Professional Documents
Culture Documents
Novel Simple Approaches To Modeling Composting Kinetics
Novel Simple Approaches To Modeling Composting Kinetics
A R T I C L E I N F O A B S T R A C T
Editor: Dr. Z. Wen Mathematical models have been developed over the past 40 years to describe the composting process, seeking to
ease its implementation, control, and optimization. Due to the complex nature of composting, the ability to
Keywords: simulate the processes kinetics in a simple and generalizable manner has proven to be elusive, acting as a sig
Aerobic biodegradation nificant limitation to effective environmental, model-assisted, decision-making. Current simple models are
Organic waste
ungeneralizable, while generalizable models lack simplicity, requiring information on many operating variables,
Modeling
such as temperature, moisture content, and oxygen content. The aim of this work is therefore to explore the use
Agricultural waste
Food waste of novel modeling approaches to produce generalizable and simple composting models that do not require any of
Simulation this data, while providing a more accurate representation of degradation than current simple models. Four
modeling methods are assessed in this study, all based on a first-order kinetic expression. These novel modeling
approaches split the degradation into three separate phases and only require two parameters: a degradation rate
and an estimation of the ratio between the duration of the mesophilic and thermophilic phases. The models were
assessed through their normalized root mean square error and validated over three independent datasets sourced
from the literature, covering a wide range of waste types and characteristics. The novel methods achieved errors
varying between 1.13% and 6.32% and outperformed a traditional first-order model in every case, as well as
more complex models in certain cases. Scenario analyses also demonstrated the resilience of the proposed ap
proaches to uncertainty.
* Corresponding author at: BioEngine – Research Team on Green Process Engineering and Biorefineries, Chemical Engineering Department, Université Laval, 1065
ave. de la Médecine, Québec, QC G1V 0A6, Canada.
E-mail addresses: [email protected] (E. Walling), [email protected] (C. Vaneeckhaute).
https://doi.org/10.1016/j.jece.2021.105243
Received 16 November 2020; Received in revised form 10 February 2021; Accepted 15 February 2021
Available online 18 February 2021
2213-3437/© 2021 Elsevier Ltd. All rights reserved.
E. Walling and C. Vaneeckhaute Journal of Environmental Chemical Engineering 9 (2021) 105243
Where Rdegradation is the rate of degradation (mass or concentration over scenario analyses to identify how the models react to various situations.
Through this investigation, this work highlights the potential of novel
time), [Si ] is the mass or concentration of biodegradable substrate, t is
simple mathematical modeling approaches for composting kinetics
time, kd is the (degradation/hydrolysis) rate constant (usually either at
modeling.
20 ◦ C or around 60 ◦ C), and f T , f MC and f O2 are correction functions for
This paper is divided as follows: following the context, problems
temperature, moisture content and oxygen content (dimensionless),
statement and research objectives presented in Section 1, Section 2
respectively (corrections can also be applied for pH, free airspace, and
presents the materials and methods used to undertake this assessment,
any other inhibiting variables).
followed by a detailed presentation of the results and discussions around
The challenge with composting modeling is that the correction
them in Section 3, with concluding remarks being presented in Section
functions all vary as functions of time, given the change in these vari
4.
ables throughout the process. For example, common expressions used
for these corrections include the cardinal temperature equation (Eq. 2)
2. Methods
[8,9], the Haug [10] equation for moisture (Eq. 3), and the Monod
equation for oxygen (Eq. 4):
2.1. General assumptions
(T − Tmax )(T − Tmin )2
fT = ( )[( )( ) ( )( )]
Topt − Tmin Topt − Tmin T − Topt − Topt − Tmax Topt + Tmin − 2T As detailed in the introduction, correction functions are a necessity
(2) to provide generalizable composting models that take into consideration
the impact of the dynamic composting environment on the rate of
1 biodegradation. One of the primary goals of this work is to develop and
f MC = (3)
exp( − 17.684MC + 7.0622) + 1 evaluate a modeling approach that could provide the benefits of
correction functions, without requiring the modeling or data necessary
O2 to obtain these temperature, moisture and oxygen corrections.
f O2 = (4)
O 2 + k O2 To achieve this, and considering the context for which these models
are addressed, the following assumptions have been made: (1) it is
where T is the temperature, Tmax is set to 71.6 ◦ C, Tmin is 5 ◦ C, and Topt is assumed that the process is well operated, meaning that oxygen content
58.6 ◦ C [11], MC is the moisture content, ranging between 0 and 1, O2 is is maintained above 10% and moisture content is maintained between
the oxygen content, expressed as a % v/v, and kO2 is the oxygen 50% and 65% [18–21]. In well operated processes, oxygen and moisture
half-saturation coefficient, given a value of 2% v/v following the works content limitations are not very inhibitive, limiting growth (with respect
of Haug [10] and Richard et al. [12]. to maximum degradation) by about 0–10% for each factor, within these
It is not rare to see correction functions being omitted if kd is fit to the ranges. As such, if it is assumed that the process operates within these
experimental data. Indeed, of the 209 models reviewed by Walling et al. ranges, constant correction coefficients for moisture and oxygen cannot
[1], 52% used first-order kinetics and 54% of the models did not be set. Based on the Haug [10] moisture correction and the Monod ox
implement correction functions. In many of these cases, Eq. (1) can be ygen correction, which have found strong support in the literature [1,
integrated into a simple (inverse) exponential equation [13–17]. How 22], these coefficients have been set at 0.9 for oxygen (cO2 ) and 0.95 for
ever, this can seriously limit the generalizability of such models, an moisture content (cMC ). In cases where it is known that the process will
aspect that can be very important when in a decision-making context. operate outside these ranges, the correction coefficients can simply be
Ensuring the use of correction functions and general kinetic data, such as changed accordingly, as dictated by the pertinent equations.
degradation rates at reference temperatures, allows for composting Regarding temperature, even well operated composting systems will
models to be as generalizable as possible, but also contributes to their see great variations in the temperature of the environment. This can lead
complexity. Indeed, when correction functions are applied, Eq. (1) must to correction functions ranging in value from 0.2 (20% of maximum
be solved by accounting for the time dependent nature of these func degradation) at 25 ◦ C to 1 (100%) at 60 ◦ C, as determined by the two
tions, which limits its resolution to traditional numerical processes such most popular temperature correction functions, the cardinal tempera
as the finite differences approach, while requiring the necessary data for ture [8,9] and the Haug [10] functions [11,23]. In the continued aim of
these correction functions. favoring simplicity, the authors therefore propose to consider these
Therefore, developing easily usable and generalizable composting temperature corrections as step functions instead. Composting generally
models could provide a powerful tool for decision-makers, while has a temperature profile with a consistent shape, heating up during the
addressing an important gap in composting modeling research, as first few days of composting, reaching a thermophilic phase for a while
identified in the most recent systematic review on the subject [1]. This and then cooling down during the passive phase of composting, as
work therefore seeks to address this gap by focusing on two primary presented in Fig. 1 [24–27]. Therefore, regarding temperature, the
research questions: (1) how can the biodegradation of organic matter process can be split into three major phases: initial increase from mes
through composting be modeled without relying on correction func ophilic to thermophilic ranges, thermophilic composting, and the
tions, and (2) how do these approaches compare to the simplest existing decrease back to mesophilic temperatures [24–27]. Consequently, the
composting models? It is therefore in this aim that novel simple temperature correction could be expressed as a step function, with each
modeling approaches to simulate the biodegradation of organic matter step being a constant correction coefficient (cTi ) representing an aver
during composting are proposed and evaluated. The modeling ap aged temperature correction over these ranges, expressed as in Eq. (5).
proaches developed in this work are all based around the basic ⎧
first-order equation (Eq. 1) and seek to model the degradation process by ⎨ cT1 if 0 ≤ t < t1
circumventing the use of correction functions and temperature, mois f T cT2 if t1 ≤ t < t2 (5)
⎩
cT3 if t2 ≤ t
ture and oxygen data, relying solely on degradation rates and process
durations. In total, four methods are evaluated, with some methods Therefore, using these assumptions and simplifications, the need for
having one or two different resolution approaches, resulting in the temperature, oxygen and moisture data can be removed, as will be
assessment of six different modeling approaches. These models were described in Section 2.2.
evaluated using three independent datasets taken from the literature
and encompassing a wide range of feedstocks and operating conditions. 2.2. Modeling methods
The resulting outputs were then analyzed numerically using the
normalized root mean square error (NRSME), as well as through An overview of the various proposed modeling approaches is
2
E. Walling and C. Vaneeckhaute Journal of Environmental Chemical Engineering 9 (2021) 105243
Fig. 1. General composting temperature profile. Phase 1 represents the initial mesophilic phase of composting, while phase 2 represents the highly active ther
mophilic phase of the process. Phase 3 represents the passive phase of composting, where biodegradation slows down and returns to mesophilic ranges.
3
E. Walling and C. Vaneeckhaute Journal of Environmental Chemical Engineering 9 (2021) 105243
( ( ))
S = S0 exp − kd cT,i+1 cMC,i+1 cO2 ,i+1 t + ti,i+1 (10) Where n is the amount of data points, Smodel,i is the amount of substrate
predicted by the model at time i, Sexp,i is the experimental value for
where S is the amount/concentration of substrate at time t, S0 is the substrate amount/concentration at time i, and S0 is the initial quantity
initial amount/concentration of substrate, kd is the rate constant, and of substrate.
cT,i , cMC,i , and cO2 ,i are correction coefficients for the phase i (dimen
sionless). 3. Results and discussion
This pathway is then repeated for every subsequent step of the pro
cess. Note that, in this specific case, t refers to the duration within a 3.1. Modeling results
phase, and therefore resets to 0 at the beginning of each phase. The total
composting time would therefore be a summation of the t of each phase, 3.1.1. Averaged NRMSEs over all the trials
or t1 , t2 and the t in phase 3. The averaged NRMSEs for each method are presented in Table 2,
with the results for each data set presented in Table 3. When looking at
2.2.4. Method 4: integrated first-order model with time dependent kd the averaged NRMSEs over all the trials in Table 2, one can start by
The final method, method 4, uses the integrated form of Eq. (1) by noting that all the modeling approaches provide, on average, a low
assuming that kd is a step function over the different phases, each with error, delivering accurate predictions for organic matter degradation
their own constant correction coefficient (ci), as expressed in Eq. (11). over time. Indeed, the average NRMSE of the approaches assessed in this
⎧ work varied from 1.86% to 4.98%, with the lowest and highest errors of
⎨ kd c1 if 0 ≤ t < t1
kd kd c2 if t1 ≤ t < t2 (11) individual trials being 1.13% and 11.98%, respectively.
⎩
kd c3 if t2 ≤ t Firstly, these results present a strong confirmation for the validity of
first-order kinetic models to accurately represent biodegradation during
Assuming three phases, the integrated form of Eq. (1) can be the composting process. This has been a somewhat contentious area in
expressed as follows for the three different phases: the literature, with an important divide between modelers implement
Phase1: S = S0 exp(kd (c1 t1 )) (12) ing first-order models [31–33] vs Monod-type (Monod, Contois, Tessier)
models [34–36]. The main processes involved in microbial growth
Phase2: S = S0 exp(kd (c1 t1 − c2 t2 + c2 t1 )) (13) during composting are hydrolysis, cellular uptake and cell death [1,37].
For microorganisms to be able to absorb soluble substrates, these com
Phase3: S = S0 exp(kd (c1 t1 − c2 t2 + c2 t1 − c3 t3 + c3 t2 )) (14) pounds must be made available for them. Though certain substrates (e.g.
simple sugars) can be degraded easily, many of the substrates used in
where S is the amount/concentration of substrate at time t, S0 is the composting are complex and hard to degrade, such as (hemi)cellulose
initial amount/concentration of substrate, kd is the rate constant, and ci and lignin. Consequently, hydrolysis of these hard to degrade com
are correction coefficients for the phase i (dimensionless), representing pounds has been believed to be the rate limiting step of the process,
the production of corrections for temperature, moisture and oxygen allowing for the accurate modeling of composting through first-order
content. (hydrolysis) kinetics [1]. Therefore, the results of this assessment seem
to provide further confirmation that hydrolysis is the rate limiting pro
2.3. Modeling assessment cess in the degradation of organic waste through composting [13,35,37,
38].
The data used to undertake this modeling assessment was sourced Secondly, there are some interesting results regarding the relative
from the works of Petric et al. [28], Kulcu [29], and Malamis et al. [30]. performance of the models compared to one-another. One important
These works studied a wide range of feedstocks and operating condi comparison is method 1b, given that this represents the optimal case for
tions, providing a strong and independent dataset to verify the modeling simple models using a traditional first-order equation. Therefore, one of
approaches. Indeed, these works cover a range of size of 35 L to 4 m3, the main questions that this work sought to answer was whether the
with substrates including municipal solid waste, poultry manure, cattle proposed alternative modeling approaches could outperform this
manure, sheep manure, sewage sludge (primary and secondary), agri commonly used method. From Table 2, it can be seen that the three
cultural waste, sawdust, wheat straw, zeolite and perlite. Of further novel modeling approaches managed to provide a lower average NRMSE
pertinence to this study, temperature ranges between and within than method 1b, though the NRMSE range of method 2b was compa
experimental sets were also highly variable, with ranges between 25 ◦ C rable to method 1b, while that of methods 3 and 4 was notably lower.
and 75 ◦ C, with the lowest trial occurring between 25 ◦ C and 55 ◦ C, The only two methods that had a higher averaged percent error were the
while the highest occurred between 45 ◦ C and 75 ◦ C. Initial substrate traditional first-order model using the averaged experimental kd
properties were also highly variable, contributing to the value of using (method 1a; 4.15%) and the finite difference approach with an averaged
these three datasets. Initial moisture content varied between 39% and correction coefficient (method 2a; 5.02%). These errors are likely more
72%, initial organic matter varied between 62% and 87%, while initial indicative of the errors to be expected if the models were generalized,
C/N ratios were between 25 and 40. In total, 77 (11 trials x 7 models per given that they were based off general experimental degradation rates
trial) different simulations were undertaken for the initial assessment, and not fit to case-specific data, which would be more likely of this kind
with an added 91 simulations for the scenario analysis and 108 for the of model application.
sensitivity analysis. Of the novel modeling approaches, method 3, using the multiple
The quality of fit of the outputs of the models were compared
quantitatively to the experimental results using a Normalized Root Mean Table 2
Square Error (NRMSE) using Eqs. (13) and (14): Average NRMSE over all trials.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ Modeling approach NRMSE Range (%) Average NRMSE (%)
√∑
√n ( )2
√ S − Sexp,i
√i=1 model,i (15) Method 1a: averaged kd 1.94–8.84 4.15
RMSE = Method 1b: optimized kd 1.86–6.38 3.02
n
Method 2a: averaged T coefficient 1.62–11.98 5.02
Method 2b: three phases 1.20–6.32 2.61
RMSE
NRMSE = × 100 (16) Method 3: three phases 1.13–3.56 1.86
S0 Method 4: three phases 1.34–3.18 1.90
4
E. Walling and C. Vaneeckhaute Journal of Environmental Chemical Engineering 9 (2021) 105243
Table 3 followed by method 2b, method 1b and then methods 1a and 2a. In the
NRMSE of the individual datasets. Note that the “CF model” is described in cases of data from Petric et al. [28] and Kulcu [29], method 2b has the
Section 3.1.4. lowest NRMSE in a few cases, though the results of methods 2b, 3 and 4
NRMSE (%) remain very similar. Beyond what was gleaned in Section 3.1.1, an
Dataset Modeling Trial Trial Trial Trial Average
important takeaway of Table 3 is the consistently improved performance
approach 1 2 3 4 of the novel methods (2b, 3 and 4) when compared to the traditional
models (methods 1a, 1b and 2a). In every trial, methods 1a and 2a are
Malamis et al. Method 1a 8.84 4.66 6.35 5.72 6.39
[30] Method 1b 6.38 3.40 4.36 4.54 4.67 outperformed by the three proposed approaches, while method 1b,
Method 2a 7.54 8.92 11.98 11.82 10.07 which represents the most accurate simple model currently available, is
Method 2b 3.17 5.00 6.32 3.41 4.48 outperformed in all but two of the twelve trials.
Method 3 3.56 1.13 1.34 2.28 2.08
Method 4 3.18 1.64 2.41 1.97 2.3
CF model 2.15 3.97 4.64 7.8 4.64
3.1.3. Numerical and visual quality of fit over different process durations
Kulcu [29] Method 1a 2.26 3.25 5.18 3.20 3.47 Further insight can be derived from looking at the profiles of the
Method 1b 2.11 1.87 3.13 2.01 2.28 models compared to the experimental data, as presented in Fig. 2, as
Method 2a 1.94 1.88 3.13 2.03 2.24 well as a quantitative assessment of this in Table 4, which presents the
Method 2b 1.29 1.2 2.32 1.67 1.62
average NRMSE of each modeling approach for three periods: days 1–5,
Method 3 1.38 1.25 2.4 1.43 1.62
Method 4 1.35 1.34 2.31 1.44 1.61 6–10 and 11 to the end (processes varying between 20- and 25-days,
Petric et al. Method 1a 3.27 1.94 2.66 2.62 total). Both Table 4 and Fig. 2 highlight the advantages of considering
[28] Method 1b 2.24 1.86 2.11 2.07 degradation in separate phases. In the simulations implementing
Method 2a 2.26 4.41 1.62 2.76 methods 1, an overprediction of biodegradation during the beginning of
Method 2b 1.88 1.56 1.82 1.75
Method 3 2.14 1.59 1.97 1.90
the process can be noted, followed by an underestimation of degradation
Method 4 2.12 1.63 1.62 1.79 afterwards. This is to be expected, given the impact of using an invari
CF model 1.86 0.99 1.53 1.46 able kd , and reflects the reason the authors sought to explore these
alternative modeling approaches. By using a constant degradation rate
during the entire process, regardless of whether it is averaged or
first-order expression fitted to one-another in series, provided the lowest
results (1.86%), though given experimental and data processing uncer
tainty, it appears to be on par with method 4 (integrated first-order
Table 4
equations with time dependent kd ; 1.90% error). The finite differences
Average NRMSE during early, mid and late stages of composting.
approach using three different temperature correction coefficients for
Average NRMSE (%)
each of the three phases (method 2b) also provided a better performance
than method 1b (2.61% vs 3.02%). Modeling approach Days 0–5 Days 6–10 Days 11 to end
Fig. 2. Comparison of modeling approaches with experimental data for trial 2 of Malamis et al.’s [30] dataset. Note the ability of the proposed modeling approaches
(2b through 4) to better approximate the impact of variable degradation rate when compared to method 1b.
5
E. Walling and C. Vaneeckhaute Journal of Environmental Chemical Engineering 9 (2021) 105243
optimized, degradation will inevitably be over or underestimate at (three from Malamis et al. [30]). Of these models that outperformed the
different locations. For example, if a maximum degradation rate is used simple modeling approaches, the difference in NRMSE was small, once
(e.g. kd = kmax, measured around 60 ◦ C), degradation outside of ther again pointing to the potential benefits of these new approaches. Indeed,
mophilic stages, notably at the beginning of the process, will be greatly the largest difference between a model using full correction functions
overestimated. Conversely, if a degradation rate at a lower reference with a lower NRMSE than a simple modeling approach was 0.57%
temperature (e.g. kd measured at 45 ◦ C) is used, then early degradation (Petric et al. [28], trial 2), with the “full” model providing an NRMSE of
would be overestimated, while degradation during the thermophilic 0.99% vs method 2′ s 1.56%. On the other hand, when looking at the
phase would be greatly underestimated. This latter case is what we cases where the simple approaches outperformed the more complex
observe in Fig. 2 when looking at the results for method 1b. ones, a much more significant difference is observed. The largest of these
Indeed, when looking at phase 1 (days 0–4), it can be noted that differences was observed in the case of trial 4 from Malamis et al. [30],
method 1b greatly overestimates degradation when compared to the with a difference of 5.83% (1.97% method 4 vs 7.80% with full
other modeling approaches. This is confirmed towards the end of this correction functions). Therefore, the modeling approaches proposed in
period, when experimental data shows that degradation is around 32% this paper might also benefit from being less sensitive to different
of biodegradable organic matter (BOM), versus the 40% predicted by combinations of correction functions, which can lead to significant dif
method 1b. The other three methods (2b, 3 and 4), which represent the ferences in results. For example, using the case of Malamis et al.’s [30]
novel approaches, are all closer to the real value, varying between 29% trial 4, as mentioned above, the combination of correction functions led
and 32%. Moving into phase 2, method 1b still overpredicts degradation to an NRMSE of 7.80%, while the most optimal combination of correc
early on, while quickly starting to underpredict it around day 7. During tion functions led to an error of 6.71% (in this case, a combination of
this same time period, the other three models continue to follow the Haug [10] for temperature, Stombaugh and Nokes [39] for moisture,
experimental curve very well, with methods 2b and 4 being particularly and Higgins and Walker [40] for oxygen), though this is still signifi
close, while method 3 lags behind somewhat, though it is still much cantly outperformed by method 4 in this particular case.
more accurate than method 1b. Thus far, between phases 1 and 2, the Another possible reason for the apparent benefit of the simpler
advantages of the variable degradation rate (achieved through the models is that the use of correction functions on such small timescales
correction coefficients) is clear, allowing for inhibited degradation (daily variations) might further contribute to the sensitivity of the
during phase 1, followed by unrestricted degradation during phase 2, models, producing overly reactive responses. This impact can be com
better mirroring the reality of the process. This once again comes into pounded when using the traditional finite differences resolution
play at the transition between phases 2 and 3, where the degradation approach, given that the degradation of each time step is dependent on
becomes more inhibited again and progresses slower, though the benefit that of prior steps, something that is not the case in the models of
might be less obvious on this relatively small timeframe given that the methods 3 and 4. Indeed, when implementing correction functions, a
majority of degradation has already happened in this case, and thus sharp increase or decrease in one of the controlling variables can relate
everything (models and experiments) converges towards about 80% of to an equally important stimulation or inhibition of microbial activity
BOM degraded. through these methods, not taking into consideration a potential lag in
Comparatively, in methods 2b through 4, which all demonstrated the response of the system to the environmental conditions. This will be
similar behaviors, the error during the early phase of composting was further discussed in Section 3.4.
decreased. In some cases, degradation was even underestimated, given
the slightly inhibitive moisture and oxygen coefficients used (cMC = 3.2. Values of the correction coefficients and time relationships
0.95, cO2 = 0.90). For example, in the case of trial 2 of Petric et al. [28],
the error could be decreased from 1.56% to 1.41% by increasing oxygen Given the central role of correction coefficients in the proposed
and moisture coefficients to 1 during the second phase (a period of 4 modeling approaches, the authors thought it pertinent to investigate the
days). This can always be considered and optimized by modelers, but values of the correction coefficients for the first and last phases, given
given the interest in as simple a modeling approach as possible, the that the second phase had a coefficient set to 1 in all trials. The mean
authors believe that the trade-off of using constant moisture and oxygen value of the temperature correction coefficient for phase 1, cT,1 , was
coefficients outweighs the small increase in error. Furthermore, in 0.45, with a standard deviation of 0.06. The lowest and highest observed
corroboration with the earlier findings, methods 3 and 4 generally values were 0.32 and 0.55. It is important to note, however, that not all
outperformed the other methods, including method 2b. experimental trials exhibited a first phase, with all of the experiments
from Kulcu [29] and one experiment from Petric et al. [28] starting
3.1.4. Comparison with models implementing full correction functions directly in the highly active thermophilic phase (t1 = 0). For the third
The last comparison that the authors believe is important to make is a phase, the mean temperature correction coefficient, cT,3 , was 0.64, with
comparison of the results of these simple models, using only degradation a standard deviation of 0.09, and a low and high of 0.47 and 0.87. Of the
rates and phase duration information to those of first-order models using processes exhibiting three phases, the mean duration of the second
temperature, oxygen, and moisture correction functions (and data). The phase was 1.93 times longer than the first phase, with a standard de
most precise version of these models, implementing Eq. (1), use opti viation of 0.75. However, the authors stress that this information on
mized degradation rates (kd ) alongside the cardinal temperature or Haug timing is based on a very limited set of experimental data and is in no
[10] models for temperature, Haug [10] model for moisture and Monod way intended to be a generalizable guideline; it is purely for informa
equation for oxygen (combination found through data not presented in tional purposes.
this work). In the aim of seeing how the novel simple models compare to
these models implementing correction functions, a first-order model 3.3. Sensitivity and scenario analysis
(Eq. 1) with the cardinal temperature, Haug [10] (moisture) and Monod
corrections was applied, using a finite difference numerical approach. The newly proposed modeling approaches are built around the
This assessment was undertaken with Petric et al.’s [28] and Malamis notion of splitting the degradation rate into distinct phases with
et al.’s [30] datasets, totaling in seven different trials. The results of this different degradation rates and lengths. Therefore, uncertainty on these
assessment are provided in the rows denoted as “CF model” in Table 3. parameters could potentially have a significant impact on the output of
Of the seven trials compared, four models using the full correction the models. This is especially important given how the duration of the
functions managed to slightly outperform the simple modeling ap phases were based on experimental temperature data in this paper,
proaches with the lowest NRMSEs (one from Malamis et al. [30] and something that may not be available when applying such a model. To
three from Petric et al. [28]), while three performed significantly worse assess this, a sensitivity analysis and a scenario analyses were applied.
6
E. Walling and C. Vaneeckhaute Journal of Environmental Chemical Engineering 9 (2021) 105243
7
E. Walling and C. Vaneeckhaute Journal of Environmental Chemical Engineering 9 (2021) 105243
the benefits of a three-phase structure, “general” models were applied 3.5. Further research
using the parameter values and timing relationships identified in Section
3.2, the results of which are shown in Table 6. In this case, phase 1 was The present work demonstrates the ability of a variety of simple
assumed to last 3 days, while phase 2 was set as being twice as long mathematical approaches to model organic matter degradation during
(rounding up of 1.93). The temperature correction coefficients were composting, without requiring temperature, oxygen or moisture content
0.45 for the first phase, 1 for the second, and 0.64 for the third. The data, and their associated correction functions. However, despite the
results of this application using general parameters yielded good results, strong results provided by the simple models, further research on
with the NRMSE varying between 2.13% and 5.29%, with method 4 degradation patterns during composting would be necessary. Given the
providing the lowest NRMSE in all three cases, though the difference in limitations highlighted in Section 3.4, the main challenge for simple
errors between the methods is very close. composting modeling is the development of generalizable models.
Being able to predict the transition between the composting phases
3.4. Model usability and limitations would be of the greatest benefit, both to the models developed here and
to the field of composting modeling as a whole. Given the near-inverse
All of the approaches explored and developed in this study aimed at exponential nature of degradation, determining the duration of the
being simple, both conceptually and regarding their numerical appli earlier phases is more important when focusing on organic matter. An
cation. However, there are some differences between a few of the important part of the length of composting processes is due to curing and
methods that could influence which one a modeler might want to maturation, which are poorly understood phenomena, from a modeling
implement. One of these particularities is notably regarding methods 3 perspective, and far outweigh the scope of these simple models. The
and 4 when compared to methods 2/2b. Unlike methods using finite advantages in applications of simple models for decision-making is that
differences, these methods provide an estimation of degradation with the variability of these phases can be considered in scenario analyses,
respect to time that is in a way independent of prior degradation. In the which are highly recommended when modeling for decision support [2].
case of methods 2/2b, the use of finite differences means that the However, it would be interesting to establish the duration of the sepa
degradation over a time interval is dependent on the amount of substrate rate phases or if the timing of their transition can be related to other
at the beginning of the time-step and is therefore affected by prior process conditions. Avenues that could potentially be explored include
degradation. As highlighted earlier when comparing to models imple whether a relative rate of degradation can be associated to a change in
menting correction functions, this might contribute to a propagation in phase (transition from more active to less active phases), or whether a
errors by causing an overreactive model, an issue which may be less relationship can be established between the duration of phases relative
present in methods 3 and 4 and that might offer an advantage when it to one-another. Furthermore, as highlighted in Section 3.4, separating
comes to simple modeling. the process into three phases appears to provide better results than a
It is also important to address the potential for under and overfitting single or two-phased approach. However, further validation would be
when it comes to such simple models. Given the simplicity of the models beneficial with different operating conditions. If improved performances
and how they are based on first-order kinetic expressions, overfitting are necessary, then it could be interesting to investigate separating the
was not an issue. There were fears though that the low NRMSEs obtained process into a few more phases, such as four or five. A separation into
throughout this study could be a product of underfitting. However, this more phases was not investigated given that, the more phases consid
was found to not be the case through a visual comparison of data with ered, the further a model is from being “simple” and the closer it is to
the model outputs, as highlighted in Fig. 2, as well as through com being a traditional model with correction functions.
parison of the results of the simple models with traditional models. In terms of application of this work, the proposed modeling ap
Regarding limitations, it is important to once again note that these proaches can provide a basis for future composting models, notably for
simple approaches are aimed at providing models that can give a very application in decision-making scenarios where information may be
general idea of degradation during composting and are not intended as a lacking, or the needs for quick results and optimization may make it
fundamental representation of the biodegradation process. As such, unfeasible to use more detailed methods. These kinetic models can also
these assessments have been undertaken using the aforementioned be used as a basis for simple models with a greater scope than simply
assumption that the process is well operated. It is possible that some of predicting degradation. For example, conversion/yield factors, which
these assumptions might not hold true for all systems. Indeed, despite are commonly used in composting modeling to relate many different
having a variable and representative range of substrates, temperature, phenomena to degradation, such as moisture production, oxygen con
moisture and oxygen profiles in the experimental datasets used for this sumption, gas emissions [41], heat generation, etc. [1,7], could still be
study, the models were not evaluated for extremes in conditions during used to provide a wider assessment of the composting process.
the process. Such conditions could be present even in well-operated
systems, such as semi-batch composting, where the addition of sub 4. Conclusion
strate can lead to a sharp temperature decrease, or naturally aerated or
windrow systems which can see more significant oxygen and moisture As highlighted in the results and discussion, the novel methods
depletion. provided very good estimations of the biodegradation, with the NRMSEs
of these methods varying between 1.13% and 6.32% over the entirety of
the datasets. These methods, notably methods 2b, 3 and 4, managed to
not only provide low errors, but also ensured a better visual quality of fit
with experimental data, showcasing their ability to better represent the
realities of composting. These methods also managed to outperform the
Table 6
simplest existing kinetic models (method 1b), while also performing
Results from the application of the modeling methods with the “general” pa
rameters and time relationships identified in Section 3.2. very closely to or significantly better than more complex models
implementing correction functions for moisture content, oxygen, and
NRMSE (%)
temperature. Furthermore, the scenario analysis also supports the ben
Modeling Malamis et al. [30], Kulcu [29], Petric et al. [28], efits of separating the degradation into three phases, instead of one or
approach trial 1 trial 2 trial 1
two, while demonstrating the relative resilience of the models to po
Method 2b 4.21 2.16 3.51 tential uncertainty in the timing of the phases. However, separation into
Method 3 5.29 2.13 2.96 more phases (four or five) was not assessed and may present potential
Method 4 3.48 2.13 2.59
for improvements, though it is important to balance potential benefits
8
E. Walling and C. Vaneeckhaute Journal of Environmental Chemical Engineering 9 (2021) 105243
with the goal of simplicity in this case. Furthermore, the models also [15] K. Nakasaki, A. Ohtaki, A simple numerical model for predicting organic matter
decomposition in a fed-batch composting operation, J. Environ. Qual. 31 (2002)
provided adequate results when using completely generalized data (i.e.
997–1003.
non-optimized degradation rates or temperature coefficients). [16] X. Qian, G. Shen, Z. Wang, C. Guo, Y. Liu, Z. Lei, Z. Zhang, Co-composting of
livestock manure with rice straw: characterization and establishment of maturity
evaluation system, Waste Manag. 34 (2014) 530–535, https://doi.org/10.1016/j.
CRediT authorship contribution statement wasman.2013.10.007.
[17] M. Hamoda, H.A. Qdais, J. Newham, Evaluation of municipal solid waste
The research and article were conceptualized, undertaken and composting kinetics, Resour. Conserv. Recycl. 23 (1998) 209–223, https://doi.org/
10.1016/S0921-3449(98)00021-4.
written in totality by Eric Walling, under the supervision of Céline
[18] E. Bertran, X. Sort, M. Soliva, I. Trillas, Composting winery waste: sludges and
Vaneeckhaute who provided guiding remarks and revisions of the work. grape stalks, Bioresour. Technol. 95 (2004) 203–208, https://doi.org/10.1016/j.
biortech.2003.07.012.
[19] T.L. Richard, H. Hamelers, A. Veeken, T. Silva, Moisture relationships in
Declaration of Competing Interest composting processes, Compost Sci. Util. 10 (2002) 286–302, https://doi.org/
10.1080/1065657X.2002.10702093.
[20] S. Tiquia, N. Tam, I. Hodgkiss, Changes in chemical properties during composting
The authors declare that they have no known competing financial of spent pig litter at different moisture contents, Agric. Ecosyst. Environ. 67 (1998)
interests or personal relationships that could have appeared to influence 79–89, https://doi.org/10.1016/S0167-8809(97)00132-1.
the work reported in this paper. [21] R. Haug. The Practical Handbook of Compost Engineering, Routledge, 2018.
[22] T. Richard, L. Walker, J. Gossett, The effects of oxygen on solid-state
biodegradation kinetics, Proc. IBE 2 (1999) A22–A39.
Acknowledgements [23] I. Mason, An evaluation of substrate degradation patterns in the composting
process. Part 2: temperature-corrected profiles, Waste Manag. 28 (2008)
1751–1765, https://doi.org/10.1016/j.wasman.2007.06.019.
This work has been carried out at Université Laval, supported [24] T. Abbasi, S. Gajalakshmi, S. Abbasi, Towards modeling and design of
financially by the Natural Sciences and Engineering Research Council of vermicomposting systems: mechanisms of composting/vermicomposting and their
Canada (NSERC) through the award of an NSERC Discovery Grant implications, Indian J. Biotechnol. 8 (2009) 177–182.
[25] M.F. Tortarolo, M. Pereda, M. Palma, N.M. Arrigo, Influencia de la inoculación de
(RGPIN-2017-04838) and an NSERC Collaborative Research and microorganismos sobre la temperatura en el proceso de compostaje, Cienc. Del
Development Grant (CRDPJ/509167-2017) awarded to Céline Suelo 26 (2008) 41–50.
Vaneeckhaute, as well as by the Fonds de Recherche du Québec Nature [26] G. Tortosa, A. Castellano-Hinojosa, D. Correa-Galeote, E.J. Bedmar, Evolution of
bacterial diversity during two-phase olive mill waste (“alperujo”) composting by
et Technologies through the award of an FRQNT doctoral scholarship 16S rRNA gene pyrosequencing, Bioresour. Technol. 224 (2017) 101–111.
(138536) to Eric Walling. Céline Vaneeckhaute holds the Canada [27] E. Walling, A. Babin, C. Vaneeckhaute, Nutrient and Carbon Recovery from
Research Chair in Resource Recovery and Bioproducts Engineering. Organic Wastes, Biorefinery, Springer, 2019, pp. 351–373, https://doi.org/
10.1007/978-3-030-10961-5_14.
[28] I. Petric, A. Helić, E.A. Avdić, Evolution of process parameters and determination
References of kinetics for co-composting of organic fraction of municipal solid waste with
poultry manure, Bioresour. Technol. 117 (2012) 107–116, https://doi.org/
[1] E. Walling, A. Trémier, C. Vaneeckhaute, A review of mathematical models for 10.1016/j.biortech.2012.04.046.
composting, Waste Manag. 113 (2020) 379–394, https://doi.org/10.1016/j. [29] R. Kulcu, New kinetic modelling parameters for composting process, J. Mater.
wasman.2020.06.018. Cycles Waste Manag. 18 (2016) 734–741, https://doi.org/10.1007/s10163-015-
[2] E. Walling, C. Vaneeckhaute, Developing successful environmental decision 0376-9.
support systems: challenges and best practices, J. Environ. Manag. 264 (2020), [30] D. Malamis, K. Moustakas, K.-J. Haralambous, Evaluating in-vessel composting in
110513, https://doi.org/10.1016/j.jenvman.2020.110513. treating sewage sludge and agricultural waste by examining and determining the
[3] B.S. McIntosh, J.C. Ascough, M. Twery, J. Chew, A. Elmahdi, D. Haase, J.J. Harou, kinetic reactions of the process, Clean Technol. Environ. Policy 18 (2016)
D. Hepting, S. Cuddy, A.J. Jakeman, S. Chen, A. Kassahun, S. Lautenbach, 2493–2502, https://doi.org/10.1007/s10098-016-1230-z.
K. Matthews, W. Merritt, N.W.T. Quinn, I. Rodriguez-Roda, S. Sieber, M. Stavenga, [31] K. Das, H. Keener, Numerical model for the dynamic simulation of a large scale
A. Sulis, J. Ticehurst, M. Volk, M. Wrobel, H. van Delden, S. El-Sawah, A. Rizzoli, composting system, Trans. ASABE 40 (1997) 1179–1189, https://doi.org/
A. Voinov, Environmental decision support systems (EDSS) development – 10.13031/2013.21339.
challenges and best practices, Environ. Model. Softw. 26 (2011) 1389–1402, [32] J.Y. Ge, G.Q. Huang, J.B. Li, L.J. Han, Particle-scale visualization of the evolution
https://doi.org/10.1016/j.envsoft.2011.09.009. of methanogens and methanotrophs and its correlation with CH4 emissions during
[4] R. Wieland, C. Gutzler, Environmental impact assessment based on dynamic fuzzy manure aerobic composting, Waste Manag. 78 (2018) 135–143, https://doi.org/
simulation, Environ. Model. Softw. 55 (2014) 235–241, https://doi.org/10.1016/j. 10.1016/j.wasman.2018.05.045.
envsoft.2014.02.001. [33] J. Zhang, D. Gao, T.B. Chen, G.D. Zheng, J. Chen, C. Ma, S.L. Guo, W. Du,
[5] S.A. Hajkowicz, Supporting multi-stakeholder environmental decisions, J. Environ. Simulation of substrate degradation in composting of sewage sludge, Waste Manag.
Manag. 88 (2008) 607–614, https://doi.org/10.1016/j.jenvman.2007.03.020. 30 (2010) 1931–1938, https://doi.org/10.1016/j.wasman.2010.04.004.
[6] Z. Li, H. Lu, L. Ren, L. He, Experimental and modeling approaches for food waste [34] J. Kaiser, Modelling composting as a microbial ecosystem: a simulation approach,
composting: a review, Chemosphere 93 (2013) 1247–1257, https://doi.org/ Ecol. Model. 91 (1996) 25–37, https://doi.org/10.1016/0304-3800(95)00157-3.
10.1016/j.chemosphere.2013.06.064. [35] F. Sole-Mauri, J. Illa, A. Magri, F.X. Prenafeta-Boldu, X. Flotats, An integrated
[7] I.G. Mason, Mathematical modelling of the composting process: a review, Waste biochemical and physical model for the composting process, Bioresour. Technol. 98
Manag. 26 (2006) 3–21, https://doi.org/10.1016/j.wasman.2005.01.021. (2007) 3278–3293, https://doi.org/10.1016/j.biortech.2006.07.012.
[8] L. Rosso, J. Lobry, S. Bajard, J.-P. Flandrois, Convenient model to describe the [36] G. Vidriales-Escobar, R. Rentería-Tamayo, F. Alatriste-Mondragón, O. González-
combined effects of temperature and pH on microbial growth, Appl. Environ. Ortega, Mathematical modeling of a composting process in a small-scale tubular
Microbiol. 61 (1995) 610–616. bioreactor, Chem. Eng. Res. Des. 120 (2017) 360–371, https://doi.org/10.1016/j.
[9] L. Rosso, J. Lobry, J.-P. Flandrois, An unexpected correlation between cardinal cherd.2017.02.006.
temperatures of microbial growth highlighted by a new model, J. Theor. Biol. 162 [37] Y. Wang, F. Witarsa, Application of Contois, Tessier, and first-order kinetics for
(1993) 447–463. modeling and simulation of a composting decomposition process, Bioresour.
[10] R.T. Haug, The Practical Handbook of Compost Engineering, CRC Press, 1993. Technol. 220 (2016) 384–393, https://doi.org/10.1016/j.biortech.2016.08.099.
[11] T.L. Richard, L.P. Walker, Modeling the temperature kinetics of aerobic solid-state [38] P.B. Woodford, In-vessel Composting Model with Multiple Substrate and
biodegradation, Biotechnol. Prog. 22 (2006) 70–77, https://doi.org/10.1021/ Microorganism Types, Kansas State University, 2009.
bp050176a. [39] D.P. Stombaugh, S.E. Nokes, Development of a biologically based aerobic
[12] T.L. Richard, L.P. Walker, J.M. Gossett, Effects of oxygen on aerobic solid-state composting simulation model, Trans. ASABE 39 (1996) 239–250, https://doi.org/
biodegradation kinetics, Biotechnol. Prog. 22 (2006) 60–69, https://doi.org/ 10.13031/2013.27504.
10.1021/bp050171d. [40] C.W. Higgins, L.P. Walker, Validation of a new model for aerobic organic solids
[13] B. Jolanun, S. Tripetchkul, C. Chiemchaisri, P. Chaiprasert, T.S. yoon, Effect of decomposition: simulations with substrate specific kinetics, Process Biochem. 36
moisture content on fed batch composting reactor of vegetable and fruit wastes, (2001) 875–884, https://doi.org/10.1016/S0032-9592(00)00285-5.
Environ. Technol. 26 (2005) 293–302, https://doi.org/10.1080/ [41] E. Walling, C. Vaneeckhaute, Greenhouse gas emissions from inorganic and organic
09593332608618561. fertilizer production and use: a review of emission factors and their variability,
[14] D.P. Komilis, A kinetic analysis of solid waste composting at optimal conditions, J. Environ. Manag. 276 (2020), 111211, https://doi.org/10.1016/j.
Waste Manag. 26 (2006) 82–91, https://doi.org/10.1016/j.wasman.2004.12.021. jenvman.2020.111211.