Container yard occupancy measurement from SAR
Synthetic aperture radar detects and counts intermodal containers in port yards at any hour, in any weather, using the distinctive double-bounce signature of metal boxes stacked on concrete. Temporal stacking of SAR scenes converts occupancy snapshots into throughput indices.
Sensors
- Sentinel-1 (C-band SAR, 5.405 GHz): Interferometric Wide Swath mode delivers 10 m ground range detected resolution across a 250 km swath. Revisit at mid-latitudes is 6 days with one satellite, 3 days with two. Free and open archive from 2014. Sufficient for yard-level occupancy indices at large terminals; individual 20-foot containers are below resolution floor.
- ICEYE X-band SAR constellation: Spot and Strip modes achieve 0.5–1 m resolution in Spot mode, enabling detection and rough classification of individual container stacks. Constellation revisit is sub-daily over priority sites. Commercial tasking adds cost but removes the weather and daylight constraints that limit optical alternatives.
- Capella Space X-band SAR: Spotlight mode resolution down to approximately 0.35 m (published figure), with single-look complex data available for coherence analysis. Useful for distinguishing loaded versus empty containers by radar cross-section difference, and for detecting change at stack level between passes.
- Sentinel-1 InSAR coherence (derived product): Coherence between repeat passes drops where containers move and remains high over static stacks. This temporal coherence layer provides a change mask that separates active yard zones from idle storage without requiring object detection, useful as a low-cost screening step.
What a metal box does to a radar pulse
A standard intermodal container is, from a radar's perspective, a nearly ideal corner reflector. The vertical steel wall meets the concrete apron at close to ninety degrees. An incoming microwave pulse bounces off the ground, then off the container wall, and returns almost directly back to the sensor. This double-bounce mechanism produces a radar cross-section far larger than the container's physical footprint would suggest, making stacked boxes conspicuous even at moderate resolutions.
X-band wavelengths (roughly 3 cm, used by ICEYE and Capella) interact strongly with container-scale geometry. C-band (5.6 cm, used by Sentinel-1) still produces detectable double-bounce but with lower contrast against surrounding clutter. The practical consequence: at 1 m resolution, individual container rows are separable; at 10 m resolution, only yard zones can be characterised, not individual stacks. Analysts working with Sentinel-1 typically count occupied yard blocks rather than containers, then apply a density coefficient derived from ground-truth or from a single high-resolution calibration pass.
Detection methods: from thresholding to deep learning
The simplest detection approach applies a backscatter intensity threshold to a calibrated sigma-nought image. Pixels above the threshold in a known yard polygon are candidate container returns. This works adequately for large, well-separated terminals with uniform apron material, but breaks down near crane structures, vessel superstructures and metal warehouses, all of which produce similar double-bounce signatures.
More reliable pipelines add a morphological step: containers are roughly 2.6 m wide and 6 m or 12 m long (20-foot and 40-foot units), so a matched-filter or template-matching stage tuned to those dimensions suppresses false positives from irregular metal structures. At sub-metre resolution, convolutional neural networks trained on labelled SAR chips can distinguish container rows from crane shadows and apron markings with published accuracy figures in the 85–92 percent range in peer-reviewed remote sensing literature, though performance degrades at terminals with dense crane infrastructure.
One honest limit: SAR cannot reliably distinguish a loaded container from an empty one on backscatter intensity alone at operational resolutions. Coherence methods offer a partial workaround. A container that has not moved between two passes maintains high interferometric coherence; one that has been lifted and repositioned shows decorrelation. This gives a proxy for dwell time and activity rather than weight.
Temporal stacking and what it reveals about throughput
A single SAR pass gives an occupancy snapshot. A time series gives something more useful: a throughput index. By computing mean yard occupancy across weekly or monthly Sentinel-1 passes, an analyst can construct a relative activity curve that tracks port congestion without access to any manifest data. Peaks correspond to vessel bunching or landside bottlenecks; troughs correspond to efficient dwell or reduced call frequency.
The Sentinel-1 archive extends to 2014 for most major port regions, which is long enough to establish seasonal baselines, detect structural changes in terminal capacity and identify anomalies such as the yard saturation events seen at several large transshipment hubs during 2020 and 2021. Commercial X-band data can be tasked to provide higher temporal resolution during specific events, with ICEYE's published revisit capability reaching sub-daily frequency over a given site when multiple satellites are directed at it.
Latency is a practical consideration. Sentinel-1 data typically appears in the Copernicus Data Space within a few hours of acquisition. Commercial SAR operators quote delivery windows of one to four hours from tasking for priority orders. For most port-economics applications, same-day delivery is sufficient; for real-time gate management it is not, and that boundary should be stated clearly.
Where the method reaches its limits
SAR is not optical imagery. It does not read container identification codes, ISO type designations or colour markings. A yard count derived from SAR tells you how many containers are present and, with coherence analysis, which ones have moved. It does not tell you whose containers they are or what is inside them. Integrating SAR-derived occupancy with terminal operating system data or AIS-derived vessel calls is necessary to convert a count into a commercial intelligence product.
Resolution also imposes a hard floor. At 10 m, counting individual containers is not feasible; the analyst is measuring zone occupancy. At 1 m, individual stacks are visible but stack height (and therefore total TEU count) is ambiguous because SAR is a side-looking sensor and layover effects compress tall stacks toward the sensor. Stereo SAR or lidar would be needed to resolve height precisely, and neither is routinely available at port scale.
Finally, very dense urban port environments with surrounding high-rise structures produce layover and shadow artefacts that can obscure yard areas. Incident angle selection matters: shallower angles (around 40–45 degrees) reduce layover in built-up ports compared with steeper geometries.
Building an operational monitoring programme
A practical occupancy monitoring programme for a single large terminal typically combines free Sentinel-1 data for continuous baseline tracking with periodic commercial X-band tasking for calibration and anomaly investigation. The Sentinel-1 time series establishes the occupancy index; the ICEYE or Capella passes validate the density coefficient and detect changes that fall below the C-band resolution floor.
Satellize structures analytics pipelines of this kind on open constellations with commercial tasking added on client licence, following the same architecture used in its Tonga crop-estimation programme. The deliverable is a georeferenced yard-occupancy layer, updated at the cadence the client specifies, with an accompanying change-detection alert when occupancy crosses a defined threshold. Port authorities, shipping lines and commodity traders have different threshold triggers; the alert logic is configured per use case rather than applied uniformly.
For a government client building a sovereign trade-intelligence capability, the same pipeline can cover multiple terminals simultaneously using Sentinel-1, with commercial passes reserved for the terminals that matter most at a given moment. That prioritisation logic is part of the programme design, not an afterthought.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 10 m ground range detected; yard-level occupancy only |
| Spatial resolution (ICEYE Spot mode) | 0.5–1 m; individual container row detection feasible |
| Spatial resolution (Capella Spotlight mode) | ~0.35 m published; stack-level discrimination possible |
| Revisit (Sentinel-1, two-satellite constellation) | 3–6 days at mid-latitudes; longer near equator |
| Revisit (ICEYE / Capella, tasked) | Sub-daily over priority sites when multiple satellites directed |
| Radar frequency | C-band 5.405 GHz (Sentinel-1); X-band ~9.6 GHz (ICEYE, Capella) |
| Minimum detectable target | Individual 20-ft container at ~1 m resolution; zone-level at 10 m |
| Archive depth (Sentinel-1) | 2014 to present for most major port regions |
| Data latency | Sentinel-1: hours after acquisition; commercial SAR: 1–4 hours from tasking |
| Delivery formats | GeoTIFF occupancy raster, GeoJSON zone polygons, CSV time-series index, alert feed |
Analytics Satellize can run
| Yard occupancy index (zone level) | Calibrated sigma-nought thresholding over defined yard polygons from Sentinel-1 IW scenes | Weekly GeoJSON layer with occupancy percentage per zone; CSV time series |
| Container stack count (row level) | Morphological matched-filter or CNN object detection on sub-metre X-band SAR chips | Georeferenced point layer of detected stack centroids; TEU estimate with stated uncertainty range |
| Active versus idle zone classification | Interferometric coherence differencing between repeat Sentinel-1 passes | Bi-weekly raster mask distinguishing high-turnover from static storage areas |
| Throughput trend index | Temporal stacking of occupancy snapshots; seasonal decomposition against multi-year Sentinel-1 baseline | Monthly trend report with anomaly flags; comparison against prior-year baseline |
| Congestion alert | Threshold exceedance on occupancy index; configurable per terminal and client | Near-real-time alert (email or API push) when yard occupancy crosses defined level |
| Multi-terminal comparative dashboard | Parallel Sentinel-1 pipelines across a defined port network; normalised index for cross-terminal comparison | Interactive dashboard or flat-file export; updated at agreed cadence |
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.