Cold-chain and logistics warehouse construction progress monitoring
Satellite imagery can track cold-store and logistics warehouse construction months before facilities enter official capacity databases, giving freight operators and investors a leading indicator of supply-chain infrastructure.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; 5-day revisit at mid-latitudes with both satellites. Sufficient for footprint extraction and roof-type classification on large facilities (>5,000 m²), but cannot reliably distinguish cold-store from standard warehouse construction at this resolution alone.
- Planet SuperDove: 3 m resolution, 8-band multispectral including red-edge and near-infrared, daily revisit over most land areas. Resolves individual roof panels and construction equipment, making it the primary sensor for week-on-week cadence tracking.
- Airbus Pléiades Neo stereo: 30 cm panchromatic, 50 cm multispectral, with same-pass stereo pairs that produce digital surface models accurate to roughly 0.5 m in height. Used for earthworks volume estimation and to confirm structural height consistent with refrigerated storage (typically 12–18 m clear internal height).
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral with 8 bands including shortwave infrared. SWIR bands help discriminate roof insulation materials and panel coatings that are spectrally distinct from standard corrugated steel.
Why construction cadence is a freight-capacity signal
Cold-chain capacity additions take 18 to 36 months from groundbreak to commissioning. Official statistics, planning registers and industry surveys typically capture a facility only at or after opening. That gap is long enough to matter to a shipping line choosing port hinterland, a retailer selecting a third-party logistics partner, or a government assessing food-security infrastructure.
Satellite imagery closes that gap. Ground clearance, slab pour, steel erection, roof installation and car-park surfacing are all spectrally and geometrically distinct events. Monitoring them in sequence produces a construction S-curve that can be extrapolated to an estimated completion date, with uncertainty bounds that narrow as the build progresses. The output is a leading indicator, not a lagging statistic.
What a standing-seam roof gives away in near-infrared
Refrigerated warehouses almost universally use insulated metal panel (IMP) roofing: a sandwich of steel facings around a polyisocyanurate or mineral-wool core. The outer steel facing is typically pre-painted with a high-reflectance coating, often white or light grey, to reduce solar heat gain. In Sentinel-2 band 8 (near-infrared, centred at 842 nm) and in Planet SuperDove's NIR band, these coatings produce reflectance values noticeably higher than the dark membrane or gravel-ballast roofs common on ambient-temperature warehouses.
The distinction is real but not absolute. Some modern ambient-temperature distribution centres also use high-reflectance 'cool roof' coatings for energy efficiency. Shortwave infrared bands (WorldView-3 SWIR, 1195–2365 nm) add discriminating power: the polyurethane and mineral-wool insulation beneath IMP panels has a different SWIR signature from the bitumen or EPDM membranes used on non-refrigerated roofs. At Sentinel-2's 20 m SWIR resolution the signal is marginal for a single building; at WorldView-3's 3.7 m SWIR resolution it becomes actionable. Honest caveat: spectral classification alone cannot confirm cold-chain use. It raises the probability. Ground-truth or planning data is needed to confirm.
Earthworks volume and the stereo DSM method
Before a steel frame goes up, earthworks reveal scale. Large cold-store developments require heavily compacted sub-bases to carry the floor loads of racked frozen product (floor loads of 5–10 tonnes per square metre are common in automated cold stores). The earthworks phase, visible as large areas of disturbed soil with characteristic pale spectral signatures, can be detected in Planet imagery within days of commencement.
Pléiades Neo stereo pairs allow a digital surface model to be derived by photogrammetric matching, typically achieving 0.5 m vertical accuracy under good contrast conditions. Differencing a pre-construction DSM against one taken during the earthworks phase yields a cut-and-fill volume estimate. This is useful for two reasons: it confirms the site is being prepared for a substantial permanent structure rather than a temporary compound, and it provides an early size estimate before any above-ground structure is visible. The method is well established in the published literature on construction monitoring from very-high-resolution stereo imagery.
Ambiguity at Sentinel-2 resolution: what it cannot resolve
Sentinel-2 is free, global and revisits every five days. That makes it the right sensor for screening large geographies to find active construction sites. It is not the right sensor for classifying what is being built. At 10 m resolution, a 10,000 m² warehouse footprint covers roughly 100 pixels. Roof spectral signatures are detectable in aggregate, but mixed-pixel effects at building edges, shadows from adjacent structures, and the spectral similarity between new metal roofing and other bright surfaces (concrete aprons, skylights, solar panels) all introduce false positives.
The practical workflow is a two-stage funnel. Sentinel-2 flags active construction sites across a wide area of interest, typically a country or logistics corridor. Planet SuperDove then provides 3 m daily imagery on the flagged sites, resolving individual roof panels, construction equipment and phasing. Pléiades Neo stereo is tasked selectively on the highest-priority sites for height and volume confirmation. Each sensor tier adds cost and reduces the area that can be covered economically, so the screening logic matters.
Constructing a progress index: from pixels to a completion forecast
A useful construction progress index assigns a normalised score (0 to 1) to each monitored site based on the proportion of construction phases completed. The phases are observable from imagery: site clearance, slab visible, structural steel at least 50% erected, roof deck complete, cladding complete, external hardstanding surfaced, car park and service yard marked. Each phase transition is dated from the image archive.
Phase transition dates, combined with the known duration distributions of each construction phase drawn from published industry data, feed a simple probabilistic model that outputs an estimated completion date with a credible interval. The interval is wide early in construction (plus or minus four months is realistic at groundbreak) and narrows to plus or minus three to four weeks once the roof is closed. That is still ahead of the facility appearing in any logistics directory or official capacity register.
Satellize runs this kind of multi-sensor construction cadence analysis on open and commercial constellations. The Tonga crop-estimation programme uses a comparable phase-detection logic, applied to agricultural rather than industrial land use, which gives a sense of how the underlying method transfers across domains.
Practical limits buyers should price in
Cloud cover is the principal operational constraint. Persistent cloud over equatorial and monsoonal regions can produce gaps of two to four weeks even in Planet's daily constellation, because usable imagery requires less than roughly 20% cloud cover over the site of interest. SAR can fill some of this gap for footprint detection but does not provide the spectral information needed for roof classification.
Archive depth for commercial very-high-resolution imagery varies by provider and location. Planet's SuperDove archive is dense from 2021 onwards over most land areas; earlier coverage is sparser. Pléiades stereo must generally be tasked prospectively rather than retrieved from archive. For sites where construction began before the monitoring programme started, the historical S-curve can only be reconstructed from whatever archive imagery exists, which may have gaps.
Finally, a progress index is a probabilistic forecast, not a confirmed opening date. Permitting delays, supply-chain disruptions and financing events are not visible from orbit. The satellite signal tells you what has been built; it cannot tell you when the refrigeration plant will be commissioned or when the first pallet will arrive.
Typical figures
| Spatial resolution (screening) | 10 m (Sentinel-2 visible/NIR); 20 m (Sentinel-2 SWIR) |
| Spatial resolution (classification and cadence) | 3 m multispectral (Planet SuperDove) |
| Spatial resolution (stereo DSM and detail) | 30 cm pan / 50 cm MS (Pléiades Neo); ~0.5 m vertical DSM accuracy |
| Revisit cadence | 5 days (Sentinel-2, mid-latitudes); daily (Planet SuperDove, subject to cloud); on-demand tasking (Pléiades Neo, typically 1–3 day response) |
| Key spectral bands | NIR (842 nm, Sentinel-2 B8; Planet band 7); SWIR (1610 nm, 2190 nm, Sentinel-2 B11/B12; WorldView-3 SWIR 1–8) |
| Minimum detectable footprint (reliable) | ~2,500 m² at Planet 3 m resolution; ~5,000 m² at Sentinel-2 10 m |
| DSM vertical accuracy (stereo) | ~0.5 m (Pléiades Neo, good contrast conditions); 1–2 m (Pléiades 1A/1B) |
| Archive depth | Sentinel-2: from 2015; Planet SuperDove: dense from 2021; Pléiades: variable, typically prospective tasking |
| Delivery formats | GeoTIFF change layers, GeoJSON footprint polygons, CSV progress index time series, PDF site reports |
| Latency (operational monitoring) | 24–72 hours from image acquisition to analytic output, depending on sensor and processing tier |
Analytics Satellize can run
| Active construction site detection | Change detection on Sentinel-2 and Planet time series using normalised difference built-up index (NDBI) and bare-soil spectral indices to flag newly disturbed ground | Weekly GeoJSON alert layer of newly active construction sites within a defined area of interest |
| Building footprint extraction | Supervised pixel or object-based classification on Planet SuperDove imagery; footprint polygons validated against DSM height layer | GeoJSON footprint polygons with area estimate and first-detection date, updated monthly |
| Roof material classification | Spectral unmixing and supervised classification using NIR and SWIR reflectance to discriminate high-reflectance insulated metal panel from dark membrane and other roof types | Per-building roof-type probability score (cold-store IMP / standard membrane / other), delivered as attributed GeoJSON |
| Earthworks volume estimate | Photogrammetric DSM differencing from Pléiades Neo stereo pairs; cut-and-fill volume calculated from pre- and mid-construction surface models | Site report with volume estimate (m³), uncertainty range and annotated DSM visualisation |
| Construction progress index | Phase-transition dating from image time series; probabilistic completion forecast using phase-duration priors | Monthly CSV time series per site with phase scores (0–1) and estimated completion date with 80% credible interval |
| Logistics corridor capacity pipeline | Aggregation of site-level progress indices across a defined corridor or administrative region; comparison against prior-year baseline | Quarterly PDF briefing with map and capacity-addition forecast for the corridor, suitable for board or investment committee use |
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.