Improved cookstove project fuel-consumption verification

Cookstove carbon credits rest on fuel-consumption figures that are hard to audit at scale. Satellite forest-disturbance signals around project communities offer an independent, if indirect, cross-check on reported fuelwood savings.

Sensors

Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at the equator with both satellites. The red-edge bands (B5, B6, B7) are sensitive to chlorophyll loss at the forest edge, making small-scale canopy thinning detectable even before full clearing occurs. Free and open archive from 2015.
Landsat 8/9 OLI: 30 m multispectral resolution, 16-day revisit per satellite (8-day combined). Provides a consistent archive back to 1984 for baseline construction. NDVI and NBR indices derived from OLI are well-validated for woody-biomass loss in dry and moist forest contexts.
VIIRS Active Fire (NOAA/NASA FIRMS): 375 m pixel resolution for active fire detection, twice-daily revisit from Suomi-NPP and NOAA-20. Distinguishes agricultural burn events from persistent cooking-fire signatures only at landscape scale; individual household fires are well below the detection threshold. Useful for flagging confounding agricultural clearing episodes within or adjacent to project areas.
Planet SkySat: Approximately 0.5 m native resolution, tasked on demand. Useful for ground-truth sampling of ambiguous disturbance patches identified in Sentinel-2 or Landsat time series, confirming whether a spectral anomaly is stump-field, agricultural plot or natural gap. Commercial tasking; not open access.

Why a carbon credit needs a landscape alibi

Improved cookstove projects earn credits by claiming that households burn less wood or charcoal than they would with a traditional three-stone fire. The claimed reduction is typically derived from kitchen performance tests and self-reported usage surveys, neither of which is cheap to conduct at scale or easy to audit from a registry desk. A project covering several thousand households across dispersed villages may submit fuel-saving estimates that no field team has independently verified.

The underlying logic of satellite cross-checking is straightforward: if a community genuinely shifts a meaningful share of its cooking energy away from fuelwood, the rate at which surrounding woodland is degraded or cleared should slow relative to comparable communities without the intervention. The forest carries a signal, however faint, of what is actually happening in the kitchen. That signal is indirect, noisy and confounded by other pressures, but it is observable from orbit without a field visit.

What the disturbance signal actually measures, and what it does not

Satellite sensors do not see fuelwood consumption. They see canopy reflectance change. When woodland is selectively cut for charcoal or fuelwood, the spectral response in the near-infrared drops and shortwave-infrared reflectance rises, a pattern captured by indices such as the Normalised Burn Ratio and the Normalised Difference Vegetation Index. Sentinel-2's red-edge bands add sensitivity to partial canopy loss that would be invisible at Landsat resolution. The method can flag disturbance patches as small as a few hundred square metres in open woodland, though dense humid forest canopy conceals sub-canopy cutting almost entirely.

The honest caveat is that fuelwood collection is rarely the only driver of woodland change around rural communities. Agricultural expansion, charcoal production for urban markets, livestock browsing and seasonal fire all produce overlapping spectral signatures. A project analyst cannot attribute a given patch of canopy loss to cooking fuel without additional contextual data. What the satellite record can do is test whether the rate and spatial pattern of disturbance around project villages changed after stove distribution, and whether that change differs from a matched set of control villages. That is a falsifiability test, not a measurement.

Building a defensible before-and-after comparison

The analytical workflow begins with a disturbance time series. Landsat's archive to 1984 and Sentinel-2's record from 2015 together allow construction of a pre-project baseline rate of canopy change within a defined radius of each project community, typically 2 to 10 km depending on local fuelwood-collection distances documented in the project design document. The baseline captures inter-annual variability driven by rainfall and agricultural cycles, which must be separated from the intervention signal.

Control communities matter enormously here. Selecting villages with similar woodland cover, population density, market access and rainfall history, but no stove distribution, allows a difference-in-differences comparison. If project villages show a statistically meaningful slowdown in disturbance rate post-distribution while control villages do not, that is evidence consistent with the claimed fuel savings. It is not proof. Confounders such as a concurrent agricultural extension programme or a local charcoal export ban could produce the same pattern. The analyst must document those alternative explanations and assess their plausibility.

VIIRS active-fire data adds a useful filter. Episodes of landscape-scale burning, typically agricultural clearing, can be flagged and excluded from the disturbance signal for the period in question, reducing the risk that a seasonal burn is misread as a fuelwood-driven canopy loss event.

Resolution limits and the small-holder problem

Fuelwood collection around cookstove project communities tends to be diffuse and selective rather than concentrated. A household may remove a few branches from dozens of trees across a wide area rather than clear-cutting a patch. This selective degradation produces a gradual spectral drift rather than a sharp boundary, and it sits close to or below the detection threshold of 10 m Sentinel-2 imagery in many woodland types. Humid closed-canopy forest is particularly opaque to this kind of analysis; the method works best in open woodland, miombo, dry savanna and similar biomes where canopy gaps are already common and incremental removal produces a detectable reflectance change.

Planet SkySat at 0.5 m can resolve individual tree crowns and stump scars, making it useful for targeted validation of suspicious patches. But systematic coverage of a large project area at that resolution is commercially expensive and operationally impractical as a routine monitoring tool. The realistic workflow uses Sentinel-2 for area-wide monitoring and SkySat for spot-check validation of anomalies.

Fitting this into a carbon registry framework

Carbon registries such as Gold Standard and Verra's VCS currently require periodic monitoring reports supported by field data. Satellite disturbance analysis is not yet a registry-accepted substitute for kitchen performance tests or usage surveys, but it is increasingly used as a supplementary line of evidence in verification audits and by third-party validators looking for inconsistencies between claimed savings and observable landscape change.

The practical value is asymmetric. A satellite analysis that shows no meaningful change in disturbance rates around project villages does not prove the project is fraudulent; fuelwood collection may simply be too diffuse to detect. But a satellite record showing accelerating deforestation around villages that are simultaneously claiming large fuel savings is a material red flag that warrants field investigation. The method is more useful as a fraud screen than as a positive verification tool.

Satellize applies this disturbance-monitoring workflow using open Sentinel-2 and Landsat archives, with optional commercial tasking for validation patches, and can deliver the output as a georeferenced change-rate layer with accompanying statistical comparison tables suitable for inclusion in a verification audit package. The approach draws on the same time-series methodology used in the Tonga crop-estimation programme, adapted to woodland rather than agricultural parcels.

What an auditor should ask before commissioning this analysis

The first question is biome. If the project operates in dense humid forest, satellite disturbance signals will be weak and the analysis may add little. If it operates in open woodland or dry savanna, the method is considerably more informative. The second question is spatial design: are project and control villages documented with precise boundaries and GPS coordinates, and is there a plausible fuelwood-collection radius defined in the project design document? Without those inputs, the comparison zone is arbitrary.

The third question is confounders. Has the project area experienced any concurrent land-use policy changes, agricultural development programmes or market shifts in charcoal prices during the monitoring period? Documenting those factors before commissioning the analysis is the difference between a credible supplementary evidence layer and an inconclusive one. Satellite data is not self-interpreting; the analyst needs the project context to say anything useful.

Typical figures

Primary spatial resolution	10 m (Sentinel-2 MSI visible/NIR); 30 m (Landsat 8/9 OLI)
Revisit frequency	5 days (Sentinel-2 dual-satellite); 8 days combined (Landsat 8+9)
Active-fire detection resolution	375 m (VIIRS I-band, FIRMS)
Minimum detectable disturbance patch	Approximately 0.1 ha in open woodland at Sentinel-2 resolution; sub-canopy selective cutting typically undetectable
Spectral bands used	Red, NIR, red-edge, SWIR (Sentinel-2 B4/B5/B6/B7/B8/B11/B12); OLI bands 4/5/6/7 for NDVI and NBR
Cloud limitation	Persistent cloud cover in humid tropics can reduce usable observations to fewer than 6 per year; dry-season compositing partially mitigates this
Archive depth	Landsat to 1984; Sentinel-2 from 2015; VIIRS from 2012
Validation resolution	Approximately 0.5 m (Planet SkySat, commercial tasking)
Typical analysis latency	2 to 4 weeks for a full baseline-plus-monitoring report; near-real-time disturbance alerts feasible within 5 to 10 days of image acquisition

Analytics Satellize can run

Pre-project woodland disturbance baseline	Landsat and Sentinel-2 time-series NDVI and NBR trend analysis, with inter-annual variability decomposition	GIS layer showing mean annual disturbance rate per community buffer zone, with confidence intervals, covering 5 to 10 years pre-distribution
Post-distribution change-rate comparison	Difference-in-differences analysis comparing project villages against matched control villages using Sentinel-2 seasonal composites	Statistical summary table and choropleth map showing change in disturbance rate by village, suitable for audit annexure
Agricultural burn exclusion mask	VIIRS active-fire event overlay filtered by date and spatial cluster to separate landscape burns from persistent small-scale clearing	Dated fire-event GIS layer clipped to project area, used to clean the disturbance time series
Anomaly validation sample	High-resolution SkySat imagery of flagged disturbance patches, interpreted for land-cover type (stump field, crop plot, natural gap)	Annotated image chips with classification labels for a stratified random sample of anomalous pixels
Confounder documentation report	Contextual overlay of project disturbance signal against publicly available land-use records, rainfall anomaly data and VIIRS fire history	Written assessment of alternative explanations for observed disturbance trends, formatted for third-party validator review
Ongoing monitoring feed	Rolling Sentinel-2 composite updated each quarter, with automated flagging of disturbance rate exceedances relative to baseline	Quarterly alert report with updated GIS layers, delivered to project developer and verifier portals

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.