
Concept: ground truthing deforestation alerts with community checks

Updated: 4 days ago
Ground truthing deforestation alerts means sending someone to the place a satellite flagged, to confirm whether tree cover was really lost and to record what caused it. The satellite answers "where and roughly when". A person standing in the clearing answers "what happened here, and does it matter for this project". This article looks at why that second answer matters, and explores an idea for turning alerts into field checks that community monitors could carry out.
Concept note: This article describes a feature we are exploring: turning satellite alerts into community field checks. It is not available in the CitizenClimate app today, and there is no release date or price.
What satellite deforestation alerts can and cannot tell you
Near-real-time alert systems scan new satellite images for signs that tree cover has been disturbed. The best known are combined in an integrated layer that Global Nature Watch describes as including "four alert systems: GLAD-L, GLAD-S2, RADD and DIST-ALERT". RADD is built on Sentinel-1 radar, while the GLAD alerts use optical Landsat and Sentinel-2 imagery, as the study below sets out.
Each alert carries a confidence level. In the integrated layer, "an alert begins as low confidence when the first signs of potential disturbance appear", is upgraded to high confidence "only after additional satellite observations provide further evidence or probability of loss", and reaches highest confidence "only when multiple alerts systems detect a disturbance in the same pixel location", according to the same Global Nature Watch explainer. The explainer presents confidence levels as a way for forest monitors to decide which alerts to follow up first.
Combining systems helps. A 2024 study by Reiche and colleagues in Environmental Research Letters, run at ten sites across the Amazon Basin, found that "Alert integration resulted in faster detection of new disturbances by days to months". The same abstract notes a trade-off: "nearby alerts mutually reinforced their confidence, but it also led to an increased rate of false detections".
Three limits matter for a carbon or nature project:
An alert is a change in pixels, not a cause. It cannot tell a smallholder's new field from a logging road, a landslide or a plantation harvest.
Not every disturbance is deforestation. DIST-ALERT is described as "capturing all vegetation disturbances, not only in tree cover", so the layer also picks up change outside forest, in grassland and cropland.
Some damage is too small or too slow to see. Selective logging under a closed canopy can leave little trace in a satellite pixel.
Why ground truthing deforestation alerts needs people on the ground
Field evidence fills those gaps, and the research on it is consistent.
Pratihast et al. (2012) compared forest change in central Vietnam reported by community members using mobile devices with SPOT satellite imagery. They found that "only 18% of community-reported selective logging events can be visually identified using remote sensing", "whereas around 88% of subsistence agricultural expansion was recognised in the satellite data". Their conclusion was that "communities are more effective to monitor small scale forest degradation".

Cause matters as much as detection. Researchers formerly with Forest Trends checked 209 alerts on the ground in Myanmar, out of 55,340 detected between June 2024 and April 2025. They found that 143 of the 209 (68%) were "recent expansion of smallholder subsistence agriculture", mostly by displaced families, and that "identical satellite alerts represented completely different situations". Families clearing land to survive and an illegal mine need very different responses. Only a visit could tell them apart.
Field checks can also change what happens next. In a randomised study in Loreto, Peru, published in PNAS in 2021, communities chose three members each as forest monitors who received alerts and checked them on patrol. A Global Nature Watch summary of the study reports that communities using the alerts saw deforestation fall by "52% in the first year and 21% in the second year" compared with similar territories. Alert data and GPS coordinates reached the communities monthly on a thumb drive, and monitors worked with them offline. Monitors received a daily stipend of $8 while on patrol.
What the satellite alert gives | What a field check adds | |
|---|---|---|
Location | A pixel or patch, with a date | A GPS point at the site and a walked route |
Is it real? | A confidence level | Photos showing what is on the ground |
Cause | Not available | Farming, logging, mining, fire, natural damage, or a false alarm |
Scale | Area of changed pixels | Notes on what the pixels miss, such as cut stumps under canopy |
Response | None | Information the community or the project can act on |
What the CitizenClimate app does today
The app does not import deforestation alerts, and it does not check submissions against satellite imagery. Its only satellite use is for Plan Vivo projects: the App Guide describes Pillars 4 and 5 as "calculated from Sentinel-2 satellite imagery inside your site boundary". What the app does have is most of the field side of a ground check.
A site boundary. A project developer can "Upload KML" (a KML, KMZ or GeoJSON file) or draw the boundary on a satellite map, and the area is shown in hectares. The Protocol says the boundary is part of the project record that verifiers can read.
Surveys built for the job. According to the App Guide, "Questions can be multiple choice, text, numbers, dates, photos, videos, audio, species identification or GPS. Some only appear after certain answers." A project can build a survey that asks what activity was seen and then asks for photos.

Location you can trust. Location can be required or optional per survey. Surveys that require a location "won't start without a GPS fix", and the Protocol says the app requests "the highest available GPS accuracy from the device (navigation-grade, typically < 5 metres)" and stores the accuracy with each reading. Mapped Route records a walked path, and Boundary Walkover closes a walked path into a polygon.
Offline by design. "The app is built for places with no signal." Submissions made offline queue "with the time and place you recorded them" and send when the phone reconnects. The Protocol says the original timestamp is kept and the sync date is recorded separately.
A record that stays. "Submissions can never be deleted", and "Photos and videos can never be deleted." Faulty records are archived with a reason from a fixed list. Verification bodies have a read-only role with "direct read access to all project data", and exports in CSV and JSON carry raw records.
Project-set rewards. On the field data collection page, "projects can add a currency reward, set per project and per survey". The project decides whether to offer one.
Concept: turning satellite alerts into community field checks
Concept, not a product feature. This describes an idea CitizenClimate is exploring. It is not available in the app today, and there is no release date or price.
Today a project can build a deforestation report survey and its monitors can fill it in wherever they find clearing. The idea we are exploring would start from the other end: a satellite alert inside the project's site boundary would become a field check that a monitor is asked to complete.

Here is how it could work.
Bring alerts into the boundary. The app could read public alert data for the area inside the project's site boundary. The project developer would choose which confidence levels to bring in, since newer, low-confidence alerts arrive faster but include more false detections.
Choose which alerts to check. Not every alert can or should be visited. The developer would weigh confidence and safety first, much as the Myanmar team weighed whether an alert lay in protected forest and whether the area was safe to visit. A visit should never put a monitor at risk.
Send a field check to a monitor. Each chosen alert would become a task on a monitor's phone, with the patch outlined on a map and a short survey attached: is tree cover lost, what caused it, how large does it look, and photos in set directions. The task would download in advance so the visit works with no signal, and the survey would need a GPS fix to start.
Attach the answer to the alert. The submission would link back to the alert it checks, so each alert ends up with a field status: confirmed, partly confirmed, not found, or could not be reached. Like any submission, the record would keep its original time and could be archived with a reason but never deleted.
Let a verifier trace it. A verifier with read access could follow the chain from alert to field check to photos, and export the lot. The project would see which alerts in its boundary have been checked and which are still open.
Our PV Nature walkthrough film, which follows a fictional project, already shows a version of this idea: a satellite change flag sends a monitor out for a ground check. That scene is an illustration. Building it as a working feature is what this concept would involve.
What field checks would not do
They would not replace remote sensing analysts or verifiers. A field check adds evidence about one place. Judging what that evidence means for a project's carbon accounts would stay with the standard and the independent verifier.
They would not prove that unvisited alerts are false. An unchecked alert stays unchecked. The record would need to show that plainly.
They would not make every site reachable. Some alerts sit in places that are too remote, disputed or dangerous to visit, and the record would say so rather than guess.
They would not decide the response. In the Peru study, communities decided together how to follow up on what monitors found. A field check would inform that decision, not make it.
Where field-checked alerts could fit in carbon and nature projects
Forest carbon and nature projects already pair remote sensing with field data. Community-based monitoring does not replace satellites or auditors, it complements them: satellites give wide coverage, and communities add ground-level detail and a continuous record. A field status on each alert would be a practical way to join the two, for REDD+ projects tracking degradation, and for nature projects that need to explain habitat change inside their boundary.
CitizenClimate's data is structured to align with the monitoring requirements of Verra VCS, Gold Standard, Plan Vivo, Puro.earth and Carbon Standards International. CitizenClimate is not certified, accredited or endorsed by any standard body, and it does not issue credits. We are not suggesting that any standard requires or recognises alert field checks of this kind. The point is narrower: where a project already receives alerts, a traceable record of what was found on the ground would make each one easier to act on and to explain.
Frequently asked questions
What does ground truthing a deforestation alert mean?
It means visiting the location a satellite alert flagged to confirm whether tree cover was actually lost and to record the cause. A good check leaves evidence behind: photos of the site taken at a recorded GPS point and time, plus a note of the cause. That cause might be farming, logging, mining or fire, or the alert might turn out to be a false alarm. That evidence turns a pixel-level signal into information a project can act on.
How reliable are satellite deforestation alerts?
They are useful but not final. Alerts start as low confidence and are upgraded as more satellite passes confirm the change; the integrated layer marks highest confidence when several systems agree. Reiche and colleagues found that combining systems speeds up detection by days to months, but also increases false detections. Alerts also cannot show the cause of a change, which is why many programmes check a sample on the ground.
Can community monitors detect forest degradation that satellites miss?
Research suggests they can. A study in central Vietnam by Pratihast et al. found that only 18% of selective logging events reported by community members could be identified in SPOT satellite imagery, compared with around 88% of subsistence farming expansion. Small-scale degradation under the canopy is hard to see from space, and people who walk the forest regularly are well placed to record it.
Does the CitizenClimate app import deforestation alerts?
No. The app does not import alerts or check submissions against satellite imagery. Its only satellite use is Sentinel-2 imagery for Plan Vivo Pillars 4 and 5 inside the site boundary. It does provide the field tools a ground check needs: site boundaries, photo and GPS questions, surveys that require a location fix, offline submission with the original timestamp kept, records that cannot be deleted and read-only access for verifiers.
Try the field side today
Alert-driven field checks are a concept. The tools a monitor would use on the ground are in the app now, from photo and GPS questions to offline submission and a record verifiers can read. Download the free app and build a deforestation report survey for your own monitors, or read the App Guide first.



Comments