akar.earthField Plots vs Remote Sensing for Carbon Stock
Field plots vs remote sensing for carbon stock estimation: accuracy, coverage, what methodologies require and how Indonesian projects combine both in practice.
Field plots give the most accurate carbon stock measurement at a point, while remote sensing gives consistent coverage across the whole project and over time; credible projects need both. Use plots to measure trees, calibrate models and set uncertainty; use remote sensing to scale those measurements, stratify land, detect change and monitor control areas. The real decision is how to divide effort between the two, not which one to choose.
Side-by-side comparison
| Attribute | Field plots | Remote sensing |
|---|---|---|
| What it measures | DBH, height, species, dead wood, soil samples | Reflectance, canopy height, structure, cover change |
| Carbon estimate | Direct via allometric equations and wood density | Indirect via models calibrated on plots |
| Accuracy at a location | High, if protocols are followed | Lower; errors grow without local calibration |
| Coverage | A small sample of the area | Wall-to-wall over the project and surroundings |
| Change detection | Slow; requires remeasurement | Fast and repeatable |
| Below-ground and soil carbon | Measurable (soil cores) or by root-to-shoot ratios | Not directly observable |
| Cost structure | Labour, travel, training, equipment | Imagery (often free), processing and expertise |
| Bias risk | Poor plot placement or measurement errors | Saturation in high-biomass forest, cloud gaps |
| Role in methodologies | Required for stocks and calibration | Required for some approaches (e.g. VM0047 area-based) |
| Audit view | VVBs remeasure a sample of plots | VVBs review methods, data and reproducibility |
Key differences
Direct versus modelled. A field plot measures trees and applies allometric equations, such as the Chave et al. 2014 pantropical model with wood density, to estimate biomass. Remote sensing never weighs a tree; it relates signals such as canopy height or spectral indices to biomass measured in plots. Without plots, the model has nothing to learn from.
Sampling versus census. Plots sample. The precision of the estimate depends on how many plots you place and how well strata separate different forest conditions. See plot sampling design. Remote sensing observes every pixel but with model error. Combining them, as in model-assisted estimation, can reduce overall uncertainty, which lowers uncertainty deductions.
Saturation. Optical indices and radar backscatter tend to saturate in dense, high-biomass forests such as Kalimantan lowland dipterocarp stands. Lidar helps but still needs plots. See lidar vs photogrammetry.
Methodology requirements. VM0047 v1.1, effective for all projects since 14 May 2025, has an area-based approach that uses remote sensing with a dynamic performance benchmark from control plots, plus plot sampling. The census-based approach for smaller plantings measures every tree. Both require ground data. Read the area-based approach.
When each fits in Indonesia and Southeast Asia
- Large ARR or REDD+ areas in Kalimantan, Sumatra or Papua: remote sensing for stratification and change, with plots concentrated in each stratum. Access cost makes efficient plot design essential.
- Smallholder agroforestry in Java, Sulawesi or NTT: many small, mixed plots where satellite pixels mix crops and trees. Field measurement, often using mobile tree measurement by trained local teams, carries more weight.
- Mangroves and peat: soil carbon requires coring; satellites map extent and water table proxies only.
- EUDR evidence: satellite checks against the 31 December 2020 cut-off, supported by field-collected polygons. The same dataset can support carbon work; see one dataset for carbon and EUDR.
Cost and effort considerations
We do not quote costs. Consider:
- Plots: travel in remote areas, team size, days per plot, training and quality control. Remeasurement each monitoring period.
- Remote sensing: open imagery is free, but skilled analysis is not. Commercial imagery and lidar add licence or flight costs.
- Location accuracy: linking plots to pixels needs good GNSS. We run RTK-surveyed plot campaigns so plots match imagery and can be revisited.
- Uncertainty economics: extra plots cost money, but large uncertainty deductions cost credits. Model the trade-off early.
Our field data campaigns and dMRV implementation services cover both sides.
How to split effort
Start with remote sensing to map strata and find where variability is highest. Place plots where they reduce uncertainty most, then use the plot data to calibrate maps across the whole area. Re-run this loop each monitoring period.
Common misconceptions
- “dMRV means no fieldwork.” Digital MRV speeds up data flow; it does not remove the need for ground truth. See ground truth vs satellite.
- “More plots are always better.” Better-placed plots in the right strata beat more plots in the wrong places.
- “Global biomass maps are good enough.” Global products help screening but rarely meet project-level precision.
- “Plot data lasts forever.” Plots must be remeasured to track growth and losses.
More in the dMRV hub.
Frequently asked questions
Can satellites replace field plots for carbon credits?
No. Current forest carbon methodologies rely on field measurements to calibrate and validate remote-sensing estimates. VM0047's area-based approach, for example, combines remote sensing with plot sampling.
How accurate is satellite biomass estimation?
Accuracy varies by sensor, forest type and calibration data. Vendor claims such as 70% accuracy for biomass and canopy height show that satellite estimates alone carry large errors without good ground data.
How many field plots does a project need?
Enough to meet the methodology's precision target at the stated confidence level, which depends on variability between plots and stratification. A sampling design calculation sets the number.
Do plots need precise GPS?
Precise plot location matters when plots calibrate remote-sensing models or are remeasured. Positional error of several metres can pair a plot with the wrong pixels.