{"id":90489,"date":"2026-10-09T18:15:44","date_gmt":"2026-10-09T10:15:44","guid":{"rendered":"https:\/\/wp-productionenv-bjg9h2g2bgg5b8aa.southeastasia-01.azurewebsites.net\/?p=90489"},"modified":"2026-10-09T18:16:52","modified_gmt":"2026-10-09T10:16:52","slug":"can-satellite-imagery-detect-selective-logging","status":"publish","type":"post","link":"https:\/\/starpath.global\/blog\/can-satellite-imagery-detect-selective-logging\/","title":{"rendered":"Can Satellite Imagery Detect Selective Logging?"},"content":{"rendered":"

Satellite imagery can detect some selective logging, particularly when tree removal creates visible canopy gaps, new access routes or repeated changes in the radar signal. However, detecting a large clear-cut is a different task from identifying scattered tree removal beneath a mostly intact canopy. The reliability of the result depends on the disturbance footprint, forest structure, image resolution, observation timing and terrain. A satellite alert can help direct an investigation; it cannot, by itself, establish that logging was illegal.<\/p>\n

A question posted on Reddit\u2019s r\/remotesensing illustrates this distinction. The user wanted to use Sentinel-1 radar imagery to identify logging during the wet season, when clouds limited optical observations. In a follow-up, they described areas of interest around 100 \u00d7 100 meters and potentially hilly terrain. Those details change the monitoring problem: radar\u2019s ability to observe through clouds does not guarantee that small forest disturbances will be distinguishable from other changes.<\/p>\n

Clear-Cutting, Selective Logging and Individual Tree Removal<\/h3>\n

Large clear-cuts generally produce a continuous change across many image pixels. Much of the canopy disappears, and exposed soil, logging debris or replacement vegetation may become visible. The spatial pattern is usually easier to distinguish from the surrounding forest than scattered damage.<\/p>\n

Selective logging removes selected trees while leaving much of the forest standing. Its detectable footprint can include crown openings, damage to neighboring trees, skid trails, roads and timber-loading areas. These features may be fragmented, partly concealed or too narrow to resolve individually. A logged forest can therefore remain classified as forest even though its structure has changed.<\/p>\n

Individual tree removal presents the greatest challenge. Removing a large canopy tree may create an observable opening, but removing a smaller tree beneath overlapping crowns may leave little visible change from above. Detecting a gap also does not necessarily identify the number, species or timber volume of the trees removed.<\/p>\n

These distinctions matter when interpreting a map. \u201cNo forest-loss alert\u201d does not mean \u201cno logging,\u201d and a mapped canopy gap does not automatically mean \u201ca harvested tree.\u201d Natural tree falls can create similar openings.<\/p>\n

A One-Hectare Site Is Not Necessarily a One-Hectare Disturbance<\/h3>\n

A 100 \u00d7 100-meter area covers one hectare. If almost all its canopy is removed, the change is relatively extensive. If only a few scattered trees are harvested within that hectare, the actual disturbed area may be much smaller. Monitoring should therefore be designed around the expected canopy gaps and associated infrastructure, rather than the size of the inspection boundary alone.<\/p>\n

At a 10-meter optical pixel size, a 100-meter-wide site spans roughly ten pixels. That describes its sampling, not a detection guarantee. A small opening can share a pixel with intact crowns, shade and understory vegetation, weakening its signal. Pixel alignment, image registration and the contrast between the opening and its surroundings also affect visibility.<\/p>\n

Sentinel-2 provides visible and near-infrared bands at 10-meter resolution, while its red-edge and shortwave-infrared bands are supplied at 20 meters. Resampling those bands onto a finer grid does not create additional spatial detail. Higher-resolution commercial imagery can provide more detail for reviewing small openings and access routes, but overlapping crowns, shadows and unsuitable viewing conditions can still obscure them.<\/p>\n

There is no minimum detectable gap size that applies to every forest. Any operational threshold should be tested against local reference data and accompanied by information about missed disturbances and false alerts.<\/p>\n

What Optical Imagery Can Reveal<\/h3>\n

Optical imagery can show changes in canopy texture, vegetation reflectance, exposed soil and woody debris. A new opening accompanied by a developing road or loading area provides more context than an isolated decline in a vegetation index.<\/p>\n

NDVI and other spectral indicators can help locate changes, but they do not identify the cause on their own. Seasonal leaf changes, drought, fire, storm damage and cloud shadows may also alter the observed signal. Analysts should review the original imagery and the wider spatial pattern before describing an anomaly as possible logging.<\/p>\n

Timing is particularly important. A 2019 study by Hethcoat and colleagues, published in Remote Sensing of Environment<\/em>, used detailed logging records to train a Landsat-based classification method in the Brazilian Amazon. Images obtained during the logging season performed better than images from the following dry season. The study demonstrates that even a real disturbance may become harder to identify when the next usable observation arrives too late.<\/p>\n

Repeated observations help distinguish a persistent opening from a temporary shadow or seasonal change. Nevertheless, vegetation can begin to obscure a gap while the forest\u2019s original structure and biomass remain altered. Visible recovery should not be equated with complete ecological recovery.<\/p>\n

What SAR Adds\u2014and What Can Interfere<\/h3>\n

Synthetic Aperture Radar, or SAR, provides observations through cloud cover and without sunlight. Sentinel-1\u2019s C-band radar can reveal changes in backscatter associated with altered canopy structure and the geometry of forest openings. This makes radar time series useful when optical imagery is repeatedly blocked during the rainy season.<\/p>\n

However, radar backscatter is not a direct tree-removal measurement. Moisture changes in vegetation and exposed ground can affect the signal, as can seasonal variation and viewing geometry. Speckle adds variability. A single darker or brighter observation is therefore insufficient evidence of logging.<\/p>\n

Comparable observations are essential. Radar time series should normally be evaluated within consistent relative orbits, viewing directions and polarization configurations. Ascending and descending observations can provide complementary coverage, but their different geometries should be accounted for rather than treated as interchangeable measurements.<\/p>\n

Sentinel-1 pixel spacing also needs careful interpretation. Common IW GRD products have approximately 10-meter pixel spacing, but their effective spatial resolution is coarser. A grid containing 10-meter pixels does not mean that every 10-meter-wide forest feature can be independently resolved.<\/p>\n

In mountainous forests, slopes change the local angle between the radar beam and the surface. Foreshortening compresses features, layover mixes returns from different positions, and radar shadow leaves some areas without a usable return. Terrain correction and radiometric slope correction help manage these effects, but they cannot recover information that was not observed. Unusable areas should be marked as coverage limitations.<\/p>\n

Forest canopy gaps can themselves produce radar shadows that specialized methods exploit. These small structural signals must be distinguished from the much larger terrain effects that can dominate steep landscapes.<\/p>\n

Real Before-and-After Evidence<\/h3>\n

The University of Sheffield published a real image comparison in April 2025 showing a logging concession before and after selective logging. The 2020 image shows the earlier canopy condition; the 2023 image shows openings within a forest that largely remains standing. A separate image overlays detections from the project\u2019s algorithm.<\/p>\n

\"Satellite
Before: forest canopy in 2020. Image published by the University of Sheffield; Google Earth imagery \u00a9 Airbus 2025.<\/figcaption><\/figure>\n
\"Satellite
After: canopy changes in the same forest area in 2023. Image published by the University of Sheffield; Google Earth imagery \u00a9 Airbus 2025.<\/figcaption><\/figure>\n

The comparison illustrates the appearance of selective logging, but it is not an independent accuracy assessment. Nor does a three-year image interval establish the date of each disturbance. Sheffield describes officials using the resulting detections to direct drone surveys or ground visits and investigate suspected violations.<\/p>\n

What Validation Studies Actually Show<\/h3>\n

A separate 2023 study by Dupuis and colleagues tested a Sentinel-1 approach in southeastern Cameroon using records covering more than 6,000 harvested-tree locations and an independent UAV canopy-gap dataset covering approximately 1,500 hectares. The method first identified harvesting activity areas, then searched for canopy gaps within them.<\/p>\n

For the canopy-gap stage, the study reported an overall accuracy of 89%, alongside an omission error of 39% and a commission error of 5%. The omission result is important: substantial real disturbance was still missed. Overall accuracy alone would give an incomplete impression of performance.<\/p>\n

The method also compared averages from twelve months before the month of interest with three months afterward. Its retrospective results should therefore not be interpreted as an immediate alert available at the moment a tree was felled.<\/p>\n

A 2022 study of Sentinel-1 radar shadows in Gabon validated detections against repeat UAV LiDAR measurements. Although the method showed promise for assessing canopy loss, its authors identified geolocation errors and false alarms that made it unsuitable for confirming individual disturbances.<\/p>\n

These studies support the feasibility of detecting some selective logging. Their results belong to particular forests, reference datasets and processing methods; they are not universal thresholds or performance claims for STARPATH GLOBAL.<\/p>\n

From a Change Alert to a Verified Finding<\/h3>\n
    \n
  1. Detect a candidate change.<\/strong> Use optical and\/or radar time series to flag a possible disturbance. Record its location, observation dates and confidence. Keep cloud-obscured or radar-shadowed areas separate from areas assessed as unchanged.<\/li>\n
  2. Review the imagery.<\/strong> Inspect comparable before-and-after scenes, check image alignment and look for gaps, roads or loading areas. Consider storms, natural tree falls, fire, seasonal effects and moisture variation. Use higher-resolution imagery where it can resolve the ambiguity.<\/li>\n
  3. Compare permits and boundaries.<\/strong> Check the candidate disturbance against the relevant concession, approved harvest area, authorization period and available harvesting records. Allow for positional uncertainty before treating a boundary overlap as significant.<\/li>\n
  4. Inspect the site.<\/strong> Direct authorized field teams or permitted drone surveys to priority locations. Record stumps, felled material, access routes and other physical evidence, then reconcile those observations with the imagery and documents.<\/li>\n<\/ol>\n

    An alert should initially describe a possible forest disturbance. Establishing unauthorized harvesting requires evidence about both the activity and the applicable authorization. Likewise, a site without an alert may still require inspection if the expected disturbance is below the method\u2019s demonstrated sensitivity.<\/p>\n

    Design Monitoring Around the Decision<\/h3>\n

    For scattered harvesting in a cloudy, hilly forest, begin with a local feasibility assessment. Test known logged sites and comparable undisturbed sites, including different slopes and forest conditions. Assess false alerts, missed events, location uncertainty, usable coverage and the delay between disturbance and detection. Reserve independent reference sites for validation.<\/p>\n

    Optical and SAR data can support complementary observations, but combining them does not automatically improve classification. A 2022 study by Hethcoat and colleagues found that adding Sentinel-1 to Landsat 8 did not improve selective-logging classification accuracy in its Brazilian Amazon study areas. The value of each sensor should be demonstrated for the intended task.<\/p>\n

    For broader background, see our explanation of how satellite data supports deforestation monitoring<\/a>. Selective logging requires a more specific assessment of what remains visible beneath a largely intact forest canopy.<\/p>\n

    For a forestry monitoring project, STARPATH GLOBAL can help match imagery selection to the expected disturbance scale, terrain and available budget. Explore our imagery catalog<\/a>, or discuss your target area and verification requirements with our team<\/a> to plan a suitable data approach. Teams building their own remote sensing capability can also explore the Pioneer Partner Program<\/a> for support from our Forward Deployed Engineers.<\/p>\n","protected":false},"excerpt":{"rendered":"

    Satellite imagery can detect some selective logging, particularly when tree removal creates visible canopy gaps, new access routes or repeated changes in the radar signal. However, detecting a large clear-cut is a different task from identifying scattered tree removal beneath a mostly intact canopy. The reliability of the result depends on the disturbance footprint, forest […]<\/p>\n","protected":false},"author":1,"featured_media":90491,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"footnotes":"","_links_to":"","_links_to_target":""},"categories":[3,652],"tags":[10540,10563,10108,169,157,165,3614,1309],"class_list":["post-90489","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog","category-environment","tag-change-detection","tag-forest-degradation","tag-forestry","tag-remote-sensing","tag-sar","tag-satellite-imagery","tag-sentinel-1","tag-sentinel-2"],"acf":[],"_links":{"self":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts\/90489"}],"collection":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/comments?post=90489"}],"version-history":[{"count":4,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts\/90489\/revisions"}],"predecessor-version":[{"id":90496,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/posts\/90489\/revisions\/90496"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/media\/90491"}],"wp:attachment":[{"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/media?parent=90489"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/categories?post=90489"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/starpath.global\/blog\/wp-json\/wp\/v2\/tags?post=90489"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}