A review of close-range and screening technologies for mitigating fugitive methane emissions in upstream oil and gas

Fugitive methane emissions from the oil and gas industry are targeted using leak detection and repair (LDAR) programs. Until recently, only a limited number of measurement standards have been permitted by most regulators, with emphasis on close-range methods (e.g. Method-21, optical gas imaging). Although close-range methods are essential for source identification, they can be labor-intensive. To improve LDAR efficiency, there has been a policy shift in Canada and the United States towards incorporating alternative technologies. However, the suitability of these technologies for LDAR remains unclear. In this paper, we systematically review and compare six technology classes for use in LDAR: handheld instruments, fixed sensors, mobile ground labs (MGLs), unmanned aerial vehicles (UAVs), aircraft, and satellites. These technologies encompass broad spatial and temporal scales of measurement. Minimum detection limits for technology classes range from <1 g h−1 for Method 21 instruments to 7.1 × 106 g h−1 for the GOSAT satellite, and uncertainties are poorly constrained. To leverage the diverse capabilities of these technologies, we introduce a hybrid screening-confirmation approach to LDAR called a comprehensive monitoring program. Here, a screening technology is used to rapidly tag high-emitting sites to direct close-range source identification. Currently, fixed sensors, MGLs, UAVs, and aircraft could be used as screening technologies, but their performances must be evaluated under a range of environmental and operational conditions to better constrain detection effectiveness. Methane-sensing satellites are improving rapidly and may soon be ready for facility-scale screening. We conclude with a speculative discussion of the future of LDAR, touching on integration, analytics, incentivization, and regulatory pathways.

There are two errors in this article which are a result of a misinterpretation of the author's original corrections.
In the following two sentences, the words in bold should have been removed: 1. 'In recent decades, there have been unnecessary developments in sensors, platforms, and analytics that could serve a screening role, including mobile ground labs (MGLs), fixed sensors, aircraft, unmanned aerial vehicles (UAVs), and satellites.' This should read: 'In recent decades, there have been developments in sensors, platforms, and analytics that could serve a screening role, including mobile ground labs (MGLs), fixed sensors, aircraft, unmanned aerial vehicles (UAVs), and satellites.' 2. 'Few fixed sensors have been independently evaluated for ambiguous and unnecessary leak detection in upstream O&G, and technical capabilities and limitations of this technology class are poorly constrained (table 1).' This should read: 'Few fixed sensors have been independently evaluated for leak detection in upstream O&G, and technical capabilities and limitations of this technology class are poorly constrained (table 1).'

8385-6573
Abstract Fugitive methane emissions from the oil and gas industry are targeted using leak detection and repair (LDAR) programs. Until recently, only a limited number of measurement standards have been permitted by most regulators, with emphasis on close-range methods (e.g. Method-21, optical gas imaging). Although close-range methods are essential for source identification, they can be labor-intensive. To improve LDAR efficiency, there has been a policy shift in Canada and the United States towards incorporating alternative technologies. However, the suitability of these technologies for LDAR remains unclear. In this paper, we systematically review and compare six technology classes for use in LDAR: handheld instruments, fixed sensors, mobile ground labs (MGLs), unmanned aerial vehicles (UAVs), aircraft, and satellites. These technologies encompass broad spatial and temporal scales of measurement. Minimum detection limits for technology classes range from<1 g h −1 for Method 21 instruments to 7.1×10 6 g h −1 for the GOSAT satellite, and uncertainties are poorly constrained. To leverage the diverse capabilities of these technologies, we introduce a hybrid screening-confirmation approach to LDAR called a comprehensive monitoring program. Here, a screening technology is used to rapidly tag high-emitting sites to direct close-range source identification. Currently, fixed sensors, MGLs, UAVs, and aircraft could be used as screening technologies, but their performances must be evaluated under a range of environmental and operational conditions to better constrain detection effectiveness. Methane-sensing satellites are improving rapidly and may soon be ready for facility-scale screening. We conclude with a speculative discussion of the future of LDAR, touching on integration, analytics, incentivization, and regulatory pathways.

Introduction
Methane, a potent greenhouse gas (GHG), has become a major focus of GHG reduction initiatives. Of the many sources of natural and anthropogenic methane, emissions from the oil and gas (O&G) industry have Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
received special attention (Moore et al 2014). In Canada, commitments have been made to cut methane emissions from the O&G sector by 40%-45% below 2012 levels by 2025 (Government of Canada 2017). In addition to meeting GHG targets, reducing methane emissions from O&G can save money (ICF International 2014 and improve local air quality (Roy et al 2014). Natural gas (NG) consists primarily of methane and is invisible and odorless in most upstream settings. Within regulatory and operational contexts, releases of NG to the atmosphere are often classified as either vented or fugitive emissions. Vented emissions are intentional releases of hydrocarbons, typically in a controlled manner, resulting from normal process conditions. In contrast, fugitive emissions (also called 'leaks') are unintentional releases of hydrocarbons from sources that should not be emitting (e.g. To mitigate fugitive emissions, operators commonly implement leak detection and repair (LDAR) programs. Method 21 (Determination of Volatile Organic Compound Leaks), introduced in 1983 by the US Environmental Protection Agency (EPA), was the first regulatory framework for conducting LDAR. In 2008, the EPA expanded the scope of possible compliance procedures by introducing the alternative work practice (AWP). AWP replaces concentration detectors with optical gas imaging (OGI) camerashandheld instruments that generate infrared images of methane plumes. Today, OGIs are preferred over Method 21 by regulators and operators in Canada and the US, and most LDAR programs rely exclusively on handheld methods, which can be labor-intensive. Both Method 21 and AWP require facility access, and LDAR must be applied to millions of components distributed over extensive spatial scales. For example, figure 1 illustrates the variable density of O&G infrastructure across the Canadian provinces of Alberta and British Columbia, with densities often below two wells per square kilometer. At each facility, LDAR technicians must manually survey hundreds or thousands of individual components.
Given the spatial extent and variable density of O&G infrastructure, deploying rapid screening technologies may improve LDAR efficiency. Screening technologies are systems that can quickly flag abnormally emitting facilities for directed follow-up with close-range methods. In recent decades, there have been unnecessary developments in sensors, platforms, and analytics that could serve a screening role, including mobile ground labs (MGLs), fixed sensors, aircraft, unmanned aerial vehicles (UAVs), and satellites. Typically, screening technologies can neither identify leaks at the component-level, nor distinguish vented from fugitive emissions. To diagnose and repair leaks, screening methods must be paired with close-range methods such as Method 21 and AWP, which can precisely pinpoint leaking components. We refer to a program that combines screening and close-range methods as a comprehensive monitoring program (CMP).
Regulators are moving towards policy frameworks that enable CMPs. In April 2018, Environment and Climate Change Canada finalized new regulations allowing operators to choose between prescriptive 'Regulatory' LDAR and 'Alternative' LDAR (Government of Canada 2017). In Regulatory LDAR, operators must use Method 21 or OGIs to monitor for fugitive emissions at least three times per year at select facilities. In Alternative LDAR, operators are invited to develop LDAR programs incorporating new methods and technologies. To be approved, Alternative programs must demonstrate emissions reductions equivalent to Regulatory LDAR. In Alberta, regulations released in December 2018 mandate screening for fugitive emissions using one of several methods, including UAVs and truck-mounted sensors (i.e. MGLs). In the US, measures for the approval of alternative LDAR have been implemented in 40 CFR Part 60 Subpart OOOOa (EPA 2016). Similar opportunities are outlined in section XII.L.8 of Colorado's Regulation 7 (CDPHE 2018).
Screening technologies are under various stages of development and commercialization and there are gaps in the information available to guide deployment and regulation. As a growing number of technology developers promise solutions, regulators and operators must determine how (and whether) to integrate these technologies into current LDAR programs. In this paper, we systematically evaluate and compare leak detection technologies, and discuss ways in which LDAR programs might integrate these technologies to mitigate emissions. This article is presented in 6 sections. In section 2 we establish the methodological framework for the review and discuss measurement techniques and quantification. Section 3 is an overview of methane measurement principles and the platforms used for monitoring methane in O&G. In section 4 we review current and emerging methanesensing technologies in the context of LDAR. Section 5 is a speculative discussion of the future of LDAR, touching on technology evaluation, multiscale integration, and regulatory models. Section 6 is a brief conclusion.

Review methods and framework
Our review is focused on six broad technology classes: handheld instruments, fixed sensors, MGLs, UAVs, aircraft, and satellites. We searched Google Scholar and the University of Calgary library database using the keywords: LDAR, fugitive methane emissions, and methane sensing O&G. Relevant articles were then used to identify further sources by consulting both 'works cited' and 'cited by' lists. This process was repeated until no new sources could be identified. Non-peer-reviewed sources were sometimes used, such as government reports or independent research publications. Sources published after 25 October 2018 may not be included.
Although this review is focused primarily on LDAR and screening for anomalous emissions, it is useful to consider whether candidate technologies are suitable for different monitoring programs. To frame this review, we consider four distinct motivations for measuring NG emissions from upstream O&G: • M1: Develop and refine emissions factors to improve inventories, • M2: Estimate top-down emissions from a region with multiple sources, • M3: Conventional, close-range LDAR using handheld instruments, and • M4: Rapid screening for anomalous emissions.
For each goal, equipment can be targeted at a granular scale (M1 and M3), or at an aggregate scale (M2 and M4). Different technologies and methods are required for each motivation, and different data products can be expected. For example, developing emissions factors requires accurate quantification, often at the component-level. In contrast, estimating top-down emissions requires mobile (often airborne) platforms capable of resolving small concentration enhancements dozens of kilometers downwind of a source region. Close-range methods often favor realtime imaging and generally do not require quantification. Screening should be less expensive than closerange methods, but deployment strategies can differ. First, screening methods can inform directed application of follow-up surveys. For example, a close-range survey of a facility could be skipped if there are no anomalous emissions identified through screening. Second, among detected emissions sources, screening methods can help triage follow-up and repair based on a size-ordered list of flagged facilities, reducing aggregate emissions as the largest leaks are repaired first. Third, screening methods can focus on super-emitter targeting. Given skewed leak-size distributions, early identification of super-emitters could mitigate a majority of emissions. In super-emitter targeting, screening methods should have low per-site cost, high spatial coverage, and frequent sampling. If a field contains few super-emitters (i.e. a less-skewed leak size distribution), implementing super-emitter targeting may be less effective.
We evaluate technologies across three product levels: (1) detection, (2) localization and/or attribution, and (3) quantification. For close-range methods, detection and localization are often accomplished simultaneously, and quantification is generally less important. For screening, quantification is often necessary to determine whether a follow-up survey should be conducted using close-range methods. For technologies with high detection limits, quantification could be less important, as each detection event could trigger a follow-up survey. If multiple detection events occur during screening, relative quantification can enable triaging. Quantification may also permit the separation of vented from fugitive emissions, but only where vented emissions are precisely known.
For LDAR in general, quantification may be important depending on the goals of the program. If mitigating fugitive emissions is the primary goal, quantification may be less important than detection, as quantification generally takes more time and money that could instead be invested in more frequent detection-only surveys. However, if the goal is to conduct LDAR while developing an improved scientific understanding of the root causes of emissions sources, to reduce uncertainty in inventories, or to track progress in emissions reduction initiatives, quantification and data management become increasingly important. Quantification may further help by improving accountability and trust among industry, government, and the public.

Technology overview
3.1. Measuring methane Today, most methane concentration measurements are made with optical instruments, using either laser spectroscopy or imaging spectrometry. Laser spectroscopy determines the concentration of target molecules by measuring characteristic absorption of a mid-or near-infrared laser along a path length of meters to kilometers. The laser path may be 'open,' where it goes through the immediate atmosphere, or 'closed,' using a mirrored cavity into which gas is pumped. Unlike laser-based instruments, imaging spectrometers measure spectral densities using pixelbased sensor elements. For methane, abundance can be inferred using specific infrared absorption bands. Imaging spectrometers generate a multi-pixel field of view measurement that captures column-integrated concentrations. Other sensor classes exist, such as ionization devices and differential absorption light detection and ranging (DIAL). An understanding of sensor differences is useful for comparing technologies, but a detailed description of gas-sensing principles is beyond the scope of this article.
Once a screening technology acquires concentration data, atmospheric dispersion models can help determine source location and emission rate. These vary in complexity from simple Gaussian dispersion models to complex particle-tracing Lagrangian models that account for turbulence. Results using both simple and complex dispersion models have shown good equivalency in methane-specific studies (Brantley et al 2014, Foster-Wittig 2015. Dispersion modeling has been used for O&G monitoring of pollutants for decades, and most regulators publish guidance documents mandating what models to apply and how. However, most established techniques were designed and validated for stationary sensing, and it remains unclear how transferrable they are to mobile platforms. Recent studies comparing different quantification methods suggest that more work is needed to constrain quantification uncertainties (Bell et al 2017, Caulton et al 2018).

Technology classes
The technologies reviewed in the following section encompass broad spatial and temporal scales of measurement (figure 2). In figure 2, the spatial scale refers to the order of magnitude length-scale typically covered during a single measurement campaign, while the temporal scale refers to measurement times over which a single survey is completed once the technology is deployed. For example, satellite-based monitoring systems can measure across the planet, quasi-continuously, for many years (large spatial and temporal scales). Conversely, close-range instruments such as OGIs may only cover a few kilometers, over the course of a few hours to a few days (small spatial and temporal scales). Fixed sensors tend to monitor at large temporal and small spatial scales. As technologies evolve, the size and position of the ellipses in figure 2 may change. Currently, the dotted horizontal line at y='day' divides semi-automated (above) and laborbased systems (below). As a hypothetical example, if MGLs become driverless their range could expand into broader spatial and temporal scales. Similarly, the UAV niche could expand if battery limitations are overcome (i.e. with solar power) and airspace regulations relax to permit UAV operations beyond visual line of sight. In the future, a satellite cluster with higher spatial resolution and revisit time of less than a day could move satellites towards continuous monitoring.
A summary of the technology metrics used to structure this review is presented in table 1. Certain themes were omitted from the table if limited information was available or if the metrics were only applicable to specific technologies. These themes are discussed in the text and include operational conditions (e.g. susceptibility to adverse weather), technology readiness levels for the four monitoring motivations introduced in section 2, and future potential. We note that the quantitative comparison of technology classes presented in table 1 must be interpreted with caution for three reasons. First, most sensors and platforms have not been sufficiently evaluated under a range of environmental and operational conditions, and only a handful of studies have reported detection limits and uncertainties for each technology class. Published measurement ranges may not be representative, and context-dependent detection probability distributions should be developed before different technologies can be reliably compared. Second, sensors, platforms, and analytics are all under active development, with evolving technical capabilities and modes of deployment. Third, incompatible sensing principles (e.g. point concentration versus integrated path-length measurements) and modeling approaches make quantitative comparisons difficult. Developing technology equivalence protocols would enable a systematic comparison of fundamentally different technologies; such an effort is beyond the scope of this article.
An important metric that was not quantitatively included in this study is cost. The costs of an LDAR program include equipment purchase or rental, labor, repair, and additional considerations such as insurance, training of personnel, and equipment maintenance. Program costs can be compared on a per facility basis, or cost per unit of mitigated emissions. Unfortunately, too few studies have reported deployment costs to enable a quantitative comparison of technology classes. Although estimates have been developed for OGIs (e.g. ICF International 2014, 2015, Saunier et al 2014; various regulatory estimates) and aircraft (Schwietzke et al 2018), more data are needed for these and other technologies, and a comprehensive economic analysis is beyond the scope of this review.

Technology review 4.1. Handheld instruments
The most common sensors used for Method 21 are flame and photoionization detectors, although catalytic oxidation sensors, and infrared absorption-based sensors, are also used (Envirotech Engineering 2007, Szulczynski and Gebicki 2017). Although detection limits tend to be low, Method 21 instruments are labor-intensive as the sensing probe should be in immediate proximity to the leaking component for detection (table 1). Method 21 is still favored by some operators but use is declining as OGIs are more  convenient. Furthermore, flame ionization detectors are not intrinsically safe for use in O&G. In recent years, OGIs have become the standard for LDAR because they generate easily communicable and intuitive results for reporting purposes (figure 3(a)), and are more efficient than Method 21 (table 1), as they survey components remotely (Benson et al 2006). Thus, OGIs are capable of limited screening, which is restricted by imaging distance to small spatial scales (Ravikumar et al 2018). Despite their simplicity and widespread use, OGIs have some limitations. First, their operation is labor-intensive, as technicians should be within a few meters of potential sources. Recent work by  suggests that camera-to-source distance is the most important factor in predicting OGI detection effectiveness, and  show that imaging distances in excess of~10 m suffer from significantly reduced performance. Surveying thousands of wells, from within 10 m, several times per year, could result in considerable operational cost (ICF International 2014. OGI performance is also affected by adverse environmental conditions such as high wind speeds (Footer 2015), low ambient air temperatures, and low background emissivity contrast . While environmental constraints might be less of a concern in warmer climates, the majority of O&G infrastructure at high latitudes (e.g. Canada, Norway, Siberia, etc) experiences months of below freezing temperatures, often accompanied by high wind speeds. Operator expertise also plays a role in detection effectiveness (von Footer 2015, Fischer et al 2016), meaning that appropriate training, compliance auditing, and incentivization structures are needed. Finally, most current OGIs only present a qualitative (visual) flux estimate. Recent software products claim to estimate flux rate from OGI videos of leaks, though their effectiveness is yet to be established in the peerreviewed literature.
Method 21 instruments and OGIs are not currently able to contribute to motivations M1 and M2, can provide limited support for M4, and are vital for M3. Despite their limitations, OGIs are the most widely-used technology for LDAR, and are unlikely to be replaced by screening technologies because component-level attribution is critical for repair and reporting. OGIs are well-suited for detecting large plumes under favorable imaging conditions. Current North American policies support the use of OGIs, so it is reasonable to expect that competition in a growing market will drive sensors to be (1) less expensive, (2) more reliable under unfavorable measurement conditions, and (3) better able to quantify flux. New technologies, like hyperspectral cameras (Gålfalk et al 2016) and laser-based handheld spectrometers (Wainner et al 2017), may one day offer improved detection and quantification at lower cost. Fixed sensors could potentially contribute to M1 and M2 and already contribute to M4. Continuous monitoring and potential for automation make fixed sensors appealing, especially in dense infrastructure. As a screening tool, a distributed sensor network could identify fugitive emissions nearly instantaneously, preventing extended emissions events that remain undetected between mobile screening and conventional LDAR visits. As the only non-mobile technology class, fixed sensors might be best suited for facilities with high component density (e.g. gas plants, compressor stations, multi-well pads). For sparsely distributed upstream O&G infrastructure, these sensors must either be affordably mass-produced, or able to monitor large areas. Some progress has been made in developing low-cost sensors (e.g. Patel 2017). The ARPA-E MONITOR program funded several projects that promise cost-effective solutions for detection, localization, and quantification. These include distributed sensor networks costing as low as $300/sensor, printed carbon nanotube sensors with 1 ppmv sensitivity, and affordable mid-infrared integration networks, among others (ARPA-E 2018). However, many of these technologies are still in development and are undergoing field trials; it is not known whether they will perform according to expectations. Few fixed sensors have been independently evaluated for ambiguous and unnecessary leak detection in upstream O&G, and technical capabilities and limitations of this technology class are poorly constrained (table 1). Nevertheless, the appeal of continuous, automated monitoring has the potential to catalyze progress in this area.

Mobile ground labs
MGLs are versatile platforms for conducting local-to regional-scale surveys of methane emissions. In their simplest form, MGLs consist of a vehicle equipped with a global positioning system and a methane sensor. This setup enables a survey approach called concentration mapping ( figure 3(c) . As a screening tool, MGLs may be used in close proximity (e.g. on the well pad) or on nearby roads, up to several kilometers downwind, but with detection limits increasing with distance (von Fischer et al 2017). Onsite screening can detect smaller sources but may miss others that are elevated, indoors, or otherwise inaccessible. Onsite and road-based screening methods have not been quantitatively compared. To date, mobile quantification studies have all relied on offsite measurements. As detected plumes generally originate upwind from roads, dispersion models must be used for localization and quantification estimates . Numerous models have been developed to remotely estimate mass flux from an MGL, including EPA other test method  et al 2017). The vehicles were tasked with taking photographs of street scenes, but the addition of a methane sensor allowed them to also collect methane concentration and identify locations of thousands of urban leaks. This approach is advantageous due to low implementation costs, although it is more difficult to collect data in certain locations, as routes are slaved to the primary task. Passive sensing and road-dependency make MGLs well-suited for urban settings. These capabilities have already proven to be useful for identifying emissions sources in municipalities with older NG distribution infrastructure, such as Washington D.C.  and Boston (Phillips et al 2013). Whether for rural or urban applications, most modern MGLs are equipped with a user interface that displays real time concentration data, enabling drivers to search for, locate, and repair simultaneously.
In recent years, MGLs have received considerable interest and have been deployed in various ways (table 1) and could contribute to M1 (e.g. using the tracer flux technique), M2 (e.g. with passive monitoring), and M4 (using mobile methods). Some approaches require that MGLs be stationed within the plume of interest for upwards of 15 min (e.g. OTM33A), but newer, less-precise methods use in-motion quantification (  (2017), in which over 1600 well pads were surveyed across nearly 8000 km of roads, has demonstrated the potential of MGLs to screen large areas. Approaches that do not require site access spend less time at each site, require minimal coordination with facility operators, and provide enforcement agencies and independent researchers the benefit of a blind sample. Despite these advantages, MGLs are unlikely to meet all the needs of an LDAR program. First, MGLs are limited by road access and meteorological conditions, especially wind direction. In the absence of sufficient wind, or if wind is blowing in the wrong direction, fugitive plumes may never reach a road. Second, screening-grade quantification remains difficult. Plume lofting due to atmospheric instability is a challenge, although some attempts have been made to better characterize vertical concentration gradients (Rella et al 2015). As with other screening methods, differentiating between routine venting and fugitive emissions remains unsolved, and criteria must be established for whether a measured enhancement warrants follow-up with close-range methods.

Piloted aircraft
In recent years, there has been a growing interest in using piloted aircraft for surveying site-level emissions. While helicopters may be used ( For screening, the main strength of aircraft relative to ground methods and UAVs is survey speed. Although satellite coverage is better, detection limits for aircraft are currently 1-3 orders of magnitude lower (table 1). However, piloted aircraft have several features that may limit their use for fugitive emissions screening. In addition to the relatively high cost of acquisition, maintenance, and operation, aircraft are subject to regulations that dictate when and where surveys can occur. The most limiting of these regulations is the lowest safe altitude (LSA) for flying, which is generally a minimum of 500 feet above the highest regional structure. Therefore, despite having the ability to resolve methane plumes in three dimensions, issues may result if a significant proportion of the gas remains below the LSA. At high latitudes, the stable atmospheric conditions that lead to insufficient vertical mixing can prevail for months. In cases where some mixing occurs, efforts have been made to account for missing concentration data from below the LSA (Conley et al 2017). It should be noted that this limitation only affects sampling approaches that pass through the plume, and that remote sensing techniques are not impacted, suggesting that the latter approach provides important advantages when the planetary boundary layer is shallow (Frankenberg et al  2016). However, imaging spectrometers require adequate insolation, which can be limited by clouds and low radiative flux during high-latitude winters, depending on the region and time of year (figure 4). In general, aircraft have difficulty monitoring in winter conditions; radiation inversions prevent air sampling and scattering by snow confounds remote sensing based on infrared spectroscopy. In high-latitude production areas, aircraft surveys may not be possible for several consecutive months each year. Given the necessary distance between piloted aircraft and potential methane sources, minimum detection limits tend to be much higher than with ground-based methods (table 1)

Unmanned aerial vehicles
UAVs have received considerable interest for their use in LDAR ( figure 3(f)). Like aircraft, UAVs can operate in three-dimensional space and are not restricted to roads, but have significantly lower detection limits (table 1). Compared to aircraft, UAVs are also cheaper and more flexible to operate, are semi-automated, and can complement close-range methods by reaching dangerous or inaccessible places, such as the tops of tanks. One of the primary advantages of UAVs is that they are uniquely suited to characterizing methane plumes in three dimensions while flying in close proximity to the source, improving confidence in attribution. To date, various attempts have been made to measure methane from a UAV (Khan et al 2012, Brownlow et al 2016, Barchyn et al 2017, Cossel et al 2017, Emran et al 2017, Smith et al 2017. The first comprehensive study investigating UAV methanesensing potential focused on a single compressor station, which was monitored over a one-week period by a fixed-wing UAV (Nathan et al 2015). The UAV's 3.1 kg, 25 W open-path sensor had a precision of 0.1 ppmv at 1 Hz. In their study, Nathan et al (2015) confirmed the capacity of UAVs to situate themselves inside a plume of interest, and revealed important spatial and temporal variability in concentration in close-range plumes. Overall emission rate uncertainty was 55%, consisting primarily of interpolation errors, wind speed, and mixing ratios. Simultaneous MGL and on-site audit validations were consistent with UAV observations. A more recent study deployed a much smaller rotary-wing UAV, which was found to reliably detect emissions as low as 0.04 kg h −1 (Golston et al 2018, Yang et al 2018).
To date, no study to our knowledge has deployed a UAV to screen for emissions at multiple facilities during a single flight. Both fixed-wing and rotary UAVs face platform-and sensor-related limitations. The preferred sensors on other platforms are generally too large, heavy, and consume too much power to be mounted on a UAV. UAVs are therefore unlikely to contribute to M1 and M2 and may contribute to M3 if equipped with an OGI or if high-resolution threedimensional concentration maps can be generated for individual facilities. In addition to payload and power constraints, UAVs are particularly sensitive to meteorological conditions, including high winds and rain, which not only make flying more difficult, but also make concentration measurements less reliable. Obtaining precise wind measurements with high temporal resolution is critical for generating reliable estimates of location and flux (Nathan et al 2015). This is relatively straightforward with fixed-wing UAVs that measure airspeed with pitot instruments, but more difficult with rotary wing UAVs (Neumann et al 2011). Low sensor frequency, when combined with high flight speeds in close proximity to a methane source, further increases uncertainty. For example, with 1 Hz sampling, the UAV detailed by Nathan et al (2015) had plume transects of 1-2 s, which is insufficient for generating an accurate spatial interpolation of concentration. Regulatory limitations often restrict UAVs to a maximum flying height, within line-of-sight distance, daytime-only flying, and prohibit flying near airports or above populated areas. Finally, short flying times (typically 20 min to 2 h) make UAVs unsuitable for screening in settings with low infrastructure density.
While UAVs are not yet established within LDAR, they may overcome some of the challenges they face. The development of cheap and lightweight sensors is underway (Patel 2017 However, for regional screening, a case for using UAVs as opposed to piloted aircraft has not been put forward.

Satellites
To date, satellites have only been used for M2, albeit with limited consensus on the reliability of regional attribution and temporal trends (Bruhwiler et al 2017). The field has developed rapidly over the past two decades, and numerous new satellites have been-or will soon be-launched (Jacob et al 2016). To date, four satellites have been used to measure methane emissions in the troposphere: SCIAMACHY, GOSAT, TROPOMI, and GHGSat, but the latter two are new and their capabilities have yet to be fully demonstrated. All these instruments use backscattered shortwave infrared (SWIR) radiation to infer column-integrated mixing ratios. Concentrations are calculated by comparing the reflected spectrum to a model spectrum, which can be generated using the 'full-physics' method or using CO 2 as a proxy (Jacob et al 2016). Observations can then be combined with a gridded emissions inventory in a chemical transport model, which typically uses a Bayesian inversion to produce an optimized emission field. Due to a sensor failure, SCIAMACHY only produced quality measurements between 2003 and 2005 (Frankenberg et al 2011) with a pixel size of 30×60 km and 6 d repeat coverage. GOSAT, with circular pixels (10 km diameter) and 3 d temporal resolution, observes pixels only everỹ 260 km in cross-and along-track directions, resulting in extremely sparse data (Kuze et al 2016). TROPOMI promises to improve on the precision and pixel resolution of GOSAT and the data continuity of SCIAMACHY at 1 d temporal resolution (Butz et al 2012); preliminary results are promising (Hu et al 2018).
Methane-sensing satellites outrival other technologies in spatial coverage (Schneising et al 2014). Once operational, satellites require relatively little further investment and should collect data for many years. Column-averaged concentration measurements do not require plume access, which alleviates many of the complications typical of plume characterization. However, while methane-sensing satellites have seen promising advances in recent years, several limitations must be addressed before they can be used for facilityscale screening. Most current and prospective satellites have pixel resolutions of 4-100+km 2 , meaning that individual pixels potentially contain multiple emitting sites and non-target sources. These satellites are unable to discriminate between thermogenic and biogenic sources, different facility types, operators, and so on, making source attribution complex or impossible. SWIR instruments require a reflective surface and are only effective on land, during the day, and in the absence of clouds ( figure 4). Only 9% of SCIAMACHY and 17% of GOSAT retrievals are successful, leading to useful observations only once every~67 and~18 d, respectively (Jacob et al 2016). Measurements in highlatitude regions may be prevented all winter by snow and low radiative flux densities. As all current satellites are in polar sun-synchronous orbit, there also exists a diurnal bias as satellites pass overhead at the same time each day, typically in the early afternoon. Finally, additional errors may arise from the chemical transport model, the bottom-up inventory, and from spatial variability in albedo due to land cover and topography (Jacob et al 2016).
It remains unclear whether satellites will become useful screening tools, but there is potential. The recently launched GHGSat Claire is a demonstration satellite that promises to screen for facility-scale emissions with a spatial resolution of<50 m (figure 3(e)), but emissions detection and measurement capabilities have yet to be publicly benchmarked. According to Jacob et al (2016), the detection limit for GHGSat should be 240 kg h −1 for winds of 5 km h −1 , and progress is being made to improve quantification algorithms (Varon et al 2018). This detection threshold is much higher than the average leak, encompassing the top 700 anthropogenic point sources reported in the US EPA GHG Reporting Program, which does not include fugitive emissions from upstream O&G (Jacob et al 2016). The development of active LiDAR sensors operating in the SWIR is now underway (MERLIN), with plans for a joint launch in 2020 by the German Aerospace Center and the French National Center for Space Studies (Kiemle et al 2014, Jacob et al 2016, Riris et al 2017. Using active sensing instruments to measure methane from space may improve spatial resolution, allow for 24 h data collection, and reduce backscattering effects and cloud contamination. Several geostationary satellites have also been proposed to provide continuous, high-resolution monitoring over areas of interest (e.g. O&G basins), and continental coverage at hourly intervals.
5. The future of LDAR 5.1. Integrating screening technologies Screening technologies are unlikely to replace handheld devices, which combine the sensitivity and spatial precision needed to confirm and diagnose fugitive sources. Among screening technologies, there exist notable differences in detection limits, readiness levels, flux estimation errors, spatial resolution, cost, susceptibility to adverse weather, suitability for alternative use-cases, and future potential. Thus, in selecting the appropriate sensor or platform for screening, the application must be well understood, and numerous trade-offs must be negotiated. Currently, fixed sensors, MGLs, UAVs, and aircraft have the highest readiness levels for screening, and each is suited to niche applications. Satellites, which currently have limited suitability for commercial LDAR, receive considerable investment and are likely to see innovation over coming years.
A major outstanding challenge for screening technologies is their inability to discern vented from fugitive emissions. Currently, most jurisdictions authorize facility-level venting limits, which may confound efforts to screen for fugitive emissions. False-positives during screening could mistakenly trigger follow-up surveys using close-range methods. Needless searching for these 'ghost sources' may increase the cost of screening and dissuade operators from moving beyond conventional LDAR. One solution is to reduce false positives by screening only for sources that greatly exceed venting limits. Should regulators impose stricter venting limits, screening techniques could become more sensitive to the relative presence of fugitive emissions. Similarly, screening could become a popular approach if regulators were to eliminate the distinction between vented and fugitive sources, opting instead for site-level limits on total emissions. Such outcome-oriented policies also provide operators the flexibility to limit emissions according to local opportunities and constraints.
Our study reveals that not enough work has been done to evaluate the performance of screening technologies. While MGLs, aircraft, and satellites have been used extensively for independent research, often with the goal of characterizing emissions from the O&G system, their suitability for use in LDAR is largely speculative. There is a pressing need for independent research to critically evaluate the strengths and limitations of screening methods, with attention to the development of detection probability curves that account for realistic environmental and operational conditions. Interest in such work is growing, good examples include the 'Mobile Monitoring Challenge' by the Environmental Defense Fund and Stanford University, and the Ginninderra experiment (Feitz et al 2018). Ultimately, such work could lead to a policy framework for evaluating equivalence in emissions reductions among different suites of technologies and methods. Furthermore, such efforts can filter through the excitement typically associated with new technologies and provide comprehensive and reliable assessments of their capabilities. However, there are additional hurdles to overcome before new technologies are assimilated. These include the perceived unreliability of new technologies, lack of familiarity with underlying principles of measurement, requisite changes to field and data management practice, and shifts in regulatory language (CATF 2013). To overcome these challenges, regulators should develop a decision-making process to test and approve new technologies.
This review suggests that effective LDAR solutions could leverage different technologies based on their context-dependent strengths and limitations. A promising approach to cost-effective LDAR is a CMP that integrates two or more screening and close-range technologies. To date, several studies have explored the use of multiple platforms for methane detection and quantification. For example, the Barnett Coordi- . However, most of these integrative campaigns were designed to improve methane emissions estimates without practical consideration for use in LDAR programs. Studies that integrate technologies in different configurations to optimize both cost and mitigation are needed, and they must not be blind to the practical considerations faced by operators. To date, only one study has compared a CMP to conventional LDAR (Schwietzke et al 2018). Here, the authors found that using aircraft to direct ground surveys could be at least as cost-effective as conventional LDAR; compared to ground teams, the aerial surveys detected up to 26 times more methane from half as many sources. However, Schwietzke et al (2018) warn that these results can be highly context-dependent, and their work illustrates the confounding influence of vented emissions on CMP effectiveness.
In areas with multiple interspersed companies or with low infrastructure density, screening by means of aircraft and satellite deployment may not be economically viable unless costs are shared. Operators and regulators should therefore explore different costing scenarios for screening. For example, satellites and piloted aircraft could be used for regular large-scale surveillance, financed using a subscription-based approach, and operated as part of a government program or by a third party. Regular UAV campaigns over high-density and/or high-risk areas could supplement these data, providing quick estimates of high-priority sources needing immediate attention. Finally, MGLs equipped with handheld devices could use intelligence from large-scale campaigns to determine where and when to investigate further or conduct repairs. A simpler CMP could integrate a handheld approach (Method 21 or AWP) with a single screening technology.
These general examples of possible CMP configurations are a simplified abstraction of what would be a multifaceted undertaking riddled with technological, logistical, and regulatory challenges. As technologies and methods evolve, so will the most appropriate CMP configuration for a given application. Some technologies may see considerable improvement in capabilities, while others may become obsolete. UAVs and satellites, as relatively young platforms, are particularly well-suited to overcome the limitations that currently prevent them from playing a larger role. As sensors become smaller, and UAV flight times increase, the UAV niche may grow. Full UAV autonomy for this application only requires leveraging existing robotics technology but will take time to mature. Similarly, if an increasing number of methane-sensing satellites are deployed, and if their capacity to deliver higher-resolution data improves, satellites could supplant manned aircraft for regional surveys. Passive, continuous measurement may also become prominent as sensors become less expensive and more durable, which could reduce MGL labor costs as sensing is accomplished on vehicles performing other tasks. Ultimately, the most popular programs will achieve compliance at the lowest cost. Monitoring plans, largely dictated by economics, may guide LDAR away from labor-intensive and towards automated methods. As new technologies are approved for LDAR, competition could increase innovation and reduce monitoring costs. This could lead to a win-win situation for the public and industry, as an increasingly greater proportion of the leak-size distribution can be repaired at a net-negative cost. Given the range of possible screening scenarios and technology metrics that must be considered, providing explicit guidance on when each technology should be used and in what combination is beyond the scope of this article. Ultimately, models should be developed to evaluate the most effective CMP for a given context, especially if multiple technologies are to be used. These models could consider infrastructure density, monitoring cost, detection limits, meteorology, and other factors. At present, these models are informed by a limited empirical understanding of available technologies (e.g. Kemp et al 2016). Multiplatform comparisons of not just the same region, but of the same plumes, would lead to an improved understanding of temporal variability and could greatly improve localization and quantification capabilities. Emissions data from facility operators can be difficult to obtain, as standardized collection and reporting for event-specific fugitive emissions has not yet been mandated, adding further incentive to developing CMPs with data-driven prediction capabilities.

Beyond technology integration
Improving incentivization structures at all levels of industry would help to further ensure that technologies are being used to their potential. Principal-agent problems must be identified and addressed. For example, those who own the infrastructure-especially in the case of pipelines-do not necessarily own the leaking product. Furthermore, an improved understanding of how human error influences detection likelihoods for different technologies is needed. Handheld cameras are especially problematic in this regard, due to the subjective nature of the instrument operation and data interpretation (von Footer 2015, Fischer et al 2016). Incentivizing LDAR among operators might be achieved by improving our understanding of the economic benefits (e.g. product loss prevention) of investing in mitigation. In the future, carbon pricing of fugitive emissions could also be used to incentivize mitigation.

Regulatory and economic considerations
Clear, predictable, and enforceable policies are needed to ensure the effective implementation of LDAR programs. Although the financial return on LDAR programs is often modeled to be net positive (Kemp et al 2016), as lost gas can be sold to offset LDAR costs, operators may have more pressing or promising investments to make, or insufficient capital to invest in new technology. While the economics vary by jurisdiction, we identify three broad classes of incentivization, similar to : (1) direct regulatory forcing, in which operators are forced to comply with regulations; LDAR is regulated in detail, and compliance is assessed against standards of operation; (2) indirect regulatory forcing, whereby emissions are taxed to offset associated externalities; (3) voluntary mitigation programs, where companies design and implement the LDAR program they find has the best return on investment, while complying with health and safety regulations.
With direct regulatory forcing (class 1), the regulator must be able to implement and enforce suitable protocols. Compliance is assessed according to the standards set by the regulator, not necessarily the absolute reduction in GHG emissions. New technology is difficult to implement, as there is a delay between establishing appropriate standards of operation and achieving regulatory approval. With indirect regulatory forcing (class 2), a reliable estimate of the actual GHG emissions must be acquired to achieve compliance. Class 2 has only recently become a consideration, as methods for quantifying emissions have matured , Atherton et al 2017. The way operators meet compliance is open, and there is impetus for innovation, as there is a competitive market for improving methods and technology. Without regulations (class 3), fugitive emissions may rise if infrastructure is neglected due to limited capital or more important alternative investment opportunities.
The success of a direct regulatory model is limited by unknowns, such as the mitigation effectiveness of different LDAR programs. Sustained long-term measurements at NG facilities could help to inform future mitigation policies. By measuring pre-and post-LDAR emissions factors through different implementation periods (e.g. 0, 3, 6, and 12 months), we could better constrain long-term emissions trends. Such efforts may also help identify the sources that are prone to relapse following repair. At the policy level, this can translate into more directed regulationsequipment that does not emit after repairs can be inspected at a much lower survey frequency, thereby reducing costs. As hazardous air pollutants are often co-emitted with methane, opportunities exist to improve efficiency by harmonizing monitoring efforts.

Conclusion
Current LDAR programs rely on close-range methods such as Method 21 and AWP. While close-range instruments are indispensable for identifying and documenting component-level fugitive sources, they are relatively labor intensive. Rather than relying exclusively on handheld instruments, regulations in Canada and the US are moving towards the integration of screening technologies. Given the characteristic shape of most leak-size distributions, frequent screening for super-emitters could reduce fugitive emissions and lead to a lower, albeit more targeted reliance on exhaustive close-range surveys. Fixed sensors, MGLs, UAVs, manned aircraft, and satellites, have been used for research-based applications and for monitoring other air pollutants, but are only just gaining interest as tools for LDAR. As screening technologies, each is uniquely suited to a range of environmental, economic, and operational contexts. Fixed sensors, MGLs, UAVs, and aircraft are arguably ready for integration as screening products into current LDAR programs. Satellites may soon be ready given anticipated development and innovation trajectories.
To meet emissions reduction targets and reduce monitoring costs, governments and the O&G industry should consider CMPs that integrate different technologies into a multi-scale, data-driven, methane-sensing system. CMPs could be tiered both spatially and temporally, with frequent monitoring at coarse spatial resolutions using screening technologies, and infrequent, targeted monitoring at fine spatial scales using close-range methods. In addition to cost-effective monitoring and enhanced mitigation, CMPs could improve scientific understanding of how, when, and why fugitive emissions occur, and enable dynamic inventories, regulatory accounting, and evaluation of mitigation success. Cooperation and transparency among regulators, O&G companies, monitoring agencies, and researchers will be crucial for moving towards a CMP model. Regulatory flexibility and standardized protocols for the approval of new technologies must be developed. Finally, there is an immediate need for research evaluating individual technologies and CMP configurations for their mitigation potential, economic viability, and regulatory compliance.