Sensors
- Sentinel-1 SAR (C-band, ESA): 5 × 20 m ground range resolution in Interferometric Wide Swath mode; 6-day repeat at mid-latitudes (12-day per pass). Coherence loss between two acquisitions flags disturbed soil reliably, but the 5 m range resolution means individual narrow trenches are not resolved, only the aggregate disturbance footprint.
- ICEYE Spotlight SAR (X-band): Spotlight mode delivers approximately 0.5 × 0.5 m resolution on tasked acquisitions. X-band penetrates light vegetation and is sensitive to small surface roughness changes. Revisit to a specific point depends on constellation size and tasking priority, typically 1-3 days commercially.
- Planet SkySat (optical, multispectral and panchromatic): Panchromatic resolution of approximately 0.5 m. Capable of resolving trench shadows and spoil berms at this scale, subject to cloud cover. SkySat can be tasked for same-day or next-day collection over priority areas.
- Maxar WorldView-3 (optical, panchromatic and multispectral): 0.31 m panchromatic resolution, the highest commercially available at scale. Shadow lengths in low-sun imagery allow depth estimation of excavations. SWIR bands (eight bands, 1.24–2.365 µm) can distinguish freshly turned soil from surface crust by moisture and mineral contrast.
What disturbed soil looks like to a radar
Interferometric SAR coherence measures how consistently a patch of ground scatters radar energy between two passes. Undisturbed soil, compacted roads and vegetation-free fields maintain coherence well. Excavation destroys it: the moment a digger turns over soil, the surface geometry changes at centimetre scale, and the phase relationship between two acquisitions collapses. The resulting low-coherence signature is detectable even when the trench itself is too narrow to resolve as a discrete linear feature.
Sentinel-1 pairs with a six-day baseline are the workhorse for this. The 5 × 20 m pixel footprint cannot image a 0.8 m slit trench, but a network of trenches dug across a field produces a coherence-loss patch that stands out clearly against stable agricultural background. The limitation is ambiguity: any soil disturbance, including ploughing, produces the same signature. Separating military earthworks from farming requires either temporal context (ploughing is seasonal and follows field boundaries) or corroborating optical imagery.
Shadow geometry and what it gives away
Optical sensors at sub-metre resolution resolve trenches differently. A trench is a negative feature: the walls cast a shadow whose length, at a known solar elevation angle, gives an estimate of depth. A 1.5 m deep trench illuminated at 30 degrees solar elevation casts a shadow roughly 2.6 m wide on the far wall. WorldView-3 at 0.31 m panchromatic can measure that shadow. The spoil berm on the parapet side casts its own shadow in the opposite direction, and the two together give both depth and the direction of the defensive face.
This is the method that open-source analysts, including those at the Institute for the Study of War, applied systematically to publicly available imagery throughout the 2022-2024 Ukraine conflict. Analysts traced hundreds of kilometres of Russian defensive lines in Zaporizhzhia and Kherson oblasts using WorldView and Planet imagery, cross-referenced against Sentinel-1 coherence maps. The combination is more reliable than either sensor alone. Optical is cloud-limited; SAR coherence is ambiguous without context. Together they close most of the gap.
The detection floor and what falls below it
Honesty about limits matters here. Individual fighting positions, shell scrapes and slit trenches narrower than approximately one metre are below the detection floor of Sentinel-1 in standard Interferometric Wide Swath mode. Even ICEYE Spotlight at 0.5 m struggles with a single narrow trench: the feature may occupy less than one pixel in width, and detection depends on the shadow or spoil contrast rather than the trench geometry itself.
Anti-tank ditches, which are typically 3-5 m wide and 2-3 m deep, are well above the floor. Dragon's teeth obstacles, berms and revetments are detectable in both SAR and optical at commercial resolution. The practical rule is that anything wide enough to stop a vehicle is wide enough to detect reliably. Infantry-scale earthworks require sub-metre optical under clear skies, and even then a well-camouflaged position with overhead cover may not be visible at all. No satellite sensor sees through netting or vegetation canopy at the scale of an individual fighting position.
Change detection across a campaign: the Ukraine record
The publicly documented analysis of Ukrainian and Russian defensive construction between 2022 and 2024 is the most extensive open-source application of this method to date. Analysts using commercial imagery traced the progressive southward extension of Russian defensive belts through late 2022 and 2023, identifying the sequence: anti-tank ditch first, then wire obstacles, then trench networks connecting strongpoints, then hardened command positions further to the rear. The temporal sequence itself carries intelligence value independent of any single image.
Coherence-change time series from Sentinel-1 allowed analysts to date the onset of construction to within the six-day repeat cycle. Optical imagery then characterised the type and maturity of the works. This multi-sensor, multi-temporal approach is now the reference methodology for field-fortification mapping. Archive depth matters: Sentinel-1 data is publicly available back to 2014, which means baseline coherence values for any area of interest can be established before a conflict begins, making anomaly detection faster when it counts.
From imagery to a usable intelligence product
Raw change detection is not an intelligence product. A coherence-loss map covering thousands of square kilometres requires classification: which anomalies are agricultural, which are construction, which are military earthworks, and which of those are offensive versus defensive in character. That classification draws on feature geometry (linear networks suggest trenches; isolated patches suggest vehicle parks or gun positions), spatial relationship to road networks and terrain, and comparison with known defensive doctrine.
Satellize structures this kind of analysis as a layered GIS product: a base coherence-change raster, a vector overlay of classified linear features, and an attributed point layer for individual positions with confidence scores. For clients with active requirements, commercial tasking through ICEYE or Maxar can be added to the Sentinel-1 baseline, reducing latency from six days to under 24 hours on priority areas.
Cloud, winter and the limits of any single pass
SAR is cloud-independent, which is its primary operational advantage in temperate and sub-arctic theatres. Eastern Ukraine averages significant cloud cover through autumn and winter, making optical revisit unreliable for weeks at a time. During those periods, Sentinel-1 coherence analysis carries the detection burden alone, with its attendant ambiguity problem. Snow cover complicates matters further: a fresh snowfall resets surface coherence for all ground types simultaneously, masking recent disturbance.
The honest operational picture is that no single sensor provides continuous, unambiguous coverage. A programme that combines free Sentinel-1 data with tasked commercial SAR and opportunistic optical collection is more reliable than any one of those alone. The cost of that combination, and the latency acceptable for the intelligence requirement, are the two variables that determine the right architecture for a given client.
Typical figures
| Spatial resolution (SAR coherence, Sentinel-1 IW) | 5 × 20 m ground range; coherence anomalies detectable at patch scale, not individual trench scale |
| Spatial resolution (ICEYE Spotlight SAR) | ~0.5 × 0.5 m; resolves spoil berms and wide trenches as discrete features |
| Spatial resolution (WorldView-3 optical) | 0.31 m panchromatic; shadow-depth estimation possible on features wider than ~0.5 m |
| Revisit (Sentinel-1, mid-latitudes) | 6-day combined ascending/descending; 12-day per orbital pass |
| Revisit (commercial tasking, ICEYE/SkySat) | 1-3 days depending on constellation availability and tasking priority |
| Minimum detectable feature (SAR coherence) | Aggregate disturbance patch of ~100 m²; individual trenches narrower than ~1 m not resolved |
| Minimum detectable feature (sub-metre optical) | Anti-tank ditches >3 m wide reliably; slit trenches <1 m width require shadow contrast and low solar angle |
| Cloud sensitivity | SAR: cloud-independent. Optical: fully cloud-limited; eastern European winters routinely interrupt optical series for weeks |
| Archive depth (Sentinel-1) | Public archive from October 2014; enables pre-conflict baseline coherence computation |
| Delivery formats | GeoTIFF coherence rasters, GeoPackage or Shapefile vector overlays, attributed PDF report, change-alert feed |
Analytics Satellize can run
| Coherence-loss change map | Sentinel-1 InSAR coherence differencing between two or more acquisition dates; low-coherence patches classified against seasonal agricultural baseline | GeoTIFF raster with classified disturbance polygons; updated on each Sentinel-1 overpass (6-day cycle) |
| Linear earthwork extraction | Directional morphological filtering and line-segment detection applied to coherence-change and optical panchromatic imagery; features attributed by orientation, length and continuity | Vector line layer (GeoPackage/Shapefile) with confidence score and feature-type classification (trench, berm, ditch) |
| Trench-depth estimation from shadow geometry | Solar elevation angle at acquisition time used with measured shadow width in sub-metre panchromatic imagery to estimate feature depth; method published in open-source OSINT literature | Point or polygon layer with estimated depth range and uncertainty bounds; included in site-level PDF report |
| Construction-onset dating | Coherence time series analysis across Sentinel-1 archive to identify the acquisition pair in which coherence first drops; optical imagery used to confirm if available | Timeline chart and attributed GIS layer showing estimated construction-start date per feature, with six-day temporal precision |
| Defensive-belt maturity assessment | Multi-temporal feature classification comparing early-stage (ditch only), intermediate (wire and trench network) and mature (hardened positions, covered emplacements) signatures against published doctrinal patterns | Structured intelligence report with maturity rating per sector and annotated imagery |
| Tasked SAR rapid-update alert | ICEYE or equivalent Spotlight SAR tasked on priority coordinates; change detection against previous pass using intensity and coherence comparison | Alert notification within 24 hours of new acquisition, with annotated image chip and change summary |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.