# Remote Sensing or Field Plots? What Satellites, Drones and LiDAR Can Measure for Carbon Projects

Last updated: 2026-11-02

> What satellites, drones and LiDAR can measure, what VM0047, Isometric, Gold Standard and Plan Vivo accept from each, and how they combine with field plots.

Date: November 2, 2026

> [ARR Carbon Guide](https://restorationscope.com/blog/category/carbon): take your ARR project from standard to verification, step by step.

Both. No main carbon standard for reforestation credits removals measured by remote sensing alone. Satellites, drones and LiDAR map area, tree cover and height everywhere, set the baseline and show where to put plots. Field plots measure biomass, survival and species, and calibrate every map. Verra's VM0047 measures credited biomass on the ground. Isometric accepts airborne LiDAR or a global biomass map, and Plan Vivo's PM002 accepts a satellite model, but both still need plots: Isometric's at least every five years, PM002's 30 per ecoregion at the start and 30 more at least every five years.

This guide is for the MRV or forestry lead deciding what to measure from the air and what on the ground, and for funders reading that plan. Restoration Scope, which publishes it, makes field software for plot and tree records. The free [forest inventory plot calculator](https://restorationscope.com/tools/forest-inventory-plot-calculator) works out how many plots your precision target needs.

**Checked November 2, 2026** against the methodology documents of Verra, Isometric, Gold Standard, Plan Vivo and the CDM, the GFOI and CEOS guidance, and the peer-reviewed accuracy studies linked in each section. Your methodology and your validation and verification body (VVB) decide what is enough.

## What can satellites, drones and LiDAR measure?

Each source sees some things well and others not at all. Biomass is the hard one: every remote estimate of it is a model trained on field plots.

| What you need | Free satellites (Sentinel-2, Landsat) | Global maps (GEDI, height and biomass maps) | Sub-metre satellite images | Drone camera | LiDAR (airborne or drone) | Field plots |
| --- | --- | --- | --- | --- | --- | --- |
| Tree cover and area | Yes, 10–30 m pixels | Partly, Hansen counts trees over 5 m only | Yes, larger crowns | Yes | Yes | Yes, inside the plot |
| Canopy height | No | Partly, 3–8 m mean error, 1–30 m maps | Partly, through models | Yes, with a ground model | Yes, about 8% per tree | Yes |
| Above-ground biomass | Partly, Landsat saturated at 55–159 t/ha in one study | Partly, about 50% per pixel | Partly, 55% per tree (25 cm aerial photos) | Partly, calibrated to plots | Yes, 13% at 1 ha in one Borneo forest | Yes, with allometric error |
| Young trees and survival | No | No, GEDI blind under 2–4 m | Partly, after a few years | Partly, crowns over 60 cm | Partly | Yes |
| Species | No | No | No | Partly, 73% to 90% with local training | No, not alone | Yes |

Sources: cover, [Hansen Global Forest Change v1.13](https://storage.googleapis.com/earthenginepartners-hansen/GFC-2025-v1.13/download.html) and [Brandt et al. 2020](https://doi.org/10.1038/s41586-020-2824-5) (Sahel, 0.5 m images, crowns over 3 m²); height, [Brandt et al. 2026](https://arxiv.org/abs/2603.06382) (preprint), table 1, [Yu et al. 2024](https://doi.org/10.1088/2752-664X/ad39f2) and [Dalla Corte et al. 2020](https://doi.org/10.3390/rs12050863); biomass, [Zhao et al. 2016](https://doi.org/10.3390/rs8060469), [Santoro et al. 2021](https://doi.org/10.5194/essd-13-3927-2021), section 3.1 (the 2010 GlobBiomass map), [Araza et al. 2022](https://doi.org/10.1016/j.rse.2022.112917), [Mugabowindekwe et al. 2023](https://doi.org/10.1038/s41558-022-01544-w) and [Coomes et al. 2017](https://doi.org/10.1016/j.rse.2017.03.017); young trees, [Fromm et al. 2019](https://doi.org/10.3390/rs11212585); species, [Schiefer et al. 2020](https://doi.org/10.1016/j.isprsjprs.2020.10.015) and [Veras et al. 2022](https://doi.org/10.1016/j.ecoinf.2022.101815).

Field plots carry error too. The widely used pantropical allometric model, with tree height, predicts a single tree's biomass with a mean error of 56.5% across sites, and its estimate of a site's total biomass was off by more than 30% at 8 of the 58 sites it was fitted on ([Chave et al. 2014](https://doi.org/10.1111/gcb.12629)). Random error per tree shrinks over many trees and plots; a bias at one site doesn't, which is why some standards add a model error for equations that weren't checked locally.

## What does each carbon standard accept?

| Standard | Credited biomass from | What remote sensing does | Field plots still needed |
| --- | --- | --- | --- |
| Verra VM0047 v1.1 | Field plots, or sampled trees in a census-based project | Baseline (stocking index), area, pre-project biomass disturbed before plots | Every five years or more often |
| Isometric Reforestation v1.2 | Field plots, airborne LiDAR or an approved biomass map | Baseline, run by Isometric; optional drone checks of early mortality | Start and end of each period with plots alone; at least every five years to check LiDAR or a map |
| Gold Standard A/R v2.1 (closed to new projects) | Growth models, checked by field-plot inventories | Not named in the methodology | Before every performance certification |
| Gold Standard STARR (draft) | Field plots, a full census or volume sampling | Area and a matched baseline, from year 4, once trees pass 1.5 m or 10% cover | Full inventory at least every five years |
| Plan Vivo PM002 v1.0 | A satellite model calibrated on plots | The credited biomass | 30 per ecoregion at the start and 30 more at least every five years |
| CDM AR-TOOL14 v4.2 | Field plots | A second measurement on sample plots, strata, a no-decrease check | At each verification, unless a no-decrease check applies |

Sources: [VM0047 v1.1](https://verra.org/documents/vm0047-afforestation-reforestation-and-revegetation-v1-1/), sections 8.2.1.2, 8.5 and 9.2, appendix 1; [Isometric Reforestation v1.2](https://registry.isometric.com/protocol/reforestation), sections 9.3–9.4 and 12, and its [LiDAR](https://registry.isometric.com/module/agb-lidar-quantification/1.0) and [global map](https://registry.isometric.com/module/agb-global-map-quantification/1.0) modules v1.0; [Gold Standard A/R v2.1](https://globalgoals.goldstandard.org/standards/403_V2.1_LUF_AR-Methodology-GHGs-emission-reduction-and-Sequestration-Methodology.pdf), section 3.11; [STARR consultation draft](https://goldstandard.cdn.prismic.io/goldstandard/aeiFSsBOoF08xMq4_GSARR-Meth_V1PC.pdf), sections 7, 9.2 and 15; [Plan Vivo approved methodologies](https://www.planvivo.org/projects/certify-a-project/pvclimate/methodologies/approved-methodologies), PM002 and module PU006; [AR-TOOL14 v4.2](https://cdm.unfccc.int/methodologies/ARmethodologies/tools/ar-am-tool-14-v4.2.pdf), sections 6.4 and 8.1.

Status as of November 2, 2026.

**Verra VM0047.** In an area-based project, above-ground woody biomass "must be measured via plot-based sampling", and Verra's remote-sensing biomass tool, VT0005, does not apply; a census-based project measures a sample of its planting units. Remote sensing builds the performance benchmark: a stocking index sampled at least once a year on 30 or more remotely sensed project plots and matched with remotely sensed control plots. The benchmark is updated at each verification or every five years, whichever comes first. The methodology sets no accuracy number for the index, only a published or locally validated link to biomass. Verra [vets stocking index providers](https://verra.org/verra-announces-five-new-data-service-providers-for-arr-methodology/) but lets projects use others. In December 2024, reviewing version 1.0, ICVCM noted that the benchmark had not been tested empirically yet ([Board observations](https://icvcm.org/wp-content/uploads/2024/12/Board-Observations_ARR_part-I_VM0047.pdf)).

**Isometric.** The one standard here where credited biomass can come from airborne LiDAR or an approved biomass map. A LiDAR model's R² and error must fall within the range of published LiDAR biomass models. A map must agree with benchmark plots within ±10% at 90% confidence, and a LiDAR map should; a map that fails is corrected with a regression on field plots, with an R² of at least 0.85. Version 1.1 drafts of these modules, in consultation until 26 November 2026, would drop the equivalence test.

**Plan Vivo PM002.** Built for smallholder agroforestry, it credits biomass change from a model, usually machine learning, on optical and radar satellite images. Without an R² of at least 0.7 and a normalised error of 30% or less on its validation plots, the model can't be applied; optical imagery coarser than 30 m can't be used on its own.

**Gold Standard.** Its A/R methodology v2.1 doesn't mention remote sensing, and it closed to new projects on 30 June 2026, when Gold Standard's deadline for submissions under methodologies not yet aligned with the Paris Agreement passed. STARR, its replacement, is still a draft. The STARR draft uses it for area and a matched baseline, but switches the comparison off for years 0–3 and starts it once trees pass 1.5 m or canopy cover passes 10%: optical satellites "cannot reliably detect biomass changes in seedlings or saplings (<1.5m)".

## How do remote sensing and plots work together?

There are three ways, and they treat a map's errors very differently.

1. **Strata.** Draw zones that grow alike from imagery (planting year, canopy cover, height), then sample each. Variation inside a zone is smaller, so fewer plots reach the same precision. The IPCC warns that wrong boundaries can cause "substantial bias", and GFOI passes on the advice to keep national inventories to 6–8 strata.
2. **Assist.** Measure the plots, then use a map or a satellite index to sharpen the estimate (a model-assisted or double-sampling estimate). A poor map helps little; the result stays about unbiased. AR-TOOL14's double sampling works this way: a satellite index on many plots, trees measured on some, if the two are linearly related.
3. **Replace.** Add up a map's pixels and use plots only to train and check it. Independent pixel errors average out (a global map's errors are correlated over 50–104 km, so they average out less), but a calibration that runs 10% high runs 10% high on every pixel. That share doesn't shrink with more pixels, so the map needs plots it wasn't trained on to prove it.

How much a map saves depends on how closely it tracks biomass. In Tanzanian miombo woodland, researchers compared the plots needed for the same precision as plots alone:

| Map used with the plots | Plots still needed |
| --- | --- |
| Airborne LiDAR | 29% |
| RapidEye satellite images | 31% |
| TanDEM-X radar | 40% |
| Global PALSAR radar map | 72% |
| Global Landsat map | 82% |

Sources: [Næsset (ed.) 2015](https://static02.nmbu.no/mina/publikasjoner/mina_fagrapport/pdf/mif31.pdf), NMBU report 31, p. 28; [Ståhl et al. 2016](https://doi.org/10.1186/s40663-016-0064-9), pp. 2–4; [IPCC 2006 Guidelines](https://www.ipcc-nggip.iges.or.jp/public/2006gl/pdf/4_Volume4/V4_03_Ch3_Representation.pdf), volume 4, annex 3A.3.3; GFOI Methods and Guidance 3.0 (2020), [box 26 and section 3.2.1.3](https://www.fs.usda.gov/research/treesearch/61714).

That comparison counted plots, not money; the researchers concluded that LiDAR could still pay for a current stock estimate even with Tanzania's low labour costs. A companion study over 16,000 km² found that for biomass change over two years on remeasured plots, LiDAR added little: remeasuring the same plots already compares each one with itself. The change it measured, in natural woodland, was small; ARR growth is faster, but remeasured permanent plots are the usual ARR design.

Calibration plots have their own rules. The CEOS biomass protocol plans on at least about 30 plots to train a LiDAR biomass model, at least 0.25 ha in the tropics and ideally 1 ha; a LiDAR model trained on tropical plots under 0.1 ha can miss by more than 40%. Under forest canopy, about half of GPS fixes are off by more than 5 m, so a plot centre needs many fixes averaged: in a test in Gabon, averaging 50 kept nearly all positions within 3 m ([CEOS protocol, 2021](https://doi.org/10.5067/doc/ceoswgcv/lpv/agb.001), pp. 18, 74–77 and 91).

## Why satellites miss young plantings

A planting is small for years, and most satellite products are built for forests:

- **Tree cover:** Hansen's global tree cover counts only vegetation over 5 m tall.
- **Height:** GEDI's spaceborne LiDAR can't measure canopy under about 2–4 m, and is biased by more than 3 m where canopy is under 5 m ([Yu et al. 2024](https://doi.org/10.1088/2752-664X/ad39f2)).
- **Height maps:** one widely used 10 m global map overestimates canopies under 5 m by 10–12 m on average at three temperate sites ([Moudrý et al. 2024](https://doi.org/10.1002/ecs2.70026)).
- **Biomass change:** for regrowth of about 4 tonnes of carbon per hectare a year in the western Andes, the CEOS protocol expects 5–10 years before GEDI-level accuracy can confidently detect the change.

Drones are the exception. On cleared conifer sites, a trained model found 98.8–99.5% of seedlings ([Pearse et al. 2020](https://doi.org/10.1016/j.isprsjprs.2020.08.005)), though detection is reliable mainly for crowns over about 60 cm and falls among weeds and regrowth. For the first years, survival counts come from the ground or from low drone flights, and the [survival guide](https://restorationscope.com/blog/track-tree-planting-survival) covers the field side.

## What does each cost?

| Source | Cost or effort | Catch |
| --- | --- | --- |
| Sentinel-2, Landsat, GEDI, global maps | Free | Too coarse for young trees; maps carry their own error |
| Sub-metre satellite images | $12.50–32.50 a km² (list price, May 2024) | Minimum orders of 25–100 km² |
| Drone LiDAR | Up to 2.5 km² a flight at 150 m and 20% overlap (DJI Zenmuse L2 spec) | Less at the 120 m height limit many countries set |
| Airborne LiDAR | $1.30–3.20 a hectare for US projects of hundreds of km² (2012–2014); $2.50–6 a hectare in a 2019 estimate, often less | Aircraft transit weighs most on remote jobs |
| Field plot, 0.25 ha, all carbon pools | About $1,330 a plot for field and lab work (eastern Amazon; our division of $298,000 over 224 plots), plus logistics | Leaving out species identification cut vegetation costs by about half |
| Field plot, 1 ha, high-diversity tropical recensus | About €15,000 a plot (2016 prices) | With every tree identified to species |

Sources: [LAND INFO price list](https://landinfo.com/satellite-imagery-pricing/), May 2024, for North American buyers; [DJI Zenmuse L2 specs](https://enterprise.dji.com/zenmuse-l2/specs); [USGS Open-File Report 2015-1161](https://pubs.usgs.gov/of/2015/1161/ofr20151161.pdf), table 3 (2012–2014); CEOS protocol, p. 102, citing Réjou-Méchain et al. 2019; [Berenguer et al. 2015](https://doi.org/10.1371/journal.pone.0133139); [Chave et al. 2019](https://doi.org/10.1007/s10712-019-09528-w), section 5.

GFOI's guidance measures a map's value as relative efficiency: how many times more plots you would need without it. Set the plots it saves against what it costs. A project that would need 105 plots alone, and 30 with a map that tracks biomass well, saves 75 plots; at $1,500 a plot, our illustrative figure, a map under $112,500 pays on precision alone. Run it with your own plot costs: in the Tanzanian work, LiDAR paid for a stock estimate even with low labour costs.

## Which mix fits your project?

- **VM0047, area-based:** the stocking index from a provider for the benchmark, permanent biomass plots every five years, drone flights for survival in the first years. The [VM0047 field data guide](https://restorationscope.com/blog/vm0047-arr-field-data) lists what to record.
- **Large or structurally varied sites, Isometric:** airborne LiDAR trained on at least about 30 plots (the CEOS protocol's figure), plus benchmark plots to check it at least every five years, can replace a dense plot grid.
- **Smallholder agroforestry:** Plan Vivo's PM002 credits a satellite model, calibrated on 30 plots per ecoregion at the start and 30 more at least every five years.
- **Any project, before the first plot:** free imagery and global layers to stratify and find open land. Restoration Scope's [canopy height screening](https://restorationscope.com/analyses/canopy-height-screening) reads Meta and WRI's 1.2 m height map for a property.

[Isometric vs Verra](https://restorationscope.com/blog/isometric-vs-verra) and [Plan Vivo vs Gold Standard vs Verra](https://restorationscope.com/blog/plan-vivo-vs-gold-standard-vs-verra) compare the standards on everything else.

## Where software fits

Every route above ends in field records: which plot is where, which trees are in it, who measured them and when, and which plots a crew couldn't find. A map is only as good as the plots it was trained and checked on.

Restoration Scope keeps those records: plots and trees captured offline with GPS and photos, the history of every visit on each record, drone orthomosaics and global layers such as canopy height on the same map, and CSV exports for your VVB. It doesn't estimate biomass or calculate credits. See [Restoration Scope for carbon projects](https://restorationscope.com/solutions/carbon).

**A monitoring plot on a phone.**

The plot on the map with each tree recorded inside it, and the plot's record below. Demo project: names and numbers are made up.

![Demo monitoring plot MP0009 on a phone: the plot circle on the satellite map with five tree points inside it, above the plot record with trees monitored and species counts](https://restorationscope.com/images/landing/mobile-geospatial/demo-plot-point.png)

More guides: [how many inventory plots an ARR project needs](https://restorationscope.com/blog/forest-inventory-plots-arr) and [what dMRV is](https://restorationscope.com/blog/what-is-dmrv).

**Disclosure:** Restoration Scope makes field and monitoring software for carbon projects. We are not a validation or verification body (VVB), and we don't issue credits.

## Questions about remote sensing and field plots

### Can satellites replace field plots in a reforestation carbon project?

Not under the main standards checked on November 2, 2026. Verra's VM0047 measures credited biomass on the ground, in plots or sampled trees, and uses satellites for the baseline. Isometric accepts airborne LiDAR or a global biomass map, and Plan Vivo's PM002 accepts a satellite model, but both need field plots to calibrate or check them, with new ones at least every five years.

### Does VM0047 accept remote sensing for biomass?

Only around the edges. Under VM0047 v1.1, an area-based project must measure above-ground woody biomass with plot-based sampling, and Verra's remote-sensing biomass tool VT0005 does not apply; a census-based project measures a sample of its planting units. Remote sensing sets the performance benchmark through a stocking index, may serve as an auxiliary variable in double sampling, and, where woody biomass was disturbed before plots could be measured, may estimate it using the upper 90% confidence bound of the model's prediction interval.

### When can satellites see a newly planted tree?

Usually not for years. Hansen's global tree cover counts only vegetation over 5 m tall, a validation study found GEDI's spaceborne LiDAR can't measure canopy under about 2–4 m, and Gold Standard's draft ARR methodology says optical satellites can't reliably detect seedlings or saplings under 1.5 m. For regrowth of about 4 tonnes of carbon per hectare a year in the western Andes, the CEOS biomass protocol expects 5–10 years before GEDI can confidently detect the change.

### How accurate are global biomass and canopy height maps for a single project?

Weak per pixel. ESA's 2010 GlobBiomass map, the forerunner of its CCI Biomass maps, has a standard deviation averaging 50% of the estimate per pixel. Across four global biomass maps, errors were correlated over 50–104 km, so averaging a project's pixels cancels far less of them than independent errors would. Global canopy height maps at 10–30 m are off by 5–8 m on average against airborne LiDAR, and a newer 1 m map by 3 m in its makers' tests, and one 10 m map overestimated canopies under 5 m tall by 10–12 m on average at three temperate sites.

### How many field plots does a LiDAR biomass map need?

The CEOS biomass validation protocol plans on at least about 30 plots to train an airborne LiDAR biomass model, of at least 0.25 ha in the tropics and ideally 1 ha. Isometric's LiDAR module asks for ground plots overlapping the scan to train the model, and recommends enough benchmark plots for an equivalence test within ±10% at 90% confidence; drafts of version 1.1, in consultation until 26 November 2026, would drop that test.

### Is airborne LiDAR cheaper than more field plots?

It depends on what a plot costs where you work. In Tanzanian miombo woodland, LiDAR cut the plots needed for the same precision to 29%, and the researchers found it could pay even with Tanzania's low labour costs for a current stock estimate. In a two-year study of change in natural woodland on remeasured plots, LiDAR added little. A map pays when the plots it saves cost more than the map.