Comparing estimates of fugitive landfill methane emissions using inverse plume modeling obtained with Surface Emission Monitoring (SEM), Drone Emission Monitoring (DEM), and Downwind Plume Emission Monitoring (DWPEM)

ABSTRACT

As part of the global effort to quantify and manage anthropogenic greenhouse gas emissions, there is considerable interest in quantifying methane emissions in municipal solid waste landfills. A variety of analytical and experimental methods are currently in use for this task. In this paper, an optimization-based estimation method is employed to assess fugitive landfill methane emissions. The method combines inverse plume modeling with ambient air methane concentration measurements. Three different measurement approaches are tested and compared. The method is combined with surface emission monitoring (SEM), above ground drone emission monitoring (DEM), and downwind plume emission monitoring (DWPEM). The methodology is first trialed and validated using synthetic datasets in a hand-generated case study. A field study is also presented where SEM, DEM and DWPEM are tested and compared. Methane flux during two-days measurement campaign was estimated to be between 228 and 350 g/s depending on the type of measurements used. Compared to SEM, using unmanned aerial systems (UAS) allows for a rapid and comprehensive coverage of the site. However, as showed through this work, advancement of DEM-based methane sampling is governed by the advances that could be made in UAS-compatible measurement instrumentations. Downwind plume emission monitoring led to a smaller estimated flux compared with SEM and DEM without information about positions of major leak points in the landfill. Even though, the method is simple and rapid for landfill methane screening. Finally, the optimization-based methodology originally developed for SEM, shows promising results when it is combined with the drone-based collected data and downwind concentration measurements. The studied cases also discovered the limitations of the studied sampling strategies which is exploited to identify improvement strategies and recommendations for a more efficient assessment of fugitive landfill methane emissions.

Implications: Fugitive landfill methane emission estimation is tackled in the present study. An optimization-based method combined with inverse plume modeling is employed to treat data from surface emission monitoring, drone-based emission monitoring and downwind plume emission monitoring. The study helped revealing the advantages and the limitations of the studied sampling strategies. Recommendations for an efficient assessment of landfill methane emissions are formulated. The method trialed in this study for fugitive landfill methane emission could also be appropriate for rapid screening of analogous greenhouse gas emission hotspots.

Introduction

Climate change presents enormous threats for the world. It has significant and undisputable implications for environment and global development and poses complex challenges for policy makers. The large commitment to the United Nations framework convention on climate change, known as the Paris Agreement, proves that fighting climate change has become a global consensus (UNFCCC 2015). Many nations around the world are developing innovative policies aiming at a transition to low-carbon economies and to adapt to changing climate. Major efforts are being made to control anthropogenic greenhouse gas emission from various sources. However, the most recent UN Environment report stated global greenhouse gas emissions show no signs of peaking (UNEP 2018). In contrast, global GHG emissions in 2030 need to be approximately 25 percent lower than in 2017 to put the world on a pathway to limiting global warming to 2°C. Therefore, sustained and more comprehensive actions are needed to understand, quantify, and manage greenhouse gas emissions. Actions should result in measurable reduction in greenhouse gas emissions contributing thus to stabilize their concentrations in the atmosphere at a level that would prevent the dangerous anthropogenic interference with the climate system.

Waste management is estimated to be the third largest source of methane emissions in the United States, behind enteric fermentation and natural gas systems (EPA 2019). In Europe, an estimated 30% of anthropogenic methane emissions are caused by landfills (EEA 2014). This caused attention to methane emissions from landfills to grow significantly. The reason is that methane is the second most important anthropogenic greenhouse gas after carbon dioxide. Furthermore, emission reduction from landfills is amongst the most feasible and cost‐effective measures to reduce anthropogenic GHG emissions (Oonk 2010). Although some modern landfills capture and use much of the methane gas produced, collecting system damages and cover defects could cause methane leaks to the atmosphere. Quantifying the amount of fugitive methane arising from landfills is a key point in GHG mitigation. In addition, measurements of methane emissions may represent a good way to evaluate the efficiency of landfill gas recovering systems or biocovers (Scheutz et al. 2011).

Measurement techniques for fugitive landfill methane emissions include: static and mobile plume measurement methods using tracer gas (Mønster et al. 2014, 2015; Scheutz et al. 2011), radial plume mapping (RPM) using optical remote sensing (ORS) by means of laser infrared radiation emissions (Goldsmith et al. 2011; Thoma et al. 2010), differential absorption light detection and ranging (LiDAR) (Babilotte et al. 2010) and inverse plume modeling (Mackie and Cooper 2009; Oonk 2010). A recent review by Mønster, Kjeldsen, and Scheutz (2019) presents currently used measurement techniques and discusses advantages and limitations of the different approaches.

Methane concentration measurements are used in many approaches (such as flux chambers, tracer method, etc.) to assess emission fluxes. In the specific case of inverse plume modeling, methane concentrations at the surface or downwind from the landfill are combined with a dispersion model to infer methane emissions (Figueroa et al. 2009). Mackie and Cooper (2009) used ambient air volatile organic compound (VOC) measurements and Voronoi diagrams to predict locations of potential emission sources. Emission rates are then calculated using linear regression. Kormi et al. (2017); (2018)) presented a methane emission estimation method that exploits ambient air methane concentration measurements on landfill surface. Kormi et al. (2018) used inverse plume modeling to estimate the whole-site methane emissions from a given landfill and identifies locations and emission rates of major leaks. Their optimization-based approach uses a stochastic search method (Genetic Algorithms) to solve the inverse problem that consists of identifying leakage sources (locations of hot-spots and corresponding emission rates) by having receptor locations and surface measurements along with meteorological conditions as input data. An alternative approach consists on using plume dispersion modeling where whole-site emission flux is obtained using downwind measurements of methane concentrations (Fredenslund et al. 2018). Mobile measurements of methane concentrations across downwind plumes are also a common emission estimation method in oil and gas industry (Atherton et al. 2017; O’Connell et al. 2019). Unlike, the tracer method, this approach does not rely on the controlled release of a tracer gas combined with downwind measurements of the target gas (Mønster et al. 2014, 2015).

A common landfill methane-concentration measurement typically involves a manual field survey which is often time consuming. An improvement to the traditional measurement process could be made by employing newer methodologies and technical advances offered through unmanned aerial vehicles (UAVs) such as drones. Many industries such as construction, site development and Digital Mapping have adopted drone-based data collection solutions. Utilizing unmanned aerial vehicles (UAVs) equipped with a light-weight gas sensing system can be an alternative to current methane monitoring methods (Berman et al. 2012; Khan et al. 2012). Yang et al. (2018) proposed a UAV-based remote system to investigate methane leaks in natural gas production sites. UAV platforms and compatible measurement instrumentation are increasingly investigated (Golston et al. 2017). Emran, Tannant, and Najjaran (2017) tested the integration of a low-cost laser-based methane detector into a multi-rotor UAV and tested it for landfill monitoring. Allen et al (2019, 2015) examined unmanned aerial platforms (rotary and fixed wing drones) and used carbon dioxide measurements to derive methane fluxes from a test landfill site using a mass balance model. The study yielded inaccurate results as carbon dioxide was a poor proxy for landfill emissions. Recently, Shah et al. (2019) proposed a near-field Gaussian plume inversion flux quantification method using UAV sampling. The method consists of flux inversion of plumes sampled near point sources. Measurements from a controlled release of methane served to test and refine the method.

As mentioned previously, Kormi et al. (2018) presented an optimization-based methane emission estimation method that exploits methane concentration measurements on landfill surface. A similar approach is used in this study to provide landfill methane emission estimates. Three different measurement approaches are tested and compared. The method is combined with surface emission monitoring (SEM), above ground drone emission monitoring (DEM), and downwind plume emission monitoring (DWPEM). The methodology is first trialed using a hand-generated case study. Subsequently, an actual landfill case study is presented. Sampling techniques (SEM, DEM and DWPEM) are tested and compared. The proposed method and the case studies are described in the next section. Results are presented and interpreted. The key findings of this study are discussed in the last section of this paper.

Materials and methods

Emission estimation approach

The proposed approach for estimating landfill methane emissions relies on ambient-air methane concentration measurements coupled with an optimization-based identification method. Measurements are used to infer emission rates through dispersion modeling and optimization. This is achieved through tracing dispersed methane back to emission sources. In the subsequent paragraphs, we briefly summarize the optimization-based approach, and refer the reader to Kormi et al. (2018) for a more detailed presentation of the method.

The starting point of the proposed method is an experimental survey of methane concentrations on top or downwind of the landfill. Concentration measurements could be obtained by an SEM campaign, a DEM or by mobile measurements across downwind plumes (DWPEM). These three approaches are tested and compared in this study. It should be noticed that instrumentation measurement errors are not treated in the current version of the method. However, this could be included in future improvements.

In all three tested surveying method, input data include methane concentration measurements and locations along with meteorological conditions (most important are wind speed and direction, insolation and temperature). Based on these data, the target is to identify the total emission flux of the studied landfill and to locate the major emission sources and their emission rates. This task is formulated as an optimization problem where the variables are the locations and the emission rates of the sources inside the landfill. The objective of this optimization task is to identify the configuration of emission sources (locations and leakage rates) that fits the best measured concentrations. The fitness of a defined configuration of emission sources is evaluated through calculating the corresponding methane concentrations at the same locations where measured concentrations are available. This is done using atmospheric dispersion model yielding a model-predicted value of methane concentration for each measurement location. Obviously, predicted concentration values are to be compared with the actual measured methane concentrations. Hence, the performance of a source configuration is evaluated through the difference between measured and predicted methane concentrations. The norm of absolute residuals calculated for all measurement points is the metric to be minimized in the optimization task.

Predicted methane concentrations are obtained using the Gaussian dispersion equation (1). This equation models the dispersion of a nonreactive gaseous pollutant from an elevated point source. Equation (1) predicts the steady state concentration (C) in μg/m³ at a point (x, y, z) located downwind from the source.

(1) $\begin{array}{rl}& \mathrm{C}\left(x,y,z\right)=\frac{\mathrm{Q}}{2\pi \mathrm{u}{\sigma }_{\mathrm{y}}{\sigma }_{\mathrm{z}}}exp\left(-\frac{1}{2}\frac{{\mathrm{y}}^2}{{\sigma }_{\mathrm{y}}^2}\right)\\ & \left\{exp\left(-\frac{1}{2}\frac{{\left(\mathrm{z}-\mathrm{H}\right)}^2}{{\sigma }_{\mathrm{z}}^2}\right)+exp\left(-\frac{1}{2}\frac{{\left(\mathrm{z}+\mathrm{H}\right)}^2}{{\sigma }_{\mathrm{z}}^2}\right)\right\}\end{array}$(1)

In Equation (1), Q is the emission rate (µg/s), σ_y and σ_z (m) are the horizontal and vertical spread parameters that are functions of the along wind distance x and the atmospheric stability which is a measure of the resistance of the atmosphere to vertical air motion. u is the average wind speed at stack height (m/s), y is the crosswind distance from source to receptor (m), z is the vertical distance above the ground (m), and H is the effective stack height (physical stack height plus plume rise expressed in m).

The Gaussian dispersion equation uses relatively simple calculations requiring only two dispersion parameters (σ_y and σ_z) to identify the variation of gas concentrations away from the diffusion source. These dispersion coefficients, σ_y and σ_z, are functions of wind speed, cloud cover, and surface heating by the sun. Generally, the evaluation of the diffusion coefficients is based on atmospheric stability classes. In this study, Pasquill-Gifford stability classes are employed, and dispersion coefficients are calculated using Briggs model (Pasquill and Smith 1983).

Starting from methane concentration measurements, inferring emission fluxes includes two main steps. First, measurement-data are used to estimate the number of potential emission sources. This is done by a peak-picking procedure that progressively locates and enumerates the positions associated with high measured concentrations. Concentration hotspots are likely to be located near emission sources and this helps estimating their number in the landfill. The estimated number of sources (and not their estimated locations) is the only information used in the subsequent step.

Having fixed the number of sources, the second step of the method is to identify the configuration of sources that fits the best the measurement data. To this end, several configurations of source positions and emission rates are generated and evaluated. For each configuration, source positions are generated inside the polygon that define the borders of the landfill. The corresponding emission fluxes are generated between 0 and an upper limit that is calculated based on the highest measured concentration. This upper limit is calculated as the highest emission flux for a unique source located inside the landfill that could be responsible for the highest measured concentration. Stochastic search is employed to explore the set of all possible configurations and to identify the source configuration that fits the best measurement data. The optimization task is tackled using Genetic Algorithms (GA): a global search method that belongs to the class of stochastic search algorithms. This method was originally proposed by John Holland, but its success owes much to the work of Goldberg (1989). Genetic Algorithms have been used efficiently since the early 1990s to solve hard optimization problems with both discrete and continuous variables.

As suggested by its name, a GA mimics the principles of natural evolution via biology-inspired mechanisms (selection, mutation and crossover). Natural evolution is simulated by encoding potential solutions (a population of solutions in genetic terminology) using a chromosome-like data structure and the search is guided by the results of evaluating the objective function for each solution in the population. New solutions are generated in the population through reproduction using crossover and mutation operators. Solutions that have higher fitness (i.e., represent better solutions) can be identified, and these are given more opportunity to breed. The genetic algorithm evolves, in successive generations, the composition of the population; enabling thus convergence toward near-optimal global solutions (Yang 2014).

As formulated in the optimization problem, a source configuration involves positions and emission rates of a definite number of source points. Even for a small landfill case, the number of possible source configurations can be very big. This is the main reason behind choosing the GA method since this it is generally capable of traversing large and complex search space to provide near-optimal solutions. The choice is also motivated by the outcomes of a comparative study involving three other well-established optimization methods (Kormi et al. 2017).

Lastly, to conclude the description of the optimization-based methodology, two technical details are worth mentioning. First, the GA optimization method is an iterative procedure that evolves a set of solutions. The algorithm converges and iterations terminate when no evolution in fitness is noticed for a certain number of iterations. Second, like all stochastic search techniques, the GA method does never guarantee to obtain the optimal solution for whatever an optimization problem. However, GA generally converges to near-optimal solutions that are, most of the time, satisfactory from an engineering point of view. Several runs (3 to 5 with different initial population for each run) should be performed in order to overcome the stochastic nature of the method. The parameters of the algorithm (selection, crossover and mutation mechanisms) are applied the same way for all runs. The best solution obtained over a sequence of multiple optimization runs could be chosen as the optimal configuration.

Hand-generated case study

For the first case study, a virtual landfill with hand-generated configurations of sources and receptors were analyzed. The objective is to present a proof-of-concept for the proposed emission estimation method. The parameters for this case study are inspired from actual landfill cases previously studied by the authors. However, they don’t correspond to any specific landfill case. Eighteen sources with predefined emission rates are simulated. Positions for sources are randomly distributed over a 360,000 m² landfill. Source emission rates are also randomly generated between 0 and 13 g/s. The assumed total emission flux for the virtual landfill approaches 138 g/s. Table 1 shows source positions and assumed emission rates (Q) for all sources.

Table 1. Hand-generated configuration of the 18 sources created to test the optimization-based methane emission estimation approach.

Source number	x position (m)	y position (m)	Emission rate Q (g/s)
1	387	785	9.2
2	485	870	7.6
3	210	510	5.8
4	585	801	11.3
5	563	594	8.4
6	308	412	5.9
7	407	383	12.7
8	478	534	4.6
9	450	782	6.7
10	392	305	8.3
11	650	700	8.2
12	315	581	10.9
13	420	200	6.2
14	150	417	8.7
15	422	608	4.5
16	567	303	6.5
17	340	300	4.7
18	487	409	7.8

Three sampling strategies are considered for the fugitive methane emitted by the landfill sources. The first one simulates a manual SEM; the second corresponds to an above ground drone emission monitoring (DEM) and the third configuration simulates mobile measurements of methane concentrations across downwind plumes (DWPEM).

The generated data is used to simulate a typical landfill case where the available information is the methane concentrations at defined locations in addition to the meteorological parameter values (wind speed and direction, insolation and temperature). These data are used as input for the GA-based emission estimation. Obviously, no information about source locations and emission fluxes are used. Instead, the synthetic data is used to estimate the total methane flux emitted through the landfill. Locations and emission rates of major hotspots in the landfill are also determined (for SEM and DEM) and compared with the predefined values.

Ground-based manual inspection is currently the most used method for SEM. In this approach, a properly equipped surveyor walks the entire landfill making concentration measurements at regular spacing. Methane concentration measurements should be georeferenced (through GPS coordinates). For the hand-generated case study, SEM is simulated assuming that methane concentrations are collected at 270 points. This is intended to simulate measurements taken at 35-meter spacing. To ensure coverage of the landfill site, the simulated measurement locations are randomly distributed over the landfill area (Figure 1). Based on the configuration of emission sources detailed in Table 1, methane concentrations at receptor positions are determined employing the Gaussian dispersion model. This is done assuming a wind speed of 2.68 m/s and wind direction of 22.5° (NNE). It is also supposed that the SEM is performed under moderate insolation resulting in a stability class “B”. The Briggs model is employed for determining horizontal and vertical dispersion spread parameters. It is assumed that the pseudo-measurements used here were obtained in a short time frame in order to justify the assumptions about constant wind speed and direction. This assumption holds also for the synthetic datasets obtained using DEM and DWPEM.

Figure 1. Source and receptor positions for the hand-generated SEM.

Above ground drone-based emission monitoring (DEM) is also simulated. It is supposed that a drone flies the landfill following a preplanned path. Equipped with methane detector and GPS unit, the drone captures methane concentrations and position information. Using such technology allows for a rapid and more comprehensive coverage of the landfill with larger number of measurement points than manual SEM. It should be noted that in real situation, the precision of the DEM measurements depends up on the type of the UAV platform (fixed wing or multi-rotor) and the instrumentations for measuring methane concentrations and wind. For interest, we refer the reader to Allen et al. (2019) for a discussion about unmanned aerial systems used for the measurement of methane flux from landfills.

For the DEM case, it is assumed that 1255 methane concentration measurements are obtained using the drone flying 5 meters above the landfill surface. This approximately corresponds to measurements at 15-meter spacing. Drone-based measurement points and source positions are displayed in Figure 2. Similar to the SEM case, methane concentrations are determined based on Gaussian dispersion model assuming a constant wind speed of 2.68 m/s, a wind direction of 22.5° (NNE) and a stability class “B”.

Figure 2. Source and receptor positions for the hand-generated drone emission monitoring (DEM).

A third synthetic dataset is used to simulate mobile measurements of methane concentrations across downwind plumes (DWPEM). For this, aligned 234 points are generated downwind and perpendicular to the methane plume originating from the landfill. The nearest point is located 1400 m far from the center of the landfill. As for SEM and DEM datasets, methane concentrations at the generated receptor points are determined using the Gaussian dispersion model and considering the same meteorological parameters. The locations and the model-based methane concentrations for this configuration are displayed in Figure 3.

Figure 3. Source and receptor positions for the hand-generated downwind plume monitoring (DWPEM). The figure shows the limits of the studied landfill and the model-predicted methane concentrations for the receptors.

The uncertainties related to meteorological conditions are of special importance in methane emission estimation using inverse modeling (Kormi et al. 2018). In this study, a special focus was made into the variability of the DEM-based emission estimates when the meteorological parameters are not correctly identified. First, the stability class is studied. GA-based methane estimation was performed using DEM datasets with varying stability classes (A, B, C and D). Second, the wind direction is varied. Four wind directions (NNW, N, NE and ENE) are considered and emission estimation results are compared with the reference case (NNE). At last, the wind speed is varied from 1 m/s to 5 m/s and emission estimation results are compared with the reference case (wind speed = 2.68 m/s). In order to showcase the effect of these parameters, in each case the studied parameter value is varied while keeping all other meteorological parameters constant.

The variability of the meteorological parameters, specifically wind speed, is further investigated. This is intended to simulate a variable wind speed occurring during the measurement campaign. This is very likely to happen in real landfill surveys especially for long sampling duration. In order to investigate such situations, the above ground drone collected measurements are used again. Rather than assuming a constant wind speed (2.68 m/s), methane concentrations are calculated using a variable wind speed. For each receptor point, a random wind speed ranging between 1.18 m/s and 4.18 m/s is considered. Note that the average wind speed value corresponds to the constant wind speed considered in reference case. The randomly generated wind speed is considered in the Gaussian dispersion model. All other required parameters are kept the same as in previous cases (Wind direction, insolation and dispersion parameter model).

Field study

The studied landfill is an active municipal waste disposal (located in Georgia, USA) with a total surface area of approximately 516,000 m² of even area, with sparse vegetation. A ground-based SEM, a drone-based air monitoring campaign (DEM) and a downwind plume monitoring campaign (DWPEM), were conducted at the landfill. The measurement campaign was performed in April 10 and 11, 2018. In day 1 (April 10, 2018), an SEM partially covering the landfill surface was performed. Methane concentrations were collected in 361 points. In day 2 (April 11, 2018), a second SEM was performed, and surface methane concentrations were collected at 275 points. This second SEM also partially covers the landfill surface. One part of the landfill site was surveyed on day 1 and the remaining part the following day. None of the two samplings covered the entire landfill area. However, the part that was not covered in the first day was covered the second day with some small overlap. The measurement points for the SEM are displayed in Figure 4 where different markers are used to separate the two sets of measurements (day 1 and day 2). It is important to notice that for all datasets (SEM, DEM and DWPEM) background concentrations are subtracted from measurement values prior to any treatment of the data. The level of background concentration is established experimentally before all samplings. The background concentration was determined as the average between upwind and downwind methane concentrations recorded outside of the landfill site. Furthermore, no cutoff value was considered in this study. All concentration measurements are included in the emission calculation even those associated with high concentration values.

Figure 4. Field study SEM measurements performed in the two-days sampling campaign with their corresponding methane concentrations in ppmv. Red circular markers indicate the position of the emission sources predicted using the GA-based method.

A portable Flame Ionization Detector (MicroFID from PhotoVac, inc.) was used to perform surface monitoring with data collected every 15 seconds. The microFID was calibrated at the beginning of each monitoring event, prior to use, in accordance with the regulations. Calibration is performed with zero air and 500 ppm methane gas. The precision of the methane analyzer was ±0.3 ppm (for 0.5 to 2000 ppm methane range) and ±3 ppm (for 10 to 50,000 ppm methane range) with an accuracy of ± 0.5 ppm. The response time is less than 3 seconds with a detection limit of about 0.5 ppm. In order to screen surface methane concentrations a funnel-shaped probe was directly put on the landfill surface and via an integrated pump the emitted gas was drawn through the MicroFID. A GPS unit was used to register measurement positions along the monitoring path. Furthermore, wind speed and direction were also collected for both days. An ultrasonic anemometer (Model 81000 V from Young, inc.) was placed at the highest elevation of the landfill. This station allows for wind speed measurements with a precision of 0.01 m/s and an accuracy of ±0.05 m/s. For wind direction measurements, the precision is 0.1 degrees with an accuracy is ±2 degrees. Wind speed and direction were measured twice: in the beginning and by the end of each survey for the two days. An average value is then considered for each set of measurements.

During day 1, a downwind plume monitoring was performed. Five traverses downwind perpendicular to the plume of methane originating from the landfill were performed by measuring the atmospheric methane concentrations more than 2 km far from the landfill. The mobile downwind plume measurements are done using a Cavity Ring-Down Spectrometer (CRDS) Picarro G2203 Analyzer for methane and acetylene. The CRDS measures methane and acetylene in ppb levels. The analyzer has an effective path length up to 20 km. The precision is 3 ppb for methane and less than 600 ppt for acetylene over a measurement interval of less than 2.0 seconds. The analyzer is capable of 1–5 seconds measurement time. The CRDS was mounted in an SUV fitted with an external snorkel intake for gas sample collection at an elevation of 2 m from ground surface. Methane concentration measurements and GPS positions are recorded in a time-synchronized data file. Wind speed and direction were measured twice: in the beginning and by the end of each traverse and average values are recorded. The weather station previously described was also used for these measurements.

Drone-based air monitoring (DEM) was performed on day 2. This allowed for a more comprehensive coverage of the landfill area with 1456 measurement points. All data was collected by a private company (BeamIO, Inc) using a proprietary gas sensor based on a “MQ-4” methane sensor. The employed sensor had a methane detection concentration ranging from of 10 to 10,000 ppm. Its sensitivity, expressed in terms of resistance ratio (Rs (in air)/Rs (5000 ppm CH₄)), is above 5. The sensor was attached to the bottom of a quad-copter drone. The drone was flown over the site, launching from the top surface of the landfill at various locations. The drone then autonomously followed a grid pattern at a fixed altitude of 6 m above each launch location. The measurement points for the DEM are displayed in Figure 5. The methane sensor used is a semiconductor type with average characteristics. Semiconductor sensors are also known for their sensitivity to the ambient temperature changing the sensor thermal conductivity. The quality of the methane sensor is considered when the DEM measurements are treated. Better sensing options exist (optical and laser technologies) for drone-based surveys of landfill methane (Allen et al. 2019; Emran, Tannant, and Najjaran 2017).

Figure 5. Field study DEM measurements performed in day 1 of the sampling campaign with their corresponding methane concentrations in ppmv. Red circular markers indicate the position of the emission sources predicted using the GA-based method.

The measurement datasets obtained using the three sampling techniques were used to determine independent estimates of total methane emissions from the landfill. The general data and climate conditions considered for the field campaigns are displayed in Table 2.

Table 2. Meteorological parameter values for the field campaign.

	Day 1 (April 10, 2018)	Day 2 (April 11, 2018)
Measurements	SEM + DWPME	SEM + DEM
Sky	Cloudy	Sunny
Temperature (°C)	20.0	20.5
Wind direction	NNE	WSW
Wind speed (m/s)	4.0	0.9
Stability class	C	A

Results

Results of the hand-generated case study

Three datasets of synthetic methane concentrations (SEM, DEM and DWPEM) are treated in this case-study. Receptor positions and the corresponding methane concentrations are the input data employed to estimate number, positions and emission rates of the virtual landfill emission sources. The whole-site landfill emission flux is calculated as the sum of the estimated emission rates for all identified sources.

The SEM-based concentration measurements are treated first. Using the previously described peak-picking procedure, the number of sources (18) was correctly determined. In order to evaluate the influence of the estimation of the number of sources on optimization identification efficiency, the identification procedure is performed assuming a number of sources equal to 17, 18 and 19. Fluxes obtained with the different number of sources are displayed in Table 3. The predicted total emission flux ranges between 138 g/s and 145 g/s which corresponds to an estimation error less than 5%. The estimation error is calculated based on the predefined emission flux in the landfill (138 g/s). These results show that the proposed method allows for determining methane emission flux even if the number of sources is not accurately estimated.

Table 3. Whole-site methane emission predictions for varying number of estimated sources using SEM and DEM for the hand-generated case-study. The estimation error is displayed for each case.

	Number of estimated sources
	17	18	19
SEM-based methane emission [g/s] (error)	141 (2.2%)	145 (5%)	138 (0%)
DEM-based methane emission [g/s] (error)	139 (0.7%)	142 (2.9%)	143 (3.6%)

The Fugitive methane assessment method is also used to treat DEM-based concentration measurements. As for the SEM dataset, the number of emission sources is correctly identified. For the whole-site methane emission in the landfill, optimization results are displayed in Table 3. The total emission flux in the landfill ranges between 139 g/s and 143 g/s which corresponds to an estimation error that ranges between 0.7 and 3.6%. Again, the obtained results show that the proposed procedure allows for determining the methane emission flux. This is achieved even if the number of sources is not accurately estimated.

Optimization results obtained with 18 sources as an estimated number of sources are displayed in Figures 6 and 7. Figure 6 shows positions of predicted sources based on SEM and DEM measurements compared to predefined source positions. Notice that using SEM dataset allowed for the localization of only 17 sources since two sources are overlapped (two predicted sources at the location around the point (580, 600) in Figure 3). It is also noticed that most source positions obtained using ground-based SEM are comparable to those obtained with the DEM dataset. However, source 9 was not predicted neither using SEM nor using DEM. In addition, one DEM-based predicted source does not match any assumed one.

Figure 6. Predicted source positions using the DEM dataset compared with the predefined source locations in the hand-generated case study. Results obtained using the SEM dataset are also displayed.

Figure 7. Predicted source emission rates using the SEM and the DEM datasets compared with the predefined emission rates in the hand-generated case study.

Figure 7 shows assumed and predicted emission rates for all sources. Using SEM dataset, there is no predicted emission rate for the source number 8 since this is the unique unidentified source.

Using DEM dataset, emission rates for most sources are well identified. However, the method fails to identify two emission sources (sources 9 and 11). Furthermore, compared to results obtained using ground-based SEM, it is noticed that a better match between assumed and predicted emission rates is obtained using above ground drone measurements (DEM).

The synthetic DWPEM dataset described in the previous section is also used to infer an estimate of the whole-site methane emission. In contrast with the cases of SEM and DEM, the GA-based method is used assuming that a unique emission source is responsible for the landfill emitted methane. Based on the DWPEM dataset, the optimization-based method yields an estimated emission rate of 113.6 g/s for the landfill. This underestimates the total methane flux in the landfill (138 g/s). Compared with SEM and DEM, the DWPEM yields less accurate results for the studied case.

The results of the parametric study used to test the sensitivity of the methane emission estimation to some input parameters are displayed in Table 4. For the stability class, methane emission results show that if a stability class “C” is considered, emission estimate is comparable with the reference value (obtained with a stability class “B”). However bigger discrepancies are noticed with stability class “A” and “D”. The deviation from the reference value ranges between 5% and 40%.

Table 4. Emission estimates and prediction errors using DEM dataset with variable stability class, wind direction and speed. The reference values are indicated in bold for each case.

	Parameter value and resulting emission rate in the landfill
Stability Class	A	B	C	D	-
Emission rate (g/s)	199.8	142	134.9	105.6
Estimation error	(44.8%)	(2.9%)	(2.2%)	(23.5%)
Wind direction	NNW	N	NNE	NE	ENE
Emission rate (g/s)	131.5	123.4	142	131.4	130.8
Estimation error	(4.7%)	(10.6%)	(2.9%)	(4.8%)	(5.2%)
Wind speed (m/s)	1	2	2.68	3	4
Emission rate (g/s)	87.7	140.2	142	163.7	199.5
Estimation error	(36.4%)	(1.6%)	(2.9%)	(18.6%)	(44.6)

As for the wind direction, comparable estimates for the whole-site landfill methane emission are obtained when the wind direction is varied from 337.5° (NNW) to 67.5° (ENE). The emission estimation results deviate by a maximum 13% when the wind direction is varied in the studied range. In all cases methane emission rates are underestimated when compared with the reference value (obtained for the wind direction NNE). Bigger deviations are noticed when the wind speed is varied. As suggested by the results of Table 4, a wrong appreciation of the wind speed can affect emission estimation results. For instance, the deviation from the reference value approaches 40% if the wind speed is equal to 4 m/s instead of 2.268 m/s.

The GA-based emission estimation method is also used to treat DEM monitoring data that are perturbed with a variable wind speed. Results of this case are displayed in Table 5. Predicted landfill emissions are determined assuming min, max and average values for the wind speed. It is noticed that the GA-based identification method can estimate the methane emission flux of the landfill when only average wind speed is used for calculation. Assuming an averaged wind speed during the monitoring campaign yields an emission estimation error less than 9%. In extreme cases where the input wind speed ranges between the minimum and the maximum actual wind speed, the methane emission estimation error ranges between 18 and 53%. A theoretically perfect scheme is to assign a wind speed value (and even a wind direction value) to each measurement point during an SEM or a DEM campaign. However, this appears to be difficult in practice. Results obtained in this case reveal that even an average wind speed calculated over the duration of the measurement campaign could be used to yield estimates of methane emissions.

Table 5. Whole site emission predictions using DEM with variable wind speed ranging between 1.18 m/s and 4.18 m/s. Emission estimates are displayed for the minimum, the average and the maximum wind speed values.

	Assumed wind speed (m/s)
	1.18	2.68	4.18
Predicted landfill emission [g/s]	113	149	211

Results of the field study

The GA-based methane emission estimation is employed using the three measurement datasets (SEM, DEM and DWPEM). The objective is to identify the methane emission sources for the studied landfill. Table 6 shows the estimated methane emission rates obtained using SEM and DEM collected data.

Table 6. SEM and DEM based estimated methane emission sources for the field study.

Parameter	SEM-based results	DEM-based results
Mean (g/s)	17	14
Minimum (g/s)	4.84	6.15
Maximum (g/s)	20.0	20.0
Total Flux (g/s)	350	287
Count/Number of Sources	21	20

The SEM measurements from the two days of the campaign are combined in a unique dataset. Using this combined dataset, 21 emission sources were found to be responsible for most of the methane emissions from the landfill. Emission rates of the 21 sources varied from 4.8 to 20 g/s. The mean methane emission rate from the 21 sources was 17 g/s. The total emission flux from the 21 sources is estimated to be 350 g/s. The locations of the 21 predicted emission sources are displayed in Figure 4. The same approach was used using the above ground drone collected data. The analysis of the DEM dataset yielded only 20 emission sources. The emission rates from the 20 sources varied from 6.1 to 20 g/s and averaged 14.3 g/s (Table 6). The total methane emission flux from the landfill was estimated to be 287 g/s as compared to 350 g/s obtained by the SEM method. The locations of the 20 predicted emission sources using DEM are displayed in Figure 5.

Comparing the results displayed in Figures 4 and 5, no clear correlation is noticed between the results for source locations. This holds also for the concentration measurements from SEM and DEM. Hotspot locations associated with high methane concentrations obtained with the SEM do not correspond to those obtained with the drone measurements. This could partially explain the discrepancy between the results of the SEM and the DEM. Not less important are the conditions of the surveys. The SEM campaign was performed over two days with different weather conditions. This could also explain the discrepancy in the results. Another concern is related to the sensor used for the drone-based survey. The methane sensor on the drone had smaller accuracy compared with the one used in the SEM campaign. Inevitably, less accurate methane concentration measurements lead to bigger inaccuracies in flux estimates and source positions. For all these reasons, the authors are more confident about the SEM based flux as the methodology was developed for that specific purpose. More work is needed, and better drone-specific measurement devices should be used to retool the methodology for drone-based measurements.

Finally, methane plumes downwind from the landfill monitored during the first day of the campaign are exploited using the GA-based emission estimation method. The winds during the second day were too weak to take advantage of the DWPEM, as the plume seems to be staying on top of the landfill. Total methane emissions were estimated independently for five different plumes. In this case, the GA-based method is employed assuming that fugitive methane originates from a unique source located at the center of the landfill. This assumption is justified by the fact that plume measurements are taken more than 2 km downwind of the landfill. As such, only the overall landfill methane emission flux is targeted in this case. For the five plumes measured during the DWPEM, the total estimated emissions from the landfill are displayed in Table 7.

Table 7. Total estimated emissions based on DWPEM for the field study. The wind direction for each plume is indicated. Results of the optimization-based method are compared with the results obtained using the method proposed by (Fredenslund et al. 2018).

	Estimated methane flux (g/s)
Plume and wind direction	Optimization-based estimation	Inverse plume modeling (Fredenslund et al. 2018)
Plume 1- (36.7°)	264	201
Plume 2- (38.7°)	214	240
Plume 3- (24.6°)	216	262
Plume 4- (44.1°)	190	176
Plume 5- (47.4°)	255	253
Average value	228	226

The average estimated emission was 228 g/s as compared to 350 g/s obtained by the SEM method. Results are also compared with the outcomes of an another inverse plume modeling method proposed by (Fredenslund et al. 2018). Rather than using the whole plume measurements, Fredenslund et al. (2018) used the peak downwind measured concentration to recover the emission flux assuming unique emission source in the landfill. The peak methane concentration is determined employing a Gaussian fit applied to the measured methane plume transect.

Based on the estimated methane flux for the 5 plumes displayed in Table 7, it is noticed that the discrepancy between the outcomes of the two employed methods is below 25% with an average value of 13%. The landfill total methane emission flux averaged over the five plumes shows that the optimization-based estimation method and the inverse plume modeling gave close results with a methane emission ranging between 226 and 228 g/s. These results are not surprising since, in this case, the two approaches are basically very close. It should be noticed that during plume measurements, variations in wind direction was recorded. Two wind measurements were done before and immediately after each plume traverse. The wind direction values in Table 7 are the average wind directions that are calculated based on the two measurements for each plume. Taking into consideration that plume measurements are taken more than 2200 m far from the landfill center, any deviation in wind direction could result in a bigger deviation in flux estimates. This point must be taken into consideration when treating downwind plume measurements.

Discussion

The Gaussian dispersion model appears as a useful tool for investigation of fugitive gas emissions. However, it is associated with some simplifying assumptions that could lead to uncertainties when the model is used to infer landfill methane emissions. First, emitted plumes typically display a Gaussian morphology only after a certain distance from the emission source. This could affect some measurements since the plumes near to sources are generally turbulent. Second, the Gaussian model does not consider the topography of the site. The model may not accurately reflect the disturbed gas dispersion resulting from rising or falling topography which can perturb the dispersion profile across the landfill sites. Furthermore, emission short-term fluctuations are ignored in Gaussian dispersion models. Instead, a constant emission rate is considered which is an approximation that holds only for small time period. Finally, a uniform wind field is assumed when using Gaussian models which could be associated with a high level of uncertainty related to wind speed and direction during measurement campaigns.

The hand-generated study revealed that inverse plume modeling combined with optimization showed promising results and could potentially be applied to landfills while taking into account the uncertainty raised by their topography. In fact, major sources of uncertainties were not considered for this case. Neither the uncertainties related to the dispersion model nor those related to the variability of the weather conditions are taken into account. Another source of uncertainties resides in the measurement instrumentation and their limitations. What the hand-generated case study simply reveals is that starting from measurement data obtained in stable weather conditions with efficient and accurate instruments, it is possible to have an estimate of the total emission flux in a landfill. The distribution of the major emission sources in the landfill can also be approached. Under these conditions, SEM and DEM may yield close results. The DEM could be more time-efficient than SEM which tends to minimize the variability of the weather parameters. The DWPEM yields underestimated results mainly because it adds another simplifying assumption considering the whole landfill site as a unique emission source.

The parametric study performed with the DEM dataset in the hand-generated case study revealed the relative importance of the weather conditions (stability class, wind speed and direction). Results for the studied case showed that the stability class and the wind speed impacted the emission estimation more importantly than the wind direction. These sensitivity conclusions are mainly true for the DEM datasets. The results obtained using the DWPEM dataset showed that the wind direction would probably have a bigger impact on flux estimates.

From another perspective, the emission estimation was performed with synthetic dataset simulating a variable wind affecting drone-collected concentrations. This corresponds to wind speed varying during sampling. Results obtained in this case revealed that an average wind speed calculated over the sampling duration could be used to yield estimates of methane emissions. The hand-generated case study could be improved in future investigations. One could simulate uneven distributions of emission hotspots and rates. Uncertainties of the dispersion model and the weather parameters could be deeply addressed.

The results obtained in the field case study emphasize the importance of having a coherent sampling strategy. The fact that the SEM was constituted from measurements taken in two days with different weather conditions in each day negatively affected the emission estimation results. Especially with measurements that didn’t cover the entire landfill each day. If the measurements covered the all site on both datasets, this would help the method to triangulate the location of the sources yielding a more accurate estimation of the methane emission.

The use of UAV could add to a toolkit of approaches for sampling and quantifying fugitive methane emissions from landfills. However, this is highly dependent on the availability of high precision methane sensors that are adapted to be used onboard. This is one the major challenging problem facing this approach. Onboard sensors should also be compact and lightweight. Instrumentations for wind measurement also could be adapted for onboard use allowing thus a synchronized recording of gas concentrations and wind measurements. Some other aspects of the drone-based methodology are important and could be part of future investigations. This includes the drone technology to be adopted in such applications (fixed wind or rotary systems), measuring heights and coverage needed for a better level of estimation accuracy.

As a simple and rapid screening method, DWPEM could be useful to obtain estimates of the methane emissions from landfills. The present study showed that DWPEM is less accurate than SEM and DEM. Especially, because of the simplifying assumption about emitting sources and the sensitivity to weather conditions (mostly the wind direction). However, this method could be more an estimation method than a quantification one.

Conclusion

An optimization-based methane emission quantification method was combined with three different measurement approaches for the assessment of fugitive landfill methane emissions. The method uses inverse plume modeling and stochastic search in order to infer methane emissions (total flux, emission rates and locations of major hotspots). In this paper, the method was combined with surface ambient methane monitoring (SEM), above ground drone emission monitoring (DEM), and downwind plume monitoring (DWPEM).

Results obtained from both hand-generated and actual landfill case-studies showed that the proposed method is promising when combined with above ground drone-based collected measurements. Compared with SEM, employing newer methodologies offered through Unmanned Aerial Vehicle (UAV) allows for a time-effective way for methane emission measuring and reporting activities. However, the success of this approach is dependent on the availability of high precision methane sensors that are adapted to be used onboard.

Results from an actual landfill case-study showed that the optimization-based methane emission estimation method is also useful when downwind plume monitoring is employed. The DWPEM precision level is lower when it is compared with SEM and DEM. However, it is the authors belief that the proposed method allows for a rapid and simple estimation of landfill methane emissions.

When appropriate sampling techniques are used, the optimization-based method trialed in this study for fugitive landfill methane emission could also be used for rapid screening of analogous greenhouse gas emission hotspots.

Disclosure statement

No potential conflict of interest was reported by the authors.