Camouflaged illegal arms depot and weapons-cache site detection
Concealed weapons caches hide under netting and canopy, but artificial camouflage materials betray themselves in near-infrared reflectance and SAR coherence time series. This page explains the sensor physics, honest detection limits, and how the methods combine.
Sensors
- Sentinel-1 C-band SAR (ESA): 20 m ground range resolution in Interferometric Wide Swath mode; 6-day repeat at mid-latitudes with both satellites active. Coherence between repeat passes degrades when ground surface or vegetation structure is disturbed; cache construction, buried container emplacement and berm building all produce measurable coherence loss even under cloud or canopy.
- ALOS-2 PALSAR-2 (JAXA): L-band (1.27 GHz) penetrates forest canopy more deeply than C-band, reaching 10–20 m into dense tropical growth depending on moisture. Spotlight mode achieves 1–3 m resolution. Useful for detecting subsurface or under-canopy ground disturbance that C-band misses, though revisit is 14 days in standard modes.
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral across 8 bands including a dedicated near-infrared band (NIR1, 770–895 nm) and a red-edge band. The NIR1 and SWIR bands together expose the reflectance mismatch between synthetic camouflage netting and living chlorophyll-bearing vegetation. Tasked commercially; not a free open resource.
- Airbus Pléiades Neo: 0.3 m native resolution, four spectral bands including NIR. Constellation of two satellites provides same-day stereo capability, which is useful for detecting earthen berms and low-relief structures by generating a digital surface model and differencing it against a baseline. Revisit to any point is roughly daily at mid-latitudes.
- Sentinel-2 MSI (ESA): 10 m resolution in visible and NIR bands; 20 m in red-edge and SWIR. Free, 5-day revisit. Insufficient resolution to detect individual cache structures, but useful for screening large areas for anomalous NIR signatures before tasking commercial sensors. Cloud cover remains a hard constraint with no SAR fallback at this sensor.
What a camouflage net gives away in the near-infrared
Living vegetation reflects strongly in the near-infrared, typically between 40 and 60 percent reflectance in the 750–900 nm range, because of the internal cell-wall scattering structure of healthy leaves. Synthetic camouflage netting, even the best military-grade material, is designed to match visible-spectrum green. It is not designed to replicate the internal mesophyll structure that drives NIR reflectance. Published remote-sensing studies have consistently found that camouflage netting produces NIR reflectance values 15 to 30 percentage points below those of the real canopy it is placed beneath or alongside.
This is not a subtle effect. On a false-colour composite where NIR is rendered as red, a camouflage net appears distinctly darker than surrounding forest. WorldView-3's eight-band multispectral suite, with its separate red-edge and NIR1 bands, sharpens this contrast further. The red-edge band (705–745 nm) is particularly diagnostic: chlorophyll absorption drops sharply across this range in live vegetation, producing a steep reflectance slope that synthetic dyes do not replicate. A site where the visible-band appearance matches the forest but the red-edge and NIR1 channels show a flat, suppressed response is a candidate for further investigation. It is not proof. Shadows, stressed vegetation, dead wood and certain soil types can produce similar suppressions, which is why SAR coherence is the necessary second layer.
What SAR coherence reveals that optical sensors cannot
Interferometric coherence measures how similar the radar backscatter phase is between two passes of the same satellite over the same ground. Stable, undisturbed surfaces, concrete, bare rock, dry soil, maintain high coherence over weeks. Vegetation fluctuates naturally, so forest coherence decays over days. The useful signal is the anomalous coherence loss that occurs when the ground surface is physically altered: soil is excavated, a container is buried, a berm is thrown up, or netting is repositioned.
Sentinel-1's 6-day repeat cycle, combined with the freely available Copernicus archive going back to 2014, makes it possible to run coherence time series over suspect areas and flag the specific 6-day interval when disturbance occurred. The method does not require cloud-free conditions. It works at night. It is not affected by camouflage netting at all, because the netting itself is not what is being measured; the ground beneath it is. The honest limit is spatial resolution: at 20 m, a single buried container is below the detection threshold. What registers is the ground disturbance associated with emplacement, which typically involves excavation over tens to hundreds of square metres. ALOS-2's L-band coherence extends sensitivity deeper into canopy, but at 14-day repeat the temporal resolution is coarser and short-duration activity may be missed.
Earthen berms and buried containers: the structural signatures
Above-ground earthen berms, common for blast protection around ammunition stores, create low-relief linear features typically 1–3 m high. They are invisible to coarse optical sensors and easy to miss even on 10 m imagery. At 0.3 m resolution, Pléiades Neo stereo pairs resolve them directly. Differencing a stereo-derived digital surface model against a pre-event baseline from the same sensor, or against a Copernicus DEM tile, reveals volume additions consistent with berm construction. A berm 50 m long and 2 m high represents a volume anomaly of roughly 200–400 cubic metres, well above the noise floor of modern stereo DSM differencing at sub-metre resolution.
Buried containers present a harder problem. The container itself is below the optical detection threshold once covered. What remains detectable is the excavation scar, the disturbed soil reflectance, and the coherence anomaly during and immediately after emplacement. If the site is subsequently revegetated or covered with netting, the NIR anomaly persists for months to years because the netting does not grow, and the underlying disturbed soil may retain a moisture and roughness signature in SAR backscatter intensity even after coherence recovers.
Combining the layers: a detection workflow and its honest limits
A practical screening workflow runs Sentinel-1 coherence change detection across the area of interest first. This is computationally intensive but feasible on open data. Sites showing anomalous coherence loss in a single 6-day window, particularly if the loss is spatially compact and geometrically regular, are flagged for optical review. Sentinel-2 NIR composites provide a free first look. Sites where the NIR suppression aligns with the coherence anomaly in space and time move to commercial tasking: WorldView-3 or Pléiades Neo at sub-metre resolution, with the eight-band or four-band multispectral stack analysed for red-edge and NIR1 anomalies.
The limits are real and worth stating plainly. Dense tropical canopy reduces SAR coherence baseline even for undisturbed ground, raising the false-positive rate. Seasonal vegetation change, agricultural activity and logging all produce coherence loss signatures that can resemble cache construction. NIR anomalies from camouflage netting are convincing only when the netting is not additionally covered by living vegetation growing over it, which can happen over months. The method is strongest in the weeks immediately following construction. It degrades as the site ages and vegetation re-establishes. No satellite method produces a confirmed identification of weapons; it produces a prioritised list of sites warranting ground investigation or further intelligence corroboration.
Satellize integrates this coherence-plus-NIR workflow into client analytics pipelines, drawing on the same open-constellation architecture used in the Tonga crop-estimation programme, with commercial sensor tasking added where sub-metre resolution is required.
Archive depth and the cold-case value
Sentinel-1's archive runs from April 2014 for Sentinel-1A, with Sentinel-1B adding coverage from 2016 until its failure in December 2021. This means coherence time series can be reconstructed retrospectively over any area with consistent acquisition coverage. For post-conflict accountability or treaty-verification work, this archive depth is significant: a site that was constructed in 2017 and dismantled in 2019 may still be recoverable from the coherence record, even if no contemporaneous tasking was ever performed.
Landsat 8 and 9 extend the optical archive back further, though at 30 m resolution the NIR anomaly from camouflage netting is detectable only for large sites. The USGS Landsat archive is open and freely accessible, making retrospective screening of historical periods feasible without additional data cost.
Typical figures
| SAR spatial resolution (Sentinel-1 IW mode) | 20 m range × 22 m azimuth |
| SAR spatial resolution (ALOS-2 Spotlight) | 1–3 m |
| Optical resolution (WorldView-3 multispectral) | 1.24 m (8-band MS); 0.31 m panchromatic |
| Optical resolution (Pléiades Neo) | 0.3 m panchromatic; 1.2 m multispectral |
| Sentinel-1 revisit (dual-satellite) | 6 days at mid-latitudes; 12 days with single satellite |
| Minimum detectable ground disturbance (SAR coherence) | Tens to hundreds of square metres; single buried container below threshold |
| NIR reflectance anomaly (camouflage vs. live canopy) | 15–30 percentage points suppression in 750–900 nm range (published literature range) |
| Sentinel-1 archive depth | From April 2014 (1A); 2016–2021 (1B) |
| Cloud penetration | SAR: full; optical sensors: nil |
| Typical analytic latency from tasking | 24–72 hours for commercial optical; Sentinel-1 coherence pair available within 6 days of acquisition |
Analytics Satellize can run
| Coherence anomaly map | Sentinel-1 repeat-pass interferometric coherence differencing; anomalous low-coherence patches extracted against seasonal baseline | GIS polygon layer (GeoJSON or shapefile) of flagged disturbance events with date-of-change attribution |
| NIR suppression overlay | Red-edge and NIR1 band ratio analysis on WorldView-3 or Pléiades Neo imagery; threshold-based masking against local vegetation reference | False-colour composite with anomaly polygons and per-pixel reflectance values, delivered as GeoTIFF with analyst report |
| Berm and earthwork volume estimate | Stereo DSM differencing against Copernicus DEM or prior-epoch DSM; volume computed per flagged polygon | Tabular report of site coordinates, estimated volume (cubic metres), and confidence tier |
| Retrospective archive screening | Sentinel-1 coherence time series reconstruction over defined AOI and date range; Landsat NIR anomaly screening for large-area historical review | Chronological event log with coherence-drop magnitude and spatial extent per epoch, exportable to PDF or GIS |
| Priority tasking list | Ranked fusion of coherence and NIR anomaly scores; sites scoring above analyst-defined thresholds on both layers elevated to commercial tasking queue | Tasking brief with site coordinates, recommended sensor, and acquisition window |
| Persistent monitoring alert feed | Automated Sentinel-1 coherence monitoring on fixed AOI polygons; alert triggered when coherence drops below baseline by defined sigma threshold | Near-real-time alert (email or API push) with coherence map attachment, within 48 hours of new acquisition processing |
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.