全球变化知识资源中心

Methane (CH₄) is one of the most important greenhouse gases, and an important energy carrier in biogas and natural gas. Its large-scale emission patterns have been unpredictable and the source and sink distributions are poorly constrained. Remote assessment of CH₄ with high sensitivity at a m² spatial resolution would allow detailed mapping of the near-ground distribution and anthropogenic sources in landscapes but has hitherto not been possible. Here we show that CH₄ gradients can be imaged on the 2 scale at ambient levels (~1.8 ppm) and filmed using optimized infrared (IR) hyperspectral imaging. Our approach allows both spectroscopic confirmation and quantification for all pixels in an imaged scene simultaneously. It also has the ability to map fluxes for dynamic scenes. This approach to mapping boundary layer CH₄ offers a unique potential way to improve knowledge about greenhouse gases in landscapes and a step towards resolving source–sink attribution and scaling issues.

Identifying sources and sinks of CH₄ and comparing their relative magnitudes in landscapes is challenging but important. CH₄ is the second most important greenhouse gas at a 100-year perspective¹ and has a high value for society as an energy source. Atmospheric levels of CH₄ have increased 2.5-fold since 1750 (ref. 1) but the reasons for this increase are not as clear as for carbon dioxide (CO₂). For example, although atmospheric CO₂ levels have increased steadily, the accumulation rate of CH₄ has varied for unknown reasons². Suggested explanations that are based on the balance of emissions from fossil fuels and wetlands² are difficult to verify and alternative explanations cannot be excluded because sources and sinks are too poorly constrained.

CH₄ is produced by methanogenic archaea in anaerobic systems including sediments and water-saturated soils, gastrointestinal systems of animals, biogas production and waste management systems^{2, 3}. CH₄ is also released from natural gas handling and combustion processes². The major sinks are believed to be atmospheric oxidation and microbial oxidation in soils, sediments, and water^{2, 3}. Wetland plants, or bubbling through shallow inland waters, function as gas conduits from anaerobic sediments. Similarly, there are also hot spot sources in agriculture (for example, rice paddies, waste lagoons, and ruminants), and industrial and urban environments (combustion and gas distribution leaks). Many if not most large sources, both natural and anthropogenic, are confined to local sites with a patchy distribution across landscapes. The sinks may also be scattered in the landscape on the basis of, for example, local moisture levels in soils. Because of the difficulty in quantitatively assessing the spatial variability of sources and sinks our current knowledge is probably biased and incomplete.

A fundamental limitation in our ability to identify and compare CH₄ sources and sinks is related to the spatial scales of available measurement techniques. Bottom-up methods often rely on flux chamber or point concentration measurements. Flux chamber measurements have a well-defined but very small footprint (typically sub-m²) and cannot easily be used to cover larger areas. High- frequency measurements can be obtained by eddy covariance (EC) and gradient-based flux assessments with larger footprints at ha to km² scales (ref. 4), but specific sources and sinks within the footprint cannot be resolved. EC and gradient flux footprints are based on statistical probability distributions, vary over time, and lack verifiable boundaries. A less common approach is the backward Lagrangian stochastic (bLs) technique⁵ which uses a laser and a reflector for each line of sight and can be used to locate a point source (or several sources depending on the number of lines of sight used) or estimate emission rates through dispersion model predictions. The dispersion models also have footprint uncertainties and specific infrastructure is required for each line of sight (such as the laser source and reflector), which limits the spatial distribution of the measurements.

Several satellites have been or are now mapping CH₄ on a global to regional scale, including SCIAMACHY (ref. 6), GOSAT (ref. 7), AIRS (ref. 8), IASI/AMSU (ref. 9), and the planned CarbonSat (ref. 10) and GRIPS (ref. 11), all having km-scale spatial resolutions. Satellites are very useful for their spatial coverage and have been successfully used in many projects for following regional patterns^{12, 13, 14, 15}, but two drawbacks are the low spatial resolution and difficulties in resolving CH₄ at the surface–atmosphere boundary layer where the source/sink patterns are revealed. A recent example is the four corners CH₄ hotspot, a 6,500 km² coal mining area in the US emitting enough CH₄ to be seen from space but still measuring only a few pixels in SCIAMACHY images¹⁶. Although successfully mapping atmospheric CH₄ content, the large pixel sizes limit our ability to link CH₄ levels to environmental drivers that may differ between different types of environments/land use. Remote sensing of CH₄ from aircraft is also in development. Examples are AirGRIPS (gas-filter correlation radiometer) and the MaMap spectrometer¹⁷, giving a resolution of 33 × 23 m at 1,000 m altitude. Higher-resolution (several m²) measurements of strong CH₄ sources have been made in both the shortwave^{18, 19} and thermal IR (ref. 20) from high altitudes, representing important progress. However, a technique with the ability to map lower levels of near-ground CH₄ in landscapes at very high spatial resolution (sub-m²), having a high enough spectral resolution to ensure separation of CH₄ from other gases, yet with good spatial and temporal coverage, and the ability to measure flow velocities directly from high-speed imaging, would substantially increase our capacity to identify, resolve, and compare different sources and sinks. In turn, this would lead to new possibilities for understanding the variability of CH₄ in the atmosphere, detecting and minimizing CH₄ leakage or emissions from anthropogenic processes, and also validating how environmental change (for example, land-use and climate change) affect CH₄ source attribution in climate models. We here present a new and generally applicable technique based on thermal IR hyperspectral imaging for landscapes that does not require a priori knowledge of source localization.

Hyperspectral imaging can be described as imaging that records a spectrum for each pixel simultaneously in a scene. Thereby a three-dimensional data cube is generated for each imaging sequence with two dimensions defining the scene spatially and a third dimension holding the spectral information (Supplementary Fig. 1). In the thermal IR part of the electromagnetic spectrum this is often a passive technique with patterns of absorption or emission lines, depending on whether the background is hotter or colder than the gas (the temperature contrast). Those patterns can be used as fingerprints of individual chemical compounds in the line of sight, making assessment of concentration gradients possible (Supplementary Fig. 2).

Our hyperspectral camera, described in the Supplementary Methods, was developed for optimized detection of CH₄ (7.7 μm band) allowing more sensitive quantification of concentration gradients than were previously avilable. Using all the spectra in a data cube and spectroscopic radiative transfer modelling at a high spectral resolution (0.25 or 1 cm⁻¹) a time-averaged CH₄ image can be calculated pixel-by-pixel (described in Supplementary Methods). The high imaging frequency of the camera during data acquisition (images including all spectral lines) also made it possible to separately derive air motion from H₂O and CH₄ motion and to construct air flow movies. In cases with high enough CH₄ fluxes and low humidity the CH₄ can be followed directly, while water vapour can be used as the air flow tracer in other cases. The presented technique can therefore generate sensitive static spectra by aggregating information over the image collection time (0.25–2 min per cube) to detect and measure the average amount of CH₄ with high precision, as well as construct air flow movies. By combining the gradients in CH₄ levels quantified from spectra with the information about net air movement, corresponding average CH₄ fluxes during the image collection period can then be assessed. Thus not only concentration gradients but also fluxes can be calculated from the obtained images and spectra for both hot-spot and diffuse emission sources.

After testing the system extensively in the lab, several successful field measurements were made. Below we present CH₄ images of different environments to demonstrate the ability of the system to remotely map CH₄ at ambient levels (~1.8 ppm; parts per million by volume) under field conditions (typically a few °C background-gas temperature contrast). We highlight examples of CH₄ mapping, showing the ability to map concentration gradients, find emission hot-spots in the landscape, and quantify CH₄ fluxes. A summary of the scenes and measured CH₄ fluxes are given in Table 1, including a comparison with typical fluxes found in the literature where available. In situ measurements for comparisons were also made using an infrared Off-Axis Integrated Cavity Output Spectroscopy (OA-ICOS) greenhouse gas analyser (Los Gatos Research, DLT 100 or UGGA).

Full table

The first example shows a controlled release of CH₄ on the lawn outside our lab (Fig. 1) at 45 ml s⁻¹ (100 ± 10 g h⁻¹; measured manually with a volumetric bubble flow meter). The spectroscopically calculated CH₄ image was made from one cube (200 × 90 pixels at 1 cm⁻¹ spectral resolution) and shows our CH₄ source and its surrounding average distribution (Fig. 1a). The air–background temperature contrast was found from the spectra to be in the range of 5–15 °C (air 19 °C, and the wall 24–34 °C) with a background distance of 50 m. By relating the mapped CH₄ from spectroscopy with the air flow from the large number of individual IR images collected over time during the imaging (in this scene 245 Hz, 6,320 images in 25.8 s) we could quantify and follow the CH₄ flow at a high temporal resolution (Fig. 1b and Supplementary Movie 1). Using 10 cubes, a spectroscopically modelled average CH₄ distribution, and detailed high-frequency imaging of the CH₄ motion (wind speeds 0.1–0.3 m s⁻¹) we calculated a flux of 102.9 ± 5.8 g h⁻¹. For a second test with a smaller release of 10 ml s⁻¹ (measured to 23 ± 2.3 g h⁻¹ by the volumetric gas flow meter) and higher wind speeds (0.8–2.7 m s⁻¹) we calculated a flux of 25.3 ± 2.8 g h⁻¹ using data from the camera. Thus, these controlled fluxes could be accurately determined with the camera.

Column densities of CH₄ above the ambient level are shown using different shades of red, overlaid on a thermal IR image. a, Average distribution (one cube, acquisition time 25.8 s). b, Snapshot (40 out of 6,320 images in the cube, integration time 0.16 s).

The second example is a barn with 18 cows inside. A plume of CH₄- rich air from the ventilation outlet was clearly revealed in our spectroscopic CH₄ image (Fig. 2), representing the average flow. The temperature difference was 9 °C between the plume and the background wall. The spectroscopic modelling (see Supplementary Methods) uses four layers: reflected cold sky off the building, heat radiation from the building itself (−1 to +1 °C), the plume (+8.8 °C at the outlet), and a layer of cold air between the plume and the camera (−2.9 °C) that contains CH₄ and H₂O at air temperature. Example spectra (spectral resolution 1 cm⁻¹) used for quantification of CH₄ within and outside the vent plume (points 1 and 2 in Fig. 2) are shown (Fig. 3) together with the corresponding spectroscopic model fits. In situ measurements in the outlet flow (location 3 in Fig. 2) of 15 ppm are in agreement with our column measurements (~30 ppm m/2 m thick plume = ~15 ppm). The imaging frequency of the camera in this scene was 63.8 Hz (6,320 images in 99 s, 320 × 256 pixels), with a high enough signal-to-noise ratio (S/N) to follow the flow at a high temporal resolution (Supplementary Movie 2). From structures in the plume, the flow speed was found to be 1.02 m s⁻¹ and combined with concentration and flux data generated over 26 min (16 cubes) our models gave a CH₄ flux of 43 mg s⁻¹ (160 ± 5 g h⁻¹). This is equivalent to 77.8 ± 2.1 kg yr⁻¹ cow⁻¹ if a constant flux is assumed. In a recent review²¹ it was stated that a cow emits 67.5–98.6 kg CH₄ yr⁻¹, depending on grazing, in agreement with our measurements.

The multilayered spectroscopic model gives CH₄ column densities (red) above the ambient level in the outlet flow. Crosses 1 and 2 mark locations for the example spectra in Fig. 3, and cross 3 marks the position where in situ measurements were made for validation.

As another example of the detection and quantification of anthropogenic CH₄ emissions, we mapped the chimney exhaust of a waste incineration plant (Fig. 4) using a previously published method for obtaining chimney mass flow rates^{22, 23}, with the cold sky as the background and a spectral resolution of 0.25 cm⁻¹ from a distance of 183 m. The cold sky had a radiation temperature between −55 °C and the air temperature (8.5 °C) depending on the wavelength and zenith angle, and the gas plume had an average temperature of +32.9 °C close to the outlet (measured from its water content, compensating for the CH₄ content in the air in front of and behind the plume). Three layers were used in the data modelling (cold sky, hot plume, and cold air in the foreground). The vertical flow velocity in the plume (2.50 m s⁻¹) was measured from flow structures using the IR images collected at a rate of 473 Hz. The average wind speed of 1.9 m s⁻¹ did not affect flow velocity estimates as the flow velocity is still vertical close to the chimney. With information on both the average CH₄ column density profile and the flow velocity in the plume using 40 cubes (14.3 min), a CH₄ flux of 696 ± 38 g h⁻¹ could be calculated, corresponding to 6.1 ± 0.33 t y⁻¹ if a constant flux is assumed. Such stand-off flux assessments could represent a breakthrough in cases such as this, where accurate flux measurements can be difficult to perform with traditional methods. It should also be noted that CH₄ fluxes from incineration or industrial combustion processes are often neglected or considered to be negligible²⁴, while our measurements showed that this is not the case.

a, Thermal image of chimney and plume (mostly H₂O) with the cross-section in b marked by a red line. b, Measured CH₄ flux in the cross section and the total flux.

As an example of mapping CH₄ gradients and fluxes in a scene with temperature differences less than a few °C and low–medium concentrations (ambient ~2 ppm to ~10 ppm mixing ratios) we mapped CH₄ around a sewage sludge deposit (Fig. 5). The regions shown as dark red in the lower part of the image have the highest mixing ratios because the lines of sight are directly towards the sludge deposit, while lower average mixing ratios are found along the lines of sight further up in the map, almost reaching the ambient level towards the distant trees (the ambient level was found by measuring a scene in a different direction that was unaffected by the deposits). The average wind speed was 1.7 m s⁻¹ from left to right in Fig. 5. In situ point measurements over the nearby edge of the sludge deposit showed a mixing ratio of 8 ppm above ambient, which is in agreement with the mixing ratios in the CH₄ map (~7 ppm; Fig. 5; note that some deviation between point measurements and integrated line-of-sight values from the camera is expected). The map was made from 16 cubes (320 × 100 pixels) with a spectral resolution of 1 cm⁻¹. Using sev