Environmental Technology & Innovation 8 (2017) 1–16

Contents lists available at ScienceDirect

Environmental Technology & Innovation

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

Estimating landfill leachate BOD and COD based on rainfall,

ambient temperature, and waste composition: Exploration of
a MARS statistical approach
Arpita H. Bhatt a,∗ , Richa V. Karanjekar a,b , Said Altouqi a,c , Melanie L. Sattler a ,
M.D. Sahadat Hossain a , Victoria P. Chen d
a
Department of Civil Engineering, University of Texas at Arlington, 425 Nedderman Hall, 416 Yates St., Box 19308, Arlington, TX 76019,
United States
b
Apex Companies, LLC, 2801 Network Boulevard, Suite 200, Frisco, TX 75034, United States
c
Oman Environmental Services Holding Company, Al-Azaiba, Oman
d
Department of Industrial, Manufacturing, and Systems Engineering, University of Texas at Arlington, 500 West First Street, Box 19017,
420 Woolf Hall, Arlington, TX 76019, United States

highlights
• Modeled leachate BOD and COD for 27 lab-scale landfills using cyclic incomplete block design.
• Multivariate Adaptive Regression Splines (MARS) statistical approach was utilized.
• MARS captured complex relationships among rain, temperature, & waste composition.
• 100% food reactors showed BOD and COD highest concentrations.
• Time and rainfall were the most important variables impacting BOD and COD.

article info abstract

Article history: Municipal solid waste generation is increasing rapidly globally. Increasing storage of
Received 18 March 2016 waste in landfills means increased production of liquid landfill leachate. Leachate contains
Received in revised form 9 March 2017 organic and inorganic pollutants which must be treated to reduce its potential impact on
Accepted 13 March 2017
surrounding water supplies. To design landfill leachate treatment, its chemical composition
Available online 18 March 2017
must be known; however, the composition of leachate can vary widely depending on waste
composition. Measuring leachate constituents can be expensive and time consuming, and
Keywords:
Landfill
is not possible for a landfill under construction. A global model able to forecast leachate
Leachate quality parameters based on a landfill’s waste composition, ambient temperature, and
BOD rainfall rate is critically needed. This research represents a first step in development of
COD such a model.
MARS modeling Laboratory data on leachate biochemical and chemical oxygen demand (BOD, COD) was
Solid waste collected as functions of waste composition (food, paper, yard, textile), temperature (70, 85,
100 °F), and rainfall rates (2, 6, 12 mm/day), according to a statistical experimental design.
Multivariate Adaptive Regression Splines (MARS) models were developed for BOD and COD,
with adjusted-R2 of 0.92 and 0.95, respectively. This exploratory research demonstrated
the usefulness of MARS in capturing complex relationships among waste composition,
temperature, and rainfall rate in order to forecast leachate quality parameters over time.
© 2017 Elsevier B.V. All rights reserved.

∗ Corresponding author.
E-mail address: arpitab@uta.edu (A.H. Bhatt).

http://dx.doi.org/10.1016/j.eti.2017.03.003
2352-1864/© 2017 Elsevier B.V. All rights reserved.
2 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16

1. Introduction

Integrated solid waste management practices, which include source reduction, reuse, recycling, and composting, have
decreased the use of landfills. However, landfills still remain the most common disposal route in many countries (Barton
et al., 2008; Diaz et al., 2005; Taiwo, 2011). Landfill leachate is generated mainly due to the infiltration of rain water which
percolates through the waste layers and accumulates at the bottom of landfills. Even though solid waste management
practices for landfilling of wastes have advanced, leachate generation and management remains one of the most important
issues associated with landfills (Mahmud et al., 2012; Tatsi and Zouboulis, 2002). This leachate can cause pollution of
surrounding soil, surface waters, and ground water. The risk of ground water pollution is probably the most substantial
environmental impact from landfills (Alkalay et al., 1998; Johannessen et al., 1999; Rapti-Caputo and Vaccaro, 2006; Saarela,
2003).
Even today, there are many landfills, especially in developing parts of the world like India, Bangladesh, Africa, and
Latin America, where open dump systems are used for the final disposal of solid waste rather than engineered landfills
(Johannessen et al., 1999; Tränkler et al., 2005). In the near future, regulations in developing countries will require
installation of liner systems, leachate collection systems, and treatment operations. A major requirement for successful
leachate treatment is quantifying its typical composition. Being able to forecast leachate composition from new engineered
landfills in these parts of the world is therefore important. In addition, in the developed world, it is likely that the frequency
of monitoring various leachate quality parameters such as BOD, COD, alkalinity, organic compounds, total dissolved solids
(TDS), etc. will increase, along with the number of parameters to be measured. However, a limitation in this aspect is that
the analyses of these types of parameters are very expensive. The associated cost may limit the number of analyses included
in monitoring programs. Being able to model, rather than measure, leachate parameters would reduce costs.
Leachate quality depends on many factors such as waste composition, waste age, climate (ambient temperature
and annual rainfall), and landfilling operations (Alkalay et al., 1998). Kjeldsen et al. (2002) provided a review of
leachate composition from various studies available in the literature and showed that there is a wide variation in
leachate composition. Several parameters change radically over time as landfills pass through stabilization phases. Several
researchers have observed that an increase in temperature of the waste also enhances microbial activity, thereby increasing
the rate of waste degradation and entry of degradation products into the leachate stream (Mata-Alvarez and Martinez-
Viturtia, 1986; Bingemer and Crutzen, 1987; Christensen and Kjeldsen, 1989; Lefebvre et al., 2000; Sanphoti et al., 2006).
Several researchers have pointed out that an increase in moisture content can also increase waste degradation rates (Mata-
Alvarez and Martinez-Viturtia, 1986; Barlaz et al., 1990; Chan et al., 2002; Mehta et al., 2002; Wreford et al., 2000).
Depending on the study objectives, many investigations of leachate composition have been restricted to leachate quality
data from a single landfill or from a few regional-specific landfills. Various studies have been conducted to assess leachate
quality (Tatsi and Zouboulis, 2002; Al-Yaqout and Hamoda, 2003; Kulikowska and Klimiuk, 2008; Sourmunen et al., 2008).
These studies considered general waste composition, temperature, and moisture content, or site-specific data from a few
landfills. The few attempts to model leachate quality/characteristics using statistical techniques or software have also
focused on a single landfill or few regional landfills (Eusuf et al., 2007; G´omez Marti´n et al., 1995a; G´omez Mart´in et al.,
1995b; Kylefors, 2003). Results of previous studies of correlations within one or several leachates have been presented
(G´omez Marti´n et al., 1995a; G´omez Mart´in et al., 1995b; Ettala et al., 1988; Fatta et al., 1998). However, the application
of such matrices without any models limits the use of the results. Other statistical techniques have been used on a few
occasions to categorize leachate based on landfill site or degradation phase (G´omez Mart´in et al., 1995b; Andreas et al.,
1999; Youcai et al., 2001).
Zacharof and Butler (2004) did develop a general model for estimating landfill leachate total organic carbon, biochemical
oxygen demand, and ammonia over time based on a simplified representation of waste decomposition and a hydrologic
model. Their model, however, does not allow for input of landfill temperature (although the hydrolysis rate constant can
be varied) or waste composition; carbohydrates, fats and oils and proteins are all assumed to degrade in the same rate, and
the difference between readily biodegradable and recalcitrant compounds is not included. In addition, the model relies on
inputs that would not be readily available to a city planner or landfill operator, such as mean pore water velocity before
capping and acetogenesis time to peak.
Rather than developing region-specific leachate quality models, developing a global model to predict leachate quality
from landfills anywhere would be more efficient. Such a model would ideally be based on information that is relatively
easy to obtain, such as rainfall rates, ambient temperature, and waste composition. The research presented here, and in a
companion paper, represent first steps toward development of such a model for BOD and COD (Bhatt et al., 2016). BOD and
COD were chosen as initial parameters to model because of their importance in determining leachate quality. In our previous
work, we assumed BOD and COD removal rates to be first-order, and then developed regression models for the first-order
rate constants based on rainfall rate, ambient temperature, and waste composition. In this work, we do not a priori assume
a form for the equation of BOD and COD removal vs. time, but rather search for the best model to fit the data. The research
objectives of this study were thus:
1. To measure BOD and COD as a function of time from laboratory scale landfills with varying waste components operated
at different temperatures and rainfall rates.
2. To use the data from Objective 1 to develop statistical models for predicting leachate BOD and COD as functions of time,
temperature, rainfall rate, and waste composition.
 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16 3

Collection of lab-scale leachate data over a variety of temperatures, rainfall rates, and waste compositions to determine
trends and interactions represents a contribution of our work; no one else, to our knowledge, has studied the impact of
these parameters on leachate composition in a systematic way. Collecting lab data, as opposed to developing a model based
entirely on field data, allowed us to vary temperatures, rainfall rates, and waste composition over a wide range of potential
values (from low to medium to high), and explore interactions among the variables. Finding field leachate data that spanned
the wide range of temperatures, rainfall rates, and waste compositions that we studied in the lab would have been difficult.
Exploration of the use of MARS in forecasting leachate parameters is an additional contribution of this study. Although
a standard multiple linear regression approach was first attempted, model assumptions were violated, as will be discussed
later, likely because of the complex relationships among the variables that govern leachate composition. MARs offered a
more flexible modeling approach. No one else, to our knowledge, has used MARS for modeling leachate quality parameters.
More broadly, no one else has, to our knowledge, attempted to develop a global model for forecasting leachate BOD
and COD from any landfill given information about rainfall rate, temperature, and waste composition. The lab-scale models
developed provide a basis from which field-scale models, adjusted for slower waste degradation rates that occur under
non-ideal conditions, can be built in future work.

2. Materials and methods

2.1. Experimental design

2.1.1. Rainfall rates

To study the effect of rainfall on leachate BOD and COD, three rainfall rates of 2, 6, and 12 mm/day were used, correspond-
ing to 60, 180, and 360 mm/month, which encompass monthly precipitation rates for most developing countries in Central
America, South America, Africa (with the exception of the Sahara countries), and the Far East (India, China, Thailand, and
Indonesia). The amount of tap water added to the reactors to simulate the rainfall was computed using the bucket dimen-
sions. Water addition of 100, 300, and 600 ml/day was calculated based on the rainfall intensities of 2, 6, and 12 mm/day.
Water addition was done using a three-way valve.

2.1.2. Temperatures
Experiments were carried out at three temperatures, 70, 85, and 100 °F, as representative ambient temperatures to study
the effect of ambient temperature on BOD and COD. The temperatures selection was based on annual mean temperatures for
most of South America, Central America, Africa, India, Australia, and Indonesia (except for desert and cold regions), ranging
from 68 °F–86 °F. Average monthly summer temperatures in these areas can range up to 95 °F. The temperatures also apply
to some areas in North America and Europe. Reactors were placed in constant temperature rooms at 85 °F and 100 °F.
Reactors placed in the lab room were maintained at approximately 70 °F.

2.1.3. Waste components

Waste components selected for this study were: food, paper, yard/wood, and textile, as well as inorganic wastes.

2.1.4. Experimental design

As discussed above, temperatures of 70 °F, 85 °F, and 100 °F, and average rainfall rates of 2, 6 and 12 mm/day, were
considered in this research. Thus, rainfall and temperature were the two factors of primary interest, each studied at 3 levels
and in all combinations; hence, 32 = 9 combinations of temperature and rainfall were used in the experimental design. It is
clear that the effect of these 2 factors on BOD and COD also depend on the combination of the waste, which is represented
by 5 waste components.
A cyclic incomplete block design was used for the experimental setup (Dean and Voss, 1999). This design enabled us
to keep the number of reactors to a minimum due to the time involved in measuring leachate parameters, monitoring
reactors and limited space in the constant temperature rooms. Using this experimental design, every combination of rainfall,
temperature, and composition did not need to be studied; the combinations that were selected sufficiently spanned the
ranges of rainfall, temperature, and composition of interest to allow regression models to be built. In addition, due to the
experimental design, reactor replicates of identical conditions were not needed: the reactors operated provided sufficient
data for constructing the regression models.
The design was constructed based on the following characteristics:
Number of treatments (v) = rainfall and temperature combinations, 32 = 9.
Number of blocks (b) = combinations of waste components = 9.
Block size (k) = number of times a block appears in the design = 3.
Number of times a treatment appears in the design (r ) = 3.
Total no. of reactors = vr = bk = 27.
4 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16

Table 1
Component percent by weight for each waste combination.
Component Component % by weight for each waste combination
a b c d e f g h i

Food 100 0 0 0 0 60 30 10 20
Paper 0 100 0 0 60 0 10 30 20
Textile 0 0 100 0 0 30 0 60 20
Yard 0 0 0 100 0 10 60 0 20
Inorganic 0 0 0 0 40 0 0 0 20

Table 2
Rainfall, temperature, and waste component combinations for the lab-scale landfill reactors.

The specific combined waste cases were determined by a mixture design such that each biodegradable waste component
(food, paper, yard, and textile) could range from 0% to 100% except inorganic waste, which ranged from 0% to 40%, since it
does not have a potential to degrade (Mason et al., 1989). These combined waste cases served as blocking variable levels for
a balanced incomplete block design, such as a Latin hypercube, to study the primary factors, temperature and rainfall (Chen
et al., 2006). Table 1 summarizes the 9 waste compositions used in the experimental design as discussed above.
Table 2 exhibits the matrix with treatments and block combinations used in the experimental design for setting up
each lab-scale landfill reactors. For example, Reactor 4 was operated with 2 mm/day rainfall at a temperature of 85 °F, and
contained waste component combination a, which according to Table 1 is 100% food waste.

2.2. Waste components: collection and preparation

Waste components selected for this study were: food, paper, yard/wood, and textile, as well as inorganic wastes. Wastes
were collected from individual places to represent pure and non-degraded waste. Food waste was collected from University
of Texas at Arlington’s (UTA) cafeteria and dining hall, and from Indian and Thai restaurants to encompass the waste
composition for countries with large portions of food waste in their landfills. A mixture of grass, leaves, and tree/brush
trimmings were obtained from UTA’s composting facility. Yard waste was mixed to represent 50% grass and 50% leaves and
branches. Textiles wastes were obtained from local tailors’ shops comprised of a mixture of cotton and synthetic material.
Paper wastes were obtained from UTA recycling bins (office paper), researcher personal recycling bins (newspapers, mail,
magazines, tissues and towels, diapers), and local restaurants and grocery stores (corrugated boxes and cartons); paper
wastes were mixed together to replicate the national average percentages found in the US (EPA, 1996). Inorganics such
as plastic bottles and aluminum cans were collected from UTA’s recycling bins, and construction and demolition (C & D)
wastes, which was comprised of small concrete stones, were collected from UTA’s structural testing laboratory. Food and
yard wastes were kept in a cold room at 4 °C until use to avoid any possible degradation. Various combinations of waste
components were prepared by weight percentage according to the experimental design (Table 2) and mixed thoroughly.
Larger pieces of textile and paper waste were cut into pieces (approximately 10 cm × 15 cm) to accommodate the reactor
size. However, these wastes were not finely shredded to better represent actual landfilled waste.

2.3. Reactor setup and filling operation

A total of 27–6 gallon (16 liter) size wide-mouth HDPE plastic reactors were built to simulate lab-scale landfills. The 16
liter reactor size was chosen to avoid fine shredding of waste, more closely representing landfill waste size. These plastic
reactors were filled with various proportions and types of waste and were operated at different rainfall rate and temperature
combinations, as discussed later in the experimental design section. Before filling the reactors with refuse, all reactors were
leak-checked using a simple U-tube manometer. Reactors were then filled with waste components, as described in the
 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16 5

Fig. 1. Schematic reactor setup and reactors in constant temperature room.

Table 3
Weight of waste in reactors.
Reactor number Waste composition Temperature °F Rainfall mm/day Waste mass in kg

1 100% paper 3.13

2 60% yard, 30% food, 10% paper 70 2 7.26
3 20% each 6.74
4 100% food 17.65
5 100% textile 85 2 6.33
6 60% textile, 30% paper, 10% food 5.91
7 100% paper 3.83
8 100% yard 100 2 8.90
9 20% each 6.88
12 60% paper, 40% Inorganics 70 6 4.19
13 100% paper 4.16
14 100% yard 85 6 7.11
15 60% food, 30% textile, 10% yard 11.0
16 100% textile 4.67
17 60% paper, 40% Inorganics 100 6 5.19
18 60% yard, 30% food, 10% paper 7.83
19 100% yard 7.52
20 60% food, 30% textile, 10% yard 70 12 7.87
21 60% textile, 30% paper, 10% food 4.78
22 60% paper, 40% Inorganics 4.90
23 60% yard, 30% food, 10% paper 85 12 7.86
24 20% each 6.50
25 100% food 14.84
26 60% food, 30% textile, 10% yard 100 12 10.0
27 60% textile, 30% paper, 10% food 5.53

experimental design section below. Anaerobic digested sewage sludge was used as seed to each reactor to achieve 10%–12%
weight of waste. Reactors’ final weights were measured and then the reactors were placed in the constant temperature
rooms and connected to leachate and gas bags, as shown in Fig. 1. All the reactors were properly sealed with a thick multiple
layers of sealant to ensure anaerobic conditions. Table 3 displays the weight of waste in each reactor. The reactors were
operated without leachate recirculation, representing conventional landfill operation. The lab experiments were carried out
until BOD:COD ratio was close to 0.1 or <0.1, which ranged from 139 to 370 days.

2.4. Leachate BOD and COD measurement

Volume of water added daily and leachate volume collected at the bottom in the leachate collection bag were measured
daily. Leachate samples from 27 reactors were collected on a regular basis (e.g. weekly, bi-weekly, monthly) and analyzed
promptly for BOD and COD measurement. The Standard Methods of American Public Health Association (APHA) 5210B
and 5220C were used to measure BOD and COD, respectively, which are briefly described below (American Public Health
Association, 1998).
Initially for the BOD test, leachate samples were diluted as they exceeded the concentration of dissolved oxygen available
in an air-saturated sample. Later, there was no dilution required due to degradation of waste. To begin the BOD test, a desired
6 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16

amount of dilution water was prepared according to the standard method and aerated until saturated with air. The dilution
water blank was prepared. Leachate samples for BOD measurements were seeded using Polyseed capsules (InterLab Supply)
to have sufficient populations of microorganisms. The seed control bottles with different dilutions (15, 20 25, 30 mg/l) were
also prepared. Besides, glucose–glutamic acid standard check was also conducted on a regular basis for quality control. The
desired amounts of leachate samples were filled in 300 ml mouth glass bottles and several dilutions were prepared for a
leachate sample to obtain precise results. 4 ml of seed solution was added to each sample bottle, and then remaining bottles
were filled with enough dilution water so that insertion of the stopper would displace all air, leaving no bubbles. The initial
dissolved oxygen (DO) was determined using the Intellical probe (Hach Company). After 5-days incubation period, final
DO was measured using the Intellical probe. The BOD was calculation for each bottle meeting the 2 mg/l minimum of DO
depletion and 1 mg/l residual DO.
For the COD test, a suitable volume of leachate sample was added into vials containing pre-mixed digestion solution
(Hach) along with a blank. The vials were inverted a few times to mix the sample. Then samples were placed into a preheated
digester for 2 h at 150 °C. Dichromate ion (Cr2 O− 2
7 ), an oxidant, reacts with the sample under controlled conditions and is
+3
reduced to chromic ion (Cr ). Samples were cooled down to room temperature and the absorption of each sample at 620 nm
was measured using a Spectronic-D spectrophotometer. A calibration curve was prepared using 6 standards of potassium
hydrogen phthalate solution. COD was determined using the best fit line from the calibration curve. Due to the interference of
chloride in COD measurement, the commercially available vials from Hach contained mercuric sulfate/mercury to eliminate
chloride interference. The COD test was performed on unfiltered leachate samples representing total COD.

2.5. Data analysis and model development

A standard multiple linear regression approach was first attempted for simplicity (Kutner et al., 2004). By using log
functions, linear regression can be used to fit a variety of non-linear models. The initial regression models violated both
the assumptions of linear model structure and constant variance. A variety of transformations were attempted on both
the response variables (BOD, COD) and the predictors (time, rainfall, waste composition, and temperature), and interaction
terms were considered. A log transformation on the BOD and COD response variables was successful for addressing the
non-constant variance issue, and was meaningful in physical terms: when a component has reach a maximum the dilution
effect in a more or less mixed reactor, the concentration will decrease and follow a decreasing exponential law. However, the
curvature could not be satisfactorily modeled with standard log transformations. Consequently, a more flexible statistical
modeling approach was needed.
MARS is a modern flexible statistical modeling method that builds a linear combination of basis functions (Friedman,
1991). Basis functions are chosen to bend the model approximation based on the data. MARS has been used in various
previous studies (Moore et al., 1991; Walker, 1990; White and Sifneos, 1997). For modeling BOD and COD in our study,
MARS was run using Salford Predictive Modeler (SPM) Builder, version 6.6. In MARS, the optimal model is selected in two-
stage process. In the first stage, MARS adds basis functions. In the second stage, unnecessary basis functions are removed.
Due to multicollinearity and uncertainty in real data, it is recommended to build an overly large model in the first stage, and
then allow the second stage to remove those functions that are not useful.

2.6. Model fitting and selection

Basic functions are formed using ‘‘hinge’’ functions that bend at a single ‘‘knot’’. A hinge function is a linear function that
applies over a portion of the range of an independent variable. The knot is the value of the independent variable at which the
first hinge function ceases to apply and a second hinge function applies instead. These are combined in a statistical linear
model to approximate the relationship between a response variable (BOD and COD) and multiple predictor variables (time,
rainfall, temperature, refuse components). An example univariate basis function (BF1 ) on the variable temperature may be
defined by MARS as:
bm (temperature) = max(0, temperature − 80) (1)
where this hinge function bends at a temperature equal to 80, which is the knot value. The initial model fitting begins
with a constant only (C0 ), then adds basis functions (initially, univariate terms and later interaction terms) to build up a
MARS model until a user-specified maximum number of basis functions has been reached. The final approximation has the
following form (for observations i = 1, . . . , n):
M

Yi = β0 + βm bm (xi ) + εi (2)
m=1
where Yi is the response variable, xi is the vector of predictor variables, β0 is a constant, βm is a coefficient (constant) for the
mth basis function, denoted as bm (xi ), and εi is a random error term.
Several runs were made with various maximum basis functions, ranging from 15 (minimum default value the software
recommended) up to 84 in an increment of 1 or 2 maximum basis functions, and different input parameters until the best
model was found, where the lowest predictive squared error (PSE) and adjusted-R2 values appeared to begin leveling off.
The PSE is the prediction error based on actual cross-validation.
 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16 7

Fig. 2. Leachate pH trend for 100% (a) food (b) paper (c) yard, and (d) textile reactors.

3. Results and discussion

Leachate pH, BOD, COD, and BOD:COD ratio act as indicators of microbial activities and leachate organic pollution
(Chian and DeWalle, 1997; Lee et al., 2014). This section discusses the pH, BOD, COD, and BOD:COD results from 100%
waste component reactors (reactors containing 100% of a single waste component: food, paper, yard, or textile) at different
temperatures and rainfall rates over time. Leachate pH, BOD, and COD results other than 100% waste component reactors are
described elsewhere (Bhatt, 2013). The lab experiments were carried out until BOD:COD ratio was 0.1 or <0.1. The methane
gas generation information for 100% waste reactors has been provided by Karanjekar et al. (2015).
From Fig. 2, it is observed that pH is initially acidic for all waste types, likely due to carboxylic acids accumulation in the
anaerobic acid phase. After the onset of initial methanogenic phase, pH increases as acids are consumed (Kjeldsen et al.,
2002; McBean et al., 1995). In the final stable methanogenic phase, pH continues to increase and then stabilizes.
Some leachate time courses started later, as shown in Fig. 2, as well as subsequent Figs. 3 and 4. This was the case when
less water was added. This is similar to the case of an actual landfill, in which leachate generation would be delayed if
moisture content were low. The MARS models for BOD and COD, discussed later, account for this effect of small amount of
leachate generated at early times for low rates of rainfall.

3.1. BOD and COD trend and its ratio for 100% waste reactors

BOD:COD ratio is a good indicator of the landfill phases. Fig. 3 displays BOD:COD ratio for reactors containing 100% waste
of a single kind. Initially, the BOD:COD ratio is <0.1–>0.8 for all 100% waste reactors. Toward the end of the reactors life, the
ratios are around close to or <0.1, indicating the wide variation in the BOD:COD values. For R8, the BOD:COD ratio was low
from the beginning due to high initial COD values relative to BOD. From Fig. 3(b), it is apparent that temperature had greater
8 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16

Fig. 3. BOD:COD ratio for 100% waste reactors.

effect than rainfall in lowering the BOD:COD ratio for 100% paper reactors. However, for the other 100% waste reactors there
may be an interaction effect of temperature and rainfall.
As waste decomposition begins to slow after the peak at the beginning of the methanogenic phase, the concentrations
of BOD and COD begin to decrease also, as shown in Fig. 4 for 100% waste reactors. Fig. 4(a) and (b) show that BOD and COD
concentrations for R14 were initially higher than R8 and R19, indicating that R14 has a slower leaching rate in the beginning.
R14 has a combination of intermediate temperature and rainfall rate of 85 °F and 6 mm/day, respectively which may have
contributed to higher concentration of BOD and COD initially. It is evident that R19 has the lowest concentration of BOD and
COD due to the combination of lowest temperature and highest rainfall. Lower temperatures prolong the time for microbial
activities to occur and higher rainfall dilutes the leachate due to washout effect.
It was observed that 100% food waste reactors (R4 and R25) had the highest concentrations for BOD and COD. This was
expected, as food is the most biodegradable component in the waste stream. From Fig. 4, it is seen that BOD and COD for
100% waste reactors (yard, food, paper, and textile) decreased in less time (<200 days) and followed first-order decay. The
BOD and COD trend in R5 followed a gradual increase and then declined as the acids were consumed which is the typical
trend for landfill leachate (Fig. 4(g) and (h)). The initial increase in the BOD and COD in R5 could be due to different types of
textile waste degrading at different rates. The textile in this study included different types of textiles: jeans, cotton, polyester,
spandex etc. The leaching of organic matter has been substantially faster than what was expected due to daily simulation of
rainfall in each reactor. Table 4 summarizes BOD and COD range for each reactor and the ratios. The wide variability is due
to the heterogeneity of the waste and treatments (rainfalls and temperatures).

3.2. MARS modeling for leachate BOD and COD

The Salford Systems MARS software employs a cross-validation approach to conduct model selection. Several MARS
models are obtained primarily by varying the input parameter that sets the maximum number of basis functions. The
 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16 9

Fig. 4. Leachate over time: (a) BOD and (b) COD for 100% yard; (c) BOD and (d) COD for 100% food; (e) BOD and (f) COD for 100% paper; and (g) BOD and
(h) COD for 100% textile.

predictive squared errors from cross-validation and the adjusted-R2 values for each model are compared. The final models for
BOD and COD were selected with the lower PSE values of 0.096 and 0.033, and adjusted-R2 of 0.92 and 0.95, respectively. The
MARS text output for BOD and COD (provided as supplementary information, Appendix A) includes the final model, ANOVA
decomposition table, relative variable importance table, and regression information. As an example, the basis functions and
final model equation for BOD and COD are provided as Figs. 5 and 6, respectively. Based on Figs. 5 and 6, a univariate linear
basis function (BF2 ) on the variable time will be truncated to zero for all the values of time that fall below 119 and 96 days
10 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16

Table 4
Summary of BOD and COD range and ratios for reactors.
Reactor number Range for BOD/COD ratio BOD range (mg/l) COD range (mg/l)

1 0.5–0.12 22–3024 254–6094

2 0.58–0.16 108–24,509 658–42,606
3 0.44–0.07 115–10,044 1050–22,674
4 0.56–0.21 58–46,134 327–60,760
5 0.35–0.1 81–8960 818–16,054
6 0.67–0.1 28–19,889 747–18,129
7 0.45–0.13 18–2816 309–6212
8 0.07–0.1 68–976 981–13,498
9 0.49–0.06 33–18,824 578–38,777
12 0.6–0.16 11–2124 77–5829
13 0.5–0.15 14–10,278 115–20,525
14 0.41–0.1 32–12,361 457–30,256
15 0.46–0.1 17–21,928 230–47,453
16 0.87–0.07 19–6008 227–6879
17 0.47–0.08 8.5–1260 90–2692
18 0.44–0.12 49–14,826 483–33,467
19 0.25–0.09 13–314 141–1265
20 0.49–0.25 36–3212 126–9929
21 0.39–0.13 17–689 96–1677
22 0.63–0.25 6.5–2319 62–9163
23 0.49–0.08 11–16,958 196–34,332
24 0.55–0.08 14–9082 172–16,487
25 0.47–0.06 11–29,892 75–64,032
26 0.40–0.08 10–15,755 106–39,024
27 0.46–0.05 7–2670 142–10,571
Mean 1886 3787
Median 84 694
Standard deviation 5275 9033

for BOD and COD final model equations. Although the final model equations for BOD and COD are complex, they could easily
be handled by a spreadsheet or computer program for future purposes.
MARS outputs 3-D plots which depict the interaction effect that a pair of predictor variables has on the response variable.
The contribution of the interaction effect, plotted along the vertical axis of each plot, indicates combined impact of these
two predictor variables on the response variable. Some representative graphs and their discussions are provided below, and
the detailed results are described elsewhere by Bhatt (2013). Although MARS does not show units on the axes, the units are
°F for temperature, mm/day for rainfall, and mg/L for BOD and COD.
From Fig. 7(a), BOD concentration decreases with time, the rate of waste degradation slows past its initial peak. Initially
with time and when temperature increases up to 85 °F, BOD concentration increases and reaches a maximum at 85 °F.
However, BOD decreases above 85 °F temperature. 85 °F may be the optimum temperature for the microbes to decompose
waste. In Fig. 7 (b), initially with time, BOD concentration is highest. Over time, BOD concentration decreases, as most of the
food waste would have been degraded. For initial times, the impact of %food on BOD is relatively constant. This is somewhat
surprising since food waste is readily degradable, although sometimes a significant lag period occurs before food begins to
degrade significantly.
In Fig. 7 (c), when time increases, BOD concentration decreases, as most of the biodegradable organic matter from waste
would have been degraded. As rain increases up to 6 mm/d, BOD concentration increases and reaches a peak and then it goes
down as rainfall increases. The dilution effect on leachate as rainfall increases from 6 to 12 mm/d is anticipated. It is not clear
why a similar dilution effect is not observed as rainfall increases from 2 mm/day to 6 mm/day. One reason would be that
optimal moisture content for degradation occurs with a rainfall rate of 6 mm/day. In Fig. 7(d), when %food and rainfall are
highest, BOD concentration is at peak. This is due to the presence of more organic matter at higher %food. However, the high
rainfall rate would be anticipated to produce low BOD concentrations due to washout. Only looking at %food, BOD increases
as %food increases, as expected, since food is very amenable to biodegradation. For low %s of food waste, BOD increases
when rainfall increases up to 6 mm/d, and then BOD decreases with higher rainfall due to the dilution effect.
In Fig. 7(e), initially with time and when %textile is high, BOD concentration is at peak. As time increases and %textile
decreases, BOD decreases. Most of the waste would have been degraded over time, decreasing BOD concentration. In Fig. 7(f),
BOD concentration is highest when temperature is low and %paper is higher. As temperature and %paper increases, BOD
decreases. Perhaps microbes are having difficulty degrading the lignin component of paper waste. BOD being highest at
70 °F rather than 85 °F (which seems to be the optimal temperature for the waste degradation) in this case is surprising.
In Fig. 8(a), it is observed that COD decreases with time very slowly. At later times, the COD in the leachate is lower,
because less waste remains to be degraded and enter the leachate. At early times, the COD concentration in the leachate is
highest at 70 °F, indicating faster waste degradation at this temperature. It would have been expected that 100 °F would have
produced faster waste degradation and thus higher COD values. In Fig. 8(b), at early times, COD displays a peak value which
is independent of food%. However, with later times, COD decreases as %food increases. This is due to the fact that the rate
A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16 11

Fig. 5. MARS basis functions and final model equation for BOD.
12 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16

Fig. 6. MARS basis functions and final model equation for COD.

of COD degradation likely depends on the degradable COD concentration; reactors with a high percent food waste contain
a high percent of readily degradable waste, which degrades quickly. In Fig. 8 (b), for a given % food, COD values decrease
with time; this is expected, as the rate of degradation has passed its peak, and less COD is entering the leachate with time.
In Fig. 8(c), as time increases, COD in the decreases, as was the case for Fig. 8(a) and (b). In Fig. 8(c), COD also decreases as
rainfall increases. More rainfall means a higher volume of leachate to dilute the COD. Initially with time at lower rainfall,
 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16 13

Fig. 7. BOD 3-D interaction plots for (a) time (d)-temperature (° F); (b) time (d)-food (%); (c) time (d)-rain (mm/d); and (d) food (%)-rain (mm/d); (e) time
(d)-textile; (f) temperature (° F)-paper (%).

COD concentration is at peak. The generated leachate is concentrated. In Fig. 8(d), when %food is high and rainfall is low,
COD concentration is at peak. This is due to more concentrated leachate and higher organic content. As rainfall increases,
COD decreases due to dilution effect.
In Fig. 8(e), when %paper and rainfall are low, COD is at peak and then decreases as %paper and rainfall increase. This
is due to dilution effect at higher rainfall and also with high %paper, microbes may be having a difficult time degrading
14 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16

Fig. 8. COD 3-D interaction plots for (a) time (d)-temperature (° F) (b) time (d)-food (%) (c) time (d)-rain (mm/d), and (d) food (%)-rain (mm/d); (e) paper
(%)-rain(mm/d); (f) rain (%)-yard (%).

lignin. In Fig. 8(f), as rainfall increases, COD concentration decreases due to dilution effect. Also at higher rainfall, when %
yard increases up to 60%, there is a decrease in COD concentration and then beyond 60% yard, COD concentration gradually
increases. At a 2 mm/day rainfall rate, %yard seems to have no impact.
 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16 15

4. Future work

Previous studies have found that lab waste degradation rates exceed field rates, because conditions are more ideal
(e.g. smaller waste size, more uniform moisture conditions) (Cruz and Barlaz, 2010). Hence, future work will use field
leachate data to develop a ‘‘scale-up factor’’ to adjust the BOD and COD models to the slower waste degradation rates that
occur in the field. Field data was not used to develop the models in the first place because finding field leachate data that
spanned the wide range of temperatures, rainfall rates, and waste compositions that we studied in the lab would have been
difficult. The lab experiments under controlled conditions could provide more detail about the relationships among the
variables studied.
After developing the scale-up factor, the next future step will be to obtain additional field data to validate the model.
Also, models will need to be developed to predict other leachate quality parameters, such as ammonia and chloride.
Finally, the impact of additional factors on leachate quality, such as waste age and landfill operation, will need to be
incorporated into the models.

5. Conclusions

This research represented a first step toward being able to predict leachate quality from any landfill, given information
that is relatively easy to obtain: the landfill’s waste composition, and its annual rainfall rates and ambient temperature.
Laboratory data was collected in a systematic way under controlled conditions to provide information about how leachate
composition varies as functions of rainfall rates, ambient temperature, and waste composition. This lab data was then used
to build models to capture the complex relationships among the variables. The ability of MARS to capture the complex
relationships, unlike that of traditional multiple linear regression, seems promising. MARS models were developed to
forecast BOD and COD, with R2 values of 0.92 and 0.95, respectively. Time and annual rainfall were the most important
variables in the MARS models for leachate.

Acknowledgments

This project was partially supported by Waste Management, Inc. The lead author (Arpita Bhatt) would like to thank Air
and Waste Management Association, Texas Solid Waste Association of North America (TxSWANA, Lone Star Chapter) and
Texas Environmental Health Association (North Texas Chapter) for awarding scholarships that supported the research.

Appendix A. Supplementary data

Supplementary material related to this article can be found online at http://dx.doi.org/10.1016/j.eti.2017.03.003.

References

Alkalay, D., Guerrero, L., Lema, J., Mendez, R., Chamy, R., 1998. Review: Anaerobic treatment of municipal sanitary landfill leachates: the problem of
refractory and toxic components. World J. Microbiol. Biotechnol. 14, 309–320.
Al-Yaqout, A., Hamoda, M., 2003. Evaluation of landfill leachate in arid climate—A case study. Environ. Int. 29, 593–600.
American Public Health Association (APHA) 1998. Standard Methods for the Examination of Water and Wastewater. 20th ed. Washington, DC.
Andreas, L., Ecke, H., Shimaoka, T., Lagerkvist, A. 1999. Characterizing landfill phases at full-scale with the aid of test cells. In: Christensen, T.H., Cossu,
R., Stegmann, R. (Eds.), Sardinia’99, Seventh International Landfill Symposium, S. Margheritha di Pula, Cagliari, Italy, 4–8 October, vol. I(V). CISA
(Environmental Sanitary Engineering Centre), Cagliari, Italy, pp. 145–152.
Barlaz, M.A., Ham, R.K., Schaefer, D.M., 1990. Methane production from municipal refuse: a review of enhancement techniques and microbial dynamics.
CRC Crit. Rev. Environ. Contr. 19 (6), 557–584.
Barton, J.R., Issaias, I., Stentiford, E.I., 2008. Carbon–Making the right choice for waste management in developing countries. Waste Manag. 28 (4), 690–698.
Bhatt, A.H., 2013. Development of statistical models for predicting leachate parameters from simulated landfills (Ph.D. Dissertation), University of Texas
at Arlington, Arlington, TX.
Bhatt, A.H., Altouqi, S., Karanjekar, R.V., Hossain, S., Chen, V.P., Melanie, S.S., 2016. Preliminary regression models for estimating first-order rate constants
for removal of BOD and COD from landfill leachate. Environ. Technol. Innov. 5, 188–198.
Bingemer, H.G., Crutzen, P.J., 1987. The production of methane from solid wastes. J. Geophys. Res.: Atmos. 92 (D2), 2181–2187.
Chan, G., Chu, L., Wong, M., 2002. Effects of leachate recirculation on biogas production from landfill co-disposal of municipal solid waste, sewage sludge
and marine sediment. Environ. Pollut. 118 (3), 393–399.
Chen, V.C.P., Tsui, K.L., Barton, R.R., Meckesheimer, M., 2006. A review on design, modeling and applications of computer experiments. IIE Trans. 38 (4),
273–291.
Chian, E.S.K., DeWalle, F.B. 1997. Evaluation of leachate treatment, vol 1: characterization of leachate, EPA-600/2–77-186a, US Environmental Protection
Agency, Cincinnati.
Christensen, T.H., Kjeldsen, P., 1989. Basic biochemical processes in landfills. In: Christensen, T.H., Cossu, R., Stegmann, R. (Eds.), Sanitary Landfilling:
Process, Technology and Environmental Impact. Academic Press, London, UK, p. 29. (Chapter 2.1).
Cruz, F.B.D.L., Barlaz, M.A., 2010. Estimation of waste component-specific landfill decay rates using laboratory-scale decomposition data. Environ. Sci.
Technol. 44 (12), 4722–4728.
Dean, A.M., Voss, D., 1999. Design and Analysis of Experiments. Springer-Verlag, New York.
Diaz, L.F., Savage, G.M., Eggerth, L.L., 2005. Alternatives for the treatment and disposal of healthcare wastes in developing countries. J. Waste Manag. 25
(6), 626–637.
U.S. EPA, 1996. Characterization of Municipal Solid Waste in the United States: EPA/530-R-96–001; Office of Solid Waste, U.S. Environmental Protection
Agency: Washington, DC.
Ettala, M., Rahkonen, P., Kitunen, V., Valo, O., Salkinoja-Salonen, M., 1988. Quality of refuse, gas and water at a sanitary landfill. Aquat. Fenn. 18 (1), 15–28.
16 A.H. Bhatt et al. / Environmental Technology & Innovation 8 (2017) 1–16

Eusuf, M.A., Hossain, I., Noorbatcha, I.A., Zen, I.H. 2007. The effects of climate and waste composition on leachate and emissions of gas: A case study in
Malaysian context. In: Proceedings of the International Conference on Sustainable Solid Waste Management, pp. 437–443.
Fatta, D., Voscos, C., Papadopoulos, A., Loizidou, M., 1998. Leachate quality of a MSW landfill. J. Environ. Sci. Health Part A 33 (5), 749–763.
Friedman, J.H., 1991. Multivariate adaptive regression splines. Ann. Statist. 19 (1), 1–141.
Go´omez Marti´n, M.A., Antig¨uedad Auzmendi, I., Pérez Olozaga, C. 1995a. Landfill leachate: variation of quality with quantity. In: Christensen, T.H.,
Cossu, R., Stegmann, R. (Eds.), Sardinia’95, Fifth International Landfill Symposium, S. Margheritha di Pula, Cagliari, Italy, 2–6 October, vol. I(III). CISA
(Environmental Sanitary Engineering Centre), Cagliari, pp. 345–354.
Go´omez Marti´n, M.A., Antig¨uedad Auzmendi, I., Pérez Olozaga, C. 1995b. Multivariate analysis of leachate analytical data from different landfills in the
same area. In: Christensen, T.H., Cossu, R., Stegmann, R. (Eds.), Sardinia’95, Fifth International Landfill Symposium, S. Margheritha di Pula, Cagliari, Italy,
2–6 October, vol. I(III). CISA (Environmental Sanitary Engineering Centre), Cagliari, Italy, pp. 365–376.
Johannessen, L.M., Boyer, G., Mundial, B. 1999. Observations of solid waste landfills in developing countries. In: Africa, Asia and Latin America. Urban
Development Division. Waste Management Anchor Team, The International Bank for Reconstruction and Development/The World Bank, Washington,
DC.
Karanjekar, R.V., Bhatt, A., Altouqi, S., Jangikhatoonabad, N., Durai, V., Sattler, M., Hossain, S., Chen, V., 2015. Estimating methane emissions from landfills
based on rainfall, ambient temperature, and waste composition: The CLEEN model. Waste Manage. 46, 389–398.
Kjeldsen, P., Barlaz, M.A., Rooker, A.P., Baun, A., Ledin, A., Christensen, T.H., 2002. Present and long-term composition of MSW landfill leachate: a review.
Crit. Rev. Environ. Sci. Technol. 32 (4), 297–336.
Kulikowska, D., Klimiuk, E., 2008. The effect of landfill age on municipal leachate composition. Bioresour. Technol. 99 (13), 5981–5985.
Kutner, M., Nachtsheim, C.J., Neter, J., Li, W., 2004. Applied Linear Statistical Models, fifth ed. McGrawHill. Irwin, New York.
Kylefors, K., 2003. Evaluation of leachate composition by multivariate data analysis (MVDA). J. Environ. Manag. 68 (4), 367–376.
Lee, A.H., Nikraz, H., Hung, Y.T., 2014. Influence of waste age on landfill leachate quality. Int. J. Environ. Sci. Dev. 1 (4), 347–350.
Lefebvre, X., Lanini, S., Houi, D., 2000. The role of aerobic activity on refuse temperature rise, I. Landfill experimental study. Waste Manag. Res. 18 (5),
444–452.
Mahmud, K., Hossain, M.D., Shams, S., 2012. Different treatment strategies for highly polluted landfill leachate in developing countries. J. Waste Manag. 32
(11), 2096–2105.
Mason, R.L., Gunst, R.F., Hess, J.L., 1989. Statistical Design and Analysis of Experiments: With Applications to Engineering and Science. John Wiley & Sons,
New York, Wiley.
Mata-Alvarez, J., Martinez-Viturtia, A., 1986. Laboratory simulation of municipal solid waste fermentation with leachate recycle. J. Chem. Technol.
Biotechnol. 36 (12), 547–556.
McBean, E.A., Rovers, F.A., Farquahar, G.J., 1995. Solid Waste Landfill; Engineering and Design. Prentice Hall.
Mehta, R., Yazdani, R., Augenstein, D., Bryars, M., Sinderson, L., 2002. Refuse decomposition in the presence and absence of leachate recirculation. J. Environ.
Eng. 128 (3), 228–236.
Moore, D., Lees, B., Dayey, S., 1991. A new method for predicting vegetation distributions using decision tree analysis in a geographic information system.
Environ. Manage. 15 (1), 59–71.
Rapti-Caputo, D., Vaccaro, C., 2006. Geochemical evidences of landfill leachate in groundwater. Geoenviron. Eng. 85 (1), 111–121.
Saarela, J., 2003. Pilot investigations of surface parts of three closed landfills and factors affecting them. Environ. Monit. Assess. 84 (1–2), 183–192.
Sanphoti, N., Towprayoon, S., Chaiprasert, P., Nopharatana, A., 2006. The effects of leachate recirculation with supplemental water addition on methane
production and waste decomposition in a simulated tropical landfill. J. Environ. Manag. 81 (1), 27–35.
Sourmunen, K., Ettala, M., Rintala, J., 2008. Internal leachate quality in a municipal solid waste landfill: Vertical, horizontal and temporal variation and
impacts of leachate recirculation. J. Hazard. Mater. 160 (2), 601–607.
Taiwo, A.M., 2011. Composting as a sustainable waste management technique in developing countries. J. Environ. Sci. Technol. 4 (2), 98–102.
Tatsi, A., Zouboulis, A., 2002. A field investigation of the quantity and quality of leachate from a municipal solid waste landfill in a Mediterranean climate
(Thessaloniki, Greece). Adv. Environ. Res. 6 (3), 207–219.
Tränkler, J., Visvanathan, C., Kuruparan, P., Tubtimthai, O., 2005. Influence of tropical seasonal variations on landfill leachate characteristics—Results from
lysimeter studies. Waste Manage. 25 (10), 1013–1020.
Walker, P., 1990. Modelling wildlife distributions using a geographic information system: Kangaroos in relation to climate. J. Biogeogr. 17 (3), 279–289.
White, D., Sifneos, J.C. 1997. Mapping multivariate spatial relationships from regression trees by partitions of color visual variables; CSM 57th annual
convention, ASPRS 63rd Annual Convention, Seattle, Washington, pp. 86–95.
Wreford, K., Atwater, J., Lavkulich, L., 2000. The effects of moisture inputs on landfill gas production and composition and leachate characteristics at the
Vancouver Landfill Site at Burns Bog. Waste Management and Research 18 (4), 386–392.
Youcai, Z., Zhugen, C., Qingwen, S., Renhua, H., 2001. Monitoring and long-term prediction of refuse compositions and settlement in large-scale landfill.
Waste Manag. Res. 19 (2), 160–168.
Zacharof, A.I., Butler, A.P., 2004. Stochastic modelling of landfill leachate and biogas production incorporating waste heterogeneity. Model formulation and
uncertainty analysis. Waste Manage. 24 (5), 453–462.