Deforestation satellite imagery has transformed forest monitoring from periodic ground surveys into near-real-time global surveillance. Satellites like Landsat and Sentinel-2 capture the Earth’s surface at regular intervals, and algorithms flag where forest cover has disappeared between passes. This is the foundation of every major deforestation tracking system operating today, from Global Forest Watch to Brazil’s PRODES program.
This article explains how satellite imagery detects deforestation, compares the main satellite programs and their capabilities, describes the detection methods used, and covers the satellite requirements for EUDR compliance.
How Satellite Imagery Detects Deforestation
The basic principle is straightforward: healthy forest has a distinct spectral signature that satellites can identify. Vegetation absorbs red light for photosynthesis and strongly reflects near-infrared light. When forest is cleared, the spectral signature shifts because bare soil, crops, and pasture reflect light differently than a forest canopy.
Satellites capture this spectral information across multiple bands (visible, near-infrared, shortwave infrared) at regular intervals. By comparing images from two dates, algorithms detect where the forest signature has been replaced by something else. This is called change detection, and it works at scales from individual farm plots to entire continents.
The accuracy depends on three factors: spatial resolution (how small a clearing can be detected), temporal resolution (how often the satellite revisits the same location), and cloud cover (which blocks optical sensors). Tropical regions, where most deforestation occurs, have persistent cloud cover that can delay detection by weeks or months with optical sensors alone.
Key Satellite Programs for Deforestation Monitoring
Landsat (USGS/NASA)
Landsat is the longest-running Earth observation program, with continuous coverage from 1972. Landsat 8 and 9 capture imagery at 30-meter resolution with a 16-day revisit cycle. The Hansen Global Forest Change dataset, which tracks annual tree cover loss worldwide, is built on Landsat data. The 30-meter resolution means clearings smaller than about 0.1 hectares may go undetected, but for large-scale deforestation tracking, Landsat remains the gold standard because of its 50+ year archive.
Sentinel-2 (ESA)
The European Space Agency’s Sentinel-2 satellites provide 10-meter resolution optical imagery with a 5-day revisit cycle at the equator. The higher spatial resolution compared to Landsat allows detection of smaller clearings, and the faster revisit reduces the window of undetected change. Sentinel-2 data powers Google’s Dynamic World land cover product, which classifies the Earth’s surface into nine categories in near-real-time. Dynamic World is one of the most recent additions to the land cover analysis toolkit and is worth understanding alongside the longer-established programs.
MODIS (NASA)
MODIS sensors on the Terra and Aqua satellites capture imagery at 250-meter to 1-kilometer resolution but revisit every 1-2 days. The low resolution cannot detect small clearings, but the daily revisit makes MODIS valuable for fire detection and rapid alerts. The MODIS-based GLAD (Global Land Analysis and Discovery) system provides weekly alerts for probable forest disturbance across the tropics.
SAR (Synthetic Aperture Radar)
SAR satellites (Sentinel-1, ALOS-2, ICEYE) transmit microwave pulses and measure the signal that bounces back. The critical advantage: radar penetrates cloud cover, which blocks Landsat and Sentinel-2 in tropical regions for weeks at a time. SAR detects changes in forest structure by measuring differences in radar backscatter between forested and cleared areas. The tradeoff is that SAR imagery is harder to interpret than optical data, and distinguishing forest types is less precise. Programs like RADD (Radar Alerts for Detecting Deforestation) combine SAR with optical data to provide alerts even during cloudy seasons.
Resolution Comparison
Choosing the right satellite data depends on what you need to detect. Here is how the main programs compare:
| Program | Spatial Resolution | Revisit Time | Sensor Type | Best For |
|---|---|---|---|---|
| Landsat 8/9 | 30 m | 16 days | Optical/thermal | Long-term trends, annual reporting |
| Sentinel-2 | 10 m | 5 days | Optical | EUDR compliance, small-scale clearing |
| MODIS | 250 m-1 km | 1-2 days | Optical/thermal | Fire detection, rapid alerts |
| Sentinel-1 | 10 m | 6 days | SAR (radar) | Cloud-penetrating monitoring |
| Dynamic World | 10 m | ~5 days | AI + Sentinel-2 | Near-real-time land cover classification |
A resolution of 10 meters means each pixel covers a 10m x 10m area on the ground. At this resolution, clearings as small as a few hundred square meters become visible. For context, a typical smallholder cocoa or coffee farm in West Africa is 1-5 hectares, so 10-meter imagery can detect clearing at the individual farm level.
Change Detection Methods
Pixel-Based Comparison
The simplest approach compares the same pixel across two dates. If a pixel classified as “tree cover” in the baseline image is classified as “cropland” or “bare ground” in the later image, the system flags it as change. This is the basis of the Hansen Global Forest Change dataset, which has tracked annual tree cover loss globally since 2000 using Landsat imagery at 30-meter resolution.
Time-Series Analysis
Rather than comparing just two images, time-series analysis tracks the spectral values of each pixel over months or years. This approach catches gradual degradation that a two-date comparison would miss: selective logging, for example, removes individual trees without creating the clear-cut signature that pixel comparison relies on. Platforms like Google Earth Engine make it possible to process decades of satellite imagery for any location, generating land cover change trajectories at scale.
Machine Learning Classification
Modern systems like Dynamic World use deep learning models trained on millions of labeled satellite images to classify each pixel into land cover categories. The model processes each new Sentinel-2 image automatically, producing a near-real-time land cover map without manual interpretation. The advantage over rule-based methods is that the model adapts to regional differences in vegetation, soil color, and agricultural patterns that would require separate calibration in traditional approaches.
Real-World Applications
Global Forest Watch combines Hansen data, GLAD alerts, and RADD radar alerts into a single monitoring platform used by governments, NGOs, and companies to track deforestation as it happens. Brazil’s PRODES program uses Landsat imagery to publish annual Amazon deforestation figures that directly inform enforcement operations by IBAMA, Brazil’s environmental agency.
In the private sector, companies with tropical commodity supply chains use deforestation satellite imagery to verify that sourcing locations are deforestation-free. A palm oil buyer can overlay supplier GPS coordinates onto satellite time-series data and check whether forest cover existed at that location on a specific date and whether it has since been cleared.
EUDR Geolocation Requirements
The EU Deforestation Regulation creates specific requirements for satellite verification that go beyond general monitoring. Under the EUDR, companies placing palm oil, soy, coffee, cocoa, cattle, rubber, or wood products on the EU market must provide:
- GPS coordinates of the plot where the commodity was produced (polygon for plots >4 hectares, single point for smaller plots)
- Evidence that the plot was not deforested after December 31, 2020 (the EUDR cutoff date)
- Evidence that production complied with local laws in the country of origin
Satellite imagery is the primary tool for verifying the second requirement. The process involves comparing land cover at the GPS coordinates on the cutoff date (December 31, 2020) against current land cover. If the coordinates show forest in 2020 and cropland today, the plot fails deforestation due diligence.
Sentinel-2’s 10-meter resolution and 5-day revisit make it the most practical satellite source for EUDR verification. Its archive extends back to 2015, covering the cutoff date with multiple clear images for most locations. For areas with persistent cloud cover, SAR data from Sentinel-1 provides backup verification.
Frequently Asked Questions
What satellite imagery is used to detect deforestation?
The main programs are Landsat (30m resolution, USGS/NASA), Sentinel-2 (10m resolution, ESA), MODIS (250m-1km, NASA for fire and rapid alerts), and SAR satellites like Sentinel-1 (10m radar that penetrates clouds). Global Forest Watch and the Hansen dataset use Landsat as their primary data source.
How accurate is satellite deforestation monitoring?
Accuracy depends on resolution and cloud cover. At 10-meter resolution (Sentinel-2), clearings as small as a few hundred square meters can be detected. The Hansen Global Forest Change dataset, using 30-meter Landsat data, has an overall accuracy above 95% for detecting annual tree cover loss. Cloud cover in the tropics can delay detection by weeks with optical sensors alone.
Can satellites detect illegal deforestation in real time?
Near-real-time, not instant. GLAD alerts provide weekly updates using Landsat data. RADD uses Sentinel-1 radar to deliver alerts even during cloudy periods. Brazil’s DETER system provides daily alerts for the Amazon. The typical detection lag is 1-4 weeks depending on the system and cloud conditions.
What resolution do you need to verify EUDR compliance?
The EUDR does not specify a minimum resolution, but 10-meter resolution (Sentinel-2) is the practical standard. At 10 meters, individual farm plots of 1-5 hectares are clearly distinguishable. The regulation requires GPS coordinates and evidence that the plot was not deforested after December 31, 2020.
How does radar (SAR) detect deforestation differently from optical satellites?
Optical satellites measure reflected sunlight and detect deforestation through changes in vegetation spectral signatures. SAR satellites transmit microwave pulses and measure the returned signal, detecting changes in forest structure. The advantage of SAR is cloud penetration: radar works through clouds that block optical sensors for weeks in tropical regions.
What is Dynamic World and how does it relate to deforestation monitoring?
Dynamic World is a Google-developed land cover dataset that uses deep learning to classify every Sentinel-2 image into nine land cover categories in near-real-time. It provides 10-meter resolution maps updated every 5 days, tracking land cover change at individual plot level. It represents a shift from annual assessments to continuous monitoring.
Deforestation satellite imagery has evolved from a research tool into the enforcement backbone of global forest protection. The data is freely available, the algorithms are proven, and the scale of the problem is well documented. What has changed is the regulatory requirement: the EUDR now mandates that companies verify their supply chains against satellite evidence at specific GPS coordinates. The technology exists; the question is whether it gets applied systematically enough to shift commodity production away from standing forest.
