Refrigerated container plug-point utilisation at cold-chain ports
Refrigerated containers cluster near power pedestals in predictable patterns, and their thermal signatures are detectable from orbit. Combined optical and thermal analysis turns reefer density into a real-time proxy for cold-chain trade volumes.
Sensors
- Airbus Pléiades Neo: 30 cm panchromatic, 50 cm multispectral (blue, green, red, red-edge, NIR). At this resolution individual container rooftops are resolved and white-top reefers are separable from dark-painted dry boxes by spectral brightness in the visible and NIR bands. Tasking revisit at a given site is typically 1 to 3 days depending on latitude and off-nadir tolerance.
- Planet SkySat: 50 cm panchromatic, 1 m multispectral (blue, green, red, NIR). Sufficient to distinguish container colour classes and count units in rows; spacing regularity near pedestals is detectable. SkySat can be tasked for same-day or next-day collect windows, useful for tracking weekly export surges.
- Landsat 8/9 TIRS: Thermal Infrared Sensor bands 10 and 11 at 100 m native resolution (resampled to 30 m in products). At this scale individual reefer rows are not resolved, but dense reefer yards produce a measurable surface-temperature anomaly above ambient terminal surfaces. 16-day repeat at the equator; both satellites together give 8-day repeat.
- NASA ECOSTRESS: 70 m thermal infrared from the International Space Station, covering latitudes to roughly 54 degrees. Irregular revisit (ISS precession gives variable overpass times) but higher spatial resolution than Landsat TIRS, making it better suited to mid-sized terminals where the heated footprint is smaller. Land surface temperature products are publicly archived.
What a white roof and a power pedestal give away
A refrigerated container is not subtle. Its roof is almost universally white or light grey, a deliberate design choice to reduce solar heat gain and protect the refrigeration unit from thermal stress. In Pléiades Neo or SkySat imagery, a yard full of reefers appears as a bright rectangular grid, noticeably lighter in the visible and NIR bands than the rust-brown or dark-green rooftops of adjacent dry boxes. That spectral contrast is the first discriminator.
The second is geometry. Reefers in active use must be plugged into electrical pedestals, so they are parked in dedicated rows with fixed inter-unit spacing dictated by the pedestal grid, typically every 2.4 metres along a row. This regularity is detectable in sub-metre imagery as a repeating pattern that differs from the irregular stacking of dry cargo. Neither cue is conclusive alone. White-painted dry containers exist, and some terminals park unplugged reefers in pedestal rows during off-season periods. The method works best as a count of plausible reefer units rather than a confirmed count of operating ones.
Thermal infrared: what the refrigeration units cannot hide
Every operating refrigeration unit rejects heat. A standard 40-foot reefer running at minus 18 degrees Celsius to maintain frozen cargo dissipates heat through its condenser coils, raising the local surface temperature of the container roof and the air immediately above it. A yard with several hundred active units creates a measurable thermal anomaly above the ambient temperature of surrounding concrete and dry-box rooftops.
Landsat 8 TIRS has detected surface temperature differences of 1 to 3 degrees Celsius between industrially active and inactive port zones in published studies, though the 100 m native resolution means the signal is an aggregate of many containers rather than a unit-level reading. ECOSTRESS at 70 m offers a modest improvement and, crucially, its ISS-driven overpass times vary across the day, occasionally providing night-time acquisitions when solar heating of concrete has subsided and the refrigeration signal is cleaner. Neither sensor can confirm electrical connection state. A reefer running on its own diesel generator produces an identical thermal signature to one drawing shore power, and an unplugged unit sitting in the sun may be warmer than an active one in shade.
Southern Hemisphere fruit calendars and what they predict
Cold-chain port activity is not flat across the year. South African citrus exports peak between May and September. Chilean table grapes move heavily from January through March. Peruvian blueberries surge in the October to December window. These calendars are published by industry bodies and agricultural ministries, and they create predictable seasonal signatures in reefer yard density at the relevant load ports.
Tracking reefer counts at Durban, San Antonio or Paita across multiple years establishes a baseline seasonal curve. Deviations from that curve, a reefer count 20 percent below the May average at Durban during peak citrus season, carry real signal: drought-reduced harvest, logistics disruption, or a shift to a competing port. The seasonal structure also helps distinguish genuine reefer activity from noise. A spike in apparent white-roof objects in January at a Southern Hemisphere fruit port is far more likely to be reefers than the same spike at an inland dry-goods depot.
Honest limits of the method
Cloud cover is the first practical constraint. Major cold-chain ports in Chile, South Africa and New Zealand sit in climatic zones with meaningful cloud frequency, and optical imagery cannot penetrate cloud. A run of overcast days during peak season can create a gap in the time series precisely when the data is most commercially valuable. Thermal infrared has slightly better cloud penetration than visible imagery in some conditions but is not immune.
Spectral confusion between white reefers and white dry containers is a genuine ambiguity. Some shipping lines paint their entire fleet white regardless of container type. At terminals handling mixed fleets, the false-positive rate from colour alone can be significant without a secondary filter such as geometric regularity or thermal anomaly co-location. Resolution is also a ceiling: Landsat TIRS cannot count individual units, and even Pléiades Neo at 50 cm multispectral cannot reliably read container identification codes to confirm type. The method produces a count of probable reefers, not a manifest. Buyers of this analysis should treat outputs as directional trade-volume indicators rather than certified inventory figures.
Turning counts into a trade signal
The analytic pipeline has three layers. First, optical classification: a supervised classifier trained on labelled Pléiades Neo chips distinguishes white-top containers from dark dry boxes and from terminal infrastructure, producing a per-image reefer candidate count with a confidence score. Second, thermal confirmation: a co-registered TIRS or ECOSTRESS acquisition flags zones where surface temperature exceeds the ambient terminal baseline, weighting the optical candidates in those zones more heavily. Third, time-series construction: counts from successive acquisitions are assembled into a weekly or bi-weekly index, normalised against port-specific seasonal baselines derived from the archive.
The output is an index number, not a container count with false precision. It moves directionally with cold-chain throughput and can be compared across ports and across years. Satellize applies a similar multi-source time-series approach in its Tonga crop-estimation programme, where the analytic challenge is also one of extracting a commodity-volume signal from imperfect imagery under variable cloud. The methodology transfers directly to port reefer monitoring, with the substitution of thermal confirmation for the vegetation-index stack used in the agricultural context.
What this analysis cannot replace
Port operator data, when available, is more accurate. A terminal management system with live plug-point telemetry knows exactly how many reefers are connected and at what set-point temperature. Satellite analysis is not competing with that data; it is providing an independent, external view for parties who do not have access to the terminal's internal systems, commodity traders, freight forwarders, agricultural ministries, and lenders with exposure to perishable supply chains.
The method also cannot distinguish the commodity inside the box. A reefer set to 2 degrees Celsius carrying South African citrus looks identical from orbit to one set to minus 25 degrees carrying Norwegian salmon. Commodity attribution requires combining the reefer count with vessel AIS data, port call records, and published export calendars, none of which are satellite products but all of which are open-source and can be fused at the analysis layer.
Typical figures
| Best optical resolution (reefer discrimination) | 30 cm pan / 50 cm MS (Pléiades Neo) |
| Optical revisit at a given port | 1 to 3 days (Pléiades Neo / SkySat, tasked); 16 days Landsat (8-day with both satellites) |
| Thermal resolution | 100 m native TIRS (Landsat 8/9); 70 m ECOSTRESS |
| Minimum thermal anomaly detectable (aggregate yard) | Approximately 1 to 3 °C above ambient; unit-level thermal not resolved |
| Spectral bands used | Visible (blue, green, red), NIR, red-edge for optical; TIR bands 10/11 (10.6–12.5 µm) for thermal |
| Minimum reefer cluster detectable optically | Approximately 5 to 10 units in a row at 50 cm MS resolution |
| Archive depth | Landsat: 1972 to present; Pléiades Neo: 2021 to present; ECOSTRESS: 2018 to present |
| Cloud limitation | Optical unusable under cloud; thermal partially degraded; no SAR thermal equivalent exists |
| Positional accuracy | Sub-metre CE90 for Pléiades Neo; 30 m for Landsat products |
| Delivery latency (tasked optical) | Typically 24 to 72 hours from acquisition to processed output |
Analytics Satellize can run
| Reefer candidate count per terminal zone | Supervised spectral classification of white-top objects in sub-metre multispectral imagery, filtered by geometric regularity consistent with pedestal-row spacing | GIS polygon layer with per-row container counts and confidence scores, updated per acquisition |
| Thermal anomaly index for active reefer yards | Land surface temperature retrieval from Landsat 8/9 TIRS or ECOSTRESS, differenced against terminal-wide ambient baseline derived from non-reefer surfaces in the same scene | Raster anomaly map and scalar index per acquisition, flagging zones exceeding the 1.5 °C anomaly threshold |
| Weekly reefer-density time series | Multi-source optical count fused with thermal confirmation, assembled into a normalised weekly index using all available cloud-free acquisitions | CSV time-series feed with cloud-cover flag per observation and rolling 4-week average |
| Seasonal deviation alert | Statistical comparison of current reefer index against port-specific multi-year seasonal baseline; alert triggered when deviation exceeds two standard deviations | Automated alert report naming the port, the deviation magnitude, and the historical context for the relevant calendar week |
| Multi-port cold-chain comparison dashboard | Parallel time-series construction across a user-defined set of ports, normalised to a common index scale to allow cross-port comparison of seasonal peaks and troughs | Interactive dashboard export (GeoJSON + CSV) covering up to ten ports per subscription period |
| Commodity-calendar alignment report | Overlay of reefer-density index against published Southern Hemisphere export calendars (citrus, stone fruit, table grapes, blueberries) to attribute density peaks to likely commodity flows | Quarterly PDF report with annotated time-series charts and commodity attribution notes |
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.