Natural wetland methane flux estimation via atmospheric inversion
Natural wetlands emit more methane than all fossil fuel sources combined, yet their fluxes remain poorly constrained. Top-down atmospheric inversion using GOSAT and TROPOMI column data, anchored by Sentinel-1 inundation mapping, is closing that gap.
Sensors
- GOSAT / GOSAT-2: Provides column-averaged dry-air mole fraction of methane (XCH4) at roughly 10 km footprint diameter, with a 3-day repeat cycle at the equator. Single-sounding precision is approximately 8–12 ppb (1-sigma); bias-corrected retrievals are available via NIES and the University of Leicester. Sparse spatial sampling means regional rather than patch-scale flux attribution.
- Sentinel-5P TROPOMI: Delivers daily global XCH4 at 5.5 km × 7 km pixel size (reprocessed to 5.5 km × 5.5 km from August 2019 onward), with single-pixel precision around 13–18 ppb. Daily revisit at any latitude makes it the workhorse for temporal flux variability. Cloud fraction thresholds typically discard 50–70 % of tropical observations, a non-trivial loss over seasonally flooded regions.
- Sentinel-1 SAR (C-band): C-band backscatter at 10 m resolution (IW mode) maps open-water inundation and flooded vegetation at 6–12 day revisit per satellite, 6 days with both Sentinel-1A and 1B when both are operational. Dense emergent vegetation attenuates the signal, so inundation under closed forest canopy is systematically underestimated. Used here to build the spatial wetland-area prior for the inversion.
- SMAP (Soil Moisture Active Passive): L-band passive radiometry at 36 km resolution provides daily soil-moisture estimates that serve as a process proxy for methanogenesis rates in non-inundated wetland soils. Resolution is too coarse for patch-level attribution but useful for regional parameterisation of flux models such as ORCHIDEE or CLM-Methane.
Why wetland methane is so hard to pin down
Natural wetlands contribute an estimated 150–200 Tg of methane per year globally, a range that has barely narrowed in two decades of study. The uncertainty is not ignorance about the chemistry. Methanogenesis is well understood: anaerobic decomposition of organic matter in waterlogged soils, accelerated by warmth and slowed by sulphate competition. The problem is spatial and temporal heterogeneity at a scale that no ground-based network can adequately sample.
A single Amazonian floodplain can contain open water, floating macrophytes, flooded forest and seasonally dry savanna within a few hundred metres of each other. Each surface type has a different emission factor, a different seasonal pulse, and a different response to temperature anomalies. Bottom-up inventories built from chamber measurements and land-cover maps carry structural errors that compound across continental scales. Top-down inversion from satellite column measurements offers a check on those inventories, but it introduces its own set of problems, principally the mismatch between a 5–10 km satellite footprint and a patchwork of emitting surfaces below.
What the column measurements actually see
GOSAT and TROPOMI do not measure surface flux directly. They measure the column-integrated dry-air mole fraction of methane, XCH4, in parts per billion. A 10 ppb enhancement above background in a TROPOMI pixel could represent a strong local source or a weaker source that has been accumulating under stable atmospheric conditions for several days. Separating those two cases requires an atmospheric transport model, typically a chemical transport model such as GEOS-Chem, TM5 or FLEXPART, run in inverse mode.
The inversion adjusts a prior flux estimate until the modelled XCH4 field matches the observed one within measurement and transport-model uncertainty. The result is a posterior flux map, usually at 1–4 degree resolution for global inversions, though regional inversions using TROPOMI can resolve down to roughly 25–50 km. That is still far coarser than the wetland patches driving the signal. The spatial prior, the assumed distribution of emitting area within each inversion grid cell, therefore matters enormously. An incorrect prior propagates directly into the posterior flux.
What a floating roof gives away: using Sentinel-1 to constrain the prior
Sentinel-1 C-band SAR is insensitive to cloud cover and acquires at night, which makes it the practical tool for mapping inundation extent across the wet tropics. Open water returns very low backscatter (specular reflection away from the sensor), while flooded vegetation shows a characteristic double-bounce enhancement as the radar signal bounces between water surface and vertical stems. These two signatures allow automated classification of inundated area at 10 m resolution, updated every 6–12 days.
Feeding a time-series of Sentinel-1 inundation maps into the inversion replaces a static, climatological wetland-area prior with one that tracks actual flooding. When the Amazon floods earlier than average, the inversion prior shifts accordingly, and the posterior flux reflects that rather than attributing the XCH4 enhancement to a spurious source elsewhere. Published work using this approach has reduced prior uncertainty in tropical wetland flux estimates by a meaningful fraction, though precise figures depend heavily on the transport model and the inversion configuration used.
The honest caveat: Sentinel-1 underestimates inundation under dense forest canopy because C-band penetration is limited. L-band SAR, such as ALOS-2 PALSAR-2 or the forthcoming NISAR mission, penetrates better and would improve the prior further. For now, Sentinel-1-based inundation maps should be treated as a lower bound on flooded area in forested wetlands.
Resolving the mismatch: downscaling and ensemble approaches
The resolution gap between a TROPOMI pixel and a wetland patch is roughly two orders of magnitude. No inversion closes that gap directly. What practitioners do instead is use high-resolution land-cover and inundation data to disaggregate the coarse posterior flux onto finer grids, applying emission factors from process models or chamber studies to each surface class. The result is a downscaled flux map that carries the large-scale constraint from the satellite but distributes it spatially using ancillary information.
Ensemble inversions run the same XCH4 observations through multiple transport models and multiple prior flux datasets, then compare the spread of posterior estimates. Where the ensemble agrees, confidence is higher. Where it diverges, the disagreement itself is informative: it usually points to either transport model error over a specific region or a prior that is structurally wrong. The Copernicus Atmosphere Monitoring Service publishes ensemble inversion products that include wetland flux components, providing a useful public benchmark against which regional analyses can be validated.
Seasonal signals and what they reveal about process
Wetland methane flux has a strong seasonal signature that TROPOMI's daily revisit is well placed to capture, cloud permitting. In the tropics, the XCH4 enhancement over the Congo Basin and Amazon typically peaks a few weeks after maximum inundation extent, reflecting the lag between flooding and peak methanogenesis. In boreal wetlands, the seasonal pulse is tighter and more temperature-driven, with Siberian and Canadian peatlands showing sharp spring onset as soils thaw.
Tracking these seasonal cycles year-on-year provides a proxy for process change. A year in which the Amazonian XCH4 anomaly peaks earlier or later than the Sentinel-1 inundation peak suggests a shift in either temperature or organic substrate availability, both of which are relevant to carbon-cycle projections. This kind of multi-signal phenological analysis is where satellite data moves from inventory support into genuine process understanding. It also creates a natural audit trail: if a government or carbon-market registry claims a wetland restoration project has altered methane flux, the satellite record either corroborates or contradicts that claim.
Satellize's analytics stack runs this kind of multi-sensor time-series analysis on open constellations, with the Tonga crop-estimation programme being an example of how agronomic flux signals can be extracted from the same Sentinel family of sensors in a completely different context.
Honest limits and what they mean for buyers
A buyer commissioning wetland methane flux analysis should understand three hard constraints. First, TROPOMI cloud filtering removes the majority of observations over tropical wetlands during wet seasons, precisely when fluxes are highest. Gap-filling using reanalysis meteorology introduces additional uncertainty. Second, atmospheric transport model error is not small: published inter-model spread in tropical methane inversions can exceed 30 % of the regional flux estimate. Third, the inversion cannot attribute flux to individual wetland patches smaller than roughly 25–50 km across; it produces regional estimates, not site-level ones.
What the approach does well is provide an independent, physics-based check on bottom-up inventories at national to continental scales. For a government preparing a national greenhouse gas inventory, or a conservation organisation seeking to verify that a wetland protection programme is delivering measurable climate benefit, that regional constraint is genuinely useful. It will not replace ground-based eddy-covariance towers, but it scales in a way that towers never can.
Typical figures
| XCH4 pixel size (TROPOMI) | 5.5 × 5.5 km (from August 2019 reprocessing onward) |
| XCH4 pixel size (GOSAT) | ~10 km diameter footprint; sparse cross-track sampling |
| TROPOMI revisit | Daily global coverage; effective cloud-clear revisit over tropics 3–7 days seasonally |
| GOSAT repeat cycle | 3-day sub-cycle; ~56 km along-track spacing at equator |
| Inundation mapping resolution (Sentinel-1 IW) | 10 m; 6-day revisit with dual-satellite constellation |
| Inversion output resolution (typical regional) | 25–100 km, depending on transport model and domain |
| Single-sounding XCH4 precision | ~13–18 ppb (TROPOMI); ~8–12 ppb (GOSAT, bias-corrected) |
| Minimum detectable regional flux anomaly | Approximately 10–20 Tg CH4/yr at continental scale; sub-national detection harder |
| Archive depth | TROPOMI from October 2017; GOSAT from April 2009; Sentinel-1 from April 2014 |
| Soil moisture context (SMAP) | 36 km passive L-band; daily global; from April 2015 |
Analytics Satellize can run
| Regional wetland XCH4 anomaly time series | TROPOMI Level-2 XCH4 filtering, bias correction and spatial aggregation over user-defined wetland domains | Monthly CSV and GeoTIFF anomaly maps with cloud-fraction-weighted uncertainty bands |
| Sentinel-1 inundation extent time series | Thresholded backscatter classification and double-bounce detection in IW GRDH scenes; change detection against dry-season baseline | Bi-weekly inundation GeoTIFF at 10 m, aggregated to inversion grid cells for prior construction |
| Top-down posterior flux estimate | Bayesian atmospheric inversion (analytical or ensemble) using TROPOMI XCH4 and a Sentinel-1-constrained spatial prior; transport modelled via published CTM climatology | Quarterly gridded flux posterior (NetCDF) with prior/posterior uncertainty comparison report |
| Seasonal flux phenology report | Cross-correlation of inundation extent time series with XCH4 anomaly to extract lag structure and inter-annual variability | Annual PDF report with time-series figures and anomaly attribution narrative |
| Inventory validation assessment | Comparison of national or project-level bottom-up wetland emission estimates against posterior satellite-derived flux at matching spatial scale | Structured discrepancy report suitable for national greenhouse gas inventory review or carbon-registry submission |
| Wetland restoration flux change detection | Pre/post intervention XCH4 anomaly differencing with inundation-area normalisation to separate area-driven from rate-driven flux change | Before/after flux comparison GIS layer and statistical significance assessment |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.