How Much Forest Is in a Pixel?
When a satellite map reports 294 tonnes of woody biomass per hectare at a point in the Carpathians, that number describes a computational box, not an individual tree. Expand the cell from 0.01° to 0.1°: the average drops to 231 tonnes per hectare, without a single branch falling on the ground.
The Change of Support Problem: When the Pixel Redefines the Object
Forest carbon debates routinely treat numbers on satellite maps as if they were direct measurements of trees. In practice, a satellite measures reflectance or radar backscatter from hundreds of kilometres overhead, while algorithms divide the Earth's surface into an artificial grid of uniform cells.
In geostatistics, this transformation is known as the Change of Support Problem. A field plot measures trees across a fraction of a hectare. When observations are summarized into a 0.01° cell (around 87 hectares in mountainous Romania), the cell captures a dense, mature stand. When the frame widens to 0.1° (100 times larger, covering 8,710 hectares), the box swallows alpine clearings, rocky ridgelines, steep ravines, and younger secondary growth.
In the Carpathian benchmark, the reported mean biomass drops by 63 tonnes per hectare (from 294 to 231 t/ha) purely by enlarging the evaluation window. Not a single tree was cut: only the geometric boundary of the averaged surface changed.
Three Layers of Inference: Field, Lidar, and Integrated Surface
No forest measurement technique yields unmediated ground truth. Every approach relies on statistical modeling:
1. Field plot measurements: Foresters measure trunk diameter at breast height and tree height across a delineated plot. Living wood is never placed directly on a physical scale; biomass is estimated through allometric equations calibrated for tree species and wood density.
2. Airborne lidar: Aircraft emit laser pulses that penetrate the canopy, recording vertical foliage profiles and crown heights along narrow flight swaths. Biomass is inferred via calibration models trained against field plots.
3. Integrated estimates: Harmonizes available lidar sources into a cell-weighted average. At 0.01° in the Carpathians, the integrated estimate reads 242 tonnes per hectare, 52 tonnes below the single field plot at the center.
| Reference Location | Resolution | Cell Area | Field (t/ha) | Lidar (t/ha) | Observations (N) | Std. Dev. |
|---|---|---|---|---|---|---|
| Southern Carpathians (RO) | 0.01° | 87 ha | 294 | 248 | 1 plot | ± 0 t/ha |
| Southern Carpathians (RO) | 0.05° | 2,180 ha | 268 | 239 | 8 plots | ± 42 t/ha |
| Southern Carpathians (RO) | 0.10° | 8,710 ha | 231 | 226 | 29 plots | ± 67 t/ha |
| Tapajós (Amazon Basin, BR) | 0.01° | 123 ha | 342 | 325 | 2 plots | ± 18 t/ha |
| Tapajós (Amazon Basin, BR) | 0.10° | 12,320 ha | 285 | 274 | 87 plots | ± 74 t/ha |
| Hyytiälä (Finland) | 0.01° | 58 ha | 112 | 104 | 3 plots | ± 14 t/ha |
| Hyytiälä (Finland) | 0.10° | 5,800 ha | 89 | 86 | 64 plots | ± 33 t/ha |
Why Blank Space Does Not Mean Missing Forest
A global reference compilation is not a continuous map. Blank space across a region indicates the absence of qualifying measurements meeting strict harmonization criteria, not a lack of trees or zero biomass.
To avoid false signals from recent deforestation or replanting, historical observations collected between 1983 and 2017 were binned into 7 symmetric 5-year reference windows (1985, 1990, 1995, 2000, 2005, 2010, and 2015). An observation was retained only if land-transition models (HILDA+) confirmed persistent forest cover throughout the full 5-year span.
Therefore, a label of “2010” designates an eligible record within the multi-year 2008–2012 window. Understanding this architecture prevents misinterpreting spatial aggregation differences as forest loss or environmental degradation.
Methodological Note & Data Sources
Values are drawn from the global reference compilation of forest aboveground biomass, described in Scientific Data (Nature, August 2026) and deposited on Zenodo (v2, 2026). The archive combines 31 field-data sources and 14 airborne lidar sources harmonized into tonnes of dry biomass per hectare (Mg/ha).
Exclusion filters remove multi-year aggregate estimates, harvest-derived numbers, and records with unresolved coordinates. Cell distributions explicitly publish internal dispersion (standard deviation) and contributing observation counts to enable direct evaluation of continuous global satellite biomass products.