Biomass energy feedstock density mapping from satellite vegetation data

SAR backscatter and optical vegetation indices can estimate above-ground biomass density across large areas, but saturation limits, cloud cover, and the gap between total biomass and harvestable residue all constrain what the data can honestly deliver.

Sensors

Sentinel-1 SAR (C-band, ESA): 10 m ground range resolution in IW mode, 6-day repeat at mid-latitudes from the two-satellite constellation. C-band backscatter (5.4 GHz) responds to canopy structure and moisture but saturates against biomass at roughly 50–80 Mg/ha in dense woody vegetation, making it most reliable for grasslands, shrublands, and sparse energy-crop stands.
ALOS-2 PALSAR-2 (L-band, JAXA): 25 m resolution in ScanSAR mode, approximately 14-day repeat. L-band (1.27 GHz) penetrates deeper into canopy than C-band, extending the useful biomass sensitivity range to roughly 100 Mg/ha in forests before saturation. JAXA releases annual global forest/non-forest mosaics derived from PALSAR data.
GEDI (Global Ecosystem Dynamics Investigation, NASA/UMD): Spaceborne lidar operating from the ISS, delivering full-waveform canopy height profiles at 25 m footprints along narrow orbital tracks between 51.6° N and S. Not a wall-to-wall sensor: coverage is track-based and sampling density varies. Published above-ground biomass density (AGBD) products (GEDI L4A/L4B) use height-to-biomass allometric models and provide a critical correction layer where SAR saturates.
Sentinel-2 MSI (ESA): 10–20 m resolution, approximately 5-day revisit at mid-latitudes. Multispectral bands support NDVI, EVI, LAI and red-edge indices that track green biomass accumulation in agricultural energy crops such as miscanthus, switchgrass, and sugarcane. Cloud contamination in humid tropical zones can reduce effective monthly coverage to below 30 % of acquisitions.

What backscatter actually measures, and where it stops

Radar backscatter from a vegetated surface is a function of canopy geometry, moisture content, and the depth to which the signal penetrates before it scatters back. C-band, used by Sentinel-1, interacts primarily with leaves and small branches. In a young miscanthus stand or a grassland, that is sufficient to track biomass accumulation through a growing season. In a mature eucalyptus plantation or a secondary tropical forest, the signal saturates: additional biomass adds no measurable backscatter increment. Published studies place the C-band saturation threshold in woody vegetation at roughly 50–80 Mg/ha, depending on species and moisture regime.

L-band, used by ALOS-2 PALSAR-2, penetrates further because its longer wavelength (roughly 23 cm versus 5.6 cm for C-band) scatters off larger woody elements, including stems and large branches. This pushes the practical saturation ceiling to approximately 100 Mg/ha. Above that, both sensors are effectively blind to additional biomass. For feedstock assessment in dense forests, this is not a minor caveat. It is the central limitation of the radar approach, and any analysis that does not state it plainly is selling you a number it cannot support.

GEDI fills the gap, but not everywhere

NASA's GEDI instrument, mounted on the International Space Station, fires laser pulses at 1064 nm and 1550 nm and records the full waveform of returned energy. From the vertical structure of that waveform, it derives canopy height and, through allometric models, above-ground biomass density. The published GEDI L4B gridded product provides mean AGBD estimates at 1 km resolution with associated uncertainty bounds, and the L4A shot-level product gives individual footprint estimates at 25 m.

The limitation is geometry. GEDI's ISS orbit restricts coverage to latitudes between roughly 51.6° N and S, and the instrument samples along narrow tracks rather than sweeping continuous swaths. In any given month, large areas receive few or no GEDI footprints. Analysts typically aggregate GEDI shots over six to twelve months, then use that sparse sample to calibrate SAR-based wall-to-wall maps. The result is a hybrid product: spatially complete from SAR, height-corrected where GEDI provides sufficient shot density. Uncertainty is highest in structurally complex forests with few GEDI shots and biomass above the SAR saturation threshold.

Optical indices for agricultural energy crops

For dedicated energy crops grown in open agricultural settings, the SAR saturation problem rarely applies. Miscanthus, switchgrass, sorghum, and sugarcane seldom exceed 30–40 Mg/ha of above-ground dry matter at peak standing crop. In these systems, Sentinel-2 vegetation indices do most of the useful work. NDVI and the red-edge chlorophyll index (CIre) track canopy greenness through the season; LAI retrievals from radiative-transfer model inversion can be related to dry-matter accumulation through published crop-growth relationships.

Temporal compositing matters here. A single cloud-free Sentinel-2 scene at peak biomass gives a reasonable standing-crop estimate. A time series through the growing season, fitted to a crop-growth curve, is considerably more informative and allows separation of inter-annual variability from structural differences between fields. In the Satellize crop-estimation programme for the Kingdom of Tonga, a similar time-series approach underpins yield modelling for agricultural parcels, demonstrating that the method transfers across crop types and geographies given appropriate local calibration data.

The residue fraction problem: what satellites cannot see

Feedstock assessments for agricultural residues, straw, bagasse, corn stover, rice husks, and similar, require knowing not just total above-ground biomass but what fraction is economically harvestable without degrading soil carbon or erosion protection. That fraction depends on crop type, tillage practice, local soil conditions, and regulatory or agronomic constraints. Satellite data cannot resolve it.

The correct workflow is to use satellite-derived biomass or yield maps as the spatial input, then apply published residue-to-product ratios and sustainability harvest fractions from agronomic literature or local surveys. This requires a crop-type map as an intermediate layer: without knowing whether a field is wheat or rice, the correct residue ratio cannot be assigned. Crop-type maps themselves carry classification uncertainty, typically 10–20 % misclassification at field scale depending on landscape fragmentation and the number of cloud-free training observations available. Propagating these uncertainties through to a feedstock density estimate is not optional. It is the difference between a number a developer can act on and one that will embarrass them at due diligence.

Putting a number on accuracy: what the public record says

The GEDI L4B global biomass product reports a relative uncertainty of roughly 16–20 % at 1 km resolution in tropical forests, rising sharply in structurally heterogeneous or high-biomass stands. SAR-only biomass maps in temperate and boreal zones, calibrated against field plots, typically achieve root-mean-square errors of 20–40 Mg/ha depending on forest type and the density of ground-truth plots. For agricultural energy crops, Sentinel-2 LAI-based biomass estimates in published studies show correlations with field measurements of r² = 0.7–0.85, with errors that widen at the ends of the biomass range.

These figures are not discouraging. They are honest. A feedstock assessment covering tens of thousands of hectares cannot be done cost-effectively any other way. The satellite approach replaces expensive field campaigns with a spatially continuous first-pass estimate, then directs ground sampling to the areas of highest uncertainty or highest commercial interest. That is the right use of the technology.

From data to a feedstock map a developer can use

A practical biomass feedstock mapping workflow runs in three stages. First, wall-to-wall SAR backscatter mosaics from Sentinel-1 and PALSAR-2 are combined with GEDI-derived canopy height to produce an above-ground biomass density layer, with explicit uncertainty estimates flagging where saturation is likely. Second, Sentinel-2 time series classify crop types and track seasonal biomass accumulation in agricultural zones, with residue fractions applied from published agronomic tables matched to the classified crop types. Third, a logistics overlay, road access, existing infrastructure, competing land uses, incorporates terrain and land-cover data to produce a net accessible feedstock density map at the spatial resolution relevant to plant siting decisions.

The output is a GIS layer with biomass density in Mg/ha, an uncertainty band at each pixel, and a tabular summary by administrative unit or catchment zone. Satellite archives going back to 2014 for Sentinel-1 and Sentinel-2 allow multi-year variability to be characterised, which matters for feedstock supply contracts that must be bankable across a decade. No single scene should anchor a commercial decision. The time series should.

Typical figures

Spatial resolution (SAR biomass mapping)	10–25 m (Sentinel-1 IW: 10 m; PALSAR-2 ScanSAR: 25 m); typically aggregated to 100 m–1 km for biomass products to reduce speckle
Spatial resolution (optical indices)	10–20 m (Sentinel-2 MSI bands)
Revisit period	Sentinel-1: 6 days (two-satellite); PALSAR-2: ~14 days; Sentinel-2: ~5 days at mid-latitudes
SAR biomass saturation threshold	C-band (Sentinel-1): ~50–80 Mg/ha; L-band (PALSAR-2): ~100 Mg/ha in woody vegetation
GEDI coverage	Track-based sampling, 51.6° N–S latitude limit; 25 m footprint diameter; L4B gridded product at 1 km
Typical biomass estimation uncertainty	16–20 % (GEDI L4B, tropical forests); 20–40 Mg/ha RMSE (SAR-only, calibrated); r² 0.70–0.85 (Sentinel-2 LAI vs. field, agricultural crops)
Cloud-cover impact	SAR unaffected; Sentinel-2 optical usable scene fraction can fall below 30 % per month in humid tropical zones
Archive depth	Sentinel-1 and Sentinel-2: from 2014/2015; PALSAR-2: from 2014; GEDI: from April 2019
Deliverable formats	GeoTIFF biomass density raster, vector summaries by parcel or administrative unit, uncertainty layer, CSV feedstock tables
Latency (retrospective assessment)	2–6 weeks from commission, depending on cloud-free scene availability and GEDI shot density in the area of interest

Analytics Satellize can run

Above-ground biomass density map	SAR backscatter fusion (Sentinel-1 C-band + PALSAR-2 L-band) with GEDI canopy-height calibration; random-forest or water-cloud model inversion	GeoTIFF raster at 100 m resolution with per-pixel uncertainty band; PDF summary report
Crop-type classification for residue fraction assignment	Sentinel-2 multitemporal NDVI and red-edge time-series classification (random forest or support vector machine) against regional training data	Vector crop-type map by field parcel; confusion matrix and overall accuracy estimate
Seasonal biomass accumulation curve	Sentinel-2 LAI time-series retrieval via radiative-transfer model inversion (PROSAIL or equivalent); fitted crop-growth model	Per-field time series CSV; peak standing biomass estimate with confidence interval
Net harvestable residue density layer	Satellite yield or biomass estimate multiplied by published crop-specific residue-to-product ratios and sustainability harvest fractions; uncertainty propagated from classification and biomass errors	GIS layer of harvestable residue (Mg/ha/year) by parcel or grid cell; tabular summary by catchment or administrative zone
Multi-year feedstock variability assessment	Annual biomass or yield maps from Sentinel-1/2 archive (2015 to present); inter-annual coefficient of variation calculated per pixel	Variability map and statistical summary suitable for bankable feedstock supply analysis
Accessible feedstock density with logistics overlay	Biomass density layer intersected with road-network buffers, land-cover exclusion zones, and slope constraints from satellite-derived terrain	Net accessible feedstock GIS layer; tabular catchment-zone summary for plant-siting decisions

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.