Peatland subsidence as a proxy for carbon loss

Drained peatlands subside as organic matter oxidises, releasing CO₂ at rates detectable by InSAR. Combining millimetre-scale surface deformation with drainage canal mapping and published bulk-density factors produces defensible carbon-flux estimates for MRV.

Sensors

Sentinel-1 C-band SAR (ESA): 5.405 GHz, 5–20 m ground range resolution in IW mode, 6-day repeat at equator with both satellites. Coherence is maintained over months on stable peat surfaces, enabling differential InSAR (DInSAR) and time-series (SBAS, PS) analysis. Cloud-transparent, day/night capable. Coherence degrades over dense tropical canopy, so works best on bare or sparsely vegetated drained peat.
ALOS-2 PALSAR-2 L-band SAR (JAXA): 1.27 GHz, 3–10 m resolution in stripmap mode, 14-day repeat. Longer wavelength penetrates light vegetation canopy better than C-band, preserving coherence in partially vegetated peat swamp. Particularly useful for subsidence mapping under oil-palm or acacia plantations on peat. Archive extends to 2014.
TanDEM-X / TerraSAR-X bistatic pair (DLR): X-band, ~2 m resolution. The TanDEM-X global DEM (12 m and 90 m public releases) provides a baseline elevation surface. Repeat-pass TanDEM-X acquisitions can detect subsidence at sub-centimetre precision under controlled conditions, though commercial tasking is required and coherence is sensitive to soil moisture change.
Sentinel-2 MSI (ESA): 10–20 m resolution, 13 spectral bands, 5-day revisit at equator. Used here not for subsidence but for drainage canal delineation, land-cover classification and vegetation-index time series that contextualise where and when drainage intensity changed. SWIR bands (1610 nm, 2190 nm) distinguish wet from dry peat surface.

Why the ground tells you what the atmosphere cannot

Peat oxidation is chemically straightforward: drain a peatland, expose organic matter to oxygen, and microbial respiration converts centuries of stored carbon to CO₂. The problem for MRV is that gas flux is invisible from orbit at the project scale. Subsidence is not. As peat oxidises and compacts, the surface drops. Published field studies from Sumatra and Borneo report subsidence rates of 2–5 cm per year on actively drained agricultural peat, with cumulative losses of a metre or more over decades. That signal is well within InSAR's detection range.

The key conversion is volumetric: measure how much surface area has dropped by how much, multiply by published peat bulk density (typically 0.08–0.13 g/cm³ for tropical peat, per values reported in the scientific literature), apply a carbon-fraction estimate (around 55–60% of dry peat mass), and you have a carbon-mass flux estimate. This is not a perfect proxy. Consolidation from water-table drawdown contributes to subsidence independently of oxidation, and the two processes are difficult to separate from deformation data alone. Honest MRV treats the InSAR-derived figure as a conservative lower bound on oxidative loss, not a precise emission inventory.

How InSAR resolves millimetres across a landscape

Interferometric SAR compares the phase of two radar acquisitions over the same ground. Phase difference encodes range change between satellite and surface to a fraction of the radar wavelength: roughly 2.8 cm for Sentinel-1 C-band, 23.6 cm for PALSAR-2 L-band. Sub-centimetre vertical sensitivity is achievable when atmospheric artefacts are corrected using external tropospheric models (GACOS is the standard open resource) and a sufficient number of interferograms are stacked.

Two processing approaches dominate. Small Baseline Subset (SBAS) analysis stacks many short-temporal-baseline interferograms to suppress decorrelation noise and recover a displacement time series at each coherent pixel. Persistent Scatterer (PS) InSAR anchors on point targets, such as building corners or exposed soil, that maintain phase coherence over years. On bare or lightly vegetated drained peat, SBAS typically achieves spatial coherence over broad areas; PS is more useful where plantation crops introduce decorrelation. Neither method works well under dense tropical forest, which is why this page covers drained peatlands specifically and why forest-carbon stock estimation is handled separately.

Atmospheric correction is the dominant error source. A single uncorrected tropospheric anomaly can introduce several centimetres of apparent deformation across a scene. Processing pipelines that integrate ERA5 reanalysis or GACOS corrections reduce this to sub-centimetre residuals in most conditions, but users should treat any single interferogram with scepticism. Time-series analysis over 12 months or more is the minimum credible baseline for MRV purposes.

Drainage canals as the causal fingerprint

Subsidence alone does not prove that a specific project or land manager caused carbon loss. Drainage canal mapping closes that gap. Canals are spectrally and geometrically distinct in Sentinel-2 imagery: linear features 5–30 m wide, with high SWIR reflectance contrast against wet peat and a characteristic straight geometry that distinguishes them from natural watercourses. Automated extraction using morphological filtering on SWIR bands, combined with manual QA, can map canal networks to sub-pixel accuracy at 10 m resolution.

Overlaying canal density and orientation with InSAR-derived subsidence rates allows spatial attribution. Peat adjacent to recently excavated canals subsides faster, and the gradient of subsidence with distance from canal edge follows patterns consistent with published drainage-influence models. This spatial correlation is the evidentiary chain that connects a land-use decision to a carbon flux, which is precisely what carbon registries and national MRV systems need.

Honest limits of the method

Several failure modes deserve explicit acknowledgement. Tropical cloud cover does not affect SAR acquisition, but it does affect Sentinel-2 canal mapping. In persistently cloudy regions, optical revisit for cloud-free imagery may stretch to months, limiting the temporal resolution of canal-change detection. PALSAR-2's 14-day repeat, while adequate for annual subsidence rates, may miss rapid drainage events. Sentinel-1's 6-day repeat is better on this count.

Coherence loss over vegetated peat is the hardest constraint. If a project area retains significant peat-swamp forest, InSAR coherence collapses and subsidence cannot be measured. L-band PALSAR-2 extends the usable vegetation density somewhat, but not indefinitely. Areas with standing water also decorrelate rapidly. In practice, InSAR-based MRV is most defensible on drained agricultural peat, degraded peat swamp and plantation-on-peat landscapes, not on intact or rewetted peatlands where inundation is variable.

The bulk-density conversion introduces its own uncertainty. Tropical peat bulk density varies with depth, humification state and drainage history. Published ranges span roughly 0.06–0.15 g/cm³. Without site-specific soil cores, applying a literature mean introduces an uncertainty of perhaps ±30% in the carbon-flux estimate. That uncertainty should be reported explicitly in any MRV submission rather than absorbed into a point estimate.

Turning deformation maps into MRV deliverables

A credible MRV product for a peatland carbon project typically requires three layers: a subsidence rate map with uncertainty bounds, a canal-network change map with dates of new excavation, and a carbon-flux estimate with a documented uncertainty budget. The subsidence rate map is derived from a time-series InSAR analysis over the project boundary and a defined buffer zone. The canal map is derived from multi-date Sentinel-2 classification. The flux estimate combines the two with published conversion factors and a stated confidence interval.

Archive depth matters here. Sentinel-1 data is available from 2014 and PALSAR-2 from the same year, giving up to a decade of retrospective analysis. That archive allows baseline subsidence rates to be established before a project's crediting period begins, which is a requirement of most carbon standards for additionality demonstration. Satellize runs this processing stack on open constellations and can add commercial PALSAR-2 tasking under client licence for areas where Sentinel-1 coherence is insufficient.

The output format for registry submission is typically a GIS layer package (GeoTIFF and shapefile) accompanied by a methodology report citing the processing chain, atmospheric correction approach, coherence thresholds applied and the bulk-density values used. Some standards also require a monitoring report with annual updates. Designing the processing pipeline to be repeatable from the outset, with version-controlled scripts and documented parameter choices, is far less painful than reconstructing it at verification time.

Typical figures

Vertical displacement sensitivity (InSAR)	Sub-centimetre after atmospheric correction; typically 3–10 mm/year detection limit for time-series analysis over 12+ months
Spatial resolution (Sentinel-1 IW mode)	5 × 20 m (range × azimuth); geocoded products typically 10–20 m
Spatial resolution (PALSAR-2 stripmap)	3–10 m depending on mode; 25 m ScanSAR available for wide-area coverage
Revisit period	6 days (Sentinel-1 dual satellite, equatorial); 14 days (PALSAR-2)
Optical canal mapping resolution	10 m (Sentinel-2 10 m bands); canals ≥5 m wide detectable with morphological filtering
Radar frequency	C-band 5.405 GHz (Sentinel-1); L-band 1.27 GHz (PALSAR-2); X-band 9.65 GHz (TanDEM-X)
Archive depth	Sentinel-1 from April 2014; PALSAR-2 from 2014; TanDEM-X global DEM baseline ~2011–2015
Atmospheric correction	GACOS or ERA5-based tropospheric correction; typical residual error <1 cm per scene after correction
Peat bulk-density range (literature)	0.06–0.15 g/cm³ for tropical peat; 0.08–0.13 g/cm³ most commonly cited for Sumatra/Borneo
Delivery formats	GeoTIFF displacement/velocity grids, shapefiles (canal network), PDF methodology report, CSV uncertainty budget

Analytics Satellize can run

Annual subsidence rate map	SBAS time-series InSAR on Sentinel-1 or PALSAR-2 stack; GACOS atmospheric correction; coherence masking at threshold ≥0.3	GeoTIFF velocity map (mm/year) with pixel-level uncertainty, clipped to project boundary plus 5 km buffer
Cumulative displacement time series	PS-InSAR or SBAS with temporal unwrapping; displacement plotted per coherent pixel from archive start date	CSV time series per zone, GeoTIFF cumulative displacement stack, PNG charts for registry reporting
Drainage canal network map and change detection	Multi-date Sentinel-2 SWIR band morphological filtering and supervised classification; change between user-defined epochs	Shapefile of canal centrelines with excavation-date attribution; area statistics per land-cover class
Carbon flux estimate with uncertainty budget	Volumetric subsidence × published bulk-density range × carbon fraction; Monte Carlo propagation of input uncertainties	PDF methodology report with tonne CO₂e/year estimate, 90% confidence interval, and documented conversion factors
Spatial attribution of subsidence to drainage proximity	Regression of subsidence rate against distance from canal network; spatial correlation analysis	GIS layer showing subsidence gradient by canal-buffer zone; statistical summary table for MRV narrative
Baseline subsidence rate (pre-project period)	Retrospective InSAR time-series over archive period prior to project start date	Baseline velocity map and summary statistics for additionality documentation under applicable carbon standard

Sources

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.