Land border crossing vehicle traffic as bilateral trade-flow and sovereign-risk indicator
Very-high-resolution optical imagery over land border crossings turns vehicle queues and truck counts into a physical proxy for bilateral trade volumes, sanctions compliance and sovereign economic stress, updated far faster than official customs statistics.
Sensors
- Planet SkySat: 50 cm native resolution (resampled product at 50 cm), capable of resolving individual truck cabins and car rooftops; tasked on demand, with video mode available at roughly 30 frames per second over a 1 km² footprint for short-interval counting
- Maxar WorldView Legion: 30 cm panchromatic resolution, sufficient to distinguish vehicle class (HGV, passenger car, tanker); six-satellite constellation targeting up to 15 revisits per day over priority sites, though actual revisit depends on tasking competition and cloud
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral; stereo and tri-stereo collection modes allow height estimation that can help separate stacked cargo from vehicles; four-satellite constellation with same-day revisit feasible at mid-latitudes
- Sentinel-2 MSI: 10 m resolution in visible bands, 5-day revisit at the equator with both satellites; individual vehicles are below the detection floor, but queue shadow extent and lane saturation can be estimated at scale as a coarse activity index, free of charge under Copernicus open-data policy
What a truck queue actually measures
A border crossing is a physical bottleneck. When trade volumes rise, queues lengthen. When a crossing closes, queues dissolve or migrate to adjacent points. Neither event is invisible from orbit. A 20-metre articulated lorry casts a shadow roughly 3 to 5 metres long at mid-morning sun angles at temperate latitudes, and its rooftop occupies a 2.5 m by 8 m rectangle in panchromatic imagery at 30 cm resolution. That is enough to count, classify and time-stamp.
The relationship between queue length and crossing rate is not perfectly linear. Inspection regimes, lane configurations and seasonal patterns all introduce noise. But over weeks and months, the correlation between satellite-derived truck counts and customs-reported bilateral trade volumes has been demonstrated in published research using crossings on the US-Mexico border and several European land frontiers. The satellite signal is not a substitute for customs data; it is a leading indicator that arrives days to weeks earlier, and it cannot be revised retroactively by the reporting government.
From pixels to counts: the object-detection pipeline
Modern vehicle detection on very-high-resolution imagery uses convolutional neural networks trained on labelled satellite imagery. Publicly documented approaches, including those described in the DOTA benchmark dataset literature, achieve mean average precision figures above 0.75 for large vehicles in open scenes at 30 to 50 cm resolution. Performance degrades in shadow, under canopies, at oblique collection angles above roughly 30 degrees off-nadir, and wherever vehicles are tightly packed so that bounding boxes overlap.
Classification into vehicle type (passenger car, light goods vehicle, heavy goods vehicle, tanker) is feasible at 30 cm with well-trained models, but tanker versus dry-bulk trailer distinction is unreliable without additional spectral or contextual cues. Lane assignment matters for crossing-rate estimation: vehicles in the outbound lane are economically distinct from those in the inbound lane, and the crossing point geometry must be mapped precisely before any automated pipeline runs.
Time-series analysis requires consistent collection geometry. A crossing imaged at 15 degrees off-nadir on Monday and 28 degrees off-nadir on Thursday will show apparent queue-length differences that are partly geometric artefact. Analysts normalise for collection angle and sun elevation before constructing any index. Archive depth matters too: Planet's SkySat archive and Maxar's WorldView archive both extend back several years over frequently tasked sites, enabling baseline construction against which current counts are compared.
Sanctions compliance and the divergence signal
The most financially consequential application is not measuring normal trade; it is detecting abnormal trade. A crossing that officially reports near-zero bilateral commerce but shows sustained heavy-vehicle activity is a material discrepancy. That divergence is the signal that matters to sanctions monitors, sovereign-risk analysts and compliance officers at correspondent banks.
Optical imagery has a hard limit here. Night crossings are invisible to passive optical sensors. Covered staging areas, where trucks park under roofed structures before crossing, break the counting chain. A sophisticated evasion operation will route sensitive cargo through hours of darkness or under cover. SAR can detect vehicles at night and through cloud, but current commercial SAR systems, including Sentinel-1 at 5 by 20 metres in IW mode and ICEYE at roughly 1 metre in spotlight mode, do not reliably distinguish a lorry from a large car at the resolutions available. The honest position is that optical counting gives a strong daytime signal and SAR gives a coarse all-weather presence indicator; neither alone closes the evasion gap.
Combining satellite counts with AIS vessel data at nearby ports, night-light radiance trends from VIIRS and official trade statistics creates a multi-source picture that is considerably harder to game than any single data stream.
Building a crossing-rate index: method and honest caveats
A crossing-rate index is constructed by dividing observed vehicle count by the time window of observation, adjusting for lane capacity and applying a seasonal correction derived from the historical archive. The result is expressed as vehicles per hour or vehicles per day, normalised to a baseline period. Expressed as a z-score against the trailing 90-day mean, it becomes a stress indicator: crossings running more than two standard deviations above baseline suggest demand surge; crossings running below suggest restriction, closure or diversion.
Cloud cover is the dominant operational constraint at optical wavelengths. Major land border crossings in monsoonal climates can be obscured for weeks at a time. Revisit frequency from commercial constellations is high in principle, but usable cloud-free acquisitions over a specific 2 km² area may arrive only every few days even with aggressive tasking. Analysts should report the number of cloud-free observations per month alongside the index value, so that a thin observation record is not mistaken for a stable signal.
Satellize applies this pipeline as part of its broader sovereign-analytics work, drawing on open Sentinel-2 data for baseline construction and commercial tasking for high-resolution event-driven collection, in the same pattern used across its analytics engagements including the Tonga crop-estimation programme.
Applications in sovereign-risk and supply-chain finance
For a sovereign-risk analyst, a border-crossing index adds a dimension that GDP nowcasts and PMI surveys do not capture: physical evidence of cross-border economic integration or its breakdown. A country that is quietly redirecting trade through a secondary crossing, or whose primary crossing shows a step-change decline in heavy-vehicle traffic coinciding with a currency crisis, is showing its hand in the imagery before the quarterly trade statistics are published.
For commodity supply-chain monitoring, the application is more specific. A crossing known to handle agricultural exports can be watched during harvest season to estimate export pace. A crossing on a refined-product corridor can be monitored for tanker-truck frequency as a proxy for downstream fuel distribution. Neither application gives tonne-level precision, but both give directional signals with latency measured in days rather than months.
The key deliverable for financial clients is not a raw count but a structured time series with uncertainty bounds, anomaly flags and a written interpretation of what the physical signal implies for the bilateral relationship in question. A number without context is just a number.
Typical figures
| Best available spatial resolution | 30 cm panchromatic (Maxar WorldView Legion, Airbus Pléiades Neo) |
| Minimum detectable vehicle length | Approximately 4 m at 30 cm resolution; reliable HGV detection from roughly 10 m vehicle length |
| Revisit frequency (commercial VHR) | Up to 15 passes per day over priority sites (WorldView Legion); practical cloud-free acquisition typically every 1 to 5 days depending on latitude and season |
| Revisit frequency (Sentinel-2 free tier) | 5 days at equator, 2 to 3 days at mid-latitudes; 10 m resolution insufficient for individual vehicle detection |
| Spectral bands used | Panchromatic (450 to 800 nm) for detection and counting; multispectral (4 to 8 bands) for vehicle-type classification support |
| Night-time optical capability | None; passive optical sensors require daylight. Night crossings are not captured. |
| SAR vehicle discrimination | Coarse presence detection only; ICEYE spotlight at ~1 m, Sentinel-1 IW at 5 × 20 m. Vehicle-class distinction unreliable at current commercial resolutions. |
| Archive depth | Planet SkySat and Maxar WorldView archives extend 5 to 10+ years over frequently tasked sites; Sentinel-2 archive from 2015 |
| Typical analytic latency | 24 to 72 hours from satellite acquisition to delivered count, depending on cloud screening and QA workflow |
| Delivery formats | GeoJSON vehicle detections, CSV time-series index, PDF interpretation report, API feed for systematic monitoring |
Analytics Satellize can run
| Daily vehicle count by lane and vehicle class | CNN-based object detection (DOTA-class architecture) applied to VHR panchromatic imagery; bounding-box classification into passenger, LGV and HGV categories | GeoJSON detection layer with vehicle-class attribute and timestamp, delivered per acquisition |
| Crossing-rate index with anomaly flags | Time-series construction from repeated counts; z-score normalisation against 90-day rolling baseline; threshold-based anomaly flagging at ±2 standard deviations | CSV time series updated on each cloud-free acquisition; automated alert when index breaches threshold |
| Bilateral trade-flow proxy compared to official statistics | Regression of satellite-derived crossing rates against published customs data to calibrate conversion factor; divergence monitoring for anomaly detection | Monthly PDF report with divergence chart, observation-count quality indicator and written interpretation |
| Sanctions-evasion divergence signal | Multi-source fusion: optical vehicle counts, VIIRS night-light radiance at the crossing zone, AIS vessel activity at proximate ports; logical consistency check against declared bilateral trade | Structured risk report flagging crossings where physical activity materially exceeds or contradicts declared trade, with supporting imagery |
| Seasonal crossing-pattern baseline | Multi-year archive analysis using available VHR and Sentinel-2 imagery to characterise normal seasonal variation by crossing and vehicle class | GIS layer of crossing-specific seasonal profiles; used as reference for in-season anomaly scoring |
| Cloud-gap interpolated activity estimate | Gaussian process interpolation across cloud-obscured periods using adjacent clear observations; uncertainty bounds widen proportionally to gap length | Continuous daily index with explicit confidence interval; gap periods flagged in data feed |
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.