Satellite-based project-finance covenant monitoring
Lenders and development-finance institutions increasingly use satellite imagery as an independent check that physical milestones have been reached before releasing loan tranches. This page explains what imagery can and cannot prove, and which sensors are fit for evidentiary use.
Sensors
- Planet SuperDove (PlanetScope): 3 m native resolution, 8 spectral bands including red-edge and NIR, daily global revisit. Excellent for change detection between milestone dates, but the resolution sits at the boundary of resolving individual structural elements on most civil-infrastructure sites.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral, tasked on demand with same-day or next-day collection windows over most latitudes. The resolution is sufficient to distinguish poured concrete from formwork, identify crane presence, and count completed structural bays, making it the most commonly cited sensor for evidentiary milestone imagery.
- Maxar WorldView Legion: 30 cm panchromatic resolution, up to 15 revisit passes per day over targeted sites when the full constellation is operational. High revisit matters when a lender needs imagery within a narrow certification window rather than accepting whatever archive date happens to be available.
- Sentinel-2 MSI: 10 m resolution in visible and NIR bands, 5-day revisit at mid-latitudes with two satellites. Free and openly archived. Useful for establishing a long baseline change record and for monitoring large earthwork footprints, but insufficient on its own to verify structural completion at the element level.
Why lenders are paying for pixels they used to pay engineers to see
Traditional project-finance covenant verification relies on an independent engineer, appointed by the lender, who visits the site before each disbursement and certifies that the works described in the drawdown certificate actually exist. For a single large project this is manageable. For a portfolio of infrastructure loans spread across multiple countries, it is expensive, slow, and subject to the practical reality that an engineer can only be in one place at one time.
Satellite imagery does not replace the independent engineer. What it does is give the lender a continuous, datestamped, third-party record of site conditions that sits entirely outside the borrower's control. That independence is the point. A contractor can produce a progress photograph from any angle on any date. They cannot reposition a Pléiades Neo satellite.
What a floating roof gives away, and what a finished façade hides
The evidentiary value of optical imagery is highest for works that are visible from directly above or at shallow off-nadir angles, and that change the spectral or geometric signature of the site in a measurable way. Completed concrete pours, roofing membranes, earthwork volumes, paved aprons, installed tank shells, and the presence or absence of construction plant are all detectable at 30 cm resolution. A floating roof on a storage tank, for instance, is unambiguous evidence that the tank shell is structurally complete and the roof has been installed, which is a standard milestone in petroleum-storage project finance.
The limits are equally important to state clearly. Optical imagery cannot assess the quality of a concrete pour, the depth of a foundation, the integrity of a weld, or any internal fit-out behind a completed wall or roof. A building that appears structurally complete from above may be an empty shell with no mechanical, electrical or plumbing works installed. Subsurface works, including piling, drainage, and buried utilities, are invisible to optical sensors entirely. Any covenant that requires certification of works below grade or behind cladding still needs a physical inspection. Imagery narrows the scope of that inspection; it does not eliminate it.
Evidentiary standards: what makes an image admissible to a credit committee
A screenshot from Google Earth is not evidence. For imagery to function as a covenant-verification document it needs a chain of custody that a credit committee or dispute arbitrator can follow. That means: a named sensor with published specifications, a precise acquisition timestamp and ground-sample distance recorded in the image metadata, an orthorectification report confirming positional accuracy, and delivery through a licensed reseller or data provider whose terms of service confirm the imagery has not been altered.
Pléiades Neo and WorldView Legion imagery delivered through their respective licensed distribution channels carries all of this. The acquisition metadata is embedded in the product and is independently verifiable against the satellite's published orbital parameters. For high-value or contested disbursements, some lenders also request a stereo pair, which allows a photogrammetric height model to be derived and used to confirm that a structure has reached a specified elevation above a surveyed datum. That is a stronger evidentiary standard than a single nadir image alone.
Sentinel-2 archive imagery, while free and well-documented, is generally treated as corroborating context rather than primary evidence at the milestone level, because its 10 m resolution cannot resolve individual structural elements. It is, however, very useful for establishing that no significant activity occurred on a site during a period when a borrower claims works were advancing.
Structuring a monitoring programme around disbursement schedules
The practical design of a satellite monitoring programme for project finance starts with the loan agreement itself. Each disbursement condition that references a physical milestone needs to be translated into an observable proxy: what does that milestone look like from 500 km up, and at what resolution does it become unambiguous?
For most civil infrastructure, the answer is a tasking schedule tied to the expected milestone dates, with a buffer of two to four weeks to allow for cloud cover. Cloud is the most common operational problem in tropical and monsoon-affected markets, which are also the markets where development-finance institutions are most active. A single tasking attempt over a site in West Africa or South-East Asia during the wet season may fail entirely. A properly designed programme tasks repeatedly across a window, uses Sentinel-2 to track whether cloud-free opportunities have occurred, and specifies a fallback protocol, typically a physical inspection, if no usable image is obtained within the window.
Revisit frequency from WorldView Legion, when fully operational, can reach 15 passes per day over a targeted site, which dramatically reduces the cloud-gap risk compared with older single-satellite architectures. Even so, persistent convective cloud over equatorial sites can block collection for days at a stretch. Honest programme design acknowledges this and writes it into the verification protocol from the outset.
From raw image to disbursement memo: the analytic workflow
Raw satellite imagery is not a disbursement document. The workflow between image acquisition and a lender-ready milestone report involves orthorectification, radiometric calibration, comparison against a baseline image and the agreed milestone definition, annotation of the specific features that confirm or deny the milestone condition, and a written assessment signed by a qualified analyst.
Change detection between a pre-construction baseline and the current image is straightforward for large earthwork or structural changes. More nuanced assessments, such as distinguishing a completed structural frame from one that is still missing key elements, require an analyst with enough civil-engineering literacy to interpret what they are seeing. Automated classification can flag candidate changes, but the evidentiary document requires human sign-off. Satellize structures its covenant-monitoring deliverables as dated milestone reports with annotated imagery panels, a written determination against each covenant condition, and a confidence rating that distinguishes between what the imagery confirms directly and what it can only suggest.
For lenders managing a portfolio, a GIS dashboard showing current milestone status across all sites, updated after each successful collection, is more operationally useful than a stack of individual PDF reports. Both formats serve different audiences within the same institution.
Where the method is weakest, and how to design around it
Three failure modes deserve explicit attention. First, a sophisticated borrower who knows the monitoring schedule could stage visible works for the collection window and pause them immediately after. Irregular tasking, combined with the Sentinel-2 baseline record that reveals activity patterns between formal collection dates, makes this significantly harder to sustain without detection.
Second, imagery resolution affects the minimum detectable milestone. At 30 cm, a completed concrete column is visible; at 3 m, it is not. The sensor choice must match the milestone granularity written into the loan agreement. If the covenant specifies completion of individual structural bays, Pléiades Neo or WorldView Legion is the appropriate sensor. If it specifies completion of a major earthwork phase covering several hectares, Planet SuperDove is adequate and considerably cheaper to task.
Third, night-time construction, common on accelerated infrastructure programmes in the Gulf and parts of Asia, is invisible to optical sensors. Thermal infrared imagery, available from Landsat 8 and 9 at 100 m resolution, can detect large heat signatures from active concrete curing or industrial processes, but this is a coarse indicator rather than a milestone-level verification tool. For programmes where night-shift progress is material to the disbursement timeline, the monitoring design needs to account for this gap explicitly.
Typical figures
| Best available spatial resolution (panchromatic) | 30 cm (Pléiades Neo, WorldView Legion) |
| Multispectral resolution | 1.2 m (Pléiades Neo); 1.2 m (WorldView Legion); 3 m (Planet SuperDove); 10 m (Sentinel-2) |
| Revisit frequency (commercial tasking) | Up to 15 passes/day over a target site (WorldView Legion, full constellation); 1 pass/day typical (Pléiades Neo, depending on latitude and off-nadir tolerance) |
| Revisit frequency (open archive) | 5 days at mid-latitudes with two Sentinel-2 satellites |
| Image delivery latency | Typically 24–72 hours from tasking to orthorectified product delivery for commercial sensors; Sentinel-2 Level-2A available within hours of acquisition |
| Positional accuracy (orthorectified) | Pléiades Neo: CE90 ≤ 3 m without GCPs, sub-metre with GCPs; WorldView Legion: CE90 ≤ 5 m without GCPs |
| Stereo capability | Available from Pléiades Neo and WorldView Legion; enables photogrammetric DSM for structure-height verification |
| Minimum detectable structural element | Individual columns, beams, tank shells at 30 cm; building footprints and large plant at 3 m; earthwork extents at 10 m |
| Archive depth | Sentinel-2: from 2015; Maxar archive: from 2001 for some areas; Planet: from 2016 for most land areas |
| Delivery formats | GeoTIFF (orthorectified), NITF, annotated PDF milestone report, GIS vector overlays (GeoJSON, Shapefile), dashboard feed |
Analytics Satellize can run
| Milestone determination report | Analyst-led visual interpretation of high-resolution imagery against covenant-defined milestone criteria, with annotated image panels | Dated PDF report with annotated imagery, written determination per covenant condition, and confidence rating; suitable for credit-committee submission |
| Baseline-to-current change map | Pixel-level and object-based change detection between pre-construction or previous-milestone image and current acquisition | GeoTIFF change layer and summary statistics showing area and type of change; delivered as GIS layer or embedded in milestone report |
| Structure-height verification | Photogrammetric digital surface model derived from stereo Pléiades Neo or WorldView Legion pair, differenced against pre-construction terrain model | DSM raster and tabular height readings at specified structure locations, referenced to surveyed datum |
| Activity timeline reconstruction | Time-series analysis of Sentinel-2 and Planet imagery across the full construction period to establish activity chronology and detect gaps inconsistent with claimed progress | Annotated timeline chart and written narrative; supports dispute resolution or audit |
| Cloud-gap risk assessment | Historical cloud-frequency analysis using Sentinel-2 scene classification layers over the project location and season | Monthly cloud-probability table for the site, used to design tasking windows and fallback protocols in the monitoring agreement |
| Portfolio milestone dashboard | Automated ingestion of collection results across multiple sites, with analyst-reviewed status flags per milestone condition | Web GIS dashboard showing current milestone status, last-image date, and pending collection flags across a lender's full infrastructure portfolio |
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.