Investigating E ﬀ ects of Landﬁll Soil Gases on Landﬁll Elevated Subsurface Temperature

Featured Application: This research was designed to understand the e ﬀ ects of landﬁll soil gases on the events of subsurface landﬁll ﬁres or elevated temperature which is experienced by many landﬁlls during their operational and post-operational period. A framework proposed in this article can assist landﬁll authorities in controlling or preventing such hazardous incidents related to subsurface elevated thermal conditions. Abstract: Subsurface temperature is a critical indicator for the identiﬁcation of the risk associated with subsurface ﬁre hazards in landﬁlls. Most operational landﬁlls in the United States (US) have experienced exothermic reactions in their subsurface. The subsurface landﬁll area is composed of various gases generated from chemical reactions inside the landﬁlls. Federal laws in the US mandate the monitoring of gases in landﬁlls to prevent hazardous events such as landﬁll ﬁre breakouts. There are insu ﬃ cient investigations conducted to identify the causes of landﬁll ﬁre hazards. The objective of this research is to develop a methodological approach to this issue. In this study, the relationship was investigated between the subsurface elevated temperature (SET) and soil gases (i.e., methane, carbon dioxide, carbon monoxide, nitrogen, and oxygen) with the greatest inﬂuence in landﬁlls. The signiﬁcance level of the e ﬀ ect of soil gases on the SET was assessed using a decision tree approach. A naïve Bayes technique for conditional probability was implemented to investigate how di ﬀ erent gas combinations can a ﬀ ect di ﬀ erent temperature ranges with respect to the safe and unsafe states of these gases. The results indicate that methane and carbon dioxide gases are strongly associated with SETs. Among sixteen possible gas combinations, three were identiﬁed as the most probable predictors of SETs. A three-step risk assessment framework is proposed to identify the risk of landﬁll ﬁre incidents. The key ﬁndings of this research could be beneﬁcial to landﬁll authorities and better ensure the safety of the community health and environment. Writing—Review J.W.E., R.W.P., S.S.; Supervision, J.W.E. R.N.; R.W.P., R.N. M.K.; Project Administration, R.N.; Funding Acquisition, R.N.


Introduction
To date, landfills have been the most commonly used solid-waste disposal option. Solid-waste landfills have been designed to contain municipal solid waste in a large scale containment system to prevent and control the harmful effects on the environment and surrounding communities. A total temperatures in the subsurface environment besides observing the effects of individual gas parameters. This paper focuses on a methodical evaluation of risk factors affecting the corresponding SET that indicates possible fires. The aim of this paper is to discuss the unsafe ranges of subsurface temperatures with potential to cause substantial damage to the landfill consistency based on safe and unsafe soil gas ranges in the landfill gas collection system.

Background
Volumetrically, landfill soil gases usually comprises 40 to 60% carbon dioxide (CO 2 ), 45 to 60% methane (CH 4 ), 2 to 5% nitrogen (N 2 ), and 0.1 to 1% oxygen (O 2 ) along with low fractions of ammonia, CO, hydrogen, sulfides, and nonmethane organic compounds (NMOCs) such as benzene, trichloroethylene, and vinyl chloride [35]. Landfills produce gases in three processes: most gases are produced by aerobic and anaerobic bacterial decomposition; solid or liquid wastes convert into vapor via volatilization, e.g., nonmethane organic compounds (NMOCs) from disposal chemicals; chemical reactions between the remaining chemicals in waste [35]. Over a period of decades, landfill waste goes through bacterial decomposition in five phases: Phase I: Refuse matters are converted into water and CO 2 in the aerobic decomposition process by aerobic bacteria that live in an oxygen enriched environment.
Phase II: In the absence of oxygen, anaerobic bacteria initiate anaerobic decomposition. Phase III: Some anaerobic bacteria produced in Phase II consume the organic acids in this decomposition phase. Organic acids formed via a microbial hydrolysis process convert to soluble acids and further break down to nitrous oxide, CO 2 , CH 4 and a small amount of VOCs [36,37].
Phase IV: This phase is relatively steady with the production rates and composition of the landfill gases. These gases are produced at a steady rate for 50 or more years after the waste pile is placed in the landfill [38].
Phase V: Some organic matters are available to be decomposed and oxygen is introduced in the environment. Figure 1 shows a flow chart of the chemical decomposition processes and the by-products generated during the different phases of waste decomposition. Factors such as waste characteristics (age and composition of the refuse), disposed chemicals in landfills, and some environmental factors (moisture content, temperature, the presence of oxygen) influence the volume and rate of generated landfill gases (e.g., CO 2 , CH 4 , N 2 , and hydrogen sulfide) [39]. Bacterial activity increases as the landfill's temperature rises resulting in increased gas production. Changes in gas compositions have been observed to occur in advance of increases in wellhead temperatures, which can be an indicator of an approaching SET event [34].
MSW landfills produce CO 2 , water, and heat in the process of aerobic decomposition (Jafari et al., 2017a). In the absence of oxygen, anerobic decomposition initiates after aerobic decomposition with the resultant production of CO 2 , CH 4 , and heat. Equations (1) and (2) can be used to express the aerobic and anaerobic transformation of organic waste, respectively [34]. A comparison between the enthalpies of both reactions indicates that the heat generated in aerobic decomposition is approximately 19 times higher than the heat produced from an anaerobic reaction [34]. As a result, the temperature typically ranges between 60 and 80 • C for a waste pile in aerobic conditions [34,40], whereas the temperature in the anaerobic landfills ranges from approximately 25 to 45 • C [31,33]. Heat accumulation by some exothermic reactions with the intrusion of oxygen or aerobic degradation creates a suitable condition to initiate and continue the smoldering combustion of MSWs [20,39]. The tetrahedron of combustion theory describes the four conditions required for a combustion to happen [39,41]: (1) an oxidizing agent (e.g., oxygen via air intrusion); (2) a source of fuel in MSWs; (3) a source of energy (e.g., heat generated from any exothermic reaction or aerobic decomposition); (4) chain reaction of combustion that is self-sustaining (e.g., charred waste). In MSW landfills, it is essential to limit any air intrusion that can be readily controlled. Subsurface combustion in landfills usually spreads through smoldering Appl. Sci. 2020, 10, 6401 4 of 19 combustion directly occurring on the solid fuel surface [10]. CO, CO 2 , water vapor, and heat are yielded from the partial smoldering combustion of cellulose [42], as shown in Equation (3). (1) Appl. Sci. 2020, 10, x FOR PEER REVIEW 4 of 20 MSW landfills produce CO2, water, and heat in the process of aerobic decomposition (Jafari et al., 2017a). In the absence of oxygen, anerobic decomposition initiates after aerobic decomposition with the resultant production of CO2, CH4, and heat. Equations (1) and (2) can be used to express the aerobic and anaerobic transformation of organic waste, respectively [34]. A comparison between the enthalpies of both reactions indicates that the heat generated in aerobic decomposition is approximately 19 times higher than the heat produced from an anaerobic reaction [34]. As a result, the temperature typically ranges between 60 and 80 °C for a waste pile in aerobic conditions [34,40], whereas the temperature in the anaerobic landfills ranges from approximately 25 to 45 °C [31,33]. Heat accumulation by some exothermic reactions with the intrusion of oxygen or aerobic degradation creates a suitable condition to initiate and continue the smoldering combustion of MSWs [20,39]. The tetrahedron of combustion theory describes the four conditions required for a combustion to happen [39,41]: (1) an oxidizing agent (e.g., oxygen via air intrusion); (2) a source of fuel in MSWs; (3) a source of energy (e.g., heat generated from any exothermic reaction or aerobic decomposition); (4) chain Smoldering ignition does not proceed to completion due to the limited O 2 available but can propagate at a low level of oxygen, e.g., less than 3% volume-to-volume ratio (v/v) [43][44][45]. Smoldering combustion within an MSW landfill has been documented to persist at temperatures between 100 and 120 • C [46]. In some cases, the temperature during smoldering combustion has been detected in the range of 200-300 • C and even as high as 700 • C in MSW landfills [23,28]. Bergström and Björner [47] measured a deep subsurface fire with a range of 80-230 • C. Research has shown that the integrity and service life of landfill leachate control systems, gas-extraction systems, materials and cover of the composite liner systems can be impacted by a sustained temperature as low as 85 • C [48]. During a Appl. Sci. 2020, 10, 6401 5 of 19 SET, the landfill gas composition rapidly changes from CO 2 (40-55% v/v) and CH 4 (50-60% v/v) to a gas composition of CO 2 (60-80% v/v), hydrogen (10-35% v/v), CO (>1500 ppmv) along with the ratio of CH 4 to CO 2 falling below 1 [34]. Jafari et al. [49] observed the subsurface temperature rising from 55 to 75 • C in their case study, when CH 4 to CO 2 ratio reduced from 1.2 to 0.3 in a course of 50 days [49].
The technical procedures for detecting, evaluating, and mitigating elevated temperatures or landfill fires vary in the literature. The elevated temperature considered in this study refers to an increase in gas-well temperature beyond a certain threshold. The United States Environmental Protection Agency (US EPA) recommends a normal level of O 2 , CH 4 , and temperature ranges for landfills to be, respectively, <5%, 45% to 60%, and less than <55 • C, and a CO level of 2000 ppm is regarded as an action level (the concentration level in excess requires regulatory or remedial action) [50]. The Solid Waste Association of North America (SWANA) suggests temperature <52 • C, O 2 levels <1%, and CH 4 levels from 45 to 58% as normal ranges, and traces of CO in excess of 25 ppm calls for an action level to take precautionary steps against SETs [39,51]. The difference in the ranges for O 2 , CH 4 , and temperature recommended by the SWANA from those of the US EPA are minimal. The level for CO suggested by the US EPA is more tolerant, up to 2000 ppm, whereas the SWANA has a strict interpretation of the action level with the presence of only 25 ppm CO. Unlike the US EPA, the SWANA considers CO and residual nitrogen (RN 2 ) to be possible indicators of smoldering events. Table 1 displays the ranges of subsurface RN 2 level along with their indications as described by Estrabrooks [52]. The SWANA emphasizes the monitoring of CO to assess landfill fires. Table 2 simplifies the details regarding landfill operations and fire prevention. Other organizations such as the Ohio Environmental Protection Agency (Ohio EPA), International Solid Waste Association (ISWA), U.S. Army Corps of Engineers (USACE), Republic Services, Inc., Cornerstone, Geosyntec Consultants, Conestoga-Rovers and Associates provide guidance for landfill gas management practices and the management of smoldering events in the form of standard operating procedures. Some of the criteria in their documents overlap and others vary from organization to organization. However, these recommendations can be only used in the presence of a fire incident already endangering the whole landfill system and surrounding environment, which implies incompetency in the current system. A risk assessment method is crucial in associating these parameters to the future risks of SETs so that fire outbreaks can be predicted and prevented. Table 1. Ranges of residual nitrogen (RN 2 ) in the landfill soil gases composition [52].

Percentage of Residual Nitrogen (RN 2 ) Indications
0-12% The internal extraction system contains this range in most operating landfills.

16-20%
Deemed essential to regulate perimeter migration, side slope emission, or where other compromises are required.

>20%
Indication of an aggressive landfill gas-extraction system with potential to initiate aerobic conditions.

Methodology
This research was conducted in the following steps that include the data collection of landfill gases (such as CH 4 , CO 2 , O 2 and balance gas), the categorization of parameters (gases and wellhead temperature) in terms of their safe ranges, and the performance of statistical tests to assess each parameter's influence on temperature. The statistical tests included the conditional inference trees algorithm to assess the effects of all parameters in different temperature ranges [53]. The significance levels of the soil gas parameters affecting the SET were assessed using this decision tree approach, which can be used to detect the most significant parameters affecting landfill subsurface fires. Next, the influence of various gas parameter combinations on the subsurface temperatures was analyzed. A naïve Bayes technique [54] was utilized to determine how different gas combinations can affect different temperature ranges in terms of safe and unsafe gas states. The likelihood of different temperature ranges corresponding with the possible parameter combinations was investigated to identify the combinations primarily responsible for SETs. Both the conditional inference trees algorithm and the naïve Bayes conditional probability tests were performed using "R", a programming language for statistical computing. Lastly, a three-step risk assessment framework is proposed to identify the risk of landfill fire incidents. Figure 2 shows a flow chart of the step-by-step research procedure and outputs. temperature ranges corresponding with the possible parameter combinations was investigated to identify the combinations primarily responsible for SETs. Both the conditional inference trees algorithm and the naïve Bayes conditional probability tests were performed using "R", a programming language for statistical computing. Lastly, a three-step risk assessment framework is proposed to identify the risk of landfill fire incidents. Figure 2 shows a flow chart of the step-by-step research procedure and outputs. Step-by-step research procedures and outputs.

Data Collection
This research used archived data relating to the abovementioned factors collected for the Bridgeton Sanitary Landfill, Missouri. The landfill was permitted to operate in 1985 and is regulated under the Missouri Department of Natural Resources' Solid Waste Management Program (SWMP). It stopped receiving waste in 2004, when the refuse mass covered almost 210,436.5 square meters at a depth of approximately 73.2 m beneath the surface with a total 97.54 m waste thickness. The landfill first notified the SWMP about the SET events along with smoldering and odor concerns in some gasextraction wells in 2010. Since 2013, public access was granted for commonly requested data files and reports related to subsurface smoldering events and odors in and around Bridgeton Landfill [55]. Step-by-step research procedures and outputs.

Data Collection
This research used archived data relating to the abovementioned factors collected for the Bridgeton Sanitary Landfill, Missouri. The landfill was permitted to operate in 1985 and is regulated under the Missouri Department of Natural Resources' Solid Waste Management Program (SWMP). It stopped receiving waste in 2004, when the refuse mass covered almost 210,436.5 square meters at a depth of approximately 73.2 m beneath the surface with a total 97.54 m waste thickness. The landfill first notified the SWMP about the SET events along with smoldering and odor concerns in some gas-extraction wells in 2010. Since 2013, public access was granted for commonly requested data files and reports related to subsurface smoldering events and odors in and around Bridgeton Landfill [55].
The box in the upper left-hand corner in Figure 3 shows a geographical site overview of the Bridgeton Landfill and the locations of the gas-extraction wells used to collect subsurface gas samples. The three other boxes show the names of the gas-extraction wells within the A, B and C areas outlined in the overview box. Weekly and monthly sample data are available for these gas-extraction wells and were used in this study for in-depth analysis. Weekly gas-well data comprise the concentrations of the basic gas parameters (CO 2 , CH 4 , O 2 , and balance gas) and temperature data, but the monthly data contains only CH 4 , CO 2 , O 2 , N 2 , hydrogen (H 2 ), and CO without the wellhead temperature data. We used weekly available datasets for the time period of June 2013 to October 2016 since the temperature is the most important parameter for our data analysis. We collected 18,469 observations from the gas-interceptor wells, gas-collection wells, and temperature monitoring probes. Table 3 shows a sample of the collected data.
As discussed in Section 2, RN 2 and the CH 4 -to-CO 2 ratio are two significant gas parameters to predict gas-well temperatures and can be calculated using the collected gas data. RN 2 is the percentage of N 2 during aerobic decomposition that stays unused. The overpulling of gas due to air infiltration through a gas collection system can cause there to be excess N 2 . The O 2 in the intruded air pulled by the vacuum in the gas-collection system kills methanogens and initiates aerobic conditions. O 2 is consumed during the decomposition process and the N 2 present in the air remains in the landfill. A report provided by the SWANA considers O 2 , CH 4 , and CO 2 as crucial parameters to estimate the concentration of the balance gas, which primarily indicates the amount of N 2 , and emphasizes that the usual ratio of N 2 to O 2 is approximately 3.76 [52]. Using a simple gas equilibrium equation, we estimated the RN 2 concentration-e.g., if a gas-well measures CO 2 (28.1%), CH 4 (32.5%), O 2 (3.7%), then the balance gas would be the rest of the gas portion (100 − 32.5 − 28.1 − 3.7 = 35.7%). The normal N 2 was measured by multiplying the typical ratio (3.76) with the oxygen composition (3.7%), 3.76 * 3.7 = 13.912%. The RN 2 was then calculated by subtracting the normal N 2 composition (13.912%) from the balance gas (35.7%) which yields an RN 2 composition of 21.8% [52]. the concentrations of the basic gas parameters (CO2, CH4, O2, and balance gas) and temperature data, but the monthly data contains only CH4, CO2, O2, N2, hydrogen (H2), and CO without the wellhead temperature data. We used weekly available datasets for the time period of June 2013 to October 2016 since the temperature is the most important parameter for our data analysis. We collected 18,469 observations from the gas-interceptor wells, gas-collection wells, and temperature monitoring probes. Table 3 shows a sample of the collected data.

Categorization of Variables
Before we analyzed the temperature pattern in the Bridgeton Landfill, it was important to determine the correlation of the selected factors to the temperature as well as their influence on temperature. Therefore, we categorized the collected gas data in terms of the safe and normal ranges for landfill soil gases according to the operational standards mentioned in title 40 of the Code of Federal Regulations (CFR) §60.753 operational standards and considered any temperature less than 176 • F or 80 • C as a safe limit [10,39]. A SET was considered to occur when the gas wellhead temperature exceeded 80 • C, which is the highest temperature limit in any aerobic and anaerobic decomposition process. The safe limit for RN 2 was considered to be less than 20% [52]. Thus, variable thresholds were established to categorize each parameter into two categories: safe and unsafe. The parameters selected for our analysis and their categorization rules are listed in Table 4. For each parameter-CH 4 , CH 4 -to-CO 2 ratio, O 2 , RN 2 , temperature, and CO-there are two possibilities. We described each sampling event with a combination of safe or unsafe states for five factors listed in Table 4. Therefore, the total number of possible sample events in our analysis can be calculated by raising two to the number of parameters. For the five parameters in our case, the possible event numbers will be 2 5 or 32.

Testing the Effect of Each Variable on Temperature
In the absence of CO data for the Bridgeton Landfill, we analyzed the other available gas parameters. First, we determined how individual gas conditions can affect the temperature. Figure 4 demonstrates the temperature ranges in boxplots, when the four factors (CH 4 , O 2 , RN 2 , CH 4 -to-CO 2 ratio) meet the criteria for safe and unsafe conditions (safe = 0 and unsafe = 1). The boxplot shows the distribution of temperature datapoints with five measures: the minimum, median, maximum, first and third quartile. The middle quartile marks the median value (midpoint of the data) and is indicated by the middle line dividing the box into two. The samples within the first to third quartiles contain 50% of the total samples and is represented as the interquartile range (shown as the colored boxes in Figure 4). A total of 25% of the total datapoints fall below the first quartile, while the remaining 25% fall above the third quartile. The data points considered as outliers are located above the third quartile and below the first quartile, 1.5 times outside the interquartile range and are shown as open circles in the figure. Three factors other than O 2 were observed to have a high median and third quartile values for temperature in the unsafe condition than in the safe condition. We noticed that the safe O 2 range recommended by the EPA was associated with higher temperature values typically referred to as an unsafe temperature range, which contrasts with the literature described above. To summarize Figure 4, when the gas conditions are unsafe, the temperature should be higher. Milosevic et al. [56] observed O 2 concentrations of 15 to 21.2% v/v (more than double the reference O 2 volume of 5%) in gas-wells, whereas the gas-well temperature ranged from 24.9 to 48.9 • C and the fire probability ratio decreased by 0.836 with an increase in the CH 4 -to-CO 2 ratio concentrations. Another study [16] also found similarly high O 2 levels (15-21.2% v/v) corresponding with gas-wells in both cool temperature and SET areas. The reason for this difference in O 2 s impact on temperature could be the categorization of O 2 with reference to the recommended range by the US EPA, or its relationship to other parameters, or the operational status of this particular landfill. O 2 s effect on temperature should be reexamined in light of its categorization levels of varying ranges. Its reference value (5% vol ) by the US EPA should also be reevaluated.
Another study [16] also found similarly high O2 levels (15-21.2% v/v) corresponding with gas-wells in both cool temperature and SET areas. The reason for this difference in O2′s impact on temperature could be the categorization of O2 with reference to the recommended range by the US EPA, or its relationship to other parameters, or the operational status of this particular landfill. O2′s effect on temperature should be reexamined in light of its categorization levels of varying ranges. Its reference value (5%vol) by the US EPA should also be reevaluated.

Decision Tree Analysis
We applied the conditional inference trees algorithm [53] to 18,469 observation datapoints of wellhead temperatures and gas parameters to identify the factors most closely associated with subsurface temperatures. The algorithm approximates a regression relationship in a conditional inference framework using the binary recursive partitioning method. First, it tests the null hypothesis of independence between the response (temperature) and input variables (gas variables), and stops if the test fails to reject the hypothesis that temperature is an independent variable. Otherwise, the input variable with the strongest association is selected based on a p-value (significance level, p < 0.05) corresponding to the hypothesis test. Following this test, a binary partition of the selected variable is implemented. The algorithm repeats these two steps recursively at each split of a variable. The first parameter selected for the binary split represents a significant parameter generally linked to subsurface temperature.
We classified all the parameters as to whether they were in a safe or unsafe range before applying the algorithm, classifying the temperatures as "under 55 • C", "55-80 • C", "80-93 • C" or "93-149 • C". Figure 5 demonstrates all the possible splits of a decision tree at a significance level of <0.05 and the parameters with the best splits in circles with their corresponding p-values. The decision tree branches indicate the levels of parameters and the bar plots at the bottom show the proportions of the four temperature ranges in each end node containing all observations with a combination of features. Among the four parameters, the CH 4 -to-CO 2 ratio is the covariate showing the strongest association with the temperature ranges with a significance level less than 0.001. Using a univariate logistical regression to study the synergistic effect of fire indicators, Milosevic et al. [56] found the CH 4 -to-CO 2 ratio concentration to be a statistically significant fire indicator. The first tree branch starting with a ratio of <1, shows a strong association with O 2 (significance level, p < 0.001), whereas CH 4 was strongly associated (significance level, p < 0.001) with the branch of a ratio of >1. The tree shows that the branch with CH 4 -to-CO 2 ratio < 1, O 2 < 5%, and CH 4 < 45% and >60% yields 8888 events from the total 18,469 data points. This branch results in the highest number of events in the temperature ranges of "80-93 • C" and "93-149 • C" than any other branches. Hence, the branch can be interpreted as the most unsafe one. The second highest datapoints in the temperature range 80-93 • C is noticed in the branch with a CH 4 -to-CO 2 ratio of <1 and O 2 > 5%. The third highest incidents (approximately 720 observations) in the temperature range of 55-80 • C were observed within the branch of CH 4 -to-CO 2 ratio > 1 and CH 4 with 45-60%. Therefore, the unsafe ranges of the gas parameters are not always associated with high temperature ranges, but rather different combinations of parameters.

Effect of Variable Combinations on Temperature
Although different documents or regulations regarding landfill operation suggest safe ranges for the mentioned parameters, it is essential to study how well these safe ranges corresponding with the subsurface temperature. We studied how the SET is influenced by various gas parameter combinations with a boxplot of temperature versus different events shown in Figure 6, where 0 is denoted as safe and 1 as unsafe. The observations from the figure are as follows:
Increasing the volume of sample data from recent years could firmly establish the truth of this statement. We used a group histogram to plot the variance and frequency of temperature in different combinations. Figure 7 presents the temperature distribution in each of the above cases, including their spread, peaks, and symmetry. The histogram shows temperature data with different frequencies, peaks, frequently non-normal values, and outliers. Histograms include multiple peaks We used a group histogram to plot the variance and frequency of temperature in different combinations. Figure 7 presents the temperature distribution in each of the above cases, including their spread, peaks, and symmetry. The histogram shows temperature data with different frequencies, peaks, frequently non-normal values, and outliers. Histograms include multiple peaks in some cases that indicate multimodal distributions. The question remains whether the mean temperature range statistically varies with respect to the combinations. The Wilcoxon rank-sum test [57], a nonparametric test method, is able to interpret the interpretation of the difference between these events. The test is robust against the presence of outliers and the non-normality of a sample distribution. Tests for a null hypothesis test (Ho) of equal means in two independent samples were performed. All pairs of the combinations were tested with the null hypothesis of no difference between any two events. Figure 8 shows the median values of the difference between a pair of events that resulted from the Wilcoxon rank-sum tests shown in a matrix of events. Each event is represented in a sequence of the safe and unsafe condition of temperature, CH4, CH4:CO2 ratio, and residual N2 and O2. The color ranges from green, white, and red in the figure represent a high negative difference, no difference and high positive difference between two events, respectively. For example, a pair of events (temperature = 1, CH4 = 1, CH4:CO2 = 1, RN2 = 1, O2 = 0) and (temperature = 0, CH4 = 1, CH4:CO2 = 1, RN2 = 1, O2 = 1) have the highest positive difference, while the (temperature = 0, CH4 = 0, CH4:CO2 = 0, RN2 = 0, O2 = 1) events and (temperature = 1, CH4 = 1, CH4:CO2 = 1, RN2 = 1, O2 = 0) events have the highest negative difference. In the figure, (***) represents an insignificant p-value <0.05 for that pair of events. The nondirectional two-sided test resulted in a significant p-value < 0.05 for all pairs except three pairs which indicates that the means for the three pairs of events share a significant similarity in their means and variance spreads.  The question remains whether the mean temperature range statistically varies with respect to the combinations. The Wilcoxon rank-sum test [57], a nonparametric test method, is able to interpret the interpretation of the difference between these events. The test is robust against the presence of outliers and the non-normality of a sample distribution. Tests for a null hypothesis test (H o ) of equal means in two independent samples were performed. All pairs of the combinations were tested with the null hypothesis of no difference between any two events. Figure 8 shows the median values of the difference between a pair of events that resulted from the Wilcoxon rank-sum tests shown in a matrix of events. Each event is represented in a sequence of the safe and unsafe condition of temperature, CH 4 , CH 4 :CO 2 ratio, and residual N 2 and O 2 . The color ranges from green, white, and red in the figure represent a high negative difference, no difference and high positive difference between two events, respectively. For example, a pair of events (temperature = 1, CH 4 = 1, CH 4 :CO 2 = 1, RN 2 = 1, O 2 = 0) and (temperature = 0, CH 4 = 1, CH 4 :CO 2 = 1, RN 2 = 1, O 2 = 1) have the highest positive difference, while the (temperature = 0, CH 4 = 0, CH 4 :CO 2 = 0, RN 2 = 0, O 2 = 1) events and (temperature = 1, CH 4 = 1, CH 4 :CO 2 = 1, RN 2 = 1, O 2 = 0) events have the highest negative difference. In the figure, (***) represents an insignificant p-value <0.05 for that pair of events. The nondirectional two-sided test resulted in a significant p-value < 0.05 for all pairs except three pairs which indicates that the means for the three pairs of events share a significant similarity in their means and variance spreads. Furthermore, the probability of four temperature ranges (under 55 °C, 55-80 °C, 80-93 °C, 93-149 °C) were analyzed with respect to the different combinations of four gas parameters in the series of CH4, CH4-to-CO2 ratio, RN2, and O2. The results of this analysis help us to identify possible gas combinations linked to SETs. Figure 9 shows a gradual upward trend for the 80-93 °C range in the four gas combinations of 1_1_0_1, 1_1_0_0, 1_1_1_0, and 1_1_1_1; whereas a decreasing trend is observed for the "under 55 °C" range. The 1_1_1_0 combination has a likelihood of only 3% in the 93-149 °C range. Hence, the graph shows that the 1_1_1_0 combination has the greatest potential to associate with high temperature ranges, rather than the 1_1_1_1 combination in which all the gas parameters fall in the unsafe range. Temperatures in the 55-80 °C range occur with a 15% probability in the combinations with 1_1_1_0, 1_1_1_1, 1_1_0_0, 1_1_0_1, 1_0_1_0, 1_0_1_1. Therefore, gas combinations with 1_1_0_0, 1_1_0_1, 1_1_1_0, 1_1_1_1 should be considered high risk; 1_0_1_0, 1_0_1_1 combinations correspond to a medium risk that indicates a tendency to proceed to a high risk level. Table 5 summarizes the potential risk levels (low, moderate, or high) corresponding with different gas combinations based on the 80 °C safe limit discussed in Section 5.1, which could be informative for the assessment of an SET risk.  Furthermore, the probability of four temperature ranges (under 55 • C, 55-80 • C, 80-93 • C, 93-149 • C) were analyzed with respect to the different combinations of four gas parameters in the series of CH 4 , CH 4 -to-CO 2 ratio, RN 2 , and O 2 . The results of this analysis help us to identify possible gas combinations linked to SETs. Figure 9 shows a gradual upward trend for the 80-93 • C range in the four gas combinations of 1_1_0_1, 1_1_0_0, 1_1_1_0, and 1_1_1_1; whereas a decreasing trend is observed for the "under 55 • C" range. The 1_1_1_0 combination has a likelihood of only 3% in the 93-149 • C range. Hence, the graph shows that the 1_1_1_0 combination has the greatest potential to associate with high temperature ranges, rather than the 1_1_1_1 combination in which all the gas parameters fall in the unsafe range. Temperatures in the 55-80 • C range occur with a 15% probability in the combinations with 1_1_1_0, 1_1_1_1, 1_1_0_0, 1_1_0_1, 1_0_1_0, 1_0_1_1. Therefore, gas combinations with 1_1_0_0, 1_1_0_1, 1_1_1_0, 1_1_1_1 should be considered high risk; 1_0_1_0, 1_0_1_1 combinations correspond to a medium risk that indicates a tendency to proceed to a high risk level. Table 5 summarizes the potential risk levels (low, moderate, or high) corresponding with different gas combinations based on the 80 • C safe limit discussed in Section 5.1, which could be informative for the assessment of an SET risk. in the combinations with 1_1_1_0, 1_1_1_1, 1_1_0_0, 1_1_0_1, 1_0_1_0, 1_0_1_1. Therefore, gas combinations with 1_1_0_0, 1_1_0_1, 1_1_1_0, 1_1_1_1 should be considered high risk; 1_0_1_0, 1_0_1_1 combinations correspond to a medium risk that indicates a tendency to proceed to a high risk level. Table 5 summarizes the potential risk levels (low, moderate, or high) corresponding with different gas combinations based on the 80 °C safe limit discussed in Section 5.1, which could be informative for the assessment of an SET risk. . Naïve Bayes conditional probability [54] of four temperature ranges corresponding with diffent gas combinations.

Discussions
Our main goal of this study was to investigate the effects of soil gases on the SET over a certain threshold. There are several regulations recommended by the regulatory agencies regarding acceptable ranges of these landfill soil gases and subsurface temperatures. A temperature threshold of 80 • C was selected in this study since temperatures up to 80 • C have been observed during normal biological decomposition processes. We selected the thresholds for some of the gas parameters according to those reported in the 40 CFR §60.753 operational standards-Thalhamer [39] and Estabrooks [52]. Our observations from the statistical analysis conducted above, can be drawn:

•
Unsafe temperature events have occurred more in the unsafe range for CH 4 , the CH 4 -to-CO 2 ratio, and RN 2 , but not for O 2 . The combination associated with the highest SET did not include an unsafe O 2 range, a finding which contradicts the suggested safe O 2 range.

•
The conditional inference tree algorithm shows that the CH 4 -to-CO 2 ratio of the four parameters has the strongest association with temperature. Ratios less than 1, O 2 concentrations less than 5% and CH 4 <45% and >60% resulted in the highest probability of events ranging from 80 to 149 • C. However, the safe O 2 range was not associated with a safe temperature range, which is also consistent with the observation above. . Using the conditional probability results, the gas combinations have been summarized according to their potential risk levels (low, moderate, or high) with respect to SETs.

•
The unsafe ranges of the gas parameters are not always associated with high temperature ranges, but rather different combinations of parameters.
• A three-step procedure can be followed to evaluate the risk of a landfill subsurface fire. The diagram in Figure 10 provides a general idea of this step-by-step procedure for preventing subsurface fire incidents.
Step 3 proposes to check gas-wells showing similar combinations of (CH 4 = 1, CH 4 :CO 2 = 0, O 2 = 0, RN 2 = 1) and (CH 4 = 1, CH 4 :CO 2 = 0, O 2 = 1, RN 2 = 1) that correspond to a medium risk, and then monitor these gas-well locations more carefully or more often to ensure that they do not become one of the riskiest combinations. Landfill authorities can repeat this three-step action plan as part of their regular monitoring practices of the gas collection system.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 17 of 20 Figure 10. A three-step action plan for preventing landfill subsurface fires.
The dataset used for this research does not contain CO data, which may be a significant parameter affecting landfill subsurface fire incidents. Jafari et al. [16] reported observations of CO concentrations above 1000 ppmv and up to 10200 ppmv at wellhead temperatures above 68 °C and as high as 95 °C. The reviewed literature also acknowledges CO as a significant indicator for identifying SETs. The analysis of gas variables without the incorporation of CO data might overlook scenarios that could be linked to SETs. Hence, adding CO to the gas combinations will lead to a more precise identification of high-risk scenarios. Similarly, incorporating parameters such as H2 concentrations, leachate collection, and pressure could improve the analysis results.

Conclusions
The objective of this article relates to identifying the causes of landfill fire hazards associated with the major landfill gases. Currently, the system lacks appropriate and standardized measures in The dataset used for this research does not contain CO data, which may be a significant parameter affecting landfill subsurface fire incidents. Jafari et al. [16] reported observations of CO concentrations above 1000 ppmv and up to 10200 ppmv at wellhead temperatures above 68 • C and as high as 95 • C. The reviewed literature also acknowledges CO as a significant indicator for identifying SETs. The analysis of gas variables without the incorporation of CO data might overlook scenarios that could be linked to SETs. Hence, adding CO to the gas combinations will lead to a more precise identification of high-risk scenarios. Similarly, incorporating parameters such as H 2 concentrations, leachate collection, and pressure could improve the analysis results.

Conclusions
The objective of this article relates to identifying the causes of landfill fire hazards associated with the major landfill gases. Currently, the system lacks appropriate and standardized measures in the forecasting and controlling of subsurface landfill fires and lets the individual landfill authority or owner deal with this hazard on their own. This research developed a methodological approach to this issue and attempted to evaluate the recommended safe ranges of the landfill gases and gas-well temperatures. The relationships were investigated between the subsurface elevated temperature (SET) and soil gases, with the greatest influence from landfill gases (i.e., methane, carbon dioxide, carbon monoxide, nitrogen, and oxygen). Our works observed that SET events do not usually occur in the unsafe oxygen range; similarly, the safe oxygen range does not always associate with the safe temperature range. Rather than focusing on the individual gas effects, we observed temperature behaving differently with the different combinations of gases in safe or unsafe ranges. Our research identified some gas combinations highly associating with SET events and proposed a three-step procedure to control the SET events before they progress into subsurface fires. The research methodology used in this study can be repeated with respect to the parameter thresholds established by different regulatory agencies, i.e., the United States Environmental Protection Agency (US EPA), Solid Waste Association of North America (SWANA), International Solid Waste Association (ISWA), and United States Army Corps of Engineers (USACE). The effect of unsafe temperature conditions can be examined by comparing the results obtained with respect to these regulatory agencies. However, our proposed methodology might be a useful tool for the landfill authorities in regulating subsurface landfill temperatures.