalyze the correlation between the RDI trends with the trends inPM exposure and meteorological factors from an investigationundertaken in the state of Himachal Pradesh, part of the WesternHimalayan region, which has reported high incidences of RDI cases(NHP, 2005e2013).
An exploratory investigation spanned over fourteen monthwas undertaken at Jawali, a rural site in Kangra district(31�20e32�50N and 75�00e77�450E) of Himachal Pradesh (Fig. S1,supplementary material) to understand a putative associationbetween environmental factors and reported RDI cases. The statehas recorded 1,514,082 ± 78,576 year�1 acute respiratory infec-tion cases between 2005 and 2013 (NHP, 2005e2013). The regionsurrounding the site spreads over 36 km2 and has a population of25,000; the surrounding region is devoid of any industrial ac-tivity and has low vehicular traffic. Geophysical attributes of thesampling site are representative of the state's rural regions. TheJawali region has three state government funded primary healthcenters (HC) catering to the population; the health centersmaintains daily outpatient data records in conformity with ICD10code (Narayana et al., 2010; CBHI report, 2005). The prevalence ofstate-wide practice of biomass combustion as a primary energysource exists in the Himalayan region, which to a large extentcontributes to the poor ambient air quality and high concentra-tion of PM in the environment (Kumar and Attri, 2016). Thepredominant emissions from biomass combustion, fine PM(PM2.5), Elemental Carbon (EC), Black Carbon (BC), Organic Car-bon (OC) and UV absorbing Organic Compounds (UVOC) affectsthe ambient environment quality and may in turn have a role inthe reported RDI from this region; at the same time the role ofother environmental factors (e.g. meteorological variables)cannot be ruled out (Goswami and Baruah, 2014).
The manuscript presents the collection and analysis of timeseries data-sets of Respiratory Disease Infections (RDI) counts,ambient PM (proxy for air pollution) and meteorological variablesover fourteen months. The counts included cases of respiratorydistress syndrome, cough and cold, bronchial asthma, bronchio-litis, pneumonitis, pharyngitis, laryngitis and tonsillitis as classi-fied by Central Council of Indian Medicine conforming to ICD10code. In addition to the consideration of mass concentration ofcollected PM (PM2.5 and PM10) their composition in terms ofassociated carbon forms (TC, EC, BC, OC and UVOC contents) werealso taken into account as explanatory variables. Average meteo-rological variables conforming to the PM sample collection timespan were considered as a part of the analysis. The analysis of thetime series data-sets (RDI with PM, PM-composition and meteo-rological factors) were subjected to descriptive exploratory sta-tistical analysis (Merrill, 2010). The appraisal of the time scale ofmulti-cyclic influences in the respective time series data-setswas detected by measuring AutoCorrelation Function (ACF). Therespective time series data was analyzed using Ensemble Empir-ical Mode De-composition (EEMD) algorithm to extract and eval-uate the present multi-cycles in the respective time series data-sets as internal mode functions (IMF) and determines non-lineartrends. Multiple linear regression analysis (MLR) was done toevaluate the association of trend in RDI with the trends in therespective explanatory variables (PM, PM-composition andmeteorological variables).
2. Material and methods
2.1. Ambient PM10 and PM2.5 load and meteorological variables
The sampling of PM in two size fractions were initiated from Jan2012 to Feb 2013 over fourteen months in a time series, thecollection of each sample was done over 24e30 h at a height of 20feet above the ground. The collection of PM2.5 was done at a fixedtime interval of 5 days on 46.2 mm PTFE filters (EPA certified,Whatman 7592e104), using a low volume sampler operated at aconstant air flow of 16.7 l min�1 (EnvirotecheAPM 550 MFC);whereas PM10 samples were collected in a time interval of 10 dayson Quartz microfiber filters (Whatman QMA 1851e865) using ahigh volume sampler (Envirotech, Model-APM 460 BL) at a con-stant air flow rate of 1.0 and 1.1 m3 min-1. The quartz filters usedwere prebaked at 550 �C for 6 h prior to their use in the collection.The mass of the collected PM (PM10 and PM2.5) samples was esti-mated gravimetrically (mg/m3) using a Sartorius electronic micro-balance (precision ± 10 mg) and stored at�18 �C until their analysis.The corresponding time series data of meteorological variables forthe sampling locationd Dew point (DP), Planetary Boundary Layer(PBL), Relative Humidity (RH), ambient temperature (T), Precipi-tation (PPT), Wind Direction (WD) and Wind speed (WS)d wasaccessed from Air resource laboratory (http://ready.arl.noaa.gov/EADYcmet.php).
2.2. The analysis of PM10 and PM2.5 associated carbon forms: TC,OC, EC, BC and UVOC
The mass of the collected PM (PM10 and PM2.5) samples wereestimated gravimetrically (mg/m3). Carbon species (TC, OC and EC)present in the samples were estimated by using Thermal/opticalcarbon analyzer (DRI Model 2001A, Atmos-lytic, Inc., Calabasas, CA,USA) following IMPROVE_A protocol (Chow et al., 2007). The esti-mation of BC and UVOC (UV absorbing organic carbon) present inthe PM2.5 samples was done using a dual wavelength measure-ments (880 and 370 nm) Transmissometer (Magee scientific, USA).The measured values were appropriately corrected with referenceto the field blanks.
2.3. Statistical analysis of time series data-sets: RDI, PM10, PM2.5,PM associated carbon forms and meteorological variables
Time series data-sets of RDI cases, PM10, PM2.5, PM associatedcarbon forms (TC, OC, EC, BC, UVOC) and ofmeteorological variables(DP, PBL, RH, T, PPT, WD andWS) were subjected to descriptive andanalytical statistical analysis (skewness, kurtosis, mean, medianand mode). The average values of the collected PM10 and PM2.5samples and the associated carbon forms, obtained over the dura-tion of investigation, are given in Table 1. Calculated estimates ofnon-parametric Spearman's r-correlation (Table S1) and Autocor-relation Function (ACF) was used to detect the presence of persis-tence (multi-cycles) in the respective time series data-sets.Presence of multi-collinearity between the time series data-sets ofall variables was calculated by estimating their Variation InflationFactor (VIF). Selection of explanatory variables having VIF < 3.0were considered as an independent variables for MLR modeling toexplain the association of RDI trends.
2.4. Determination of time dependent cyclic variability and trend intime series data-sets
The temporal profiles of all data-sets, RDI cases and predictorvariables manifested a large time dependent variation; attributescommon to almost all environmental geophysical proxy temporal
Table 1The Annual average mass concentrations of PM10, PM2.5, and of carbon forms present in PM10 (TC, OC, EC); the average data, % proportion of PM2.5, TC, OC and EC in PM10. Thegiven average values are over 14 months, the duration of this investigation.
Sample/Species Annual average load (mg m�3) Species Percent (%) present in PM10 load
PM10 79.8 ± 45.3 PM2.5 68PM2.5 52.4 ± 18.7 Total CAa 34.9TC 18.5 ± 10.3 TC 24.3OC 13.6 ± 7.3 OC 17.7EC 4.9 ± 3.7 EC 6.5BC 2.2 ± 1.0 in PM2.5 BC 4.9 of PM2.5
UVOC 2.0 ± 1.2 in PM2.5 UVOC 3.9 of PM2.5
a Total Carbonaceous aerosols (CA) was calculated on the basis of established relation: Total_CA ¼ 1.6�OCþEC(Cao et al., 2005).
A. Kumar, A.K. Attri / Atmospheric Pollution Research 7 (2016) 858e864860
data, including the presence of non-linearity and non-stationarity.The general representation of the respective time series data set ofvariables, Y(t), was conceptualized in model form as:
YðtÞ ¼ TimelineTrendðtÞ þXk
i¼1
IMFiðtÞ þ RandomNoiseðtÞ (1)
In Equation (1), t represents time, IMFi designates ith timedependent cyclic modes in the time series, and k represents num-ber of modes in the data set; each mode (IMF) is distinguished fromeach other on the basis of their respective magnitude and fre-quency (period). The de-convolution of the additive components ofthe time series (1) were obtained by Ensemble Empirical Mode De-composition (EEMD) analysis, a method widely applied to extractembedded cyclic variations (IMFs) (Huang et al., 1998; Wu et al.,2011). EEMD analysis of the respective temporal data sets (PM,associated carbon forms and meteorology variables) allowed theappraisal of the statistical significance of IMFs and determination ofthe trend at 95% confidence level (Huang et al., 1998; Hyndman andKostenko, 2007); details of EEMD analysis are available in literature(Chen et al., 2013; Huang et al., 1998; Kuo et al., 2013; Tandon et al.,2013; Wu and Huang, 2009; Wu et al., 2011; Yadav et al., 2014).
3. Results and discussion
The results of exploratory statistical analysis of the time seriesdata-sets (RDI cases, PM10, PM2.5, TC, BC, OC, EC, and UVOC) indicatethat all data-sets deviated from normal distribution. The averageconcentration of PM (PM10 and PM2.5) and associated carbon formsrepresentative of the duration of this investigation are summarizedin Table 1. By proportion 68% of PM2.5 accounted for the PM10 load,35% of the associated carbonaceous contents accounted for PM10load. The average mass concentrations of both PM10 and PM2.5exceeded the ambient PM pollution standards threshold levels(WHO, 2014).
3.1. Correlation between RDI, PM and meteorological variables
The non-parametric Spearman (r) multi correlation estimatesbetween the RDI cases and PM10, PM2.5, associated carbon formsand meteorological variables are given in Table S1. The absence ofstatistically significant correlation between RDI and PM (PM10 andPM2.5) load was observed, i.e. contrary to the widely reported andwell established link between the RDI and PM (Mehta et al., 2013).At the same time the correlation of RDI was statistically significantwith PM associated carbon forms: TC, BC, UVOC and EC (Table S1).Correlation of RDI with meteorological variables (DP, PBL, T andWD) was significant, but eve. The absence of correlation of RDIwith PM size fractions reflected that the presence of multi-scalecyclic influences in the respective data-sets to a large extentwould obscure the expected association of RDI cases with PM
(Birmili et al., 2010). The support for this observation comes fromthe presence of autocorrelation in the respective data-sets, ameasure of the pattern of time dependent signals present in therespective time series. The calculated autocorrelation function(ACF) with lag for RDI, PM10, PM2.5 and PM associated carbon formsis shown in Fig. 1 [A]. ACF for RDI and meteorological variables (DP,PBL, RH, T, WD and WS) are shown in Fig. 1 [B]. The extent of ACFvaried and persisted over considerable time span, which indicatesthat the presence of cyclic influences (monthly, seasonal, annual) inthe respective time series data set. Statistically, reliable estimationof trend profile from the respective time series required anappropriate accounting of the cyclic modes (IMFs) embedded in therespective time series (Tandon and Attri, 2011; Weatherhead et al.,1998). The most likely explanation for the absence of correlationbetween RDI cases with PM10 and PM2.5 mass concentration mayarise due to the presence of significant autocorrelation/persistence.
However, even in the presence of autocorrelation, the RDI timeseries data manifested statistically significant correlation with TC,BC, EC (p-value <0.05), and with UVOC (p-value <0.01); whereas,the correlation was not significant with OC contents associatedwith PM (Table S1). Carbon forms associated with ambient PM arisefrom combustion of fossil fuel and biomass, the emissions of carbonforms exhibit a bimodal size distribution (0.05e0.12 mm and0.5e1.0 mm); their size is much smaller than the PM2.5(Venkataraman and Friedlander, 1994). In general, the reportedaverage ratios of PM1:0
PM2:5and PM1:0
PM10loads reported from rural regions
respectively are 0.84 and 0.60 (Gomi�s�cek et al., 2004), which sug-gests that the major bulk in PM2.5 load is constituted by PM1.0 sizefraction. Consequently, it stands to reason that RDI syndromes maycorrelate more with PM2.5 as this fraction's (84%) load is predom-inantly constituted by PM1.0. Most of the carbon forms from com-bustion would be smaller and be part of PM1.0 fraction. Thisexplains why the carbon forms (TC, BC, EC and UVOC) manifestcorrelationwith RDI even in the presence of multi-cyclic influencesin the data, but not with PM2.5 load. It stands to reason thatPM < 1.0 mm and associated carbon forms would have an easy ac-cess deeper into the lungs.
3.2. EEMD analysis of RDI, PM10, PM2.5, and PM associated carbonforms, and meteorological variable to determine their trend profiles
All time series data-sets (RDI, PM, PM associated carbon formsand meteorological variables) of same size were de-convoluted todetermine the presence of embedded IMFs and trend. Total of 4IMFs, indicative of the multi-cyclic influences of varying time pe-riods, were extracted from each time series and their statisticallysignificant was tested as per the EEMD algorithm (Wu et al., 2011).The plot of the time series data for RDI, PM10 and PM2.5 loads, andPM associated TC, EC and UVOC are shown in separate panels ofFig. 2. Each panel of the Fig. also plots the corresponding variable'sestimated ensemble average trend (blue curve) at 95% confidence.
Fig. 1. Panel [A] of the Fig. plots autocorrelation function (ACF) with lag in days for RDI cases, PM10, PM2.5, TC, BC, EC, OC and UVOC in their respective time series data set. Panel [B]plots ACF with lag in days for RDI and meteorological variables (DP, PBL, RH, T, WD and WS) in their respective time series data set.
A. Kumar, A.K. Attri / Atmospheric Pollution Research 7 (2016) 858e864 861
The spread of the confidence region is demarcated by the sur-rounding red curves. The trend profiles displayed a non-linearcharacter. The trends in RDI, PM2.5, TC, EC and UVOC were higherduring winter and declined during summer and monsoon period.The close similarity in the trends among RDI, PM2.5, TC, EC andUVOC is also reflected in the calculated correlation estimates(Table 2). The absence of statistically significant correlation be-tween PM10 and RDI is also evident from the trend (blue curves)
Fig. 2. Time series data of RDI, PM10, PM2.5, TC, EC and UVOC are plotted in separate panelsEEMD analysis is superimposed on the corresponding time series data in each panel. Eachcurves. (For interpretation of the references to colour in this figure legend, the reader is re
obtained for PM10 (Fig. 2). Trends associated with BC, OC data-setswere also estimated, but are not shown in the figure as the esti-mated correlation of these two variable's trend with RDI was notstatistically significant (Table 2). Only one IMF was statisticallysignificant in RDI, PM10 and PM2.5 time series data; whereas twoIMFs were significant in TC, EC and UVOC; the ensemble averageprofile of these IMFs (blue curve) are plotted in different panels ofFig. S2 (Supplementary Material). Confidence region (95% CI) of the
of the Fig. The respective variable's ensemble average trend (blue curve) obtained fromplotted trend is surrounded by the confidence region (95% CI) demarcated by the redferred to the web version of this article.)
Table 2Spearman r-correlation between timeline trends of RDI and explanatory variables; the values given in bold are statistically significant at p-value <0.05*and <0.01**.
*Correlation is significant at the 0.05 level (2-tailed).** Correlation is significant at the 0.01 level (2-tailed).
A. Kumar, A.K. Attri / Atmospheric Pollution Research 7 (2016) 858e864862
ensemble average IMF plots are shown by the surrounding redlines.
Time series data-sets of RDI and meteorological variables (DP,PBL, RH, T and WD) are plotted in separate panels of Fig. 3. TheEEMD analysis based ensemble average trends of these time seriesdata-sets are superimposed on their respective data (blue curve)along with the confidence region (95% CI) demarcated by the sur-rounding red curves. The temporal profiles of the trends of themeteorological variables were significantly different than thatdetermined for RDI, PM10, PM2.5 and PM associated carbon forms.The plots of the statistically significant ensemble average IMFs(blue curve) in DP, T, PBL and RH are shown in Fig. S3.
The average periods of the obtained significant IMFs in therespective time series data-sets were calculated by using DiscreteFourier Transform (Tandon et al., 2013). The obtained periods cor-responded to the cyclic influence (IMFs) in the data (monthly,
Fig. 3. Time series data of RDI, DP, PBL, RH, T and WD are plotted in separate panels of the Fanalysis is superimposed on the corresponding time series data in each panel. Each plotted trinterpretation of the references to colour in this figure legend, the reader is referred to the
seasonal and annual) for the respective time series data set aregiven in Table S2. The time span spread in the periods of statisticallysignificant IMFs ranged between 87 and 365 days. This indicatesthat the affecting environmental factors to which the population isexposed to, differ significantly in terms of their magnitude andperiodicity over time; this time dependent variability would havebearing in establishing their association with RDI.
3.3. Multiple linear regression (MLR) of RDI trends with PM, PM-composition and meteorological variables
MLR fit between RDI trends as dependent variable was fittedwith the explanatory variables PM2.5 and RH trends. The selectionof independent variables in the model was determined by takinginto consideration the statistical significance (p-value <0.001) ofthe estimated correlation (Table 2), and by ensuring that the
ig. The respective variable's ensemble average trend (blue curve) obtained from EEMDend is surrounded by the confidence region (95% CI) demarcated by the red curves. (Forweb version of this article.)
Table 3MLR fitted timeline trends in RDI as a function of PM2.5 and RH trends.
Model R R2 Std. Error of estimate Sig F change VIF
A. Kumar, A.K. Attri / Atmospheric Pollution Research 7 (2016) 858e864 863
calculated collinearity (VIF) between the independent variableswas <3.0. The general form of the MLR fit explaining the RDI trendover the period of this investigation, in terms of PM2.5 and RHis represented as RDIðtÞTrend ¼ b0 þ b1 � PM2:5ðtÞTrend þ b2�RHðtÞTrend. The fit statistics of MLR and estimated values of thecoefficients (b0, b1, b2) in the equation are given in Table 3. The R2
value for the fit was 0.991, and the calculated VIF between PM2.5and RH was 1.038. No other combination of explanatory variablessatisfied the above stated twin criteria (correlation and collin-earity). The absence of collinearity between PM2.5 and RH implytheir independent effect on the RDI trend. The relationship estab-lished fromMLR, between RDI trend with the corresponding trendsin PM2.5 and RH is represented by Equation (2)
The small intercept value suggests that the trend in RDI isessentially determined by the trend in two environmental factors(PM2.5 and RH). The fitted coefficients (intercept, PM2.5 and RH)werewell determined as reflected by a small standard error in theirrespective fitted values (±0.392, ±0.001 and ± 0.008). Of the totalvariation in RDI trends, PM2.5 explained 77% and RH explained 22%.The fitted Equation (2) suggests that even in the absence of lowambient PM load, the RH variation can impact RDI trend. This initself is an important finding, which suggests that the ambient PMmass concentrations, per se, may not be the only environmentfactor in determining the trend profile of RDI. This explanation isfurther substantiated by the inferences of few reported in-vestigations that indirect effects of RH on human health induce RDIsyndromes (Arundel et al., 1986; Lowen et al., 2007). The findingsfurther indicate that the influence of confounding factors can be asignificant and their magnitude may vary with location specificclimate regime.
4. Conclusion
Time series investigation, over 14 months, was undertaken at arepresentative rural location in the state of Himachal Pradesh tofind correlation between Respiratory Disease Incidences (RDI) withambient air pollution and other environmental factors. Presence ofmulti-cycles having variable period in the time series data-setsconstrained the determination of straightforward correlation be-tween RDI time series with ambient PM and meteorological vari-ables. De-composition of time series data-sets as a sum of trend andmulti-cyclic influences by using Empirical Ensemble Mode De-composition (EEMD) method allowed the determination of corre-lation between the RDI trends with corresponding trends in PM2.5
concentrations and Relative Humidity (RH). EEMD analysis of timeseries is not constrained by the presence of non-linearity and non-stationarity in the time series data-sets and allows the extraction ofmultiple cycles as IMFs without any assumption about the scale ofthe embedded cycles and determines trend, which may be linear ornon-linear. The role of other environmental factors, in present caseof ambient Relative Humidity (RH), to induce RDI cases was
established through MLR modeling; i.e., in addition to the airpollution exposure. The mechanism of inducing RDI throughambient RH is indirect, whereas the RDI from air pollution exposuredirectly involve respiratory system.
Time series studies involving the de-composition of data-setscan assist in establishing correlation between RDI cases andexplanatory variables at different time scales: trend, weekly orseasonal. This requires access of high resolution time series data(hourly or daily data) of health outcome, air pollution and otherenvironmental factors to have a better understanding about thereported high RDI cases from the Himalayan region.
Acknowledgments
Authors acknowledge Dr. Sarita Rana, Dr. Surinder Sharma andMs Roshani Katnoria for providing RDI data of public health centers.Ajay Kumar extends his appreciation to the Council of Scientific andIndustrial Research (CSIR), India for the award of Research fellow-ship. Authors acknowledge the contributions of anonymous re-viewers of this manuscript as their in depth review, suggestionsand comments have helped a great deal in improving thismanuscript.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.apr.2016.05.005.
References
Arundel, A.V., Sterling, E.M., Biggin, J.H., Sterling, T.D., 1986. Indirect health effects ofrelative humidity in indoor environments. Environ. Health Perspect. 65, 351.
Balakrishnan, K., Ghosh, S., Ganguli, B., Sambandam, S., Bruce, N., Barnes, D.F.,Smith, K.R., 2013. State and national household concentrations of PM2.5 fromsolid cookfuel use: results from measurements and modeling in India forestimation of the global burden of disease. Environ. Health 12, 77.
Birmili, W., Heinke, K., Pitz, M., Matschullat, J., Wiedensohler, A., Cyrys, J.,Wichmann, H.-E., Peters, A., 2010. Particle number size distributions in urbanair before and after volatilisation. Atmos. Chem. Phys. 10, 4643e4660.
Brauer, M., Amann, M., Burnett, R.T., Cohen, A., Dentener, F., Ezzati, M.,Henderson, S.B., Krzyzanowski, M., Martin, R.V., Van Dingenen, R., 2012.Exposure assessment for estimation of the global burden of disease attributableto outdoor air pollution. Environ. Sci. Technol. 46, 652e660.
Cao, J.J., Wu, F., Chow, J.C., Lee, S.C., Li, Y., Chen, S.W., An, Z.S., Fung, K.K., Watson, J.G.,Zhu, C.S., 2005. Characterization and source apportionment of atmosphericorganic and elemental carbon during fall and winter of 2003 in Xi'an, China.Atmos. Chem. Phys. 5, 3127e3137.
CBHI report, Government of India, 2005. Improving and Strengthening the Use ofICD10 and Medical Record System in India: a Case Study (2004e2005). http://cbhidghs.nic.in/writereaddata/linkimages/Combined107166151888.pdf.
Chen, X., Zhang, Y., Zhang, M., Feng, Y., Wu, Z., Qiao, F., Huang, N.E., 2013. Inter-comparison between observed and simulated variability in global ocean heatcontent using empirical mode decomposition, part I: modulated annual cycle.Clim. Dyn. 41, 2797e2815.
Chow, J.C., Watson, J.G., Chen, L.W.A., Chang, M.C.O., Robinson, N.F., Trimble, D.,Kohl, S., 2007. The IMPROVE_A temperature protocol for thermal/optical carbonanalysis: maintaining consistency with a long-term database. J. Air WasteManag. Assoc. 57, 1014e1023.
Cleveland, R.B., Clevenland, W.S., McRae, J.E., Terpenning, I., 1990. STL: a seasonal-trend decomposition procedure based on loess. J. Off. Stat. 6, 3e73.
A. Kumar, A.K. Attri / Atmospheric Pollution Research 7 (2016) 858e864864
Ezzati, M., Kammen, D.M., 2001. Indoor air pollution from biomass combustion andacute respiratory infections in Kenya: an exposure-response study. Lancet 358,619e624.
Franzke, C., 2012. Nonlinear trends, long-range dependence, and climate noiseproperties of surface temperature. J. Clim. 25, 4172e4183.
Gomi�s�cek, B., Hauck, H., Stopper, S., Preining, O., 2004. Spatial and temporal vari-ations of PM 1, PM 2.5, PM 10 and particle number concentration during theAUPHEPdproject. Atmos. Environ. 38, 3917e3934.
Goswami, P., Baruah, J., 2014. Quantitative Assessment of Relative Roles of Drivers ofAcute Respiratory Diseases. Scientific reports. 4.
Huang, N.E., Shen, Z., Long, S.R., Wu, M.C., Shih, H.H., Zheng, Q., Yen, N.-C., Tung, C.C.,Liu, H.H., 1998. The empirical mode decomposition and the Hilbert spectrum fornonlinear and non-stationary time series analysis. In: Proceedings of the RoyalSociety of London a: Mathematical, Physical and Engineering Sciences. TheRoyal Society, pp. 903e995.
Kumar, A., Attri, A.K., 2016. Biomass combustion a dominant source of carbonaceousaerosols in the ambient environment of Western Himalayas. Aerosol Air. Qual.Res. 16, 519e529. http://dx.doi.org/10.4209/aaqr.2015.05.0284.
Lim, S.S., Vos, T., Flaxman, A.D., Danaei, G., Shibuya, K., Adair-Rohani, H.,AlMazroa, M.A., Amann, M., Anderson, H.R., Andrews, K.G., 2013. A comparativerisk assessment of burden of disease and injury attributable to 67 risk factorsand risk factor clusters in 21 regions, 1990e2010: a systematic analysis for theglobal burden of disease study 2010. lancet 380, 2224e2260.
Mehta, S., Shin, H., Burnett, R., North, T., Cohen, A.J., 2013. Ambient particulate airpollution and acute lower respiratory infections: a systematic review and implica-tions for estimating the global burden of disease. Air Qual. Atmos. Health 6, 69e83.
Merrill, M., 2010. Environmental Epidemiology: Principles and Methods. Jones andBartlett Publishers.
Narayana, A., Padhi, N.M., Srikanth, N., Venkateshwarlu, B., Rajashekharan, R.,Srinivasulu, B., Kumar, P.V., Prashanti, Y.S., 2010. Ayush Research portal on In-dian systems of medicine, homoeopathy and related sciences research-a needof hour. J. Ind. Med. Harit. XL 281e286.
NHP, 2005e2013. Health Status Indicators. Reports 2005e2013. Central Bureau ofHealth Intelligence reports. http://cbhidghs.nic.in/index1.asp?linkid¼267.
Peng, R.D., Dominici, F., 2008. Statistical Methods for Environmental Epidemiologywith R, R: a Case Study in Air Pollution and Health. Springer.
Pope III, C., Burnett, R.T., Turner, M.C., Cohen, A., Krewski, D., Jerrett, M.,Gapstur, S.M., Thun, M.J., 2011. Lung cancer and cardiovascular diseasemortality associated with ambient air pollution and cigarette smoke:shape of the exposure-response relationships. Environ. Health Perspect.119, 1616e1621.
Pope III, C.A., Burnett, R.T., Krewski, D., Jerrett, M., Shi, Y., Calle, E.E., Thun, M.J., 2009.Cardiovascular mortality and exposure to airborne fine particulate matter andcigarette smoke shape of the exposure-response relationship. Circulation 120,941e948.
Pope III, C.A., Dockery, D.W., 2006. Health effects of fine particulate air pollution:lines that connect. J. Air Waste Manag. Assoc. 56, 709e742.
Smith, K.R., Bruce, N., Balakrishnan, K., Adair-Rohani, H., Balmes, J., Chafe, Z.,Dherani, M., Hosgood, H.D., Mehta, S., Pope, D., 2014. Millions dead: how do weknow and what does it mean? Methods used in the comparative risk assess-ment of household air pollution. Annu. Rev. public health 35, 185e206.
Tandon, A., Attri, A.K., 2011. Trends in total ozone column over India: 1979e2008.Atmos. Environ. 45, 1648e1654.
Tandon, A., Yadav, S., Attri, A.K., 2013. Nonelinear analysis of short term variationsin ambient visibility. Atmos. Pollut. Res. 4, 199e207.
UNEP, 2014. Air Pollution: World's Worst Environmental Health Risk. United Na-tions Environment Programme. http://www.unep.org/yearbook/2014/PDF/chapt7.pdf.
Venkataraman, C., Friedlander, S.K., 1994. Size distributions of polycyclic aromatichydrocarbons and elemental carbon. 2. Ambient measurements and effects ofatmospheric processes. Environ. Sci. Technol 28, 563e572.
Weatherhead, E.C., Reinsel, G.C., Tiao, G.C., Meng, X.L., Choi, D., Cheang, W.K.,Keller, T., DeLuisi, J., Wuebbles, D.J., Kerr, J.B., 1998. Factors affecting thedetection of trends: statistical considerations and applications to environ-mental data. J. Geophys. Res. Atmos. 103, 17149e17161 (1984e2012).
WHO, 2014. Ambient (Outdoor) Air Quality and Health. World Health Organization.http://www.who.int/mediacentre/factsheets/fs313/en/.
Wu, Z., Huang, N.E., 2009. Ensemble empirical mode decomposition: a noise-assisted data analysis method. Adv. Adapt. data analysis 1, 1e41.
Wu, Z., Huang, N.E., Wallace, J.M., Smoliak, B.V., Chen, X., 2011. On the time-varyingtrend in global-mean surface temperature. Clim. Dyn. 37, 759e773.
Yadav, S., Tandon, A., Attri, A.K., 2014. Timeline trend profile and seasonal variationsin nicotine present in ambient PM10 samples: a four year investigation fromDelhi region, India. Atmos. Environ. 98, 89e97.
