Chemical weapons precursor production facility activity monitoring
Open-source satellite imagery can reveal chemical weapons precursor activity through tank-farm changes, effluent discolouration, thermal signatures and vehicle traffic. This page documents the published OSINT methods, sensor capabilities and honest detection limits.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at mid-latitudes with two satellites. The 12-band multispectral suite covers the red-edge and SWIR bands useful for detecting chromatic shifts in effluent ponds caused by pH changes or organochlorine loading. Free archive from 2015.
- Landsat 8/9 OLI-TIRS: 30 m multispectral, 100 m thermal infrared (Band 10/11). The thermal band is the key differentiator: continuous exothermic processing at a reactor or distillation column produces a persistent surface-temperature anomaly detectable at roughly 0.1–0.3 °C sensitivity against background. 16-day revisit per satellite, 8-day combined. Archive to 1972 for baseline.
- Maxar WorldView-3: 30 cm panchromatic, 1.24 m multispectral, 16-band including SWIR. At this resolution, individual tanker vehicles, pipe connections between storage tanks, and changes to bund-wall geometry are legible. Tasked commercially; not on a fixed revisit cycle, so cost and latency depend on tasking priority.
- Planet SuperDove: 3–4 m resolution, 8 spectral bands including red-edge, near-daily revisit over most land areas. Best used to build a high-cadence change-detection time series that flags activity dates for follow-up with higher-resolution tasking. Temporal density compensates for modest spatial resolution.
What the surface gives away
A facility producing organophosphate or chlorine-family precursors cannot fully hide its physical footprint. Storage tank farms expand when feedstock throughput rises. Bund walls are added or enlarged. New hardstandings appear for road tankers. These are slow changes, but a monthly Sentinel-2 time series captures them reliably at 10 m resolution, and Landsat's archive stretches back far enough to establish a pre-suspicion baseline.
The most distinctive indicator is effluent pond colour. Chlorination byproducts, heavy-metal catalysts and pH-shifted wastewater all shift the spectral reflectance of standing water in ways that are visible in the SWIR and red-edge bands of Sentinel-2. Bellingcat's documented work on Syrian facilities at Al-Safira used exactly this approach: comparing multitemporal pond-colour signatures against known production-pause periods to infer operational tempo. The method does not identify the specific compound, but it does distinguish between an idle facility and an active one.
Thermal infrared as a process indicator
Many precursor synthesis routes are exothermic. Chlorination, phosphorylation and certain sulphonation reactions release heat continuously during production runs. Landsat 8/9's TIRS band records land-surface temperature at 100 m resolution, which is coarse, but persistent anomalies above background are detectable when averaged across multiple cloud-free acquisitions.
The published detection sensitivity for Landsat TIRS is approximately 0.1–0.3 °C against a stable background, though real-world clutter from sun-facing roofs and adjacent industrial heat sources can raise the practical threshold considerably. A single anomalous scene proves little. A pattern of elevated temperature coinciding with other activity indicators, and dropping during known shutdown periods, is a different matter. WorldView-3's SWIR bands can also detect thermal emission from very hot surfaces, though they are not a calibrated temperature sensor.
Vehicle and tanker traffic as a throughput proxy
Road tankers are the connective tissue of precursor logistics. Feedstocks arrive; intermediate or finished product leaves. At Planet SuperDove's 3–4 m resolution, individual vehicles are detectable as objects, though not legible as text. A time series of daily imagery can count vehicle presence in hardstanding areas and detect changes in access-road dust signatures that correlate with heavy traffic periods.
WorldView-3 at 30 cm makes vehicle classification feasible: tankers versus flatbeds versus personnel carriers have distinguishable length-to-width ratios and shadow profiles. OPCW technical secretariat reports on Syrian facilities have cited vehicle-traffic patterns as corroborating evidence alongside chemical sampling. The OSINT community's replication of those assessments using commercial imagery demonstrated that the vehicle-count approach is reproducible without access to classified data.
One honest limit: a facility can move product by pipeline rather than road. If no surface tanker traffic is visible, that does not confirm inactivity. It shifts the inference burden back to thermal and spectral indicators.
Construction and expansion signatures
Facility expansion is among the clearest signals. New building footprints, fresh concrete aprons, additional tank pads and earthworks all produce high-contrast change signals in multispectral difference imagery. Sentinel-2's 10 m resolution is sufficient to detect a new tank with a diameter above roughly 20 m. Smaller tanks require Planet or Maxar tasking.
The Libyan chemical weapons storage and production sites documented by OPCW inspectors and subsequently monitored by open-source analysts provide a published case for what expansion looks like in imagery. Berm construction around storage areas, in particular, is a distinctive signature: the raised earthwork creates a shadow pattern and a spectral contrast against surrounding terrain that persists across seasons.
Construction also implies a timeline. If earthworks begin in month one and a tank farm is complete by month eight, that sequence constrains when production capacity could have come online. Time-series analysis converts ambiguous snapshots into a coherent operational narrative.
Honest limits of the method
Cloud cover is the most mundane and most serious constraint. Sentinel-2 and Landsat are optical sensors. Facilities in humid climates or under persistent cloud can have weeks-long gaps in usable imagery. SAR can penetrate cloud, but SAR coherence change detection is covered in a sibling page and does not directly address the spectral and thermal indicators discussed here.
Attribution is the deeper problem. Spectral anomalies in effluent ponds and thermal signatures above background are consistent with chemical weapons precursor production, but they are also consistent with legitimate industrial chemistry: fertiliser plants, pharmaceutical manufacturing and petrochemical processing produce similar signatures. OSINT-grade imagery analysis can establish that a facility is active, expanding and handling liquids at elevated temperatures. It cannot, on its own, identify the compound being produced. That distinction matters enormously in any context where the analysis might inform policy or legal proceedings.
Revisit frequency also limits detection of short-duration events. A batch process running for 48 hours may fall entirely between Sentinel-2 passes. Planet's near-daily cadence reduces but does not eliminate this gap.
Satellize's analytics work draws on Sentinel-2, Landsat and commercial tasking within the OSINT framework described here.
Building a defensible analytical record
The value of this monitoring is cumulative. A single scene is an observation. A two-year time series with annotated change events, cross-referenced against reported production pauses, is evidence. The OPCW's fact-finding missions and the open-source work that has corroborated or challenged their findings both rely on this principle: the pattern across time is more defensible than any individual image.
Practically, that means maintaining a consistent archive of cloud-free acquisitions, logging the date and sensor for every usable scene, and recording negative observations, periods when the facility appeared inactive, as carefully as positive ones. A gap in the record is not the same as a gap in activity. Treating them as equivalent is the most common analytical error in facility monitoring.
Typical figures
| Spatial resolution (optical) | 3–4 m (Planet SuperDove); 10 m (Sentinel-2); 30 m (Landsat OLI); 30 cm panchromatic (WorldView-3) |
| Thermal infrared resolution | 100 m (Landsat 8/9 TIRS); WorldView-3 SWIR not a calibrated thermal sensor |
| Revisit cadence | Near-daily (Planet); 5-day (Sentinel-2, two satellites); 8-day combined (Landsat 8+9); on-demand (WorldView-3) |
| Thermal detection sensitivity | Approximately 0.1–0.3 °C above background (Landsat TIRS, multi-scene average); practical threshold higher in cluttered industrial settings |
| Minimum detectable tank diameter | ~20 m at Sentinel-2 10 m resolution; ~6 m at Planet 3 m; ~1 m at WorldView-3 30 cm |
| Spectral bands relevant to effluent analysis | Sentinel-2 Bands 5–7 (red-edge, 740–783 nm), Band 11/12 (SWIR 1565–2190 nm); Landsat OLI Bands 5–7 |
| Archive depth | Sentinel-2: 2015–present; Landsat: 1972–present; Planet: 2016–present (coverage varies) |
| Cloud cover limitation | Optical sensors unusable under cloud; SAR required for persistent-cloud environments (separate capability) |
| Latency (open data) | Sentinel-2 and Landsat: 1–3 hours post-acquisition to public archive; Planet and WorldView-3: licence-dependent |
Analytics Satellize can run
| Facility activity index | Multitemporal change detection on Sentinel-2 and Planet time series; normalised difference water index applied to effluent ponds for colour-shift quantification | Monthly PDF report with annotated imagery, activity score per observation date, and trend chart |
| Thermal anomaly time series | Landsat 8/9 TIRS land-surface temperature retrieval; anomaly scoring against rolling 90-day background; multi-scene compositing to reduce single-pass noise | GIS layer (GeoTIFF) of temperature anomaly magnitude per acquisition date, with flagged exceedance events |
| Tank-farm footprint change log | Object-based image analysis on WorldView-3 or Planet scenes; tank-pad and bund-wall delineation; area-difference calculation between baseline and current epoch | Annotated change map (GeoPackage) with timestamps and estimated capacity additions |
| Vehicle-presence count | Manual and semi-automated vehicle detection on Planet SuperDove daily series; shadow-length validation for object height estimation; count normalised to scene acquisition time | Time-series spreadsheet of daily vehicle counts by hardstanding zone, with flagged surge events |
| Effluent pond spectral profile | Sentinel-2 SWIR and red-edge band ratio time series; comparison against reference spectra for known industrial effluent types from published remote-sensing literature | Spectral trend chart per pond polygon, with qualitative classification (clear, turbid, discoloured) and dates of significant transition |
| Consolidated OSINT-grade assessment | Structured analytic technique integrating all indicator streams; confidence-level scoring per OPCW and Bellingcat published methodological frameworks | Analytical report suitable for policy briefing, with source imagery citations, confidence ratings and explicit statement of what the imagery cannot determine |
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.