Performance evaluation of a hybrid-passive landfill leachate treatment system using multivariate statistical techniques

Waste Management xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Waste Management

journal homepage: www.elsevier .com/locate /wasman

Performance evaluation of a hybrid-passive landfill leachate treatmentsystem using multivariate statistical techniques

http://dx.doi.org/10.1016/j.wasman.2014.10.0110956-053X/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author. Tel.: +1 613 533 3053; fax: +1 613 533 2128.E-mail addresses: [email protected] (J. Wallace), champagne@civil.

queensu.ca (P. Champagne), [email protected] (A.-C. Monnier).

Please cite this article in press as: Wallace, J., et al. Performance evaluation of a hybrid-passive landfill leachate treatment system using multivaritistical techniques. Waste Management (2014), http://dx.doi.org/10.1016/j.wasman.2014.10.011

Jack Wallace a, Pascale Champagne a,⇑, Anne-Charlotte Monnier b

a Department of Civil Engineering, Queen’s University, Ellis Hall, 58 University Avenue, Kingston, Ontario K7L 3N6, Canadab National Institute for Applied Sciences – Lyon, 20 Avenue Albert Einstein, 69621 Villeurbanne Cedex, France

a r t i c l e i n f o a b s t r a c t

Article history:Received 26 October 2013Accepted 10 October 2014Available online xxxx

Keywords:LandfillsLeachate treatmentLeachate chemistryPrincipal components analysisRegressions

A pilot-scale hybrid-passive treatment system operated at the Merrick Landfill in North Bay, Ontario,Canada, treats municipal landfill leachate and provides for subsequent natural attenuation. Collectedleachate is directed to a hybrid-passive treatment system, followed by controlled release to a naturalattenuation zone before entering the nearby Little Sturgeon River. The study presents a comprehensiveevaluation of the performance of the system using multivariate statistical techniques to determine theinteractions between parameters, major pollutants in the leachate, and the biological and chemical pro-cesses occurring in the system. Five parameters (ammonia, alkalinity, chemical oxygen demand (COD),‘‘heavy’’ metals of interest, with atomic weights above calcium, and iron) were set as criteria for the eval-uation of system performance based on their toxicity to aquatic ecosystems and importance in treatmentwith respect to discharge regulations. System data for a full range of water quality parameters over a 21-month period were analyzed using principal components analysis (PCA), as well as principal components(PC) and partial least squares (PLS) regressions. PCA indicated a high degree of association for mostparameters with the first PC, which explained a high percentage (>40%) of the variation in the data, sug-gesting strong statistical relationships among most of the parameters in the system. Regression analysesidentified 8 parameters (set as independent variables) that were most frequently retained for modelingthe five criteria parameters (set as dependent variables), on a statistically significant level: conductivity,dissolved oxygen (DO), nitrite (NO2

�), organic nitrogen (N), oxidation reduction potential (ORP), pH, sul-fate and total volatile solids (TVS). The criteria parameters and the significant explanatory parameterswere most important in modeling the dynamics of the passive treatment system during the study period.Such techniques and procedures were found to be highly valuable and could be applied to other sites todetermine parameters of interest in similar naturalized engineered systems.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Municipal Solid Waste (MSW) is a major concern for cities andcommunities around the world and presents a persistent manage-ment challenge. It is estimated that 1.3 billion tonnes of MSW isgenerated in cities around the world, and will only rise as urbani-zation continues (Hoornweg and Bhada-Tata, 2012). Landfill dis-posal still overwhelmingly remains the primary method formanaging MSW in both high- and low-income countries, andbrings with it the toxic by-product of landfill leachate. This waste-water is often composed of metals, organic matter, chlorinatedchemicals, and high levels of nutrients (Speer et al., 2012). Due

to the complexity of the solution and the high concentrations ofindividual pollutants, leachate treatment can be relatively expen-sive in terms of energy requirements and chemical inputs. Addi-tionally, leachate production continues for many years after alandfill is closed, creating a long-term management burden forthe operator (Mulamoottil et al., 1999; Rew and Mulamoottil,1999). Landfill operators in low- and high-income countries canboth benefit from sustainable treatment technologies that mini-mize the capital and operating costs of leachate treatment, whilesufficiently treating effluents to minimize ecological impacts andhuman health risks.

Passive treatment systems are one such technology that can beoperated effectively at a lower cost while performing comparablyto more active and conventional systems. They have beensuccessfully implemented at numerous facilities and are responsiveto changing leachate composition and carry lower energy and

ate sta-

http://dx.doi.org/10.1016/j.wasman.2014.10.011

mailto:[email protected]





http://www.sciencedirect.com/science/journal/0956053X

http://www.elsevier.com/locate/wasman


2 J. Wallace et al. / Waste Management xxx (2014) xxx–xxx

maintenance costs (Rew and Mulamoottil, 1999; Mehmood et al.,2009; Speer et al., 2012). An active treatment stage can be incorpo-rated to improve performance in cold climates by providing a level ofpre-treatment that removes a large fraction of the oxygen demandand metals (Speer, 2011), with the combined train termed ahybrid-passive treatment system. The complexity of the leachatenecessitates the monitoring of a large number of water qualityparameters to evaluate performance, where focused analyses canthen be performed depending on the broad system dynamics. Mul-tivariate statistical analysis is a useful set of techniques that canindicate interactions between variables and identify trends in treat-ment performance both spatially and temporally.

This study identifies the principal parameters affecting the per-formance of a novel hybrid-passive treatment system receivingleachate collected from an operating MSW landfill and providingfor the controlled release of treated leachate to a natural attenua-tion zone prior to entering the receiving environment. The hybrid-

Fig. 1. Hybrid-passive treatment system schematic implemented at the Merrick Landfillthe configurations of the pre-treatment, PW, and AWL subsystems shown (Speer, 2011)

Please cite this article in press as: Wallace, J., et al. Performance evaluation of atistical techniques. Waste Management (2014), http://dx.doi.org/10.1016/j.was

passive system is considered novel due to the incorporation of dos-ing and rest cycles and active pre-treatment, and the evaluation ofits performance in the challenging conditions of a cold northernclimate, as reported by Speer et al. (2012). A comprehensive setof water quality parameters were monitored bi-weekly over a21-month period, from December 2009 to August 2011. The datawere analyzed using PCA, as well as PC and PLS regression analysesin order to understand the primary variables influencing systemperformance and the interactions and trends between them, withrespect to treatment objectives that include minimizing theimpacts of the treated leachate on the receiving environment.While these techniques have been applied to specific sets ofparameters of interest, a principal objective of the study was todemonstrate the ability to identify and understand which parame-ters influence specific treatment targets or processes in natural-ized, open systems such as the hybrid-passive system, throughmultivariate statistical analysis. Previous studies by Speer et al.

in North Bay (Ontario, Canada) as designed and implemented by Speer (2011), with. Plan views are shown above side profile views for each component.

hybrid-passive landfill leachate treatment system using multivariate sta-man.2014.10.011


J. Wallace et al. / Waste Management xxx (2014) xxx–xxx 3

(2012) Wallace et al. (2012) have evaluated treatment perfor-mance through examination of time series patterns and removalefficiencies. This study expands on this work by identifying theparameters influencing this performance and providing a method-ology for predicting key system parameters for future studies ofsimilar systems.

2. Experimental methods

2.1. System overview

The Merrick Landfill (site) serves as the primary MSW disposalsite for the City of North Bay (Ontario, Canada) and surroundingtownships, and is located approximately 25 km north of the City.First opened in 1994, the site has an expected service life of40 years and was initially designed with monitored natural atten-uation within the site boundaries as the proposed leachate man-agement strategy. This design was based on the expectation thatthe site hydrogeochemistry could allow physical, chemical, andbiological mitigation of contaminants. After three years of opera-tion, leachate-impacted surface and groundwater was observedin a wetland and river 500 m west of the site and resulted in con-sistent exceedances of the regulatory approvals for the facility. Inresponse to the need for leachate treatment prior to discharge intothe natural attenuation zone, a hybrid-passive leachate treatmentsystem was developed and implemented. It consisted of a constantflow attached-growth aerated cell (pre-treatment system), which

Table 1Parameters of interest for the three year sampling program of the hybrid-passive treatmanalytical methods employed indicated.

Classification Parameter Frequency Me

Water quality pH Bi-weekly WTAlkalinity, as CaCO3 Bi-weekly AmTemperature Bi-weekly WTOxidation reduction potential (ORP) Bi-weekly WTConductivity Bi-weekly YSIDissolved oxygen (DO) Bi-weekly YSISulfate Bi-weekly IonSulfide Bi-weekly Ind

speChloride Bi-weekly IonPhenols Bi-weekly Ga

Solids Total solids (TS) Bi-weekly AP– Total volatile solids (TVS) Bi-weekly AP– Total fixed solids (TVS) Bi-weekly AP– Total suspended solids (TSS) Bi-weekly AP– Total dissolved solids (TDS) Bi-weekly AP

Nitrogenspecies

Ammonia Bi-weekly APUnionized ammonia Bi-weekly CalNitrate, as N (NO3

�) Bi-weekly IonNitrite, as N (NO2

�) Bi-weekly IonTotal Kjeldahl nitrogen (TKN), total Bi-weekly AP

– TKN, soluble Bi-weekly AP

Phosphorusspecies

Total phosphorus (P) Bi-weekly AP– Total P, soluble Bi-weekly AP

Orthophosphate Bi-weekly Ion– Orthophosphate, soluble Bi-weekly AP

Hydrolysable P Bi-weekly Pho– Hydrolysable P, soluble Bi-weekly Pho

Oxygen demand Carbonaceous biochemical oxygen demand(cBOD), total

Bi-weekly AP

– cBOD, soluble Bi-weekly APChemical oxygen demand (COD), total Bi-weekly AP

– COD, soluble Bi-weekly APBiochemical oxygen demand (BOD), total Bi-weekly AP

– BOD, soluble Bi-weekly AP

Metals Refer to text Bi-weekly ICP


fed two passive treatment systems: a peat and wood shaving bio-logical trickling filter (PW), and a sand and gravel wetland (AWL).The effluent of each passive treatment system was released to thenatural attenuation zone in a controlled manner. An overall systemdesign diagram is presented in Fig. 1, and the following subsectionssummarize the system design which was previously detailed bySpeer (2011) and Speer et al. (2012). The values for the systemoperating parameters presented in the following subsections weretested, optimized, and adopted through a previous study of thesystem, as conducted and presented by Speer (2011) and Speeret al. (2012).

2.1.1. Pre-treatment systemRaw leachate from the leachate collection system continuously

fed the pre-treatment system at a flow rate of 4 m3/d. The pre-treatment system consisted of: (1) a 3-day HRT, 12 m3 aeratedfixed-film reactor; and (2) a 1-day HRT, 4 m3 mixing and storagetank. Aeration was maintained at a rate of 0.43 m3/min. Biologicalattachment was encouraged through the use of packing in thefixed-film reactor composed of an inert plastic medium (1 percent(%) of the total reactor volume).

2.1.2. Passive treatment subsystemsPre-treated leachate from the pre-treatment system fed each

passive treatment subsystem with identical flow rates of 2 m3/d.The PW filter consisted of a two-cell vertical subsurface flow con-figuration, with cell media consisting of a 25% peat and 75% wood

ent system designed and installed by Speer (2011), with sampling frequencies and

thod

W� Sensolyt� DWAerican Public Health Association (APHA) Standard Method 2320BW� Sensolyt� DWAW� Sensolyt� DWA� Tetracon� 700 IQ� FDO� 700 IQchromatography

uctively coupled plasma (ICP) mass spectrometry (MS) and optical emissionctrometry (OES)chromatography

s chromatography

HA Standard Method 2540DHA Standard Method 2540DHA Standard Method 2540DHA Standard Method 2540DHA Standard Method 2540D

HA Standard Method 4500-NH3-Dculationchromatographychromatography

HA Standard Method 4500-Norg-BHA Standard Method 4500-Norg-B

HA Standard Method 4500-P-EHA Standard Method 4500-P-E

chromatographyHA Standard Method 4500-P-E

tometrictometric

HA Standard Method 5210B

HA Standard Method 5210BHA Standard Method 5220DHA Standard Method 5220DHA Standard Method 5210BHA Standard Method 5210B

MS and ICP OES




shavings mixture, selected from a previous bench-scale study(Speer, 2011). Cell 1 provided for aerobic treatment in a downwardflow configuration. Treatment was achieved through intermittentfeed-and-rest dosing cycles, with six 0.33 m3 doses per day andas described by Chazarenc et al. (2009). Cell 2 provided for anaer-obic treatment in an upward flow configuration.

The AWL wetland consisted of four cells with sand and gravelmedia. Each cell contained salt tolerant vegetation, consisting ofcommon reeds and manna grass obtained from a nearby naturalwetland system to ensure their tolerance to the regional climate.Cells 1, 2, and 4 provided for aerobic treatment in a downward flowconfiguration. A dosing cycle similar to the one applied to the PWfilter was used, with eight 0.25 m3 doses per day. Cell 3 providedfor anaerobic treatment in an upward flow configuration.

Table 2Statistical data subsets and the sampling points included in each subset for PCA andregression analysis of the performance of the hybrid-passive treatment system.

Subset name Sampling points included Ratio betweenobservationsand variables

All LE, FF, ST, PW1, PW2, WE1, WE2, WE3, WE4 15.3PW-system LE, FF, ST, PW1, PW2 8.6AWL-system LE, FF, ST, WE1, WE2, WE3, WE4 12Pre LE, FF, ST 5.3PW PW1, PW2 3.4WE1 WE1, WE2, WE3, WE4 6.7

2.2. Parameters of interest

Parameters of interest for the study, along with sampling fre-quency and analytical methods, are presented in Table 1. The waterquality parameters monitored during study were selected based onprevious work by Speer (2011), Speer et al. (2011), Speer et al.(2012), where sample collection and analyses are provided indetail. These parameters of interest are frequently reported aseither: fundamental measures of water quality for wastewaters,in the case of pH, alkalinity, temperature, ORP, DO, conductivity,solids, and organic matter (Snoeyink and Jenkins, 1980); toxic con-stituents of landfill leachate, in the case of metals of interestdeemed to be ‘‘heavy’’ in terms of human health and environmen-tal impact, as defined below, sulfate, sulfide, chloride, phenols, andorganic matter (Chiemchaisri et al., 2009; Sawaittayothin andPolprasert, 2006; Ribé et al., 2012); or, nutrients that may causeeutrophication, in the case of the N and P species (Barsanti andGualtieri, 2006).

A suite of metals of interest were analyzed using inductivelycoupled plasma (ICP) mass spectrometry (MS) and optical emissionspectrometry (OES): silver (Ag), aluminum (Al), arsenic (As), bar-ium (Ba), calcium (Ca), cadmium (Cd), cobalt (Co), chromium(Cr), copper (Cu), iron (Fe), potassium (K), magnesium (Mg), man-ganese (Mn), molybdenum (Mo), sodium (Na), nickel (Ni), lead(Pb), tin (Sn), strontium (Sr), titanium (Ti), thallium (Th), uranium(U), vanadium (V), and zinc (Zn). ‘‘Heavy’’ metals were defined asthose with atomic weights above calcium and thus included allmetals of interest except aluminum, calcium, potassium, magne-sium, and sodium (Venugopal and Luckey, 1975; Hale andMargham, 1988; Hawkes, 1997; Hogan, 2010). Concentrations ofheavy metals were calculated as the sum of the individual concen-trations of the constituent metals as defined.

In evaluating treatment performance, key parameters related totoxicity of the treated leachate and performance of the hybrid-pas-sive treatment system were isolated in the analysis as treatmentobjectives. Of these parameters, ammonia, heavy metals, andorganic matter (indicated as COD), are generally considered theprimary pollutants in landfill leachate (Chiemchaisri et al., 2009;Sawaittayothin and Polprasert, 2006; Ribé et al., 2012). Theseparameters are often important targets for management in dis-charge regulations. Alkalinity is an essential indicator of the receiv-ing body’s health as a measure of pH buffering capacity (Neal,2001). Additionally, iron, while classified as a heavy metal, war-ranted individual examination due to the potential for cloggingfrom iron precipitates (Speer, 2011; Speer et al., 2013). Thus, theevaluation of treatment performance was framed around thedynamics contributing to changes in the concentrations of the fol-lowing criteria parameters, set as the dependent variables inregression analyses: ammonia, alkalinity, COD, heavy metals (asdefined), and iron.


2.3. Field sampling and laboratory analysis

Effluent samples were taken from each stage of the hybrid-pas-sive treatment system (Fig. 1) on a bi-weekly basis and analyzed,as specified by Speer et al. (2012), and according to APHA (2005).The eight sampling points in order of leachate flow are indicatedas follows: raw leachate influent (LE), fixed-film reactor effluent(FF), storage tank effluent (ST), AWL cell 1 effluent (WE1), AWL cell2 effluent (WE2), AWL cell 3 effluent (WE3), AWL cell 4 effluent(WE4), PW cell 1 effluent (PW1), and PW cell 2 effluent (PW2).For all parameters, values below the detection limit (DL) werereported as one half of the DL (Speer, 2011).

2.4. Statistical analysis

Statistical analyses for all sampling points were performed on adataset consisting of the measured concentrations using JMP�

10.0.0 software, produced by SAS�. The multivariate modelingpackage includes PCA, time series, multiple regression, and PLScapabilities. At times, the dataset from the study contained missingdata and the JMP� imputation tool was used to estimate missingvalues prior to PCA. In some cases, imputation estimated negativenumbers for missing values. These estimates were set to one half ofthe individual DLs for all parameters, with the exception of ORP, forwhich negative values are possible. Data from individual samplingpoints were converted to loadings through multiplication with theaverage daily flow rates for each point: 4 m3/day for LE andthe pre-treatment system points; and 2 m3/day for all points inthe AWL wetland and PW filter systems. A log-transformationwas applied to the dataset to increase the normality of the data dis-tribution and the suitability of parametric analytical methods suchas linear regression. For ORP, a constant equal to the lowest nega-tive number plus one was added to observations to allow log-transformation, as suggested by Wicklin (2011). Normality wasassessed using the Shapiro–Wilk test for a normal distribution fitover a histogram of the data for each parameter. This test wasapplied to both a non-transformed dataset and a log-transformeddataset, with higher Shapiro–Wilk W values indicating better fitsto the normal distribution (SAS, 2013). Summing the W valuesfor each parameter and dividing by the number of parametersresulted in average W values of 0.647 and 0.812 for the un-trans-formed and log-transformed datasets, respectively; indicatingincreased normality of the distribution for the log-transformeddata. Hence, the log-transformed data was incorporated furtherinto the analysis.

Statistical analyses were performed on data subsets capturingthe range of conditions in the system as noted in Table 2. Due tothe large number of interacting parameters, and the limited num-ber of observations, two sets of statistical analyses were performedfor each subset: one with the total parameter fractions (whereapplicable) and one with the dissolved parameter (soluble) frac-tions. For all subsets, PCA runs were performed with each one




excluding a criteria parameter, with each criteria parameter set asthe dependent variable in PLS and PCA regressions.

For all PCAs, the ratio between the number of observations andvariables in a dataset is an important consideration in PCA accu-racy, with Suhr (2005) and Allen (2002) citing a minimum of 5–10 observations per variable, with more observations increasingreliability of the results. In the data from this study, the best ratioof observations to variables that was possible was 15.3 when allsampling points were included, and only the PW subset had a ratiobelow 5 (Table 2). From the first PCA results for all subsets, evalu-ation of the principal variables contributing to variance in the data-set was performed using the Methods B2 and B4 presented byJolliffe (1986) and used by King and Jackson (1999). In methodB2, each variable most highly associated with the last K-p compo-nents, determined by highest absolute loading value, is removedsequentially. In method B4, each variable most highly associatedwith the first p components, determined by highest absolute load-ing value, is kept sequentially (Jolliffe, 1986; King and Jackson,1999). Both methods first require the selection of a number p ofPCs to retain based on a given criteria. King and Jackson (1999) rec-ommended the following criteria for choosing p: (1) all PCs witheigenvalues greater than 0.6; (2) all PCs that contributed to acumulative percentage of variation explained of greater than90%; and (3) utilizing the broken-stick procedure as presented byKing and Jackson (1999) and Peres-Neto et al. (2003). The highestp value generated from application of each of the three criteriawas used, to remain conservative. The p variables retained are con-sidered to be the ones that contribute most to the variation in thedata, while the K-p variables are considered redundant in explain-ing the data (King and Jackson, 1999).

After performing PCA, regression analyses were performed todetermine the most representative quantitative relationships fromthe water quality parameters and PCs to model the concentrationsof the criteria parameters in the system. PLS regressions were per-

Table 3Physical and chemical characteristics of the raw leachate entering the pre-treatment syste

Water quality Cations

Parameter Average + Std. Dev. Parameter

pH 7 ± 0.54 AgDO 0.97 ± 0.8 AlTemperature (�C) 14.8 ± 2.8 AsConductivity (mS/cm) 13.7 ± 16.3 BORP (mV) �25.3 ± 110.3 Basulfate 93.4 ± 174.4 CaChloride 966.7 ± 321.5 CdAlkalinity 4289.5 ± 1319.8 CoPhenols (lg/L) 63.3 ± 127.8 CrOxygen demand CuBOD5 444.9 ± 750.6 FecBOD5 279.2 ± 447.6 KCOD 1223.2 ± 737.4 MgSolids MnTSS 116.4 ± 266.1 MoTDS 5309.8 ± 1536 NaNitrogen (mg/L-N) NiTKN 482.7 ± 140.8 PAmmonia 523.7 ± 156.3 PbOrganic N 76.5 ± 74.4 SNitrite 1.9 ± 1.5 SnNitrate 2.2 ± 2.2 SrTotal nitrogen 525.8 ± 115.8 TiPhosphorous TlTotal P 2.8 ± 1.8 UTotal P, sol. 1.7 ± 1 VOrthophosphate 2.3 ± 1.5 ZnOrthophosphate, sol. 1.4 ± 1.3Hydrolysable P 2.1 ± 1.1Hydrolysable P, sol. 1.1 ± 0.9


formed on the p variables selected for each subset, with the indi-vidual criteria parameters as the dependent variable in each case.PC standard least squares regressions were performed on the PCscores generated by PCA runs on the p variables for each subset,with the individual criteria parameters as the dependent variablein each case. Both regression techniques are suitable for multivar-iate situations, as they eliminate the problem of multi-collinearity(Abdi, 2003). The combination of PCA method and regression typethat produced the highest R2 value was deemed the most suitablemodel for each individual subset and dependent variable.

3. Results

3.1. Removal efficiencies

The physical and chemical characteristics of the raw leachateare presented in Table 3. Concentrations of the criteria parametersat the influent and effluent sampling points for the each treatmentsystem (LE, ST, PW, and WE), are presented in Fig. 2. Averageremoval efficiencies for each of the criteria parameters for the indi-vidual treatment components and across the entire AWL and PWsubsystems are presented in Table 4 and were calculated accordingto Eq. (1). Total ammonia was most successfully removed, and theAWL subsystem exhibited higher removal efficiencies than the PWsubsystem for all parameters with the exception of heavy metals.

%Cavg;m�n ¼Avg:Cm � Avg:Cn

Avg:Cm� 100% ð1Þ

where, %Cavg,m-n represents the average removal efficiency of thegiven parameter across sampling points m-n, Avg. Cm the averageconcentration at sampling point m (mg/L) and Avg. Cn the averageconcentration at sampling point n (mg/L).

m; all units in mg/L unless noted otherwise.

Dissolved cations

Average + Std. Dev. Parameter Average + Std. Dev.

0.02 ± 0.03 Ag 0.02 ± 0.030.44 ± 0.75 Al 0.22 ± 0.25

0.1 ± 0.1 As 0.1 ± 0.17.2 ± 1.9 B 7.6 ± 2.90.3 ± 0.2 Ba 0.3 ± 0.1

255.2 ± 76.1 Ca 229 ± 95.60.1 ± 0.1 Cd 0.1 ± 0.10.1 ± 0.1 Co 0.1 ± 0.10.1 ± 0.1 Cr 0.1 ± 0.10.1 ± 0.2 Cu 0.1 ± 0.2

22.6 ± 27 Fe 16.2 ± 25.2417.3 ± 397.7 K 366.2 ± 98.1149.5 ± 34.8 Mg 143.2 ± 40.3

2 ± 1.5 Mn 1.7 ± 1.50.1 ± 0.1 Mo 0.1 ± 0.1

918 ± 268.2 Na 898.3 ± 284.30.2 ± 0.1 Ni 0.2 ± 0.22.1 ± 1.8 P 1.4 ± 1.30.1 ± 0.1 Pb 0.1 ± 0.1

20.1 ± 34.8 S 19.2 ± 35.40.1 ± 0.2 Sn 0.1 ± 0.21.7 ± 0.3 Sr 1.6 ± 0.40.1 ± 0.1 Ti 0.1 ± 0.10.1 ± 0.2 Tl 0.1 ± 0.20.2 ± 0.4 U 0.2 ± 0.30.1 ± 0.1 V 0.1 ± 0.10.4 ± 0.4 Zn 0.2 ± 0.3



0

5

10

15

20

25

30

35

Nov-09 Feb-10 May-10 Sep-10 Dec-10 Mar-11 Jul-11

Heavy

Metals(m

g/L)

LE ST WE4 PW2

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000


COD(m

g/L)

LE ST WE4 PW2

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000


Ammon

ia(m

g/L)

LE ST WE4 PW2

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

45,000

50,000


Alkalinity(m

g/L)

LE ST WE4 PW2

200

250LE ST WE4 PW2

0

50

100

150


Iron

(mg/L)

Fig. 2. Concentrations of criteria parameters (alkalinity, ammonia, COD, heavy metals, iron) over the study period at the LE, ST, PW, and WE sampling points.

Table 4Average removal efficiencies and standard deviations of criteria parameters for each hybrid-passive treatment system component and the overall system.

Parameter Pre-treatment system (%) Passive treatment subsystems Overall

AWL (%) PW (%) AWL (%) PW (%)

Alkalinity 27 ± 30 84 ± 13 64 ± 12 87 ± 12 73 ± 12Ammonia 15 ± 76 87 ± 16 47 ± 132 91 ± 12 74 ± 17COD 22 ± 30.5 62 ± 13 50 ± 6 70.5 ± 13 62 ± 14Heavy Metals 47 ± 17 29 ± 31 51.5 ± 14 63 ± 19.5 75.5 ± 8.5Iron 64 ± 44.5 82 ± 10.5 83 ± 11 94.5 ± 5.5 93.5 ± 9.5Average 38.5 ± 36.3 64.6 ± 17.3 60 ± 25.8 79.3 ± 12.5 76.1 ± 11.6


3.2. PCA loading plots

Score and loading plots for PCA runs on the All, AWL system, andPW system data subsets (total fractions) are presented in Fig. 3 andshow that most parameters, with the exception of ORP and NO3

�,were generally well aligned and related, with the first PCs in eachrun accounting for 41.3%, 43.8%, and 39.5% of the variation in thedatasets for the All, AWL system, and PW system subsets, respec-tively. ORP and NO3

� were characterized as unique since they werethe only parameters to be negatively loaded onto the first PC, sug-gesting that they were negatively correlated to the other parame-ters and were expected to be important in describing system


performance. The strong rightward directions of the loadings ofall other parameters on the first PCs, for each dataset (a, b, and c)suggests that they are all strongly associated with these first PCs,which in turn had high percentages of variation explained. Thus,the parameters are well captured by the PCA conducted as theircontributions to changes in the data are accounted for to a largedegree by these first PCs. These trends are witnessed across theAll, AWL system, and PW system datasets shown in Fig. 3. The vari-able reduction procedure applied to each PCA run generallyretained approximately 50% of the variables, reducing the parame-ters describing the system from 33 to 15, in the case of the All sub-set for the total fraction.



(a)

(b)

(c)

FeHeavy

AlkNH3

COD

FeHeavy

COD

NH3 Alk

AlkCOD

Heavy

NH3

Fe

Fig. 3. PCA score (left) and loading (right) plots showing the PC scores ofobservations and loadings of variables on the first and second PCs, with percentagesindicating the proportion of variation in the data explained by the PCs, for: (a) Allsubset (total fraction); (b) AWL-system subset (total fraction); (c) PW-system subset(total fraction).

Fig. 4. Maximum R2 values for regression models performed on each data subset(total fraction) for each criteria parameter as a dependent variable in the modelfrom all PCA and regression methods tested, with (�) indicating the best regressionmodel for the specified criteria parameter within each data subset and (⁄)indicating the best regression model for the specified data subset within eachcriteria parameter.

Fig. 5. Maximum R2 values for regression models performed on each data subset(soluble fraction) for each criteria parameter as a dependent variable in the modelfrom all PCA and regression methods tested, with (�) indicating the best regressionmodel for the specified criteria parameter within each data subset and (⁄)indicating the best regression model for the specified data subset within eachcriteria parameter.


3.3. PLS and PC regressions

Results of PLS and PC regressions on each of the data subsets arepresented for the total and soluble fractions in Figs. 4 and 5,


respectively, in which the maximum R2 values from each combina-tion of PCA method (B2, B4) and regression type are shown graph-ically. 80% of the regression models had R2 values above 0.6,indicating relatively good fits on an overall basis.

The regression models were examined for suitability from twoperspectives according to maximum R2 values: the determinationof criteria parameter(s) that best modeled the system within eachdata subset, with the criteria parameters as dependent variables inthe regression models (as represented by (�) in Figs. 4 and 5); and,the determination of the data subset that represented the bestregression model determined for each individual criteria parame-ter (as represented by (⁄) in Figs. 4 and 5). From these two perspec-tives, the best regression models, as measured by the highest R2

value, indicated which criteria parameter was best explained bythe data within each data subset, and which data subset providedthe best regression model for each criteria parameter. Generally,these regression models allowed for the understanding of theimportance of the criteria parameters with respect to each compo-nent of the overall treatment system. For the total fraction (Fig. 4),the criteria parameters that generated the best regression models




(with criteria parameter as the dependent variable) for each subsetwere found to be: COD for All; iron for PW-system; alkalinity forAWL-system; alkalinity for Pre; heavy metals for PW; and COD forAWL. The datasets for which the best regression models were gen-erated for each criteria parameter were found to be: PW-system forheavy metals and iron; AWL-system for alkalinity and COD; and Prefor ammonia. For the soluble fraction (Fig. 5), the most appropriateregression models for each subset were noted as: COD for All;heavy metals for PW-system; alkalinity for AWL-system; alkalinityfor Pre; heavy metals for PW; and COD for AWL. The datasets forwhich the best regression models were generated for each criteriaparameter were found to be: All for COD; AWL-system for alkalinityand iron; PW for heavy metals; and AWL for ammonia.

The regression models considered to be appropriate for furtheranalysis were those that matched the two evaluation perspectivesutilized: those that had the highest R2 value within a data subset ofall regression models tested for a particular criteria parameter, orthe highest R2 value within a criteria parameter of all regressionmodels tested for a particular data subset; as indicated for total(Fig. 4) and soluble (Fig. 5) fractions. From these best-fit models,the frequency that each individual parameter was noted to be anexplanatory variable in the models was determined, providing anindication of which parameters were most often used and werethus describing much of the overall system (Table 5). The signifi-cance of the frequencies was assessed by a t-test relative to theexpected frequency (average number of counts for all entrieswithin a subset), with p-values below a = 0.05 (95% confidencelevel) indicating significant frequencies. Based on this test of thefrequencies, the following 8 parameters were considered keyexplanatory variables with respect to modeling the criteria param-eters: conductivity, DO, NO2

�, organic N, ORP, pH, sulfate and TVS.

4. Discussion

4.1. System performance

The AWL subsystem was found to be more effective at removingall criteria parameters than the PW subsystem, with the exceptionof heavy metals. Notable changes in treatment efficiency due toseasonal conditions and temperature were observed for ammonia(Fig. 2) primarily during the first winter period, which can beattributed to the limited cold temperature dosing protocols imple-mented during the start up of the systems, particularly recircula-tion rates for the AWL system. (Speer, 2011; Speer et al., 2012).For winter operation (October–March), both systems weredesigned with submerged (30 cm below ground surface) dosingmanifolds (Fig. 1) to utilize the natural insulating capacity of thematerial above the dosing manifolds. The submerged manifoldswere noted to play a key role in maintaining nitrification and over-all treatment performance in the systems during cold weather

Table 5Frequency that parameters (those considered statistically significant based on a t-test of thbest fit models, as determined by having the highest R2 value within a set of models for a

Parameter Selection based on variable

Total Soluble

Cond. 5 4DO 5 5NO2� 5 4

Org. N 5 5ORP 5 5pH 4 5Sulfate 4 5TVS 5 5

a p-Value not calculated because standard deviation is zero.


Speer et al. (2012), and likely minimized the effects of seasonalchanges on treatment efficiency (Fig. 2).

The types of media employed in each of the systems were alsoreported to be responsible for the differences between the two sys-tems in terms of nitrification and COD removal, as the high poros-ity and particle surface area of the peat in the PW system allowedfor greater Ca and Fe sorption and ion exchange than in the AWLsystem (packed with inert sand and gravel), thus depriving hetero-trophic nitrifying bacteria of key nutrients and lowering nitrifica-tion rates (Speer et al., 2012). Speer et al. (2012) also noted thatit was expected that once the PW system had maturity withrespect to sorption capacity, nitrification would be expected toincrease as these nutrients would become more readily available.

The higher PC percentages in PCA (Fig. 3) and higher R2 values inregressions (Figs. 4 and 5) for the AWL wetland compared with thePW filter suggests that the dataset better described the dynamicsoccurring in the former and may have reflected the higher overallperformance achieved by the AWL wetland compared with the PWfilter, with respect to the five criteria parameters. No distinct dif-ferences were noted between the results obtained for the totaland soluble fractions, as the PCA and regressions for both fractionsproduced similar loadings and R2 values.

4.2. Criteria parameters for treatment

The criteria parameters were generally well explained by theregression models consisting of the retained parameters, suggest-ing that leachate treatment in the overall system was associatedprimarily with these parameters. The regressions in the AWL sys-tem subset had the highest average R2 values, which may indicatethat treatment goals were best met in this system, as the data arebetter modeled by the regressions than in the PW system. High R2

values indicate that the criteria parameters (dependent variable)were well explained by the retained parameters (independent vari-ables), which in turn suggests that the changes in the data werewell captured by the models. Since we attribute changes in the cri-teria parameters to treatment of them, then we can say that theregression models help indicate how treatment goals were met.As with the average removal efficiencies, the exception to thistrend was associated with the heavy metals, which were betterremoved in the PW filter (Speer, 2011).

The leachate-impacted groundwater entering the wetlands andriver adjacent to the site was noted to have alkalinity concentra-tions 1000 times higher than that generally found in natural sur-face waters of the area (CRA, 2003). While alkalinity was reducedconsiderably in the overall system (87% in the AWL wetland and73% in the PW filter), the concentrations in the effluents of theAWL and PW subsystems remained relatively high with averageconcentrations of 2069 and 4469 mg/L as CaCO3, respectively(Speer, 2011). As expected, regression analyses explained alkalinity

e means of the frequencies, with p-value shown) appear as predictor variables in thecriteria variable or for a data subset.

Selection based on subset t-Test p-value

Total Soluble

5 4 0.00255 5 –a

4 3 0.01255 5 –a

4 3 0.01334 5 0.00233 4 0.01255 5 –a




well in the AWL-system data subset, where the highest removalefficiencies were observed. In the PCA loading plot for the AWL-sys-tem subset (total fraction), it was found that alkalinity was highlyloaded to the first PC along with most other parameters (Fig. 3b),with a loading coefficient of 0.915. Alkalinity was weakly nega-tively loaded with the second PC (�0.093), along with total ammo-nia, BOD, cBOD, organic N, orthophosphate, total P, hydrolysablephosphorus (P), phenol, TSS, Al, Ca, Fe, and heavy metals.

Ammonia is extremely toxic to receiving environments, is amajor constituent of landfill leachate (Chiemchaisri et al., 2009;Sawaittayothin and Polprasert, 2006), and is the primary N formin landfill leachates (Speer et al., 2012). It is often reported to bethe most toxic component of landfill leachate (Ribé et al., 2012).In the PCA loading plot for the AWL-system subset, it was foundthat ammonia was highly loaded to the first PC (Fig. 3b), with aloading coefficient of 0.788. Ammonia was weakly negativelyloaded with the second PC (�0.124), along with alkalinity, BOD,cBOD, organic N, orthophosphate, total P, hydrolysable P, phenol,TSS, Al, Ca, Fe, and heavy metals. The opposite sign for NO3

� andNO2� loadings on the second PC compared to ammonia suggests

that nitrification was likely an important ammonia removal mech-anism in the system (Speer et al. 2012). Additionally, the negativecorrelation between ammonia and Ca and Fe would indicate theconsumption of these cations, possibly by nitrifying bacteria, inthe systems as noted in Section 4.1.

COD is a measure of the amount of organic matter in the leach-ate. In the PCA loading plot for the AWL-system subset, it was deter-mined that COD was highly loaded to the first PC (Fig. 3b), with aloading coefficient of 0.932. COD was weakly positively loaded withthe second PC (0.176), along with pH, ORP, conductivity, chloride,NO3�, Total N, total solids (TS), TVS, total fixed solids (TFS), total dis-

solved solids (TDS), K, Mg, and Na. Additionally, Speer et al. (2012)noted a strong statistical link between COD and nitrification in thepassive treatment subsystems, with a high level of nitrificationreducing COD, and COD consumption corresponding to ammoniaconversion into NO3

� (nitrification). This was noted in the PCA bythe opposite loadings between COD and NO3

� on the first PC, sug-gesting a negative statistical relationship, and the aligned loadingsbetween COD and ammonia on the first PC, suggesting a positivestatistical relationship.. In a study of leachate quality from an activelandfill, Durnusoglu and Yilmaz (2006) reported that COD was anessential parameter for explaining variation in the data, which wehave determined to be the case for this system, as it was highlyloaded onto the first PC and thus contributed to a high level of var-iation in the dataset. In terms of removal efficiencies, the higherCOD removals observed in the AWL system compared to the PWsystem are likely related to the higher nitrification and ammoniaremoval also observed in this system which favoured the growthof heterotrophic bacteria as noted by (Speer et al., 2012).

While generally considered a ‘‘heavy’’ metal, iron was examinedindividually due to its potential effect on biofilter media clogging(Venugopal and Luckey, 1975; Hale and Margham, 1988;Hawkes, 1997; Hogan, 2010; Speer et al., 2012). Iron was reducedvery effectively in both passive systems, with average reductions of94.5% and 93.5% for the AWL and PW subsystems, respectively(Speer, 2011; Speer et al., 2012). For the total fraction, the regres-sion on the PW-system subset explained iron the best; while theregression on the AWL-system subset best explained the solublefraction. In the PCA loading plot for the AWL-system subset, ironwas found to be highly loaded to the first PC (Fig. 3b), with a load-ing of 0.805. Iron was weakly negatively loaded with the second PC(�0.302), along with pH, ORP, conductivity, chloride, NO3

- , Total N,TS, TVS, TFS, TDS, K, Mg, and Na.

Unlike the other criteria parameters, the ‘‘heavy’’ metals werereduced more effectively in the PW filter, with average reductionsof 76% compared to 63% in the AWL wetland (Speer, 2011). In the


PCA loading plot for the PW-system subset, heavy metals werehighly correlated to the first PC (Fig. 3c), with a loading of 0.855.Heavy metals were weakly negatively correlated with the secondPC (�0.26), along with BOD, cBOD, orthophosphate, TP, hydrolysa-ble P, phenols, TSS, Al, Ca, and Fe. The higher removal in the PWsystem compared to the AWL system was likely due to the devel-opment of higher biologically mediated sorption and ion exchangecapacity in the peat media of the PW system, compared to the inertsand and gravel media in the AWL system (Speer et al., 2012). Thiswas expressed in the PCA through the positive correlations to BODand TSS, both representing bacteria and suspended organic matterthat may be associated with an active biofilm. Removal efficienciesfor BOD and TSS were approximately 6% and 17% higher in the AWLsystem than the PW system, respectively, suggesting either thereduced retention or new generation of solids and microbial bio-mass in the peat and woodchip media.

4.3. Identification of significant parameters

Through the PCA and regression analyses, eight significantexplanatory parameters were identified on a statistically signifi-cant level (p-values < 0.05): conductivity, DO, NO2

�, organic N,ORP, pH, sulfate, and TVS (Table 5). These parameters appearedas retained variables across all subsets for frequencies consideredstatistically significant according to a t-test of the means of the fre-quencies, and may be considered the key parameters in describingsystem treatment performance with respect to the criteria param-eters. pH is a vital measure of any water quality system, and drivesall acid-base equilibria (Snoeyink and Jenkins, 1980). Conductivityis a measure of the amount of charged ions in the leachate. In astudy by Ribé et al. (2012), where landfill leachate was treatedwith a pine bark biosorbent followed by exposure of organismsto the treated leachate to determine toxicity, it was found thatthe leachate samples highly associated with pH, conductivity,and heavy metal concentration in PCA were most strongly corre-lated to an increased toxic response in daphnia and algae, suggest-ing an important role in toxicity of leachate. These parameterswere confirmed to be important in this study in terms of interac-tions with the criteria parameters. ORP, which was observed tobe an outlier on PCA score plots (Fig. 3), was identified as one ofthe key parameters from the regressions. Leachate contains highconcentrations of charged metal species, which can be solubilizedor precipitated based on the redox conditions (as measured by thevariable ORP) and pH (Snoeyink and Jenkins, 1980; Durnusogluand Yilmaz, 2006). NO3

�, which was the other parameter that wasobserved to be an outlier on the PCA loading plots (Fig. 3), wasnot identified to be important, but the closely associated NO2

�

was identified as a key parameter from the regressions. NO2� is

an intermediary compound in biological nitrification, which wasnoted to be a dominant mechanism for ammonia removal in thepassive treatment systems (Speer, 2011; Speer et al., 2012). Nitri-fication in the passive treatment subsystems was also noted tobe strongly correlated to COD, another criteria parameter, bySpeer (2011) and Speer et al. (2012). DO is required for nitrificationand other aerobic biological processes that were supported in Cell1 of the PW filter and Cells 1, 2, and 4 of the AWL wetland (Speer,2011). Sulfate in landfill leachate could contribute to the reductionof COD by sulfur-reducing bacteria and to the precipitation of met-als through the formation of metal sulfides which may in turn leadto the generation of alkalinity and a resulting pH increase (Nedwelland Reynolds, 1996; Speer et al., 2012). In a study on the quality ofa river and wetland receiving untreated landfill leachate, Nyameet al. (2012) reported that sulfate was a statistically significantconstituent, as determined through strong loadings to the firstand second PCs in relation to other physical and chemical param-eters. TVS was likely retained as a significant parameter as they




constitute volatile fatty acids (VFAs), which formed up to 20% ofthe overall oxygen demand of the leachate at the site andaccounted for up to 18% of the COD removed in the system(Speer et al., 2012). VFAs are most closely associated with theanaerobic phase of landfill waste degradation, which consists ofthe breakdown of the waste to amino acids (containing organicN) and VFAs (Speer, 2011). The importance of organic N in landfillleachate was likely related to the anaerobic phase of landfill wastedegradation and the hydrolyzing of organic N to ammonia (Speer,2011).

4.4. Appropriateness of multivariate techniques

PCA and other multivariate statistical techniques to interpretand evaluate landfill leachate environmental effects and treatmentperformance have been applied in a number of studies (Clémentet al., 1997; Durnusoglu and Yilmaz, 2006; Ziyang et al., 2009;Nyame et al., 2012; Ribé et al., 2012). This study has shown theapplicability of PCA and PLS and PC regressions in determiningkey variables for treatment performance with respect to individualcriteria parameters and reducing large datasets to more manage-able parameter listings. These methods may be used as a screeningtool in long-term leachate treatment monitoring, whereby themost important parameters with respect to treatment goals maybe identified and sampling and monitoring programs may bedeveloped accordingly.

5. Conclusions

This study identified the principal parameters affecting the per-formance of a novel hybrid-passive treatment system receivingleachate collected from an operating MSW landfill near NorthBay, Ontario, Canada. A comprehensive set of water chemistryparameters were monitored bi-weekly over a 21-month period.The data were analyzed using PCA and PC and PLS regression anal-yses in order to understand the primary parameters influencingperformance and the interactions and trends between them, withrespect to treatment objectives to minimize the impacts of thetreated leachate on the receiving environment. The principalparameters and their effect on treatment objectives were exam-ined in terms of five criteria parameters: ammonia, alkalinity,COD, heavy metals, and iron.

The removal percentages and multivariate analyses confirmedthat the AWL wetland subsystem provided better treatment forfour of the criteria parameters: ammonia, alkalinity, COD, and iron;while the PW filter subsystem provided better treatment for heavymetals. PCA indicated a high degree of association with the first PCfor all parameters except NO3

� and ORP, and a high percentage ofvariation (>40%) was explained by this PC, suggesting strong statis-tical relationships among the parameters. A variable reduction pro-cedure reduced the dataset from 33 to between 15 and 17 variablesthat were considered the most important in explaining the varia-tion in the data. Of these 15–17 variables within each PCA, regres-sion analyses further identified 8 explanatory variables that weremost frequently retained for modeling the criteria parameters, ona statistically significant level (p-values < 0.05): conductivity, DO,NO2�, organic N, ORP, pH, sulfate, and TVS. These significant param-

eters are often associated with landfill leachate as noted in otherstudies (Nedwell and Reynolds, 1996; Durnusoglu and Yilmaz,2006; Speer, 2011; Ribé et al., 2012; Nyame et al., 2012), and theirretention in the modeling was found to be related to treatmentperformance as reported in previous studies (Speer, 2011; Speeret al., 2012; Wallace et al., 2012).

Multivariate statistical tools, including PCA and PLS and PCregressions, are valuable in determining key parameters and their


relationships and may be successfully applied to landfill leachatedata, as shown in this study. The long-term monitoring burdenassociated with MSW landfills and leachate treatment, and thewidespread implementation of landfills throughout the worldhighlight the importance of statistical evaluation in identifying sig-nificant parameters and treatment trends (Mulamoottil et al.,1999; Rew and Mulamoottil, 1999; Durnusoglu and Yilmaz,2006). The criteria parameters and the significant explanatoryparameters were most important in modeling the dynamics ofthe hybrid-passive treatment system during the study period,and such techniques and procedures may be applied to other sitesto determine parameters of interest in similar fashion.

Acknowledgements

The authors thank the City of North Bay, SNC-Lavalin, Cones-toga-Rovers & Associates, and AQUA Treatment Technologies, fortheir technical and financial support throughout the project. Addi-tional financial support from the Ontario Centers of Excellencethrough the Interact program, the Natural Sciences and Engineer-ing Research Council (NSERC) thorough the Collaborative Researchand Development (CRD) and the Systems Training and Education inWater Assets Research and Development (STEWARD) CollaborativeResearch and Training Experience Program (CREATE) programs,and the Canada Research Chair (CRC) program for Dr. Champagne’sChair in Bioresource Engineering, are gratefully acknowledged. Theauthors also acknowledge the considerable work conducted by Dr.S. Speer that formed the basis for the results presented in thispaper.

References

Abdi, H., 2003. Partial least squares regression. In: Lewis-Beck, M., Bryman, A.,Futing, T. (Eds.), Encyclopedia of Social Science Research Methods. Sage,Thousand Oaks, California, USA.

Allen, J.B., 2002. Principal components analysis. PCA Soundbyte, 4/29/02. http://apsychoserver.psychofizz.psych.arizona.edu/JJBAReprints/PSYC501A/pdfs2008/PCAEXPL.pdf (accessed March 27.2013).

American Public Health Association (APHA), 2005. In: Clesceri, L.S., Greenberg, A.E.,Rhodes Trussell, R. (Eds.), Standard Methods for the Examination of Water andWastewater. American Public Health Association, Washington, DC, twenty first ed..

Barsanti, L., Gualtieri, P., 2006. Algae: Anatomy, Biochemistry, and Biotechnology.CRC Press, Boca Raton, Florida, USA.

Chazarenc, F., Gagnon, V., Comeau, Y., Brisson, J., 2009. Effect of plant and artificialaeration on solids accumulation and biological activities in constructedwetlands. Ecol. Eng. 35 (6), 1005–1010.

Chiemchaisri, C., Chiemchaisri, W., Junsod, J., Threedeach, S., Wicranarachchi, P.N.,2009. Leachate treatment and greenhouse gas emission in subsurfacehorizontal flow constructed wetland. Bioresour. Technol. 100 (16), 3808–3814.

Clément, B., Janssen, C., Le Du-Delepierre, A., 1997. Estimation of the hazard oflandfills through toxicity testing of leachates; 2. Comparison of physico-chemical characteristics of landfill leachates with their toxicity determinedwith a battery of tests. Chemosphere 35 (11), 2783–2796.

Conestoga-Rovers and Associates (CRA), 2003. Updated Design and Operations Plan,Merrick Landfill Site, North Bay, Ontario. Report to the City of North Bay, May2003.

Durnusoglu, E., Yilmaz, C., 2006. Evaluation and temporal variation of raw and pre-treated leachate quality from an active solidwaste landfill. Water, Soil, AirPollut. 171, 359–382.

Hale, W.G., Margham, J.P. (Eds.), 1988. Collins Dictionary of Biology. Collins,Glasgow, UK.

Hawkes, S., 1997. What is a ‘‘Heavy Metal’’? J. Chem. Educ. 74 (11), 1374.Hogan, C.M., 2010. Heavy metal. The encyclopedia of earth. <http://

www.eoearth.org/view/article/153463/> (accessed June 27.2013).Hoornweg, D., Bhada-Tata, P., 2012. What a waste: a global review of solid waste

management. Urban development series; knowledge papers no. 15. The WorldBank, Washington D.C., USA. <http://documents.worldbank.org/curated/en/2012/03/16537275/waste-global-review-solid-waste-management> (accessedMay 21.2013).

Jolliffe, I.T., 1986. Principal Component Analysis. Springer-Verlag, New York, USA.King, J., Jackson, D., 1999. Variable selection in large environmental data sets using

principal components analysis. Environmetrics 10, 67–77.Mehmood, M.K., Adetutu, E., Nedwell, D.B., Ball, A.S., 2009. In situ microbial

treatment of landfill leachate using aerated WSPs. Bioresour. Technol. 100 (10),2741–2744.

Mulamoottil, G., McBean, E.A., Rovers, F. (Eds.), 1999. Constructed Wetlands for theTreatment of Landfill Leachate. Lewis Publishers, New York, USA.


http://refhub.elsevier.com/S0956-053X(14)00495-4/h0005



http://apsychoserver.psychofizz.psych.arizona.edu/JJBAReprints/PSYC501A/pdfs2008/PCAEXPL.pdf
























http://www.eoearth.org/view/article/153463/

http://www.eoearth.org/view/article/153463/

http://documents.worldbank.org/curated/en/2012/03/16537275/waste-global-review-solid-waste-management

http://documents.worldbank.org/curated/en/2012/03/16537275/waste-global-review-solid-waste-management











Neal, C., 2001. Alkalinity measurements within natural waters: a standardisedapproach. Sci. Total Environ. 265, 99–113.

Nedwell, D.B., Reynolds, P.J., 1996. Treatment of landfill leachate by methanogenicand sulfate-reducing digestion. Water Res. 30 (1), 21–28.

Nyame, F.K., Tigme, J., Kutu, J., Armah, T., 2012. Environmental Implications of thedischarge of municipal landfill leachate into the Densu River and surroundingRamsar Wetland in the Accra Metropolis, Ghana. J. Water Resour. Protect. 4,622–633.

Peres-Neto, P.R., Jackson, D., Somers, K.M., 2003. Giving meaningful interpretationto ordination axes: assessing loading significance in principal componentanalysis. Ecology 84 (9), 2347–2363.

Rew, S., Mulamoottil, G., 1999. A cost comparison of leachate treatmentalternatives. In: Mulamoottil, G., McBean, E.A., Rovers, F. (Eds.), ConstructedWetlands for the Treatment of Landfill Leachate. Lewis Publishers, New York,USA, pp. 261–270.

Ribé, V., Nehrenheim, E., Odlare, M., Gustavsson, L., Berglind, R., Forsberg, A., 2012.Ecotoxicological assessment and evaluation of a pine bark biosorbent treatmentof five landfill leachates. Waste Manage. 32, 1886–1894.

SAS Institute Inc., 2013. User Guide. JMP10 [software].Sawaittayothin, V., Polprasert, C., 2006. Nitrogen mass balance and microbial

analysis of constructed wetlands treating municipal landfill leachate. Water Sci.Technol. 54 (11), 147–154.

Snoeyink, V., Jenkins, D., 1980. Water Chemistry, first ed. Wiley, Hoboken, NewJersey.

Speer, S. 2011. Passive and naturalized landfill leachate treatment systems forsource water protection. Ph.D. dissertation. Kingston, Canada: Department ofCivil Engineering, Queen’s University.


Speer, S., Champagne, P., Anderson, B., 2011. Treatability study of two hybrid-passive treatment systems for landfill leachate operated at cold temperature.Water Qual. Res. J. Can. 43 (3), 230–238.

Speer, S., Champagne, P., Anderson, B., 2012. Pilot-scale comparison of two hybrid-passive landfill leachate treatment systems operated in a cold climate.Bioresour. Technol. 104, 119–126.

Speer, S., Champagne, P., Anderson, B., 2013. Using fundamental hydrogeologicalequations to monitor the effects of clogging and media consolidation on thehydraulic regime of a vertical subsurface flow treatment system. J. Environ.Manage. 118, 11–20.

Suhr, D.D., 2005. Principal component analysis vs. exploratory factor analysis. Paper203–30, SUGI 30 Conference Proceedings, Philadelphia, Pennsylvania.

Venugopal, B., Luckey, T.D., 1975. Toxicology of nonradio-active heavy metals andtheir salts. In: Luckey, T.D., Venugopal, B., Hutcheson, D. (Eds.), Heavy MetalToxicity, Safety and Hormology. Georg Thieme Verlag, Stuttgart, Germany.

Wallace, J., Champagne, P., Speer, S., Plourde, R. 2012. Environmental interactionsand impacts of treated landfill leachate following treatment with a hybrid-passive system. In: Presented in the proceedings for the 12th InternationalEnvironmental Specialty Conference of the Canadian Society for CivilEngineering.

Wicklin, R., 2011. Log transformations: How to handle negative data values? The DOLoop – SAS Blog. <http://blogs.sas.com/content/iml/2011/04/27/log-transformations-how-to-handle-negative-data-values/> (accessed April24.2013).

Ziyang, L., Youcai, Z., Tao, Y., Yu, S., Huili, C., Nanwen, Z., Renhua, H., 2009. Naturalattenuation and characterization of contaminants composition in landfillleachate under different disposing ages. Sci. Total Environ. 407, 3385–3391.






































http://blogs.sas.com/content/iml/2011/04/27/log-transformations-how-to-handle-negative-data-values/

http://blogs.sas.com/content/iml/2011/04/27/log-transformations-how-to-handle-negative-data-values/





Documents

Performance evaluation of a hybrid-passive landfill leachate treatment system using multivariate statistical techniques