Grassland above-ground biomass monitoring for carbon programmes

Grassland carbon projects demand periodic above-ground biomass estimates, but shallow canopies, rapid phenological swings and dry-season senescence defeat simple optical approaches. Getting to registry-grade numbers requires combining vegetation indices, SAR backscatter and careful seasonal sampling.

Sensors

Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 20 m in red-edge and shortwave infrared. Five-day revisit at the equator with both satellites. Red-edge bands (B5, B6, B7 at 705, 740, 783 nm) are particularly sensitive to chlorophyll content in sparse canopies. Cloud cover and dry-season senescence both degrade optical retrievals significantly.
Sentinel-1 SAR (C-band): C-band backscatter at 5.405 GHz, 10 m ground range detected resolution in IW mode, 6-day repeat with both satellites. Backscatter from short grass canopies is dominated by soil moisture and surface roughness, which introduces ambiguity in biomass retrieval. Useful as a phenology-independent complement to optical data, especially during cloud cover or senescence.
PRISMA hyperspectral: ASI's PRISMA sensor provides 239 spectral bands across 400–2500 nm at roughly 30 m spatial resolution and a 30 km swath. Narrow-band indices derived from PRISMA can separate green and dry grass fractions and estimate leaf area index more precisely than broadband sensors, but tasking is on-request and revisit is irregular, typically days to weeks.
Planet SuperDove: Eight spectral bands including two red-edge channels, 3–4 m spatial resolution, near-daily revisit over most land areas. High temporal density supports phenological curve fitting, which is essential for integrating biomass over a growing season rather than relying on a single date. Costs are commercial; archive depth and continuity depend on client licence.

Why grass canopies resist the standard playbook

Above-ground biomass estimation in forests rests on canopy height and structure. Grass offers neither. A mature savanna grass sward may be 0.5–1.5 m tall and structurally uniform, giving SAR almost nothing to work with in terms of volume scattering. Optical sensors see a mixture of green leaf, dry stem, bare soil and, in savannas, scattered woody cover, all within a single 10 m pixel. The signal is real but easily confounded.

The core difficulty is that spectral reflectance in the near-infrared and red-edge responds to chlorophyll and leaf area index, not directly to dry matter mass. Green biomass and dry biomass have very different spectral signatures: green grass absorbs strongly in the red and reflects in the near-infrared, while dry grass behaves more like bare soil, with high reflectance across the shortwave infrared. A single-date image taken mid-dry-season may dramatically underestimate total seasonal biomass production because most of that production has already senesced.

Vegetation indices: what they measure and where they fail

NDVI (Normalised Difference Vegetation Index) remains the most widely used proxy, but it saturates at moderate green leaf area index values and is sensitive to soil background, which is particularly problematic in sparse savanna grass. EVI (Enhanced Vegetation Index) partially corrects for soil and atmospheric effects and is less prone to saturation. For dry biomass, the Cellulose Absorption Index and indices using shortwave infrared bands (Sentinel-2 B11 at 1610 nm, B12 at 2190 nm) are more informative because dry plant matter absorbs strongly in those regions.

Red-edge indices, particularly those using Sentinel-2's B5 and B7 bands, improve chlorophyll estimation in low-biomass canopies relative to broadband NDVI. Published studies using Sentinel-2 red-edge bands in South African grasslands have reported R² values in the 0.6–0.75 range for green biomass estimation against field measurements, which is useful but not precise enough for high-integrity carbon accounting without dense calibration plots. Hyperspectral data from PRISMA, with its narrow contiguous bands, can resolve specific absorption features associated with cellulose, lignin and water content, improving dry-matter estimates, but only where tasking has been arranged in advance.

Phenology is not noise: it is the measurement

A single-date biomass estimate is almost meaningless for carbon accounting in seasonal grasslands. Peak green biomass in a temperate system may occur over a two-to-four-week window. In a savanna, the green flush after the first rains can double NDVI within ten days. Annual net primary productivity, which is what carbon programmes actually need, requires integrating the biomass signal across the full growing season.

This is where dense time-series matter. Planet SuperDove's near-daily revisit allows phenological curve fitting using methods such as double-logistic or harmonic regression, extracting peak biomass timing, green-up rate and senescence rate as explicit parameters. Sentinel-2's five-day revisit is sufficient for many temperate systems but can miss the narrow green peak in semi-arid savannas. Cloud cover compounds this: in humid savanna zones, a five-day revisit may yield only two or three usable images per growing season, making single-sensor reliance risky. Combining Sentinel-2 and Planet observations, gap-filled with SAR-derived phenological indicators, is the practical approach for high-revisit requirements.

Dry-season SAR backscatter from Sentinel-1 does not track biomass directly in short grass, but it does track soil moisture changes and surface roughness, which correlate with the onset of the wet season and therefore with the start of the green-up period. Used as a phenological trigger rather than a biomass estimator, C-band SAR adds genuine value to the seasonal reconstruction.

Calibration, allometry and the ground-truth problem

Remote sensing indices must be converted to tonnes of dry matter per hectare, and that conversion requires field calibration. For grasslands, this typically means destructive harvesting of 0.25–1 m² quadrats, oven-drying and weighing. The carbon fraction of dry grass biomass is commonly taken as approximately 0.45 g C per g dry matter, though this varies by species and phenological stage. Published allometric relationships for specific grass communities are available in the literature, but applying a relationship developed in Kenyan savanna to Mongolian steppe is not defensible.

The minimum number of calibration plots needed to support a regression model with acceptable uncertainty depends on the spatial variability of the grassland, but published grassland biomass remote sensing studies typically use 30–100 field plots per study area. For carbon programmes seeking Verra VCS or Gold Standard verification, the uncertainty bounds on biomass estimates must be explicitly quantified, usually requiring bootstrapped confidence intervals on the regression. This is not a limitation of the satellite data alone; it is a sampling design problem that satellite data cannot fully substitute for.

What the numbers can and cannot support

Realistically, optical and SAR methods for grassland above-ground biomass can detect relative change in green biomass at the field scale with reasonable confidence when calibration data exist and phenological timing is well sampled. Detecting absolute stock levels at the precision required for carbon credit issuance, typically better than ±10–15% at the project level, is harder and depends heavily on calibration plot density and spatial heterogeneity.

Dry-season estimates are the weakest point. Once senescence is complete, optical indices collapse toward soil-background values and carry almost no biomass information. SAR backscatter in C-band similarly loses sensitivity to standing dry biomass because the short wavelength interacts mainly with the soil surface once the green canopy is gone. L-band SAR, used by ALOS-2 PALSAR-2, penetrates further and retains some sensitivity to dry standing biomass, but that system is not in the sensor set discussed here. Programmes relying on a single dry-season overpass will underestimate annual production substantially.

Satellize's analytics work, including the seasonal biomass reconstruction methods developed for the Kingdom of Tonga crop-estimation programme, applies similar phenological curve-fitting logic to grassland contexts, with explicit uncertainty propagation at each processing step.

Typical figures

Optical spatial resolution	10 m (Sentinel-2 visible/NIR), 20 m (red-edge/SWIR), 3–4 m (Planet SuperDove), ~30 m (PRISMA)
SAR spatial resolution	10 m ground range detected, Sentinel-1 IW mode
Revisit frequency	5 days (Sentinel-2, both satellites); 6 days (Sentinel-1, both satellites); near-daily (Planet SuperDove); irregular on-request (PRISMA)
Key spectral bands for biomass	Red-edge 705, 740, 783 nm; SWIR 1610, 2190 nm (Sentinel-2 B5–B7, B11, B12); 400–2500 nm continuous (PRISMA)
SAR frequency	C-band 5.405 GHz (Sentinel-1)
Minimum detectable biomass change	Approximately 20–30 g dry matter m⁻² for optical index methods with adequate calibration; lower precision in sparse or senesced canopies
Cloud cover impact	Optical methods fully blocked by cloud; SAR unaffected. Humid savannas may yield fewer than 3 usable optical images per growing season without multi-sensor fusion
Archive depth	Sentinel-2 from 2015; Sentinel-1 from 2014; Planet from approximately 2016 (varies by region and licence)
Typical latency	Sentinel-2 and Sentinel-1 data available within 1–3 hours of acquisition via Copernicus Data Space; Planet typically same-day
Delivery formats	GeoTIFF biomass rasters, CSV/GeoJSON plot-level estimates, seasonal phenology parameter layers, uncertainty rasters

Analytics Satellize can run

Seasonal green biomass time-series	Phenological curve fitting (double-logistic or harmonic regression) on dense NDVI/red-edge index stacks from Sentinel-2 and Planet SuperDove	Annual biomass trajectory raster stack with peak-biomass and integrated-season layers, GeoTIFF
Dry-matter fraction mapping	SWIR-based dry-biomass indices (Sentinel-2 B11/B12) combined with green-fraction decomposition using spectral mixture analysis	Seasonal dry-matter raster with per-pixel uncertainty estimate, GeoTIFF
Annual net primary productivity proxy	Integrated NDVI or EVI over the growing season (sum of cloud-gap-filled index values), calibrated against field harvest data	Annual NPP proxy map in tonnes dry matter per hectare, GeoTIFF with confidence intervals
SAR-assisted phenological onset detection	Change-point detection on Sentinel-1 VV/VH backscatter time-series to identify wet-season onset and green-up trigger dates	Per-pixel green-up date layer used to align optical sampling windows, GeoJSON
Hyperspectral biomass component separation	Narrow-band indices and partial least squares regression on PRISMA data to separate green leaf, dry stem and soil fractions	Fractional cover and dry-matter density map for tasked acquisitions, GeoTIFF
Carbon stock change report for verification period	Multi-date biomass difference with bootstrapped uncertainty quantification, applying published carbon fraction (0.45 g C g⁻¹ dry matter)	Tabular stock-change report with uncertainty ranges suitable for Verra VCS or Gold Standard audit submission, PDF and CSV

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.