Precursor chemical storage and transshipment site detection
Bulk precursor chemicals move through warehouses and tank farms that are visually indistinguishable from legal agrochemical storage. Change detection on very-high-resolution imagery, combined with AIS vessel calls and road-traffic density, builds the corroborating layers that imagery alone cannot.
Sensors
- Maxar WorldView-3: 30 cm panchromatic, 1.24 m multispectral (8 bands including SWIR), revisit roughly 1–4.5 days depending on latitude and tasking priority. At 30 cm, individual tanker trucks, drum pallets and secondary containment berms are resolvable. SWIR bands can distinguish some liquid-surface reflectance signatures from solid goods.
- Airbus Pléiades Neo: 30 cm panchromatic, 70 cm native multispectral (4 or 6 bands), daily revisit over most latitudes with the four-satellite constellation. Stereo pairs allow 3-D volume estimation of open-top storage, useful for tracking inventory change at tank farms.
- PlanetScope: 3 m multispectral (8-band SuperDove), near-daily global revisit. Insufficient resolution to count drums or read truck plates, but excellent for detecting new impervious surfaces, roof additions and access-road construction that precede facility activation. Acts as a wide-area tripwire before VHR tasking.
- Sentinel-1 SAR: C-band SAR at 5 m (IW mode), 6-day revisit per satellite, 12-day combined. Cloud-independent. Coherence-change detection flags ground disturbance and new metal-roofed structures even when optical imagery is blocked by monsoon cloud. Backscatter intensity from tank roofs and container stacks is distinctive and stable enough to track over time.
- Spire GNSS-RO / AIS: Spire operates a commercial AIS constellation providing vessel position data globally. Correlating AIS port calls, vessel type and dwell time against facility activity seen in imagery addresses the ship-side leg of transshipment chains.
What a warehouse cannot hide
Precursor chemicals for narcotics synthesis, acetic anhydride for heroin, potassium permanganate for cocaine, ephedrine and pseudoephedrine for methamphetamine, are internationally controlled under the 1988 UN Convention against Illicit Traffic and tracked by the International Narcotics Control Board (INCB). The control regime creates an incentive to route shipments through jurisdictions with weak inspection capacity, which in practice means secondary ports, free-trade zones and border-crossing depots that are geographically accessible but institutionally opaque.
The structures that store these chemicals are not exotic. They are standard industrial buildings: cylindrical tanks with secondary containment berms, corrugated-metal warehouses with roller-door loading bays, and hardstand areas capable of taking heavy tanker trucks. That ordinariness is the analytical challenge and, paradoxically, also the analytical opportunity. Because the infrastructure is standard, it leaves standard signatures in satellite imagery, and departures from expected patterns become detectable.
Change detection as the primary signal
No single image confirms illicit activity. The method is temporal: comparing a baseline archive against current imagery to flag construction of new structures, expansion of existing hardstand, appearance of secondary containment (the low concrete berm surrounding a tank farm, required by many national fire codes and therefore a reliable indicator of bulk liquid storage), and changes in truck congregation density.
At 30 cm resolution, WorldView-3 or Pléiades Neo imagery resolves tanker trucks clearly enough to count axles and distinguish road tankers from flatbed freight. A loading dock that shows two trucks per week in the baseline and then shows twelve per week in a six-month change stack is a candidate for further investigation. This is not proof; legitimate fertiliser or solvent distributors show identical patterns during peak agricultural seasons. The change signal is a filter, not a verdict.
Sentinel-1 SAR coherence change detection adds a cloud-independent layer. When ground surface or structure changes between two SAR passes, coherence drops. New metal-roofed buildings, freshly poured concrete hardstand and disturbed soil all produce coherence loss that is detectable even under continuous cloud cover, which matters enormously in tropical border regions where optical revisit can be blocked for weeks.
The ambiguity problem, stated honestly
A legitimate agrochemical depot serving smallholder farmers and an illicit precursor transshipment hub are, from above, the same object. Both have tanks, berms, trucks and warehouses. Both show seasonal activity spikes. Spectral analysis helps at the margins: WorldView-3's SWIR bands can differentiate some liquid types by surface reflectance, and certain chemical vapour plumes produce detectable signatures, but the concentrations involved in sealed storage are typically below any published remote-sensing detection threshold for passive optical sensors.
The honest answer is that imagery alone does not resolve the ambiguity. What imagery does is narrow the candidate set. A facility that appears in a known INCB-flagged transshipment corridor, shows rapid construction not preceded by building-permit activity visible in municipal records, and sits within five kilometres of a port receiving vessels with anomalous AIS behaviour (gaps, identity changes, unusual dwell times) is a very different analytical object from a facility that has operated continuously for a decade in an agricultural zone.
Corroborating layers matter. AIS data from Spire or similar providers can show which vessels called at a nearby port, how long they stayed and whether their declared cargo class is consistent with the volumes implied by the tanker-truck activity seen in imagery. Road-traffic density derived from PlanetScope time series can quantify throughput. None of these layers is conclusive individually; their convergence is what produces an actionable intelligence product.
Building the corroborating stack
A practical workflow runs in three tiers. First, PlanetScope daily mosaics cover a defined area of interest at 3 m, flagging any new impervious surface or roof construction. This is cheap and fast. Second, when a flag appears, VHR tasking on WorldView-3 or Pléiades Neo provides the structural detail needed to characterise the facility: tank diameter and height (which gives approximate volume via geometry), number of loading bays, presence of secondary containment, vehicle types. Third, Sentinel-1 coherence change fills the cloud-gap periods and provides an independent confirmation of the optical findings.
AIS correlation runs in parallel. A port call by a vessel that previously called at a known precursor-exporting port, followed within days by elevated truck activity at a nearby depot, is the kind of temporal linkage that shifts a facility from 'possible' to 'probable' in an analytical assessment. The INCB publishes annual reports naming source countries and trade routes for the major precursors; these documents provide the geographic priors that make the imagery analysis tractable rather than a global needle-in-haystack exercise.
Resolution floors and what they mean for confidence
At 30 cm, individual drums on a pallet are at the edge of detectability but not reliably resolvable. Drum counts from VHR imagery carry significant uncertainty and should not be reported as precise inventory figures. Tank volumes estimated from geometric measurement of diameter and roof height are more reliable but still carry errors of 10–20% depending on roof type and viewing angle.
Revisit is the other constraint. Even with aggressive commercial tasking, a specific facility may be imaged only every few days under clear-sky conditions, and tanker-truck loading events that last two to three hours can fall entirely between passes. The practical implication is that imagery-based analysis characterises patterns over weeks and months rather than capturing individual events. Enforcement agencies that need real-time vehicle tracking require ground-based or airborne assets; satellite analytics provide the strategic targeting layer that directs those assets.
Satellize runs this kind of multi-source change-detection workflow on open and commercial constellations, structuring the output as GIS-compatible alert layers and periodic written assessments. The approach is similar in principle to the crop-estimation work done for the Kingdom of Tonga, where the analytical value came from combining multiple data sources rather than from any single sensor.
What a useful product looks like
An actionable output for a government enforcement client is not a heat map of 'suspicious warehouses'. It is a ranked list of facilities with documented change histories, structural characterisations, corroborating AIS events and explicit confidence levels, delivered as a GIS layer with attached PDF assessments that a non-specialist analyst can read and act on.
Confidence levels must be stated honestly. A facility that scores high on structural indicators, is located in a documented INCB transshipment corridor and shows correlated vessel activity warrants a 'high priority for ground verification' designation. A facility that shows only one indicator warrants 'low priority, monitor'. The product's value is in the triage, not in claiming certainty that satellite data cannot support.
Typical figures
| Best spatial resolution (optical) | 30 cm (WorldView-3 pan, Pléiades Neo pan) |
| Best spatial resolution (SAR) | 5 m (Sentinel-1 IW mode) |
| Typical VHR revisit | 1–4.5 days (WorldView-3); daily (Pléiades Neo 4-satellite constellation) |
| SAR revisit | 6 days per satellite, 12 days combined (Sentinel-1 A+B) |
| Wide-area monitoring revisit | Near-daily (PlanetScope 3 m) |
| Spectral bands used | VIS/NIR/SWIR (WorldView-3 8-band); C-band 5.405 GHz (Sentinel-1) |
| Minimum detectable structure | ~3 m roof span at 30 cm resolution; secondary containment berms >0.5 m height detectable in stereo |
| Tank volume estimation accuracy | ±10–20% from geometric measurement at VHR, depending on roof type and incidence angle |
| Archive depth | WorldView-3 from 2014; Sentinel-1 from 2014; PlanetScope from 2016 |
| Delivery formats | GeoTIFF change layers, GeoJSON facility polygons, PDF analytical assessments, AIS correlation tables |
Analytics Satellize can run
| New structure detection alert | Pixel-level change detection on PlanetScope time series (image differencing or NDVI/impervious surface classification); confirmed by VHR tasking | GeoJSON polygon alert with date of first detection, area, confidence flag |
| Facility structural characterisation | Object-based image analysis on WorldView-3 or Pléiades Neo: roof type, tank geometry, secondary containment presence, loading-bay count | PDF site assessment with annotated imagery and volume estimates |
| Tanker-truck congregation index | Vehicle detection on VHR optical imagery using published deep-learning object-detection methods (e.g. YOLO-class models trained on satellite imagery); temporal stack produces weekly truck-count time series | Time-series chart and GIS layer showing truck density per facility per week |
| SAR coherence change map | Sentinel-1 interferometric coherence differencing between baseline and current epoch; low-coherence zones flagged as ground or structure change | GeoTIFF coherence-loss layer, updated on each Sentinel-1 pass |
| AIS vessel-call correlation report | Temporal join of AIS port-call records (vessel type, flag, dwell time, prior port history) against facility activity windows identified in imagery | Tabular report of correlated vessel calls with confidence scoring |
| Facility priority ranking | Multi-criteria scoring combining structural indicators, location priors from INCB corridor data, change magnitude and AIS correlation; explicit weighting documented for auditability | Ranked facility list with per-criterion scores, delivered as spreadsheet and GIS layer |
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.