Methodology

How do we estimate methane emissions?

Methane comes from many activities, and there is no single survey which allows us to calculate how much comes from each activity in a particular grid area, or grid cell. Instead, we accumulate information on how much an activity generates in a region (say Australia, or a state), then that emission is distributed among grid cells.

The national estimate [Australia's National Inventory of greenhouse gas emissions is published on the Department of Climate Change website] for each activity is, itself, indirect. Instead we estimate or count the amount of an activity taking place and multiply that by how much methane is emitted by each unit of activity.

Emission factors must be estimated or measured at one place and time, then applied elsewhere. Incorrect emission factors will lead to incorrect emission estimates. This is a major source of uncertainty in methane emissions calculated in this way.

Using Surrogates to Estimate Spatial Distribution

Sometimes we don't have spatial data for the activity we want, but can measure a good surrogate.

For example, the amount of fossil fuel burnt in a region is calculated by how much is sold to end-users. This can be established by sales statistics which are often well known, but where people use the fossil fuel is not.

Previous research has shown that fossil fuel use is predicted well by the intensity of nighttime lights that are observed by satellites. We can spatialise fossil fuel sources from different sectors according to the intensity of nighttime lights.

Calculating Confidence in the Emissions Inventory

We need to calculate the level of confidence we have in the emissions inventory. This informs the emissions update process (we will be more likely to make corrections where we least trust the inventory), but also helps us use the data to answer questions like the significance of detected trends.

Uncertainties in emissions come from uncertainties in activity data, emission factors, regional totals, the data we use to spatialise them, and the assumption that our spatial surrogate is accurate for the emissions sector.

Uncertainties in numbers like emission factors are particularly serious, since they may bias emissions in a region rather than a single point.

We usually express our confidence in an uncertainty, expressed as a standard deviation . We calculate the uncertainty for each point in the inventory by adding the uncertainties from each sector.

Data flow begins with global weather observations. Most large weather agencies generate short-term weather forecasts which they correct with satellite and surface observations. This combination of forecast and observation produces the best picture of wind, temperature and pressure in the atmosphere we have. Agencies run these forecasts globally on a mesh or a grid. The weather data is pixelated at the resolution of the mesh, so the finer the mesh the better our data.

Computational restrictions limit these global forecasts to about a 100km mesh, which is not fine enough for our needs. Because of this, we run a finer mesh weather model called the Weather Research and Forecast model (WRF) in our region of interest.

We need to know what either behaviour is coming from outside the area of our focus, so we use global weather models to feed information to the edges of our region.

The WRF model produces a snapshot of pressure, wind, temperature and moisture every few minutes. It is these wind patterns which move gases like methane around in the atmosphere.

Calculating a complete picture of the weather every few minutes is extremely expensive. Moving gases around in the atmosphere is simpler; we don't need to calculate rainfall or thunderstorms, for example. Because of this, we take the weather data from WRF and use it in an air-pollution model, the Community Multi-Scale Air Quality Model (CMAQ).

CMAQ can also calculate how pollutants are formed and destroyed by chemical processes in the atmosphere, though for our purposes these processes are not very important.

By integrating our Initial Estimate with atmospheric condition models, we create a map indicating expected concentrations of methane.

This visualisation reflects the amount of methane contained in each cubic metre of air across Australia, averaged through the atmosphere.

Overview

Satellites measure light or infrared radiation (heat). When they view the Earth, they see radiation that has passed through the atmosphere and may have bounced off the surface. 

Different gases in the atmosphere absorb different wavelengths (colours) of light. So if we know or can measure the amount of light at different wavelengths entering the atmosphere from the Sun, and can measure how much makes it back out of the atmosphere towards the satellite, we can calculate how much of the various gases occur in that atmosphere.  

This is a complicated measurement of a number of reasons:

• Firstly, the light may not make it all the way to the surface. If it bounces off a cloud, we may only measure the gas in part of the atmosphere. We usually don't know how high the cloud is, so we don't know how much of the atmosphere we're measuring and can't use these measurements.
• The light may pass through dust and smoke in the atmosphere. These also affect the way light travels. While we can correct for small amounts of these, if there is too much dust or smoke we cannot calculate the gas amount.
• The ground may also absorb different wavelengths differently. We can correct for this provided that we know enough about the surface, but this is not always true.
• The use of sunlight means we can only measure during the day and can't measure too close to the poles in the winter.
• If the Sun or the satellite are too low in the sky when viewed from the place the light strikes the ground, then the light is weaker and more likely to encounter clouds or dust. Measurements from SP5 work best in places with little cloud cover, not too close to the poles, and without lots of dust pollution. Australia fits these criteria well.

TROPOMI

Open Methane uses measurements from the Tropospheric Monitoring Instrument (TROPOMI) on board the Sentinel 5 Precursor (S5P) satellite. The troposphere is the lowest 80% of the atmosphere.

S5P orbits the Earth approximately every 90 minutes at a height of 824 kilometres. Its orbit is orientated almost North-South, and as such passes near both poles. Its orbit is structured so that the satellite stays fixed relative to the Sun, and the Earth rotates beneath it. This is called a sun-synchronous orbit, which means S5P crosses the equator at the same local time in every orbit, about 1:30PM travelling from South to North, and 1:30AM when travelling from north to south.

For half of each orbit, S5P is on the sunny side of the Earth and can take measurements, while on the dark side it cannot. The length of the orbit means there are about 16 orbits each day, but these deliberately do not repeat.

Detectors on satellites are similar to those found on digital cameras. One difference is that instead of capturing an image like a camera does, the satellite captures a single line, usually at right angles to the direction the satellite is travelling. Also, like a digital camera, the satellite takes time to capture enough light to produce useful data. All this time the satellite is travelling over the Earth's surface.

Currently, the satellite travels 5.6km across the surface of the Earth as it takes each measurement. Along the line of measurements (perpendicular to the satellite's path) the satellite designers must decide how much to zoom or pan the instrument. A zoomed-in image will have smaller pixels - and thus more detailed measurements - but with a smaller area, leaving parts of the Earth that may never be measured. TROPOMI's particular resolution and orbit mean that it potentially sees every point on the globe once per day.

Impact of Clouds

Earth is, on average, about 50% covered by cloud at any given time. Due to the impact of cloud cover on visibility, TROPOMI cannot make methane measurements for Open Methane in cloudy locations. This is particularly troublesome in some regions such as the tropics in the wet season, where it can stay cloudy for long periods.

Clouds are also generally not continuous, so TROPOMI can make patchy measurements through breaks in cloud cover. One trade-off of SP5's orbit and optical design is that TROPOMI often looks well off to the side of its orbital track to make a measurement. These measurements are at high light angles, so the chance of being intercepted by a cloud is increased.

TROPOMI also cannot make a measurement when there is too much smoke or dust in the atmosphere.TROPOMI also cannot make many measurements over water. For these reasons only a few percent of the total TROPOMI measurements are available, but this still leaves 40000 measurements available over Australia each day.

Detecting Plumes

When a pollutant like methane enters the atmosphere, it usually spreads horizontally and vertically in a characteristic plume.

A methane plume will gradually mix with plumes from other sources and the background of methane already in the atmosphere. Thus we would like to identify each plume at its most concentrated and distinct. Satellite instruments are good at this because they observe so many locations, but they suffer from measuring methane through the whole atmosphere. Thus even when measuring near the source of the plume it is diluted by the background of methane already present.

Dealing with uncertainty

Every measurement is uncertain. This is particularly clear with satellites where what they measure (radiation arriving at their detectors) is not what they wish to infer (e.g. methane concentration in the atmosphere several hundreds of kilometres beneath them). However, we can use an algorithm to link the two. This algorithm is based on well-understood physics, but, as already mentioned, other aspects of the atmosphere and surface can affect the satellite measurements.

The algorithm that calculates methane concentration also calculates an uncertainty on that measurement. We check both the measurements with a network of ground stations which can themselves be checked against absolute standards. This traceability of measurements to absolute standards is critical for confidence in detection, and for being able to use multiple different instruments.

Open Methane publishes notifications when measured methane concentrations are significantly different from what is expected. We calculate the expected value by combining our emissions inventory with the air pollution model. This creates a forecast of the expected methane concentration for every hour of every day.

By carefully sampling the model forecast in exactly the same way as the satellite samples the atmosphere, we can compare our expected value with what we measure. We can only do this where the satellite makes an observation, so much of Australia will be unavailable on a single day. It's important to remember that absence of data does not imply support for the emissions inventory, simply that we cannot observe it.

Care is also needed when deciding what is a significant difference between expected and measured values. Concentrations vary naturally as winds move methane from different sources around in the atmosphere. In principle, we can predict all these variations with CMAQ, but no model is perfect, so we will certainly fail to capture some variations. 

This threshold will be refined with experience. We also cross-check significant uncertainties for potential problems with the satellite data such as incorrect environmental factors. 

There is a difference between the resolution of the concentration map produced by CMAQ (25 x 25km) and the satellite observations (5.6 x 7.2km). This means we may see concentration events occurring at a particular part of a model pixel due to the location of an emission within the pixel. This is both useful and problematic. It is useful because it can give guidance on where to look for the source of an event. It is problematic because CMAQ can only relate average emissions in pixels to average concentration in pixels. 

If we measure an uneven distribution of satellite pixels within a CMAQ pixel, we could bias this average. This is a serious problem with networks of fixed stations, but we also need to account for it in satellite data. We include this as an extra uncertainty in how well we expect our model concentrations to match the satellite data.

Once we have accounted for possible uncertainties in the satellite data and the capability of our models, we assume that other mismatches between expected and observed concentrations are caused by errors in our emissions inventory. Our task is to work backwards from the concentration mismatches to work out where and how much to correct the emissions. We do this by effectively running the atmospheric pollution model backwards.

Remember that the satellite measures through the whole atmosphere and that the wind distributes methane from an emission in three dimensions. Thus we need to backtrack concentration differences through the atmosphere, not just at the surface.

By applying these corrections from all concentration mismatches together, and accounting for all the changes in wind direction over a month, we can produce a corrected map of the emissions inventory.

A single incorrect emission will hopefully generate several concentration mismatches so that we have multiple cross-checks on the location and size of any error. However the challenges presented by clouds and dust, as discussed earlier, may mask certain erroneous emissions, lacking the necessary satellite observations to identify these errors. 

Much like the events we've previously addressed, the absence of correction to an emission could either suggest that the satellite observations confirm the emissions inventory, or indicate that we lack the necessary observations to verify it.

This uncertainty can be addressed through a series of hypothetical scenarios. By randomly adjusting emissions and calculating the expected concentrations that the satellite would observe, we can assess the sensitivity of our model. If these hypothetical adjustments in emissions result in significant differences in the calculated concentrations, it gives us confidence that actual discrepancies would also be detected by the satellite. 

Conversely, if changes in emissions don't result in noticeable differences in modelled concentrations - typically due to cloud interference - then we cannot place as much confidence in our updates to the emissions inventory for that particular region.