An Improved Calibration for Satellite Estimation of Flared Gas Volumes from VIIRS Nighttime Data

Abstract

The VIIRS Nightfire (VNF) data product is particularly useful for monitoring of global natural gas flaring and estimation of flared gas volumes. Advantages of VIIRS include the collection of nightly global coverage with the inclusion of four daytime channels in the near and shortwave infrared that cover the wavelengths of peak radiant emissions from flares. VNF calculates flare temperatures, source areas, and radiant heat using physical laws. For more than a decade, the Earth Observation Group has estimated flared gas volumes based on radiant heat with a calibration based on reported annual flared and vented natural gas volumes from Cedigaz. The calibration was tuned with an exponent of 0.7 placed on the VNF source areas to achieve the highest regression correlation coefficient. The Cedigaz calibration has wide error bars attributed to unresolvable reporting errors in the Cedigaz data. In this paper we report on the development of an empirical calibration for estimating flared gas volumes based on VIIRS observations of flares running at low, medium, and high flared gas volumes. Tests were run with both single and double flares, with and without atmospheric correction. The new calibrations were applied to VIIRS detection profiles for metered flares located in the North Sea, Arabian Peninsula, and Gulf of Mexico. The results indicate the following: (1) the exponent is unnecessary and causes flared gas volumes to be overestimated for small flares and underestimated for large flares, (2) the calibration can be applied to sites having either single or multiple flares, and (3) flared gas volume estimates can be improved by applying an atmospheric correction to account for regional difference in band-specific transmissivity levels. The new calibration has a prediction interval (error bars) seventy times smaller than the Cedigaz calibration.

1. Introduction

Upstream flaring is a widely used mechanism to dispose of natural gas produced at oil and gas extraction facilities that lack sufficient infrastructure to capture or utilize all of the gas that is produced. The term “associated gas” refers to natural gas that emerges when crude oil is brought to the Earth’s surface. This is referred to as upstream flaring, and it is the largest source of gas flaring worldwide [1]. Smaller quantities of gas flaring occur at oil refineries, natural gas processing facilities, landfills, coal mines, etc. Because upstream flaring is a waste disposal process, there is no systematic reporting of the flaring locations and flared gas volumes are rarely provided by operators. Additionally, where flare volume data are reported, the data are typically self-reported by the flare operators who often do not meter but estimate flared gas volumes based on the difference between the natural gas volume produced and the quantity used or sold. It is therefore difficult to assess the reliability and accuracy of the reported flaring data.

There are five distinct applications for site-specific estimates of flared gas volumes. First, there are carbon cycle analyses that rely on site-specific knowledge of the locations and magnitudes of greenhouse gas emissions to the atmosphere [2]. Second is the tracking of progress in the reduction of flaring [3,4,5], such as those called for by the UN–World Bank initiative to eliminate routine flaring by 2030 [6]. Third is the identification of potentially attractive locations for waste gas utilization [7,8]. Fourth is the calculation of the carbon intensity of fuels [9,10,11]. And fifth is to “tip-off” collections on unlit flares by other sensors capable of detecting methane plumes [12,13,14].

In this paper, we present a new calibration for estimating flared gas volumes based on nighttime VIIRS (Visible Infrared Imaging Radiometer Suite) data. VIIRS is particularly well suited for detecting and quantifying gas flaring due to its collection of two NIR and SWIR channels at night (Figure 1). With sunlight eliminated, the daytime channels record the noise floor, punctuated with bright radiances when an IR emitter is present. Flares can be distinguished from biomass burning and other emitters based on their high temperatures, in the range of 1300–2500 K. Flares have their peak radiant emissions in the SWIR. The new calibration is based on VIIRS observations of metered natural gas flares from low, medium, and high flared gas flows. The timing of the flare events precisely matched the VIIRS overpass times, and the gas flow rates were measured with calibrated meters. The events were monitored on the ground with broadband radiometers, video, and basic weather data. The new calibration replaces a calibration [1] based on annual flared and vented gas volumes reported for individual countries by Cedigaz [15]. The validity of the Cedigaz calibration is questionable due to untraceable flared gas volume measurement and reporting errors, a lack of atmospheric correction, and exclusion of adjacent pixels containing flare radiant emissions. Additionally, the Cedigaz calibration for estimating flared gas volumes includes an exponent applied to the source area term, which has not been tested.

2. Previous Calibrations

Each year, EOG identifies approximately 10,000 upstream flares worldwide. Satellite remote sensing is the only viable avenue for global flare monitoring due to the large number of flares, their wide spatial distribution, remote locations, lack of reliable measurement and reporting by operators, and the temporal variability of individual flares. The first reports of satellite detection of natural gas flares from space came from the U.S. Air Force’s Defense Meteorological Satellite Program (DMSP) low-light imaging data collected at night in the mid-1970s [16]. The DMSP cloud imaging sensor is the Operational Linescan System (OLS), which collects data in two spectral bands: panchromatic (0.5 to 0.9 μm) and longwave infrared (LWIR 11–12 μm). A photomultiplier tube (PMT) is used to intensify the panchromatic band at night, enabling the detection of moonlit clouds, a requirement for meteorologists. In addition to moonlit clouds, the nighttime DMSP visible band imagery detects lighting present at the Earth’s surface. DMSP data have been used to map nighttime lights with an annual time series extending from 1992 to 2020 [17]. Gas flares are mixed in with the city lights but can be distinguished by their large halos of scattered light and their tendency to remain active for only a limited number of years. EOG developed a calibration for estimating flared gas volumes by summing the DMSP brightness values for flaring sites, including the glow surrounding them (Figure 2). The reference data was reported for flared gas volumes compiled by the World Bank. The calibration was used to derive the first multi-year assessment of global gas flaring, spanning 1992–2006, with a paper published in 2009 [18].

There are several disadvantages to DMSP estimation of flared gas volumes. First, the OLS visible band data employ six-bit quantization, ranging from zero to 63. Second, the OLS visible band has no in-flight calibration. Third, the OLS visible band saturation threshold is set to cover moonlit clouds, and flares can readily exceed this level, resulting in signal saturation. Fourth, flaring site bounding vectors are hand-drawn and updated annually to incorporate new flares. With detection in a single band, there is no option to calculate flare temperature or heat output. Instead, EOG developed a “sum-of-lights” (SOL) index from annual cloud-free nighttime lights composites. The SOL index (Figure 2) incorporates grid cells affected by saturation and the halo of scattered light surrounding large flares.

In 2012, DMSP was supplanted by VIIRS as the premier sensor for global monitoring of gas flaring. As with DMSP, VIIRS collects global imagery with a 3000 km swath width and fourteen orbits per day. The VIIRS provides significant advantages over DMSP for flare monitoring. VIIRS has many more spectral bands than the DMSP-OLS and has the advantage of in-flight calibration outside the longwave infrared (LWIR). The major advantage VIIRS has over most other sensors is the nighttime collection of near-infrared (NIR) and shortwave infrared (SWIR) spectral bands. This includes spectral bands designated as M7 and M8 (NIR) and M10 and M11 (SWIR). These spectral bands are designed for daytime imaging of the Earth based on reflected sunlight. At night, these spectral bands record the noise floor of the sensor, punctuated by bright radiant emissions from hot objects present on the Earth’s surface (Figure 1). These spectral bands are also present on many Earth observation sensors (e.g., Landsat, MODIS, Planet, etc.), but are rarely collected at night. The nighttime SWIR bands are significant as they cover the peak radiant emissions from flares. Using nighttime VIIRS data, flares can be detected in up to seven spectral bands (M7, M8, M10, M11, M12, M13, and M14) spanning a range of nine micrometers. This data product is referred to as VIIRS Nightfire [19]. With sunlight eliminated, dual Planck curve fitting is performed for a hot emitter (the flare) and a low-temperature (ambient) background. Temperature is calculated using Wein’s Displacement Law [20]. The source area is calculated with Planck’s Law [21]. These two variables are input to the Stefan–Boltzmann Law [22,23] to calculate radiant heat in megawatts (MW). This is substantially more information on the flare than DMSP. The calculation of average annual radiant heat (RH) is based solely on local maxima of VNF detection pixels deemed cloud-free. Data gaps from solar contamination and cloud cover are compensated for by assuming that the flare activity levels are unchanged during the observational gaps. The initial Cedigaz-based VIIRS flared gas volume calibration was built by pairing the annual average radiant heat with the reported annual flared (assuming all flares have the same combustion efficiency) plus vented gas volumes for upstream flaring of individual countries from Cedigaz [1]. To address the mismatch between three-dimensional flare volumes and the flat two-dimensional source areas calculated from Planck’s Law, it was decided to introduce a fractional (less than 1) exponent to the VNF source area before calculation of radiant heat with the Stefan–Boltzmann Law. The exponent-modified version of RH is referred to as RH prime (RH’). The exponent value is tuned to 0.7 to obtain the highest regression coefficient (Figure 3). This choice is best to be viewed as an empirical variance-stabilizing transform to compress the dynamic range of large events; it was not derived from a specific physical law governing flare radiant output. With the exponent, the annual BCM for the largest flares is constrained. The Cedigaz calibration, shown in Figure 4, has wide confidence intervals due to the inclusion of low-quality reporting by multiple countries.

For more than ten years, EOG applied the Cedigaz calibration to estimate annual flared gas volumes for individual flares, which are aggregated to form national flared gas volumes, as shown in Figure 5. The World Bank and other organizations widely report these estimates to track increases and declines in national flared gas volumes. The shortcomings of Cedigaz data in building a flared gas volume calibration include the unknown errors present in the source data, wide error bars, and untested inclusion of an exponent.

3. Data Collections and Analysis

3.1. John Zink Metered Flare Tests

Between January 2018 and August 2019, we collected ground and satellite data from 36 test flares across three separate collection events. The flares were fired outdoors at the John Zink LLC test facility in Tulsa, Oklahoma, located at coordinates 36.1873° N, 95.8434° W. All the tests were conducted at nighttime with the same burners and the same composition of natural gas (93% methane, 5% ethane) in similar weather conditions, characterized by clear skies and low wind. The flares were lit one minute before the satellite was observed over the overpass with the VIIRS sensor in a fully developed stage. During the experiments, we varied flare configuration (single flare or double-flare cluster), flow rate, and satellite view geometry. Multiple tests at the same flow rate were conducted to incorporate satellite view angles ranging from near nadir (satellite zenith angles < 20°), medium (20–50°), and side view (50° to 70°).

In the first collection (

Table 1. Experimental Flare Test Schedule (John Zink Facility).

Collection 1 in January and February 2018, 12 single flares
			Satellite view angle
Flare size	Flow rate lb/hr	Flared volume BCM/yr	Nadir	Medium	Side
Small	750, later 1500	0.004, later 0.008	1	2	1
Medium	7500	0.04	1	2	1
Large	75,000	0.4	1	2	1
Collection 2 in October 2018, 12 double-flares
			Satellite view angle
Flare size	Flow rate lb/hr	Flared volume BCM/yr	Nadir	Medium	Side
Small + Small	1500 + 1500	0.016	1	2	1
Small + Medium	1500 + 7500	0.05	1	2	1
Medium + Medium	7500 + 7500	0.08	1	2	1
Collection 3 in August 2019, 12 single flares
			Satellite view angle
Flare size	Flow rate lb/hr	Flared volume BCM/yr	Nadir	Medium	Side
1/3 large	25,000	0.13	2	2	2
2/3 large	50,000	0.27	2	2	2

) with 12 test flares, we have varied flow rate in the gas supply to the burner (small 1500 lb/h, medium 7500 lb/h, and large 75,000 lb/h). The large flare test volume of 75,000 lb/h represents the largest flared gas volume possible at the John Zink test flare facility. Initially, the slight flare was set to 750 lb/h, which triggered detection in a single band, from which a temperature could not be calculated. As a result, we doubled the flared gas volume for the small flare category to 1500 lb/h. In the second collection, two flares were lit 100 m apart. The 100-m distance was selected to ensure that the bulk of the radiant emissions would be within a single VIIRS pixel footprint. Each flare in the doublet was either small (1500 lb/h) or medium-sized. Four tests were collected with two small flares, four with small and medium-sized flares, and four with two medium-sized flares. The goal of the second collection was to examine linearity between the ground-measured flare flow rate from multiple flares within a single satellite pixel. The third set of tests included large flares of two sizes, 25,000 and 50,000 lb/h. The goal of the third collection was to fill the logarithmic size step between the medium and the very large flare sizes in the previous tests, and to verify linearity between the ground-measured flare flow rate and the observed satellite signal.

3.1.1. Experimental Setup for John Zink Test Flares

The experimental configuration for single-flare tests (Collections 1 and 3) is depicted in Figure 6. Before the test, natural gas was accumulated in a volume tank. During the test, natural gas was pumped to the burner through pressure, temperature, and mass flow meters. The flare was observed from both a satellite and a multipoint ground-based telemetry system, which included continuous photo-video recording and broadband radiometers. Shortwave infrared spectroradiometer measurements were also collected during some tests. An on-site weather station recorded meteorological conditions. Cloud cover along the line of sight from the satellite to the flare was monitored using a modified astrophotography digital camera (for observing stars) and a thermal imaging camera (for observing patchy clouds).

In all single-flare experiments, an identical flare tip with a 36-inch inlet equipped with steam injection nozzles was employed. Steam injection was not utilized during the tests, and the tip functioned as a simple pipe flare. The discharge area and elevation of the flare tip were 0.553 m² and 15.6 m above ground, respectively.

The second flare tip, utilized in the double-flare experiments, featured a smaller 14-inch inlet and was equipped with gas injection nozzles. Similar to the single-flare experiments, gas injection was not implemented, and the tip operated as a simple pipe flare. The discharge area and elevation of the second flare tip were 0.062 m² and 4.3 m above ground, respectively. The two flare stacks were positioned 100 m apart, as illustrated in Figure 7 and Figure 8.

To facilitate the measurement of total radiative heat and the observation of the gas flare’s 3D shape from two distinct orientations, the test facility operator, John Zink LLC, installed two posts equipped with Sony HD video cameras and three-component full-spectrum 3% precision broadband radiometers. The post locations were adjusted throughout the experiments. The initial configuration for single-flare experiments is illustrated in Figure 9 (upper). The arrangement of the radiometric and imaging posts for dual-flare experiments is depicted in Figure 9 (lower). The output from the broadband radiometers is expressed in units of radiant heat intensity (RHI), which records a portion of the total heat output of the flare [24]. This is distinct from radiant heat (RH), which captures a flare’s total radiant output in all directions.

Sky conditions in the direction from the test site to the satellite were monitored using an astrophotography camera to observe stars near the satellite’s position in the sky (Figure 10). A thermal imaging camera was employed to observe patchy clouds, which exhibit temperatures higher than the background of the night sky. A digital camera, Canon EOS 750D, was modified for astrophotography by removing the IR and UV blocking “hot mirror” filter located in front of its 18-megapixel CCD sensor. This modification extended the camera’s spectral response into the VNIR region up to 1.0 µm. In conjunction with the FLIR E1 long-wave infrared thermal imaging camera, the two cameras were utilized to ensure cloud-free conditions along the view line at the time of the satellite overpass. To accurately point the cameras towards the satellite, an orientation sensor was designed and mounted on the same tripod. This orientation sensor comprises a 9 DOF MEMS inertial measurement unit, an ESP8266 microcontroller module, and an LCD display. Minutes before the test, both cameras were aimed at the satellite’s viewing coordinates (azimuth and zenith). The test proceeded only after the operator confirmed clear-sky conditions, defined by the visual detection of stars and the absence of cloud signatures in the thermal imagery. The visible and thermal images from this verification check were saved to the test log.

3.1.2. Ground-Based Spectroradiometer Collection on Flares

In several flare tests we have used the ASD FieldSpec spectrometer with bare fiber optic input to measure the near field spectrum of flares in visible (VNIR) and shortwave infrared (SWIR) wavelength range 0.4–2.5 μm, To avoid saturation, the fiber input was turned backwards facing at 45 deg. towards a 5″ × 5″ white Spectralon reflectance panel (Figure 11). Spectralon is a near-Lambertian reflector, and it gives a diffuse view of the flare. There were no outdoor lights or heat sources except the test flare, thus the ambient nighttime illumination has minimal impact on the resulting spectra.

3.1.3. Conversion from Mass Flow to Flared Gas Volume

When running the test flares, the natural gas mass flowrate in the flare was calculated in lb/h units using direct measurements of fuel pressure (±0.14% accuracy), differential pressure (±0.08% accuracy), and temperature (±0.1% accuracy) meters (Figure 6 and Figure 7) with a one second sampling interval, following the ASME MFC-3M-2004 standard for fluid flow measurement in pipes with orifice, nozzle, or Venturi devices. Uncertainty ±1.3% in the mass flowrate was estimated by propagating measurement uncertainties from the calibrated meters, accounting for errors in these parameters and geometric factors, as specified in the Equations (1)–(13) of the ASME standard [25].

However, the oil and gas community mostly use volumetric units to measure the flared volume of natural gas. The ideal gas law PV = nRT [26] can be used for mass to volume conversion. Here P, V, and T are, respectively, the pressure, volume, and temperature of the n moles of gas, and R = 8.31446 ${\displaystyle \frac{{\mathrm{m}}^3\mathrm{P}\mathrm{a}}{\mathrm{K}\mathrm{m}\mathrm{o}\mathrm{l}}}$ is the ideal gas constant in Si units. From the ideal gas law, the same mass of gas will have different volumes depending on its temperature and pressure. To compensate for the pressure and temperature variance at different locations, the natural gas volume is assumed to be measured at Standard Conditions of 60 F, 14.73 psia (15.6 C, 1 atm or 288.7 K, 101.56 kPa) [27]. The gas supply used in the testing is near 93% methane and 7% ethane. The metering reports mass flow (dm/dt) in pounds per hour (lb/h), plus temperature and pressure. Converting to standard cubic meters per year (m³/h or SCMH) consider the mass flow, temperature, and pressure:

$$\dot{m}{\displaystyle \frac{\mathrm{l}\mathrm{b}}{\mathrm{h}}}\times 0.453592{\displaystyle \frac{\mathrm{k}\mathrm{g}}{b}}\times {\displaystyle \frac{\mathrm{m}\mathrm{o}\mathrm{l}}{17.03\mathrm{k}\mathrm{g}}}\times 8.31446{\displaystyle \frac{{\mathrm{m}}^3\mathrm{P}\mathrm{a}}{\mathrm{K}\mathrm{m}\mathrm{o}\mathrm{l}}}\times 288.7 \mathrm{K}\times {\displaystyle \frac{1}{101.56 \mathrm{P}\mathrm{a}}}=Q {\displaystyle \frac{{\mathrm{m}}^3}{\mathrm{h}}}\times 0.6295$$

(1)

The formula for annual flared gas volumes multiplies the daily flared gas volumes calculated in billion standard cubic meters per year (BCM):

$$\dot{m}{\displaystyle \frac{\mathrm{l}\mathrm{b}}{\mathrm{h}}}\times 0.6295{\displaystyle \frac{{\mathrm{m}}^3}{\mathrm{l}\mathrm{b}}}\times 24 {\displaystyle \frac{\mathrm{h}}{\mathrm{d}}}\times 365 {\displaystyle \frac{\mathrm{d}}{\mathrm{y}}}\times {\displaystyle \frac{\mathrm{B}\mathrm{C}\mathrm{M}}{{10}^9{\mathrm{m}}^3}}=Q \mathrm{B}\mathrm{C}\mathrm{M} \times 5.5146\times {10}^{-6}$$

(2)

3.1.4. VIIRS Nightfire Satellite Data Processing

The schedule for the series of flare tests conducted at the John Zink LLC facility in Tulsa, Oklahoma, is provided in Table 1. After the first tests, the minimum flow rate was increased from 750 lb/h to 1500 lb/h due to the non-detection of a smaller flare by the satellite. Two smaller flares were detected exclusively in the VIIRS M11 spectral band. Consequently, their VIIRS spectra cannot be fitted with the Planck curves using the Nightfire algorithm and no temperature can be calculated. For tests having VNF detection in multiple spectral bands (Figure 12), the flare temperature, source area, and radiant heat are calculated using methods described by Elvidge et al. [19]. The source area calculation requires the size of the M band pixel footprints, shown in Figure 13.

The VNF v3 algorithm uses the top-of-atmosphere radiance (TOA) from the VIIRS M band images to estimate RH and RH’ uncorrected for atmospheric transmissivity. A second set of RH and RH’ estimates was derived using VNF v4 [28], which operates on atmospheric transmissivity corrected radiances. The Cedigaz calibration was not applied to the VNF v4 data as it is specific to the VNF v3 TOA radiances. The transmissivity of the VIIRS bands was calculated for each test using MODTRAN [29] parametrized using total precipitable water and total column ozone from NCEP [30], and aerosol optical depth from the NAAPS grids produced by the U.S. Navy [31]. The transmissivity calculation incorporates the path distance from the flare to the satellite. The correction was applied by dividing the TOA radiance by the MODTRAN calculated transmissivity. The corrected radiances are referred to as “earth surface radiances” (ESR). The atmospheric correction is designed to normalize for regional differences in transmissivity primarily through path-length effects (type of atmosphere, temperature/pressure, water vapor, ozone, and aerosol) and viewing geometry so that the calibrated relationship between apparent radiative heat and flared volume is portable.

Both the TOA and ESR detection radiances were processed with Planck curve fitting for a small high temperature emitter and a large ambient temperature background using a simplex optimization [32]. The result is a temperature (K), source area (m²), and radiant heat (megawatts) for the high temperature emitter and temperature and source area for the background.

For some flares the radiant signal is spread across two or more pixels. The original Cedigaz calibration only considers the local maximum pixel from clusters of adjacent detection pixels. In cases where the flare radiant emissions are split across two or more pixels, estimating flared gas volumes with the pixel having the maximum signal results in an underestimation of flared gas volumes. Although trimming off the weak signals that can occur along the outer edges of the flare detection cluster is desirable, we want to minimize the discard of adjacent pixels having strong radiant emissions, relative to the local maximum. To develop a selection criterion, we ran a test to determine an appropriate threshold for including radiance from adjacent pixels in the total RH estimate. In the test, the inclusion threshold was varied as a percentage of the local maximum pixel’s RH. Values considered include 0% (all detections), 25%, 50%, 75%, and 100% (local maximum pixel only). The RH from each option was paired with the flared gas volume and a regression performed. The best adjacent pixel inclusion option was selected based on the highest coefficient of determination (R²).

3.2. Metered Gas Flare Temporal Profiles from Upstream Flares

High-temporal-resolution metered upstream flare records from four sites to analyze VNF flared gas volume detection limits and to intercompare the calibration options. These records were contributed by members of the Oil and Gas Climate Initiative (OGCI). Three sites are located offshore, with one in the North Sea and two in the Gulf of Mexico. The fourth site is onshore, located on the Arabian Peninsula. The names and precise locations are not disclosed here, in keeping with the non-disclosure agreements with the data contributors. The instantaneous metered flared gas volumes are converted to daily levels in terms of million metric standard cubic meters per day (mmscmd), assuming the flare level stays constant during the day. The mmscmd flared gas volumes were matched to VNF estimates made during satellite overpasses to compile matched sets for both VNF detections and VNF non-detections. These paired sets are used in two ways: (1) to estimate the VNF detection limits for flared gas in terms of mmscmd, and (2) to assess the conformity of the field data with the ‘new’ calibration.

3.3. Comparison of Calibration and Atmospheric Transmissivity from a Single Day of VNF Data

We selected 1 September 2023 for the examination of the flared gas volume calibrations and the effects of atmospheric correction on a full day of global VNF data. The VNF v3 and v4 detections were cross matched and then filtered to only include upstream flares to facilitate the examination. VNF v4 calculates the atmospheric transmissivity in all the VIIRS moderate resolution bands collected at night (M7, 8, 10, 11, 12, 13, 14, 15, and 16) and adjusts radiances to account for atmospheric variations [27]. There are 964 detections of upstream flares used for the intercomparison of the calibration options. Additionally, a global half-degree grid of M10 atmospheric transmissivity was generated and used to examine the range of transmissivities for upstream flares worldwide.

4. Results

4.1. Flare Temporal Flutters in Near-Field Radiative Heat

Flare radiative heat intensity (RHI) for the John Zink flare tests was measured at a sampling rate of 1 s with full-spectrum wideband radiometers positioned at two near-field posts (Figure 9). These measurements started from the flare ignition one minute prior to the VIIRS overpass and continued for a second minute after the overpass. For each single-flare experiment, one post was oriented East-West and the other North-South relative to the flare. These data allow us to examine the stability of flares on the scale of seconds. For this examination, we assembled temporal profiles of RHI and metered flared gas volumes based on the time stamps. An example RHI temporal profile from one of the flare tests is shown in Figure 14. The profiles show the fluttering nature of natural gas flares at the scale of seconds.

Due to the scanning design of the VIIRS satellite radiometer and the meridional inclination of its solar-synchronous orbit, the test flare was observable from either an eastern or western perspective. Consequently, data from the East-West near-field post could be correlated with flare radiative heat and source area measurements obtained from the satellite.

To account for flare stack height (H) and variations in distance between the East-West post and the flare stack across experimental trials, the total radiative heat intensity (RHI_EW) measured by the radiometer was normalized to a standardized flare-to-post distance (R_norm) of 100 m.

$${RHI}_{norm}={\displaystyle \frac{R_{EW}^2+H^2}{R_{norm}^2}}{RHI}_{EW}$$

(3)

Figure 15a presents a scatterplot illustrating the relationship between normalized radiative heat intensity (RHI_norm) from the East-West post and gas flow rate for all single-flare experiments. The red dots mark the VIIRS sampling times. The spread in the ground-based RHI versus flow rate is attributable to the turbulent flutter present in the flare [33] during the time it was running for the VIIRS collection. Figure 15b depicts a scatterplot of the normalized near-field heat intensity (RHI_norm) against the total radiative heat (RH_75%) measured by satellite. Linear regressions through the origin for both datasets yielded R2 values of 0.97 and 0.92, respectively, confirming strong linear correlations between near-field flare radiance and both metered gas flow rate and total flare radiative heat as measured by satellite.

4.2. Examination of Outliers

Outlier removal is a standard practice used to trim out spurious observations prior the quantitative analysis [34]. We identified two styles of VNF detection outliers in the John Zink test data. In the first style of outlier, the flare radiant heat is split between two or more adjacent pixels. This is possibly due to an overlap in pixel footprints, particularly at the edge of the scan. In such cases, reliance on the VNF local maximum pixel undercounts the flare’s radiant output. We tested the improvement in estimation of flared gas volumes associated with merging two or more adjacent pixels having relatively equal RH. Another source of outliers traces back to wind speed and wind direction vis-à-vis the VIIRS scan direction.

4.2.1. Multipixel Flare Detections

To address the potential detection of the same flare in multiple pixels, we evaluated RH estimates that include varying contributions from adjacent pixels:

• RH_max = max(RH) the local maximum:
• RH_75% = sum(RH, when RH > 75% RH_max);
• RH_50% = sum(RH, when RH > 50% RH_max);
• RH_25% = sum(RH, when RH > 25% RH_max);
• RH_sum = sum(RH) for all the pixels with flare detection.

Within the calibration framework, we employed a linear regression model that establishes a relationship between the volumetric flowrate Q and the VIIRS-estimated radiative heat:

$${\overline{RH}}_i={b_0+b}_1\times Q_i+{\epsilon }_i$$

(4)

where i = 1 … n, and n is the test number, $\overline{RH}$ is one of the RH_max, RH_50%, RH_75%, or RH_sum, and the residuals ε_i are the regression errors. The magnitude of the regression errors and adherence to statistical assumptions are fundamental for determining the precision of the estimated slope (b₁) and the intercept (b0) for the reliability of any probabilistic statements (p-values) made about the regression model, as well as the proportion of variability explained (R²). Beyond individual coefficient significance and explained variance, the F-statistic (a ratio of explained variance to unexplained variance) provides a crucial overall assessment of the regression model’s utility. It tests the null hypothesis that the entire model explains no significant variation in the dependent variable. A large F-statistic and a small p-value suggest that the model provides a significantly better fit than a model with no predictors. Results of the ordinary least squares fit (OLS) in linear regression are listed in

Table 2. Results of the OLS linear regression fit between RH and Q.

	RHmax	RH75%	RH50%	RH25%	RHsum
R2	0.856	0.93	0.924	0.853	0.866
F-statistics	191	426	388	185	207
b0	−0.363	−1.38	−1.13	−1.57	−1.21
p-value for b0	0.597	0.128	0.226	0.347	0.401
b1	78.078	85.22	84.61	93.837	96.387
p-value for b1	6.45 × 10−13	2.71 × 10−19	1.03 × 10−18	1.32 × 10−14	2.1 × 10−13

.

For all the RH options in Table 2, p-values for intercept b₀ consistently exceed 0.05, indicating that the intercept is zero with greater than 95% significance. This observation aligns with the theoretical prediction that zero RH corresponds to zero flared gas volume. Consequently, regression through the origin (RTO) approach is appropriate for the calibration function.

The RH_75% option yielded the highest R² value of 0.93 indicating that the linear regression model explains 93% of the variance when the sum of RH values for pixels exceeding 75% of the maximum RH_max is used to estimate RH. The 75% threshold appeared to strike an optimal balance (for calibration purposes) between two opposing reasons for multi-pixel detections:

• Split of the signal from a flare between neighboring pixels due to the overlap of the pixel footprints, especially in the bowtie regions at the edge of scan.
• Atmospheric scattering, which adds a blurry multiple-pixel halo around a subpixel infrared emitter.

The regression model for RH_max exhibited the lowest R² value of 0.755, attributable to its inability to capture the multi-pixel cluster of detections associated with large flares. Figure 16 presents a comparison between the two regression models for RH_75% and RH_max as a function of flow rate Q. The transition from RH_max to RH_75% is particularly evident in the test corresponding to the highest flow rate on 5 February 2018, where the VNF detections are evenly distributed across two pixels within the same scan line.

4.2.2. Wind and Scanline Azimuth Effects

To investigate the potential cause of outliers, VNF pixels, we have plotted the residuals as a function of satellite zenith angle, wind direction, the number of pixels in the VNF detection cluster, and flare temperature. Satellite zenith angle influences the atmospheric transmissivity along the optical path from the satellite to the flare, as well as the visible source area. Transmissivity decreases as the zenith angle increases from nadir (0 degrees) to side view (70 degrees). The visible source area is contingent on the flare shape and orientation, but in most cases, it will increase from nadir to side view. The wind direction determines the flare’s tilt and rotation. The satellite observes the flare either from the east or west due to the VIIRS scans being approximately perpendicular to the satellite orbit. Consequently, the visible flare surface will be larger for flares tilted along the north/south wind directions compared to east/west winds. The pixel count in the cluster of VNF detections atop the flare is dependent on the flare size and atmospheric glow. According to Stefan–Boltzmann law, the satellite-observed flare RH rises with the temperature to the fourth power.

Regression residual plots of RH_75% against the fitted values and the test numbers are depicted in Figure 17. The merger of RH adjacent pixels with the 75% RH rule resolves an outlier from the 75,000 lb/h test set. Yet several outlier VNF pixels remain, which we identified using statistical methods [33]. Outliers are defined as test cases exhibiting residuals exceeding three scaled median absolute deviations (MAD) from the median.

The remaining VNF outliers originate from medium and large flares in the final test set from August 2019. They are depicted in red in Figure 18. To investigate the potential cause of these outliers, we have plotted the residuals as a function of satellite zenith angle, wind direction, the number of pixels in the VNF detection cluster, and flare temperature. The satellite zenith angle affects the atmospheric transmissivity along the optical path from the satellite to the flare, as well as the visible source area. Transmissivity decreases as the zenith angle increases from nadir (0 degrees) to side view (70 degrees). The visible source area is contingent on the flare shape and orientation; however, in most cases, it will expand from a nadir to a side view. The wind direction determines the flare’s tilt and rotation. The satellite observes the flare from either the east or west due to the VIIRS scans being approximately perpendicular to the satellite’s orbit. Consequently, the visible flare surface will be larger for flares tilted along the north/south wind directions compared to east/west winds. The pixel count in the cluster of VNF detections atop the flare is dependent on the flare size and atmospheric glow. According to the Stefan–Boltzmann law, the satellite-observed flare RH is influenced by the temperature derived from the Planck curve fit to the flare spectrum in the fourth power.

Examination of the residual plots in Figure 18 fails to identify a definitive explanation for the substantial deviations from the fitted line observed in the outlier tests numbered 22, 23, 27, 30, and 31 in Figure 18. While satellite observations for the first four outliers were acquired at large zenith angles, the outlier in test 31 was detected at nadir. The absence of a consistent underlying cause across these cases suggests that the model is not missing a major, systematic hidden variable. Instead, these isolated outliers likely represent random, unmodeled noise or unique measurement artifacts that are not indicative of a broader underlying relationship being missed by the current model.

4.3. Comparison of Near-Field and Satellite Observed Planck Curves

Near-field SWIR spectrum of the large flare on 12 January 2018 was collected with the ASD FieldSpec spectrometer following the setup shown in Figure 11. Collected and calibrated near-field irradiance spectrum is plotted as a black line in Figure 19. Center wavelengths of the VIIRS spectral bands DNB, M7, M8, M10, and M11 are shown with vertical dashed lines. To simulate near-field response of the VIIRS instrument we convolved the observed ASD spectrum with the pre-launch relative spectral response (RSR) curves. The predicted VIIRS radiances in the near-field are shown in Figure 19 with red triangles. One can see that they are located between the spectral features of water vapor and closely fit the black-body spectrum (Planck curve) with the temperature 1695°K (solid red line). That is 100 degrees less than the Nightfire temperature derived from the far-field spectrum. To compare near- and far-field VIIRS responses, we have plotted relative values of the radiances observed from space with red circles. There is a high degree of correspondence between the Planck curves derived from the near-field spectrum and those obtained using the VNF algorithm.

4.4. John Zink Flared Gas Volume Calibrations

We derived both VNF v3 and VNF v4 calibrations for estimating flared gas volumes based on the temporally matched metered flow rates and the matched VNF pixels. The 75% RH option was used to merge multiple adjacent pixels having nearly equal radiant output. Calibrations options for VNF v3 RH and RH’ are shown in Figure 20. Calibration options for VNF v4 RH and RH’ are shown in Figure 21.

4.5. Comparison of the Calibrations and Detection Limit Analysis from Upstream Metered Flares

4.5.1. Detection Limit Analysis

To estimate the minimum gas flare flow rate at which the VIIRS Nightfire (VNF) algorithm reliably detects thermal emissions, we employed the Receiver Operating Characteristic (ROC) method [35]. The method sets the detection limit based on metered flared gas volumes for VNF-detected events, in contrast to the non-detected event set. Because the ROC makes no assumptions about the functional form of the detection response, it accommodates the variability inherent in satellite observations, such as changes in viewing geometry, atmospheric conditions, and flare characteristics.

A total of 877 VNF v3 detection matches were found using the flare data from the Arabian Peninsula, which has one-minute metering. The offshore OGCI contributed metered flare temporal profiles yielded 58 VNF matches. Figure 22 shows the result of the detection limit analysis, which combines data from four contributed flare datasets. The metered flared gas volumes are plotted on the x-axis, and the tally of VIIRS events is plotted on the y-axis. The VNF detection limit is estimated to be 0.008 mmscmd at the 95% detection probability level. This estimate is supported by the Tulsa experiment, where the same minimal flare flow rate was detected.

4.5.2. Calibration Comparisons

Figure 23 shows the comparison of the Arabian Peninsula metered flow rates (x-axis) versus VNF v. three estimates of the flared gas volumes for the JZ RH, Cedigaz, and RH’ calibrations. The diagonal lines indicate the perfect match between metered and VNF-estimated flared gas volumes. Note that the JZ RH calibration (C) consistently overestimates the flared gas volumes, with the VNF pixel cluster riding above the diagonal. This effect is moderated by the Cedigaz calibration (B), with minor flare VNF clusters positioned closer to the diagonal line. The best match between the metered flared gas volumes and VNF estimated volumes is with the JZ RH calibration (A).

The results from the 58 VNF pixels from a Gulf of Mexico flare are shown in Figure 24, with metered flared gas volumes on the x-axis and VNF estimated flared gas volumes for the calibration options from VNF v3 and v4. The scattergrams show that the best match between VNF flared gas volume estimates and the metered data is with the JZ RH calibration option. Both the Cedigaz and JZ RH calibration options overestimate flared gas volumes for minor flares and underestimate flared gas volumes for the largest flares.

4.6. Comparison of Results from a Single Day

Global M10 Transmissivity Grid

VNF v4 calculates atmospheric transmissivity for each detected flare in each of the VIIRS M bands. Since flare radiant output peaks in the SWIR band 10 (1.63 μm) [1,18], we focus on M10 to examine the impacts of atmospheric correction on VNF flared gas volume estimates. The two VIIRS SWIR bands are positioned in atmospheric windows that are particularly valuable for seeing-to-the-ground observations. Figure 25 shows the M10 transmissivity at half-degree resolution, generated using MODTRAN, for 1 September 2023. Overlain are shorelines and the location of upstream flares detected in 2023. Visual examination of Figure 25 reveals substantial variability in transmissivity. Figure 26 displays a global histogram of M10 transmissivities, as shown in Figure 25. The features in the histogram correspond to the variability observed in Figure 25. Figure 26 has two vertical lines at transmissivities of 0.9 and 0.95, dividing the world into three transmissivity categories: high (0.95–0.97), medium (0.9–0.95), and low (under 0.9). Figure 27 shows the M10 histogram for upstream flares identified in 2023, using the same vertical lines as in Figure 26. While the majority of upstream flaring sites are located in zones with M10 transmissivities of 0.9 or higher, 23% of the flares are in zones with transmissivities of less than 0.9. Flares having low atmospheric transmissivity are typically in humid tropical regions.

VNF v3 and v4 calibration options were applied to test date VNF pixels from 2278 upstream flares for intercomparison of the estimated flared gas volumes in terms of mmscmd. Scattergrams of VNF RH-based flared gas volume estimates versus VNF v3 RH’, VNF v3 Cedigaz, and VNF v.4 RH’ are shown in Figure 28. This includes v3 RH versus v3 RH’ estimates (panel A), v3 RH versus Cedigaz estimates (panel B), and VNF v4 RH versus v4 RH’ in panel C. The diagonal lines on the three panels indicate equal mmscmd values. The data clouds do not align with the diagonal lines, indicating biases in the RH-derived calibration options when compared to the RH-only options. In all three cases, the RH-based calibrations yield higher flared gas volumes for small flares and lower flared gas volumes for larger flares. The tipping point between the RH and RH’ is near 0.3 mmscmd.

5. Discussion

5.1. Selection of RH-Based VNF v3 and v4 Calibrations

This is the first study to critically examine the inclusion of an exponent (0.7) applied to the source area term in the calculation of radiant heat (RH’) while estimating flared gas volumes. The exponent was introduced to compensate for the difference between the three-dimensional volume of flares versus the two-dimensional projection afforded by the Stefan–Boltzmann Law. The exponent is used in the Cedigaz calibration and was also included as an option in the John Zink calibration, referred to as the JZ-RH’ calibration.

Intercomparison of the calibration options reveals the unintended consequence of the exponent. Including the exponent, reduced flared gas volumes are calculated for large flares, whereas conversely, they increase the flared gas volumes calculated for smaller flares. This effect is observed in five locations within our study. First, in the Arabian Peninsula metered gas flare (Figure 23), second in the Gulf of Mexico (Figure 24), third in the whole day analysis (Figure 28), and finally in the global annual flared gas volume estimates. As per Zipf’s Law [36], there are many more small flares than large flares. Thus, the overestimation of flared gas volumes for small flares propagates into annual global flared gas volume estimates and country-level estimates of flared gas volumes. Based on these findings, in the current year (2025), EOG will calculate flared gas volumes using the RH-based calibrations from VNF v3 and v4, as well as the Cedigaz calibration from VNF v3. We plan to transition to the John Zink calibration entirely in 2026.

5.2. Single and Double Flares

The John Zink flare tests include 24 single flares and twelve tests with two flares 100 m apart. The dual flare test pixels plot in line with data from the single flare tests (Figure 20 and Figure 21). This indicates that radiant output from multiple flares combines additively in nighttime VIIRS data. We also found it is not possible to distinguish various flares that are closely spaced due to the footprint size of the VIIRS M bands (Figure 13). This finding is important as many flaring sites include multiple closely spaced flares.

5.3. Combining Radiant Heat from Multiple Adjacent Pixels

The original Cedigaz-based method for estimating flared gas volumes was based on the radiant heat from local maxima pixels. We found evidence in the local maxima outlier examination that in some cases, the luminous output from a single flare (or closely spaced flare cluster) is spread across multiple pixels. In some cases, there are low RH VNF pixels surrounding the local maximum, interpreted as glow induced by atmospheric scatter. The testing to explore the value of merging multiple pixels included three cases: (1) local maximum only, (2) all VNF pixels including glow, and (3) merging of VNF pixels that are within a set percentage of the local maximum RH. Examination of the merger percentage options of 25%, 50%, and 75% of the local maximum RH. Setting the merge threshold to 75% of the local maximum RH resulted in a higher regression coefficient. These results indicate that radiant heat from multiple pixels should be merged when adjacent pixels have 75% or more of the local maximum RH. In addition, glow should be excluded when calculating flared gas volumes.

5.4. Examination of Atmospheric Corrections

The John Zink test flare facility is situated in a mid-continent location with native grasslands, presenting a limited set of atmospheric conditions. VNF v4 includes an atmospheric correction that adjusts radiances upward based on band-specific transmissivities calculated using MODTRAN. Atmospheric transmissivities vary between the VIIRS spectral bands and also temporally. Figure 29 shows the effect of M10 atmospheric transmissivity on estimates of flared gas volumes from the VNF v3 RH versus VNF v4 RH calibration for 2278 upstream flare pixels from 1 September 2023. At high transmissivity levels, v3 and v4 flared gas volume estimates are pretty similar. However, the two diverge as transmissivity drops. The increase in VNF v4 mmscmd estimates tracks the decrease in M10 transmissivity (Figure 30).

5.5. Aggregating Flared Gas Volume Estimates from Instantaneous to Annual Data

EOG’s VNF product produces nightly surveys of flare locations and instantaneous flared gas volume estimates. The nightly observations of individual flaring sites are compiled into temporal profiles, from which annual flared gas volume estimates are aggregated by statistical averaging. Only observations deemed to be clear are considered when estimating annual flared gas volumes. That is because heavy clouds block flare detections, and radiant heat dims when the VNF detection is through optically thin clouds. The yearly aggregation method sums the clear sky radiant heat values. It is divided by the number of precise observations in a year to generate an annual average RH, which is converted to a flared gas volume. Clear-sky satellite overpasses with no VNF detection are included in the average as zero RH values, under the assumption that either the flare was inactive or below the VNF detection threshold.

There are two weaknesses in this approach that work in concert:

• Temporal under-sampling: The VIIRS sensor integration time period for individual M-band pixels ranges from 2–3 milliseconds [37], and the VIIRS collections occur during a three-hour time slot from midnight to three in the morning. As a result, the VNF product does not capture short-term fluctuations or diurnal variability in flaring activity. This limitation can lead to errors in estimating annual flared volumes in cases where transient flaring spikes occur outside the brief observation period afforded by VIIRS. Improved characterization of flaring activity could be achieved by integrating data from multiple VIIRS sensors, incorporating daytime thermal detections (e.g., from Landsat), and applying advanced statistical models to account for observed flare variability.
• VNF detection limits: Radiant heat emissions from low-intensity or intermittent flares that fall below the VNF detection threshold (0.008 mmscmd) are recorded as zero in the annual summation of RH. To the extent that this happens, there is a systematic underestimation of flared volumes. This is likely the case in shale basins (e.g., the Bakken and Permian formations) where small-scale episodic flaring is prevalent. We are currently researching the extension of the flare detection limit for small flares using the low-light imaging day/night band (DNB). It is well established that in the 1200–2500 K range, the DNB detection limit drops far below that of the M-bands used in VNF [38]. The DNB is famously known for its role in the detection of electric lighting [39]. We plan to utilize the DNB radiance when flares are detected by VIIRS to define a trajectory, estimating flare temperature and flared gas volume from the DNB in observations lacking VNF, even in the presence of electric lighting at a facility. This approach was used to distinguish low-level flaring versus flare outages in a 2018 study [40].

5.6. Impact of Fuel Composition on Spectral Output and Heat Release

While the Tulsa natural gas used in the flare tests was not pure methane, the presence of 5% ethane has negligible effects on the heat output of the tests. The combustion heat content of methane (single carbon) is 890.7 kJ/mol. The combustion heat content of ethane (double carbon) is 1560.7 kJ/mol [41]. When normalized to the number of carbon atoms, ethane combustion heat content is 14% lower than methane. Since ethane is 5% of Tulsa’s natural gas, the overall heat output from the flare is diminished by 0.7%, as compared to pure methane. The units of the VNF flared gas volume estimates are “methane equivalents” in order to standardize the estimates to methane, the predominant molecule in upstream natures gas flares. The precise composition of gases being flared is rarely known.

6. Conclusions

Frequent global monitoring of individual natural gas flaring provides data for calculating greenhouse gas emissions to the atmosphere, monitoring progress toward the elimination of routine flaring, prospecting for sites that may be suitable for recovery and utilization of underutilized methane, calculation of the carbon intensity of marketed fuels, and as tip-offs regarding flaring gaps in satellite sensor collection planning in the search for methane venting. It would be impossible to conduct frequent and consistent monitoring of flares worldwide with ground-based or airborne methods. This leaves satellite sensors as the only viable approach for global flare monitoring. Nighttime data collected by the NASA/NOAA VIIRS sensors are particularly valuable for the worldwide detection and tracking of flares, as data are collected nightly and include observations from four daytime near-infrared and shortwave infrared spectral bands, which cover the peak radiant emissions of flares and are free from solar contamination. The VNF algorithm runs a separate detector for the spectral bands M7, M8, M10, and M11, as well as M12–13, in pairs. Detection radiances are analyzed with physical laws to derive temperature, source area, and radiant heat, key variables for long-term flare monitoring.

In 2015, EOG developed a calibration for estimating flared gas volumes based on annual Cedigaz reporting of country-level flaring and venting. The calibration indexes were used to convert flared gas volumes to heat output, calculated from top-of-atmosphere VIIRS M-band radiances. The Cedigaz calibration includes an exponent of 0.7 applied to the source area term before calculating heat output using the Stefan–Boltzmann Law. This heat output variant is referred to as RH’ and was implemented to compensate for the difference between the flare’s three-dimensional state and the two-dimensional area term coming from Planck’s Law. The Cedigaz calibration has never been fully satisfactory due to its wide error bars, lack of ground-based verification of the slope, and the untested inclusion of an exponent. In this study, we have developed a pair of empirical flared gas volume calibrations based on nighttime VIIRS observations of test flares running at high, medium, and low flow rates. Testing was performed at the John Zink (JZ) test flare facility, which consisted of 24 single flares and 12 dual flares, with flare stacks spaced 100 m apart. The new calibration represents a significant advancement in satellite estimation of natural gas flaring volumes, with error bars reduced by a factor of 70 compared to the Cedigaz calibration.

EOG produces two distinct versions of VNF from the same source data provided by NOAA. VNF v3 is based on the top-of-atmosphere radiances in the VIIRS M-bands with no atmospheric calibration. VNF v4 is processed from atmospherically corrected radiances, where radiances are adjusted upwards to compensate for losses passing through the atmosphere (transmissivity). VNF v4 is unique in its subpixel analysis of primary (hotter) and secondary (cooler) infrared emitters. For v4 calibration development, only the primary emitter is included. For VNF v3 and v4, calibration options were developed for both RH and RH’. The Cedigaz calibration has no equivalent in the atmospherically corrected VNF v4. During the transition period v3 and v4 will be processed in parallel, allowing direct comparisons and continuity checks, particularly under low-opacity conditions.

The JZ tests confirm that, in some cases, radiant heat is spread relatively evenly across multiple pixels. This can occur due to the partial overlap of pixel footprints and the presence of closely spaced flares. The Cedigaz calibration is based solely on local maxima, with adjacent VNF pixels excluded. The JZ tests prove that the local maxima will underestimate flared gas volumes in cases where the local maximum pixel has one or more neighboring pixels with near-equal radiance and a substantial signal. Testing indicates that flared gas volume accuracies improve if temperatures and source areas from local maxima adjacent pixels are used, provided the RH of neighboring pixels is 75% or higher of the local maximum. This merger of near-equal RH pixels is only applied for adjunct pixels within the same 16-line VIIRS scan. Including all VNF-detected pixels, including dim glow, reduces the accuracy of the flared gas volume estimates. The present calibration with the 75% multiple pixel selection threshold is derived from low-wind, limited-coverage data. Its stability under high-wind and bow-tie conditions will be re-evaluated as larger metered datasets become available.

For intercomparison of the methods, flared gas volumes were calculated from the five calibration options for four types of test sets: (1) two contributed high temporal resolution metered flares, (2) a full day of pixel matched VNF v3 and v4 from upstream flaring sites, (3) VNF v3 annual global flaring, and (4) the top 19 flaring countries from 2012–2023. From the intercomparison of the five options, we conclude that flared gas volumes are better calculated from RH, without the 0.7 exponent. The inclusion of the exponent results in an underestimation of flared gas volumes for large flares and, conversely, an overestimation of flared gas volume estimates for small flares. This pattern is present in all four types of test data.

The single versus dual flares tests demonstrated that the radiant energy from multiple closely spaced flares combines additively. Inspection of high spatial resolution satellite data from Google Earth and other sources indicates that it is not uncommon for various flares to be present with individual pixel footprints or sets of adjacent pixels. This means that the coarse spatial resolution of VIIRS should not be relied upon for estimating flare numbers. However, the good news is that total flared gas volumes from multiple closely spaced flares can be calculated with reasonable confidence.

One of the primary objectives in developing an empirically based flared gas volume calibration has been to reduce the error bars, expressed as the prediction interval. For this, we report great success, with prediction intervals 70 times smaller than those from the Cedigaz calibration.

From 2012 to 2024, EOG produced annual flared gas volume estimates for individual flares based on the Cedigaz calibration. One of the primary conclusions of the study is that the inclusion of the 0.7 exponent on the source area has the unintended consequence of underestimating flared gas volumes for large flares and overestimating them for small flares. The second significant finding is that including an atmospheric correction is crucial for accurate estimation of flared gas volumes in humid tropical areas, such as Venezuela, Indonesia, and Nigeria. The third significant finding is that the VIIRS day/night band (DNB) should be further investigated for detecting flared gas volumes below the VNF detection limit. An initial examination of the DNB’s ability to estimate flare gas volumes below the VNF detection limit will be performed with the existing contributed temporal profiles of metered upstream flares. Additional temporal records of low-level flaring will be sought to develop a DNB calibration for estimating flared gas volumes below the VNF detection limit. Such a DNB calibration should include a mechanism for ignoring radiance contributed by electric lighting co-located with low-level flaring.

Based on these findings, the entire set of VIIRS nighttime flare records, dating back to 2012, will be reprocessed using the RH formulation and the VNF v4 algorithm. Fortunately, this can be done from the VNF database, which records the DNB M-band radiances, location, date, and time for individual pixels. We plan to continue producing Cedigaz estimates of flared gas volumes for several years, allowing the user community to transition to the RH-based flared gas volume estimates.