FEASIBILITY OF UTILIZING SPACEBORNE IMAGERY TO IDENTIFY LOST GAS IN A NATURAL GAS GATHERING SYSTEM by Michael E. Burgess II A thesis presented to the Department of Geography and the Graduate School of the University of Central Arkansas in partial fulfillment of the requirements for the degree of Master of Geographic Information Systems Master of Geographic Information Systems Conway, Arkansas December 2012 TO THE OFFICE OF GRADUATE STUDIES: The members of the Committee approve the thesis of Michael E. Burgess II presented on ???? X, 2012 ____________________________________ Dr. Brooks Green, Committee Chairperson ____________________________________ Dr. Brooks Pearson ____________________________________ Dr. Balraj Menon PERMISSION Title Feasibility of Utilizing Space-borne Imagery to Identify Lost Gas in a Natural Gas Gathering System Department Geography Degree Master of Science In presenting this thesis in partial fulfillment of the requirements for graduate degree from the University of Central Arkansas, I agree that the Library of this University shall make it freely available for inspections. I further agree that permission for extensive copying for scholarly purposes may be granted by the professor who supervised my thesis work, or, in the professor‟s absence, by the Chair of the Department or the Dean of the Graduate School. It is understood that due recognition shall be given to me and to the University of Central Arkansas in any scholarly use which may be made of any material in my thesis. _____________________ Michael E. Burgess II September 4, 2012 Acknowledgement iv Abstract Methane (CH4), a greenhouse gas, is released into the atmosphere by natural and anthropogenic processes such as power plants, natural gas processing, industrial areas, landfills, swamps, and rice patties. A low cost, accurate method for monitoring CH4 releases can be useful for identifying natural and anthropogenic sources, conserving CH4 as an energy source, and to assist energy production/processing/transportation companies in maximizing revenues. Numerous studies have been conducted regarding the indirect identification of terrestrial CH4 plumes utilizing airborne and spaceborne multi-spectral/hyper-spectral sensors by targeting the effects of CH4 on the spectral response of plants and ground materials. Less research has been conducted on the direct identification of CH4 by targeting specific absorption bands of CH4. This research aims to build on past studies to directly identify terrestrial CH4 plumes utilizing the Hyperion 1 hyperspectral sensor aboard NASA‟s Earth Observing (EO-1) satellite. Rather than targeting specific absorption bands of CH4 this thesis will identify pixels covering a landfill with known CH4 production and identify other pixels that have a similar spectrum. The methodology will utilize a threestep process that incrementally increases in intensity. Any positives will be field validated using flame pack equipment and the spectrum of any true positives will be compared to the traditional 1.66µm and 2.3 µm CH4 absorption bands. Keywords: methane, Hyperion, absorption, training data, remote sensing, near-infrared v Table of Contents Acknowledgments………………………………………………………………………..iv Abstract……………………………………………………………………………………v List of Figures……………………………….....………………………………………..viii List of Symbols and/or Abbreviations……………………………………………………ix 1.0 Introduction……………………………………………………………………….…..1 1.1 Background…………………………………………………………………...1 1.2 Literature Review…………………………………………………………..…3 1.2.1 Agricultural Emissions……………………………………………3 1.2.2 Wetland Emissions………………………………………….……4 1.2.3 Oil and Gas Applications…………………………………….…..5 1.2.4 Atmospheric Applications…………………………………….…7 1.2.5 Extraterrestrial Applications………………………………….….8 1.3 Methane Spectra and Properties…………………………………………..….9 1.4 Overview of Hyperspectral Platforms……………………………………....10 1.5 Discussion……………………………………………………………….…...12 2.0 Problem Statement……………………………………………………………….…..12 3.0 Research Questions…………………………………………………………………..14 4.0 Hypotheses………………………………………………………………..................14 5.0 Methodology………………………………………………………………………....14 5.1 Research Area………………………………………………………….….....14 5.2 Hyperion………………………………………………………………….….15 5.3 Software………………………………………………………………….......17 vi 5.4 Analysis………………………………………………………………………18 5.4.1 Imagery Pre-Processing………………………………………….18 5.4.1.1 Conversion to Radiance…………………………………18 5.4.1.2 Conversion to Reflectance (Atmospheric Correction) …...19 5.4.2 Anomaly Detection………………………………………………19 5.4.3 Target Detection ……..…………………………………………..20 5.4.3.1 Spectrum Derivation ...…………...……………………..20 5.4.4 Sub-pixel Analysis……………………………………………….22 5.4.5 Geographic Association………………………………………….23 5.5 Validation…………………………………………………………………….24 6.0 Expected Results…………………………………………………………………......25 7.0 Timeline……………………………………………………………………………...26 8.0 References……………………………………………………………………………27 9.0 Appendices…………………………………………………………………………...31 Appendix A…………………………………………………………………………..31 Appendix B………………………………………………………………………......31 Appendix C…………………………………………………………………………..32 Appendix D…………………………………………………………………………..35 vii List of Figures Figure 1 Equation Estimating Total Cost of Lost………………………………………..2 Gas to Production Companies Per Day Figure 2 JAXA Calculated and Observed Absorption Near the ………………………..9 1.66μm Methane Absorption Band. Figure 3 Fayetteville Shale Play in North Central Arkansas…………………………....17 Figure 4 ESRI‟s® Raster Calculator…………………………………………………….19 Figure 5 Location of Sample Spectra…………………………………………………....21 Figure 6 Photopac Micro FID Detector……………………………………………….…24 Figure 7 Research Flow Diagram………………………………………………………..24 viii List of Symbols and/or Abbreviations ADEQ-Arkansas Department of Environmental Quality AOGC-Arkansas Oil and Gas Commission BSCFD-Billion Standard Cubic Feet/Day CH4-Methane CHRIS-Compact High-Resolution Imaging Spectrometer DAR-Data Acquisition Request DN- Digital Number EPA- Environmental Protection Agency EROS-Earth Observation and Science Center ESRI-Environmental Systems Research Institute HDF- Hierarchical Data Format JAXA- Japanese Aerospace Exploration Agency MIR-Middle Infrared NASA-National Aeronautics and Space Administration NIR-Near Infrared nm- Nanometer NOAA-National Oceanic and Atmospheric Administration GOSAT- Greenhouse gas Observing Satellite OSP-Orthogonal Subspace Projection PROBA- Project for OnBoard Autonomy LIDAR- Light Detection And Ranging MAS- MODIS Airborne Simulator SWIR-Short Wave Infrared Tg-Teragram TIR-Thermal Infrared TM-Thematic Mapper μm-Micrometer VIR-Visible Infrared USDA-United States Department of Agriculture USGS-United States Geological Survey ix 1.0 Introduction 1.1 Background Considering the growing concern of greenhouse emissions and the growing popularity of natural gas as fuel for vehicles and energy generation, an efficient method of identifying surface methane (CH4), the primary component of natural gas, plumes is becoming increasingly important. A method that can routinely, at minimal cost, monitor emission sources such as power plants, industrial areas, landfills, swamps, and rice patties, is needed. More importantly such a methodology could also be used for natural gas energy conservation and production, such as leak detection, and in natural gas pipeline systems or detection of microseepages to identify gas reserves. Concerning natural gas usage, the United States Energy Information Administration (2012) projects the United States consumption of natural gas to be at 736 billion cubic meters (26.55 trillion cubic feet) per year by 2035, which is an increase of 19% since 2009. The Environmental Protection Agency (EPA) estimated that the production, processing, transmission/storage, and distribution of natural gas accounts for 19% (8.9 +- 2.9 billion cubic meters) of the total United States anthropogenic emissions (as sited in Kirchgessner, Lott, Cowgill, Harrison, & Shires, 1997). The EPA further estimated that 12% of the 19% was the result of processing. In this study, natural gas gathering was included within processing. Based on the EPA‟s conclusions and assuming a $4.00 wellhead price per 28.3 cubic meters (1,000 cubic feet or 19.26 grams) of natural gas, this is of potential lost revenue of near half a million dollars per day to natural gas production companies operating in the United States (See Figure 1). 1 Figure 1. Equation estimating total cost of lost gas to production companies per day. Considering these numbers a method that would allow for continuous monitoring of gathering systems to decrease lost gas in the form of leaks could prove extremely beneficial in reducing greenhouse emissions, maximizing revenue, and maximizing natural gas production. Several widely known methods for identifying terrestrial methane plumes are currently in use or in an experimental stage. These include, but are not limited to: 1) Light Detection and Ranging (LIDAR) 2) Thermography 3) Imaging Synthetic Aperture Radar (SAR) 4) Hydraulic modeling software 4) Acoustic modeling 5) Gas detectors that analyze the composition of the air 6) Trained scent dogs and 7) Air-borne spectroscopy. Ground and air-borne methods can be laborious and expensive as compared to spaceborne identification. This is especially the case if CH4 surveys are needed on a periodic basis. Because spaceborne imagery can be less expensive due to cost sharing, is periodic (if the orbit is not geosynchronous), and can be gathered/analyzed at the desktop, it is logical to research the feasibility of identifying methane emission sources using spaceborne imagery. 2 1.2 Literature Review 1.2.1 Agricultural Emissions Through an intensive literature review one will find that scholarly research conducted on direct remote sensing of CH4 is, while extensive, dispersed across several types of applications. Several studies have utilized indirect remote sensing of CH4 emissions such as Melack, Hess, Gastil, Forsberg, Hamilton, Lima, and Novo (2004) used Scanning MultiChannel Microwave Radiometer and the Japanese Earth Resources Satellite-1 to calculate total CH4 emissions from wetlands in the Amazon Basin. Their identification of CH4, as is the case with many similar projects, is an indirect identification of CH4. Pre-tested formulas were used that were created from results in laboratory wetland ecosystems and from other studies. The satellite imagery was only used to identify wetland or aquatic areas. Then the area totals were applied to the formula to reach the total CH4 emissions. Thimsuwan, Eiumnoh, Honda, and Tingsanchali (2000) used both Landsat Thematic Mapper (TM) and National Oceanic and Atmospheric Administration Advanced Very High Resolution Radiometer (NOAA AVHRR) for a similar study. However, this too was an indirect measure of methane. The remotely sensed imagery was only used to calculate the total biomass, or age, of rice stands whereas the biomass of a rice paddy changes as it ages, so does the electromagnetic absorption/reflectance. The researchers had previously grown rice in an isolated environment and measured the produced CH4. From this, a formula was created which was used once the area total and biomass total was calculated based on the imagery. In addition to this study, Manjunath, Panigrahy, Kundan, Adhya, and Parihar (2006) conducted a comparable study in India. 3 This methodology could be considered even more of an indirect calculation of CH4 than Thimsuwan et al. (2000) because the authors utilized SPOT imagery to identify possible rice farming areas based on elevation, irrigation, and rainfall. This is in contrast to Manjunath et al. (2006) where the actual paddies were identified. Thimsuwan et al. (2000) created a formula, similar to Manjunath et al. (2006), which was applied to the potential paddy area totals to calculate total methane emissions. The studies by Thimsuwan et al. (2000) and Manjunath et al. (2006) are not applicable to studies where the presence of methane must be determined. The titles for both of these works are somewhat misleading since space-borne imagery is not actually used to calculate/identify methane emission, but used, rather, to identify areas that are known to produce CH4. This method is only useful when it is known that an entity is a methane source and that it emits methane at a constant rate. Studies such as this are not applicable to this work since the purpose is to identify sources rather than analyze known sources. 1.2.2 Wetland Emissions Research regarding CH4 emissions of wetlands has similarly used indirect methodologies. Melack, Hess, Gastil, Forsberg, Hamilton, Lima, and Novo (2004) used microwave and radar sensors to establish inundation patterns in the Amazon Basin from 1979-1987 and 1995-1996. The two sensors, Scanning Multichannel Microwave Radiometer (SMMR) and Japanese Earth Resources Satellite-1 (JERS-1), were used because longer wavelengths were needed to penetrate predominating cloud cover, smoke, and forest canopy. However, like the aforementioned research on CH4 emissions from rice cultivation, this study also field measured the parts of methane per million parts of 4 air (ppm) then associated these values with the inundation levels to quantify total emissions. This study actually builds from an earlier study by Rosenqvist, Forsburg, Pimentel, Rauste, and Richey (2002). The studies can almost be considered identical except that Rosenqvist et al. (2002) did not utilize any imagery other than JERS-1 imagery whereas Melack et al. (2004) used SMMR to widen the temporal coverage. 1.2.3 Oil and Gas Applications Similar to the preponderance of indirect quantification of CH4 due to agriculture and wetlands emissions, research has primarily focused on indirect identification of hydrocarbons within oil and gas exploration and transmission research. One may realize upon reading this work that spatial resolution of current sensors is limited regarding the ability to directly identify point sources such as natural microseepages or pipeline leaks. As a result, research has primarily focused on isolating the spectral signatures of vegetation that is stressed by released hydrocarbons due to microseepages and pipeline leaks. Conversely, the low spatial resolution of space-borne sensors as of 2000 lacked the high spectral resolution needed to identify hydrocarbons directly (includes CH4) or the signatures created in plants that have been stressed by hydrocarbons (Yang, Zhang, & Van der Meer, 2000). To assist in CH4 pipeline leak detection Smith, Steven, and Colls (2004) “gassed” different species of grasses in a laboratory to graph how the spectral signatures would change. They were unable to establish a gassing ratio that could be definitively used without comparison to ungassed adjacent vegetation. Due to this, it is rational to research direct CH4 identification. Doing so would remove the “indirect” variable and allow for a more reliable identification that removes the need for intensive, manual interpretation. 5 This is in contrast to Van Der Meer, Van Dijk, Van Der Werff, and Yang (2002) whom indicated that CH4 strongly absorbs in the 2.2-2.4μm range and that hydrocarbon seepages can be identified with the use of additional geologic data. Conflicting statements/results such as this could explain why a widely used protocol has not been developed for identifying small CH4 point sources. Also possible is that much of the research regarding microseepages is conducted by private energy firms and therefore publishing results is obviously a competitive disadvantage in this situation. Because the transportation of natural gas (CH4 is primary component) is not viewed as competitive as exploration and production regarding identification of leaks, it is more logical to review research conducted in this sector. A more recent study by Khan and Jacobson (2008) did utilize a more advanced hyperspectral sensor with better spatial resolution. This study utilized Hyperion to detect hydrocarbon seepages in the Patrick Draw in South-central Wyoming. While they were successful in identifying hydrocarbon seepages, they used indirect methods. In their study, they collected training data utilizing a field spectrometer in areas of obvious vegetation stress in an existing oil field. While this study shows that better spectral resolution makes it possible to identify hydrocarbons on the Earth‟s surface, it relied on training data. Obtaining training data for leaks is obviously not possible due to the dynamics of gas concentration based on leak size and atmospheric conditions at the time of scene collection. Barnhouse (2005) is an example of one of the few works that has studied the feasibility of directly detecting methane produced by oil and gas operations and landfills. This study attempted to identify terrestrial methane plumes using the MODIS Airborne 6 Simulator (MAS) around 3.314μm. The study pointed to the need for increased spatial and spectral resolution as MODIS is actually a multi-spectral scanner rather than hyperspectral. These implications will be further discussed as Barnhouse (2005) is this most applicable research to this study. 1.2.4 Atmospheric Applications Conversely, CH4 has been identified directly in Earth‟s atmosphere in a multitude of ways. Frankenberg, Meirink, van Weele, and Platt (2005) used SCIAMACHY (scanning imaging absorption spectrometer for atmospheric chartography) on board the European Space Agency‟s ENVISAT satellite to detect CO2 (carbon dioxide) and CH4 in the atmosphere. ENVISAT has a very high spectral resolution at 8000 channels between 240 and 2,390 nanometers (nm) (McCarthy, Pratum, Hedges, & Benner, 1997). To detect CO2 and CH4 this study used differential optical absorption spectroscopy (DOAS) based on the assumption that these gases absorb both Earth thermal radiation and solar nearinfrared radiation proven by McCarthy, Hedges, and Benner (1998). Retrieval windows of 1562 nm to 1586 nm were used for CO2 while a retrieval window of 1630-1670 nm was used for CH4. Also integrated into this methodology is the volume mixing ratio (VMR) which is a ratio of the parts that make up a larger part. In this case, it refers to the parts of CO2 and CH4 per million parts of air. It is unclear, but assumable, whether this study utilized a sub-pixel analysis process to calculate the CO2 and CH4 volumes since the results were actually the VMRs. While this direct identification method with a passive sensor works well for atmospheric applications it is not applicable to identification of specific surface sources due to the spatial resolution of SCIAMACHY. The spatial 7 resolution varies depending the spectrum and angle, but, for example, the spatial resolution of channel 8 is 120km x 30km (Meirink, Eskes, & Goede, 2006). 1.2.5 Extraterrestrial Applications Extensive research has been conducted on extraterrestrial methane identification particularly on Mars and Titan (one of Saturn‟s moons). Even though the sensors used in these studies are not applicable to terrestrial CH4 identification, a brief discussion is needed in order to understand any strengths that may be applied or short comings that may be avoided. Three groups believe to have identified methane plumes on Mars: Krasnopolsky, Mailliard, and Owen (2004), Formisano, Atreya, Encrenaz, Ignatiev, and Giuranna (2004) and Mumma, Novak, DiSanti, and Bonev (2003). Krasnopolsky et al. (2004) and Mumma et al. (2003) utilized terrestrial telescope sepectroscopy to identify methane while Formisano et al. (2004) utilized the Planetary Fourier Spectrometer (PFS) aboard the Mars Express orbiter. To relate this to terrestrial identification of CH4, it is worth noting the spectral range that each group used for methane identification. Krasnopolsky et al. (2004) identified CH4 in the 3.31-3.52 μm wavelengths with the strongest detection at 3.7 μm. Mumma, Novak, DiSanti, Bonev, and Dello Russo (2004) observed methane in the 3030-3040 cm-1 spectral range. Using the PFS, Formisano et al. (2004) used the PFS, covering 1.2-4.5μm spectral range, to identify CH4 in the 30003030 cm-1 window. In addition to Mars, several teams have attempted to detect and quantify CH4 in Titan‟s atmosphere. Using Hubble Space Telescope Imaging Spectrograph (HST) Anderson, Young, Chanover, and McKay (2008) attempted to, not only quantify, but also identify the elevation at which methane clouds exist on all of the moons. In order to 8 isolate CH4 signature in the lower levels of the atmosphere, multiple absorption bands were used within the .6-1μm range. The authors were, however, unable to isolate surface methane. To do so, the Anderson et al. (2008) believe that collection should be beyond 1μm, perhaps between 2.0-2.35μm. Extraterrestrial research is very useful since it is an indicator of what ranges CH4 may be identified. However, as one can see, identifying terrestrial CH4 based on a pre-determined range may not be feasible. Doing so may require the researcher to test numerous ranges to determine what works best for the analysis. 1.3 Methane Spectra and Properties Understanding of the spectral response of the material which is being analyzed is necessary even if utilizing training data, which will be discussed in the methodology section of this project. If true positives of the target material are identified, knowledge of the specific absorption bands are needed to validate the results and assist in future research. After reviewing the existing research on remote sensing of hydrocarbons covered in this project, it is apparent these studies have mainly attempted to target absorption bands in the regions of 1.66 μm, 2.3 μm, and 3.3 μm. The 3.3μm region is the fundamental C-H molecular stretch such that this region shows the most intense absorption. Weaker absorptions in the NIR region from 1-2.6µm, particularly around 1.66µ and 2.3µ, are the result of combination overtones (King & Clark 1988). 9 Figure 2. JAXA calculated and observed absorption near the 1.66μm methane absorption band. Source: Modified from JAXA Press Release (2009). Because the fundamental absorption band is outside of the sensor‟s spectral range that will be utilized in this research, one of the weaker overtone combination absorption bands will be vital. This will also be discussed in more detail later in this thesis. 1.4 Overview of Hyperspectral Platforms There are many commercial and civilian airborne and space-borne multi-spectral sensors available, including, but not limited to: MODIS Airborne Simulator (MAS), Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), Enhanced Thematic Mapper (ETM+), and Moderate Resolution Imaging Spectroradiometer, and Satellite pour l‟Observation de la Terra (SPOT). However, multispectral imagery does not allow for finite separation of wavelengths for identifying substances with similar, but slightly different spectral responses (Aspinall, Marcus, & Boardman, 2002) Several options are available regarding hyperspectral airborne imagery. Two notable sensors are AVIRIS (Airborne Visible/Infrared Imaging Spectrometer) and the commercial HYMAPTM (Hyperspectral Mapper) owned by HyVista Corporation. However, this type of data collection is difficult to acquire for routine monitoring. For example, the NASA Jet Propulsion Laboratory only allows graduate students one free 10 flight-line of data per year regarding AVIRIS imagery. Furthermore, data is only made available after data intended for NASA funded researchers have been processed and delivered (Jet Propulsion Laboratory 2012). Few options are available regarding hyperspectral space-borne imagery. Shippert (2004) noted that only three space-borne sensors are currently available, including the FTHSI sensor MightySat II, the Compact High Resolution Imaging Spectrometer (CHRIS) aboard the European Space Agency‟s Project for On Board Autonomy (PROBA-1) satellite, and the Hyperion 1 sensor aboard NASA‟s Earth Observing (EO-1). MightySat II, the first true space-borne hyperspectral sensor, was inserted into orbit in 2000 (Yarbrough, Caudill, Kouba, Osweiler, & Arnold, 2002). However, MightySat II re-entered the atmosphere in November 2002 and is therefore obviously not applicable to this study since development of the study area (discussed later) did not begin until 2004 (Earth Observing Portal 2010). CHRIS may be considered a second option regarding hyperspectral sensors regarding this study. While the spatial resolution of 20 meters is superior to other sensors, it only collects within the 400-1050 nm spectral range (European Space Agency 2010). This range is insufficient because this author feels that a significant amount of atmospheric noise will be detected. In addition, aforementioned studies were only successful in detecting methane at much higher wavelengths outside this range. Since Shippert (2004), the Japanese Aerospace Exploration Agency (JAXA) launched the IBUKI (GOSAT-Greenhouse Observing Satellite) in early 2008. The issue with utilizing this sensor is the data format is vector point data with, minimally, 500 11 meter spacing. This far exceeds what is required to detect materials with varying locations on Earth‟s surface since it is not continuous. This satellite is still in the test phase, but JAXA claims that a leakage of .73 Tg CH4 per day has been detected from compressor stations in Russia (Inoue & o‟Hashi, 2008). Furthermore, Inoue and o‟Hashi (2008) state that detecting methane is dependent upon the wind carrying and dispersing the methane from its point source. This aligns with the fact that GOSAT‟s point sampling is extremely course. This author feels that it will be very difficult or impossible to identify point sources such as pipeline leaks with this sensor. In addition, leaks as large as .73 Tg are not likely to be found in gathering systems in the United States due to the prudence of natural gas pipeline operators and the strong regulatory environment. Building on Barnhouse‟s (2005) study, discussed in 1.2.3., it is apparent that a hyper-spectral sensor with, at least, moderate spatial resolution should be examined. As recommended for future research by Barnhouse (2005), a sensor such as Hyperion will be used for this study. Hyperion is discussed in more detail in section 6.2. In addition to targeting oil and gas operations, Barnhouse (2005) targeted sanitary landfills in Ohio with limited success in identifying CH4 with high confidence. This thesis will further build on Barnhouse (2005) by targeting a sanitary landfill, but, rather than targeting methane using a laboratory collected CH4 spectrum, apply the spectrum from the landfill across an area with extensive natural gas gathering infrastructure. 1.5 Discussion After reviewing current research regarding methane identification using spaceborne sensors, it is apparent many barriers exist to direct terrestrial identification of surface methane point sources. More research is needed regarding direct identification of 12 terrestrial methane or more importantly what spectral and spatial resolution is required to do so. Research regarding plume calculations and minimum mixture is needed to better understand the minimal emissions needed in order to better understand the minimum spatial resolution needed for full or sub-pixel identification of CH4. In relation to this, it appears to be unpredictable what spectral range CH4 will register when identified through Earth‟s atmosphere at, or just above ground level. To date, only CH4 signatures within laboratory, ground-based, air-borne, and extraterrestrial have been tested to the point that a spectral table could be created 2 Problem Statement Regarding detection of CH4 utilizing a space-borne sensor, many researchers have identified and/or quantified CH4 emissions from agricultural operations through indirect identification of plant species. Researchers have been able to quantify or estimate CH4 atmospheric column abundance. However, while some researchers state that it is possible, very little research exists regarding direct detection of terrestrial methane plume sources. Based on past research, the reasons for this appear to stem from difficulties due to: 1) sensor limitations regarding the inability of multi-spectral sensors to discern narrow bands of electromagnetic radiation and 2) ability to isolate wavelengths that nullify the effects of mixed atmospheric gases. Therefore, this study will investigate the detection of CH4 terrestrial methane plume sources utilizing the Hyperion hyper-spectral sensor aboard the Earth Observing-1 satellite. 13 3 Research Questions 1) Does the spectrum of training data collected from a landfill with known CH4 production have similar enough characteristics to be used to identity CH4 plumes released by a natural gas gathering system? 2) Is the spatial resolution of a Hyperion scene fine enough to detect CH4 using anomaly detection, target detection, and sub-pixel analysis methodologies? 3) Will the spectra of the training data and any detected true positives indicate absorption near the 1.66µm or 2.3µm CH4 absorption bands? 4 Hypotheses 1) Spectrum of training data collected from a landfill may be used to identify other methane plumes, but will be limited to areas with similar ground cover. 2) CH4 plume sources will not be identifiable with 30 meter spatial resolution hyperspectral Hyperion imagery. However, CH4 will be detected when the spectrum of the training data is used with the target detection and sub-pixel analyses. 3) The spectrum of the training data and any true positives will show absorption near 1.66µm or 2.3µm CH4 absorption bands but may vary from the center frequency of the utilized band due to the effects of the background signatures. 5 Methodology 5.1 Research Area The Fayetteville Shale natural gas play in North Central Arkansas was chosen as the study area. This area is one of the most active natural gas shale plays in the United States. Based on gas sales data from the Arkansas Oil and Gas Commission, as of 14 November 1, 2010, over 3,500 wells have been completed in the Fayetteville Shale since drilling began in 2004 with daily production at approximately 56.6 million cubic meters per day (2 Bcfd) (AOGC 2010). In addition, based on the Arkansas Department of Environmental Quality‟s (ADEQ) online Air Permitting Database, approximately 75 minor source permits for gas processing facilities have been submitted within the study area since 2004 (ADEQ 2010). Also, this author estimates, based on cursory review of aerial imagery, more than 3,000 kilometers (1,864 miles) of natural gas pipeline have been installed since 2004 to gather this gas. This extraordinary amount of activity should allow ample opportunities for targeting and validating CH4. The physiology of the study area can be classified as mountainous in the North and transitioning to rolling hills in the South. The southwestern area of the Fayetteville Shale Play transitions to the Arkansas River Valley while the eastern edge can be considered Mississippi River Alluvial Plain. The remainder of the study area is located within the Boston Mountains on the Ozark Plateau in north central Arkansas. This area may be further classified as having flat-topped mountains with steep slopes. The hydrology may be described as a dendritic drainage pattern (Arkansas Geological Survey 2010). Elevation of this region ranges from 107 meters (350 feet) to 640 meters (2,100 feet) above mean sea level. 5.2 Hyperion Vincent and Singleton (1995) estimated that airborne hyperspectral platforms are theoretically capable of identifying onshore CH4 plumes with a thickness of 1 meter and a concentration of at least 2000 ppm or .2% CH4 by volume, which can be considered a very lean mixture. Hyperion is a space platform and will be more susceptible to the 15 effects of other atmospheric gases. Therefore, it will have more difficulty targeting specific absorption bands of methane even if the mixture above a landfill may be up to four times the minimum mixture needed for airborne platforms. However, this project‟s aim is not to target a specific absorption band, but rather use pixels where the likelihood of CH4 is high and apply that spectrum to the remainder of the scene to identify similar responses. Scenes collected by the Hyperion hyperspectral sensor aboard the USGS‟s EO-1 will be utilized in this study. As previously mentioned in section 1.2.7, Hyperion is the best available hyperspectral space-borne sensor for this study when one considers its spectral resolution, spatial resolution, wavelength range, cost, and availability. This sensor detects 220 channels ranging from .357 to 2.576 μm (Beck 2003). Appendix A represents a cross-comparison of NASA‟s current sensors, including Hyperion, that are capable of sensing the complete VIS/NIR/SWIR/MIR/TIR spectrum. A disadvantage to using the Hyperion sensor, due to it being a relatively new sensor, is that as of February 25th, 2011, only two scenes have been collected in the study area. Therefore, it was necessary to submit a Data Acquisition Request (DAR) to the United States Geological Survey‟s Center for Earth Observation and Science Center (EROS). This was further complicated by the fact that only 12 scenes may be scheduled per day (Beck 2003). Also, scene collection may be further restricted if only nadir scenes are requested. This research, in order to increase the likelihood of collection, submitted requests for nadir or pointed mode scenes. It is estimated that pointed mode scenes may have look angles 5 to 17 degrees from true nadir. By requesting pointed mode, considering that the EO-1 follows the World Reference System-2 (WRS-2) path/row 16 system, and therefore has a 16-day repeat nadir collection opportunity, the sensor will have three opportunities per 16-day cycle to collect the scenes over the study area (Beck 2003). Regarding scenes collected by Hyperion for the study area, as previously mentioned, only two scenes had been collected since the inception of the Fayetteville Shale Gas Play. However, at the time of the collections, the study area had 70 percent and 90 percent cloud cover, therefore blocking the ground from view. See Figure 3 below. On March 1, 2012, a scene was collected based on this author‟s DAR. The scene is ideal considering it has zero percent cloud cover, and therefore, will be used for this project. Figure 3. Fayetteville Shale Play in North Central Arkansas 17 5.3 Software This study will utilize ERDAS Imagine 2010® software to perform all spectral analyses. Environmental Systems Research Institute (ESRI) ARCINFO 10® will be utilized for all spatial correlation analyses. The two software suites offer full interoperability between one another. 5.4 Analysis Specific areas within the Hyperion scenes within the study area will be targeted to maximize probability of methane presence. This includes locations of gas processing facilities, locations of gas custody transfer points, and gathering gas pipeline right-ofways. The analysis will consist of a three-step method intended to systematically and incrementally increase the intensity of the analysis of the data. If a preceding step in the methodology produces positive CH4 detection, the next steps will still be conducted for the purposes of accuracy and validation. 5.4.1 Imagery Pre-Processing Prior to the three-step analysis, Hyperion scenes require pre-processing in order to convert the digital numbers (DN) of the pixels to radiance and then true reflectance. Radiance is defined as the amount of light that reaches the sensor that includes the amount of light reflected by the atmosphere and does not account for the amount of light absorbed by the atmosphere. Reflectance can be defined as the ratio of the amount of light that is reflected from an object to the amount of light that reaches the target (Borengasser 2008). 18 5.4.1.1 Conversion to Radiance The digital values of Hyperion Level 1 scenes are scaled to maximize precision. In order to convert the digital values to radiance, it is necessary to divide the visible nearinfrared bands by 40 and the short wave infrared bands by 80 (Beck 2003). Hyperion bands 151 and 216, center wavelengths of 1.659 and 2.314µm, will be used since this study will attempt to identify CH4 absorption. A complete list of the center wavelengths of bands within Hyperion‟s range is listed in appendix D. Therefore, the digital values of the pixels in this band will be divided by 80 using ESRI®‟s raster calculator (See Figure 4). Radiance= DN Band 151 80 Figure 4. ESRI®‟s raster calculator 5.4.1.2 Conversion to Reflectance (Atmospheric Correction) USGS‟s Earth Resources Observation and Science Center recommends using the formula below to calculate reflectance for Hyperion scenes that are relatively clear (EROS 2011). ESRI®‟s raster calculator will also be used to perform this operation. Pp= π* L *גd2___ ESUN * גcosθs Where: Pp= Unitless planetary reflectance L =גSpectral radiance at the sensor‟s aperture d= Earth-Sun distance in astronomical units from nautical handbook or interpolated from values listed in appendix B ESUN = גHyperion solar irradiances listed in appendix C θs = solar zenith angle in degrees 19 5.4.2 Anomaly Detection As previously stated, this will be a three-step process which increases in intensity with each step. Anomaly detection will be the first methodology used. This process is designed to identify spectral signatures that significantly deviate from the background spectra. The base image, which is not atmospherically corrected, will be used for this process since ERDAS® designed this engine to use RAW digital number (DN) values. This step will use the Orthogonal Subspace Projection (OSP) methodology. This method was chosen because it is ideal for identification of target signatures within natural environments and requires the target to be present in low concentrations. This matches this scenario well considering that the study area is uniform, consisting mostly of graze land and mixed forests. If any CH4 plumes are present it can be assumed that the CH4 will be in relatively low concentrations. 5.4.3 Target Detection Where the presence of a certain material is known to occur in small concentrations, it is possible to use target detection methods. The target detection process attempts to nullify background spectra and identify or “target” pixels within an image that exhibits a spectrum similar to a sample that has been provided as a parameter. The sample or base signature of the target material is required in order to instruct the software what spectrum to attempt to identify. To supply sample spectra, ERDAS Imagine® will be used to pinpoint the pixels with the highest probability of containing CH4. 5.4.3.1 Signature Derivation Even after atmospherical correction, it is possible that other atmospheric gases will mask a laboratory collected spectra such as the ones contained in HITRAN. The 20 second step of analysis will be to derive a signature or sample from pixels believed to contain CH4 in the same imagery being used for target detection. It is impossible to simultaneously create a methane plume in the study area as Hyperion is collecting the scene to scheduling complications with NASA. Therefore, it is necessary to identify a source that continually produces CH4 such as farming operations that store large amounts of manure, landfills, or waste treatment facilities to create a reference spectrum or “training data”. Carman‟s (1996) (as sited in Vincent, 1997, p. 265) study estimated that the atmospheric gas concentration above three landfills in Ohio ranged between 50008000 ppm dependent upon the atmospheric pressure. This is 2.5 to 4 times the concentration that Vincent and Singleton (1995) (as sited in Vincent, 1997, p. 281) estimated was the minimum concentration required for an airborne platform to detect methane. Several landfills exist within the study area and should provide good opportunities to create spectra samples. One in particular, Waste Management Corporation‟s Two Pine Landfill, has numerous shallow natural gas wells with a gathering system that supplies on-site electricity generators fuel (AEDC 2012). The landfill is located near Jacksonville, Arkansas, approximately 40 kilometers south of the southern edge of the known natural gas production in the Fayetteville Shale Play. The scene that was collected for this project, 185 kilometers in length, contains both the landfill and the eastern flank of the Fayetteville Shale Play. This will allow for this site to be used as sample spectra that can be applied, in-scene, to facilities located well within the limits of the Fayetteville Shale Play (See Figure 5). 21 Figure 5. Location of sample spectra 5.4.4 Subpixel Analysis The third, and most scrupulous step of this methodology, will be to perform subpixel analysis to the scene in an attempt to identify the presence of methane in mixed pixels. Essentially, subpixel analysis is similar to target detection except that it is a process that discriminates between materials that may be mixed within one pixel. This is in contrast to “Target Detection” where the spectra analyzed are a result of the total spectral contributions of all the materials within the pixel combined. ERDAS Imagine 2010 Sub-pixel Classifier® utilizes a mixed pixel fraction methodology to identify “materials of interest” within individual pixels. To simplify, the analysis initiates with Preprocessing where the program attempts to identify a set of general background spectra. Next, automated environmental correction can be applied to the scenes to account for atmospheric effects and sensor irregularities. Residuals from the general background spectra and fractions are calculated that will be used later in the process. Signature Derivation follows where the program operator is responsible for 22 manually choosing which pixels should be considered the material of interest. In this case, pixels covering the Two Pines Landfill will be included to attempt to capture mixed pixels that contain some mixture of emitted methane. The signature derived in section 6.4.3.1 will be utilized. The last step in this process is MOI Classification where the program will calculate the mixed pixel fraction based on the material of interest training data spectra and the potential background spectra (ERDAS 2009). The mixed pixel fraction will be the value that will produce the intensity that the Hyperion satellite observed. After all processing is completed, any positive samples will move to the next step for geographic association. 5.4.5 Geographic Association The resultant rasters will be overlaid with natural gas gathering system data using ARCINFO 10®. The natural gas gathering system data were previously digitized for this project utilizing 3-meter resolution natural color aerial imagery collected by the United States Department of Agriculture (USDA) in 2009. According to the USGS Earth Observing and Science Center (EROS 2012), Hyperion Level 1Gst imagery is spatially registered and corrected for terrain and parallax error. However, it should be noted that the preliminary research for this project has found that Hyperion Level 1Gst level data appear to be shifted approximately 290 meters (950 feet) North 45 degrees West, using quarter quadrant bearings, from its actual position in North Central Arkansas. Any positives shown in the results will be corrected by this distance and bearing using ARCINFO 10®. After spatial refinement, all data will be loaded into a Trimble Geo XH® handheld mapping grade GPS so that it may be used for navigation to the area of the positive results. 23 5.5 Validation After navigating to the location of the spatially refined positive samples, each location will be tested utilizing a Photovac MicroFID Flame Ionization Monitor®, generically known as a “Flame Pack” (See Figure 6). Flame Packs are commonly used by the oil and gas industry for leak detection and confined space entry. The Photovac MicroFID® is capable of detecting methane in concentrations between 0.5 and 50,000 ppm (Photovac 2002). If any of the positive samples from the analysis show measurable concentrations of CH4, then the sight will be considered a validated positive sample. In addition, the concentration of the methane will be recorded and reported in the final project for future research. Figure 6. Photovac Micro FID Detector Source: Pine Environmental Available Online: http://www.pineenvironmental.com/flameionization-detector/photovac-microfid.htm 24 Figure 7. Research Flow Diagram 6 Expected Results True positives are not expected to be identified until the third and most scrupulous process, sub-pixel analysis. In addition, it is anticipated that if positives are identified they will not be limited to fugitive emissions generated by gas production and gathering. Positives may also include emissions from swamps and agricultural operations. Lastly, it is expected that any true positives identified will show absorption lines near 1.66µm and 2.3µm, but may vary slightly either side of the peak absorption due to the background contributions. Please see Figure 7 above, which depicts the entire research flow for this thesis. 25 7 Timeline 26 8 References Aspinall, R.J., Marcus,W.A., & Boardman, J.W. (2002). Considerations in collecting, processing, and analyzing high spatial resolution hyperspectral data for environmental investigations. Journal of Geographical Systems, 4,15-29. ADEQ,. (2010) Arkansas Department of Environmental Quality Air Permitting Database. Retrieved from http://www.adeq.state.ar.us/compsvs/webmaster/databases.htm Anderson, C.M., Young, E.F., Chanover, N.J., & McKay, C.P. (2008) HST spectral imaging of Titan‟s haze and methane profile between 0.6 and 1μm during the 2000 opposition. Icarus. 194, 721-745. AOGC, (2010) Arkansas Oil and Gas Commission, Arkansas Online Data System. Retrieved from http://www.aogc.state.ar.us/Fayprodinfo.htm AEDC. (2012) Arkansas Economic Development Commission. Transforming an Environmental Liability Into an Energy Asset. Retrieved from http://www.arkansasenergy.org/energy-in-arkansas/videos/transforming-anenvironmental-liability-into-an-energy-asset.aspx. Arkansas Geological Survey (2010) Retrieved from http://www.geology.ar.gov/geology/general_geology.htm Barnhouse, W.D. Jr. (2005) Methane plume detection using passive hyper-spectral remove sensing. Master Thesis. Retrieved from http://drc.ohiolink.edu/handle/2374.OX/15739?type=datecreated&focusscope=23 74.OX/4069&mode=browse Bartlett K.B, & Harriss R.C. (1993) Review and assessment of methane emissions from wetlands. Chemosphere, 26, 261-320. Beck, Richard. (2003). EO-1 User Guide-Version 2.3. Retrieved from https://edcsns17.cr.usgs.gov/eo1/documents/hyperion Borengasser, Marcus., Hungate, William S., Watkins, Russell. (2008) Hyperspectral remote sensing: principles and applications. Retrieved from http://www.crcnetbase.com/isbn/9781566706544 Carman, R.E. (1996) An Analysis of Methane Gas Concentrations for Three Sanitary Landfills in Northwest Ohio. Master‟s Thesis. Earth Observing Portal (2010). MightySat Program. Retrieved from http://www.eoportal.org/directory/pres_MightySatProgram.html 27 Encrenaz, T., (2008) Search for methane on Mars: Observations, interpretation and future work. Science Direct. 42, 1-5. ERDAS, (2009) IMAGINE subpixel classifier user’s guide. Technical Documentation. Retrieved from http://www.uwf.edu/gis/manuals/SubpixelClassifier.pdf EROS (2011) Radiance to Reflectance. Retrieved from http://eo1.usgs.gov/faq/question?id=21 EROS (2012). Products. Retrieved from http://eo1.usgs.gov/products European Space Agency (2010) CHRIS. Retrieved from http://earth.esa.int/object/index.cfm?fobjectid=4216 Formisano, V., Atreya, S., Encrenaz, T., Ignatiev. N., & Giuranna, M., (2004) Detection of Methane in the Atmosphere of Mars. Science. 306, 1758-1761. Available from http://www.sciencemag.org/content/306/5702/1758.abstract Frankenberg, C., Meirink, J.F., van Weele, M., Platt, U., & Wagner T. (2005) Assessing methane emissions from global space-borne observations. Science. 308, 10101014. Kirchgessner, D.A., Lott, R.A., Cowgill, R.M. Harrison, M R. & Shires, T.M. (1997) Estimation of Methane Emissions From the U.S. Natural Gas Industry. EPA/600/SR-96/080 Retrieved from http://www.epa.gov/ttnchie1/ap42/ch14/related/methane.pdf Jet Propulsion Laboratory (2012). AVIRIS –Free Data for Graduate Research. Retrieved from http://aviris.jpl.nasa.gov/data/grad_research.html Inoue, G., & O‟hashi, K. (2008) Workshop Proceedings from Natural Gas STAR Implementation Workshop „08: Alternative Leak Detection Technologies: GOSAT. Houston, TX Retrieved from http://www.epa.gov/gasstar/workshops/annualimplementation/2008.html Khan, S.D., Jacobson, S. (2008) Remote sensing and geochemistry for detecting hydrocarbon microseepages. Geological Society of America Bulletin. 120, 96-105 King, T.V.V. and Clark, RN., (1988) In Soils: Proceedings First International Symposium: Field Screening Methods for Hazardous Waste Site Investigation ‟89: Reflectance Spectroscopy (.2-20µm) As an analytical method for the detection of Organics. Retrieved from http://www.epa.gov/nscep/index.html. Krasnopolsky, V., Mailliard, J.P., Owen, T. (2004). Detection of methane in martian atmosphere: evidence for life?. Icarus. 172, 537-547. 28 McCarthy, M. ,Pratum, T. Hedges, J., & Benner, B. (1997) Chemical Composition of dissolved organic nitrogen in the ocean. Nature. 390, 150-154. Available from http://www.nature.com/nature/journal/v390/n6656/full/390150a0.html McCarthy, MD., Hedges, J., & Benner, R. (1998) Major bacterial contribution to marine dissolved organic nitrogen. Nature. 281(5374), 231-234. Available from http://www.sciencemag.org/content/281/5374/231.abstract Melack J.M., Hess, L.L., Gastil, M., Forsberg, B.R., Hamilton, S.K., Lima, I.BT., & Novo, E., (2004) Regionalization of methane emissions in the Amazon Basin with microwave remote sensing. Global Change Biology, 10, 530-544. Manjunath, K.R., Panigrahy, S., Kundan, K., Adhya, T.K., & Parihar, J.S. (2006) Spatiotemporal modeling of methane flux form the rice fields of India using remote sensing and GIS. International Journal of Remote Sensing. 27. 4701-4707. Meirnick, J.F., Eskes, H.J., Goede, A.P.H., (2006) Sensitivity analysis of methane emissions derived from SCIAMACHY observations through inverse modeling. Atmospheric Chemistry and Physics. 6, 1275-1292. Mumma, M.J., Novak, R.E., DiSanti, M.A., & Bonev, B.P. (2003) A sensitive search for methane on Mars. Bulletin of American Astronomical Society. 35, 937. Mumma, M.J., Novak, R.E., DiSanti, M.A., Bonev, B.P. & Dello Russo, N. (2004) Detection and mapping of methane and water on Mars. Bulletin of American Astronomical Society. 36, 1127. Photovac. (2002). MicroFID Flame Ionization Monitor. Waltham, MA. Retrieved from http://www.farrwestenv.com/Site/Adobe%20PDF%20Files/PHOTOVACmicroFI D.pdf Rosenqvist, A., Forsburg, B.R., Pimentel, T., Rauste, Y.A., & Richey, J.E. (2002) The use of spaceborne radar to model inundation patterns and trace gas emissions in the central Amazon floodplain. International Journal of Remote Sensing. 23, 1303-1328. Shippert, Peg (2004) Why use hyperspectral Imagery? Photogrammetric Engineering & Remote Sensing. April 2004. 377-380. Smith, K.L., Steven, M.D., & Colls, J.J. (2004). Use of hyperspectral derivative ratios in the red-edge region to identify plant stress responses to gas leaks. Remote Sensing of Environment. 92, 207-217. Thimsuwan, Y., Eiumnoh, A., Honda, K., & Tingsanchali, T. (2000) Estimation of methane emission from a deep-water rice field using Landsat TM and NOAA 29 AVHRR: a case study of Bangkok Plain. The Imaging Science Journal, 48, 7785. United States Energy Information Administration (2012). Annual Energy Outlook 2012: with projections to 2035 (AEO2012) Retrieved from http://www.eia.gov/forecasts/aeo/pdf/0383(2012).pdf Van Der Meer, F., Van Dijk, P., Van Der Werff, H., & Yang, H. (2002) Remote sensing and petroleum seepage: a review and case study. Terra Nova. 14, 1-17. Vincent, R.K., & Singleton, E.B. (1994). Methane Gas Concentrations Required for Infrared Imaging. Final Report for Hughes Santa Barbara Research Center Grant, Dept. of Geology, Bowling Green State University, Bowling Green, Ohio. Vincent, R.K. (1995) Fundamentals of geological and environmental remote sensing. Upper Saddle River, NJ: Prentice-Hall, Inc. Yang, H., Zhang, J. & Van der Meer, F. (2000) Imaging spectrometry data correlated to hydrocarbon microseepage. International Journal of Remote Sensing. 21, 197120. Yarbrough, S., Caudillc E., Kouba, E., Osweiler, V., & Arnold, J. (2001) Proceedings of the Conference on Imaging Spectrometry VII ‟02: MightySat II.1hyperspectral imager: summary of on-orbit performance (Invited Paper). San Diego, CA, 30 9.0 Appendices Appendix A Cross-Comparison of NASA‟s Current Sensors (EROS 2012). Appendix B Interpolated values of Earth-Sun Distance in Astronomical Units (EROS 2012). 31 Appendix C (Source: EROS 2012) 32 33 34 Appendix D (Source: EROS 2012) 35 36 37 38 39 40
© Copyright 2025 Paperzz