Sensors
- Pléiades Neo tri-stereo (Airbus): 30 cm native resolution panchromatic, three along-track viewing angles collected in a single pass. Tri-stereo geometry yields dense point clouds with vertical accuracy typically 0.3–0.5 m RMSE over textured surfaces, degrading on uniform fine-grained spoil.
- TerraSAR-X / TanDEM-X (Airbus): X-band SAR at 0.25–1 m resolution in Spotlight mode. Amplitude-based height estimation and radargrammetric stereo can resolve heap height changes of roughly 1–2 m; phase-based interferometry is less reliable on loose, temporally decorrelating spoil surfaces.
- SPOT 6/7 stereo (Airbus): 1.5 m panchromatic stereo pairs. Vertical accuracy 1–2 m RMSE under good conditions. Useful for large heaps where sub-metre precision is not required, and for archival baseline DSMs before a project begins.
- Sentinel-1 (ESA, C-band SAR): Free, 6-day repeat at mid-latitudes in IW mode, 5×20 m ground range resolution. Insufficient for precise volumetric differencing but useful for detecting gross heap boundary expansion and monitoring site activity between commercial acquisitions.
Why a spoil heap is a better progress meter than a contractor's report
Every cubic metre of rock and soil a tunnel boring machine or drill-and-blast crew removes has to go somewhere. On most large tunnelling contracts, it accumulates in licensed spoil heaps adjacent to the portal or at designated disposal sites. The relationship is simple geometry: excavated volume equals spoil volume multiplied by a bulking factor, typically 1.2 to 1.5 depending on rock type. That factor is well-documented in geotechnical literature for common lithologies. Measure the heap, divide by the bulking factor, and you have an independent estimate of how much tunnel has actually been driven.
Project owners, planning authorities, and infrastructure lenders all have reasons to want that independent check. Contractors may over-report advance to trigger milestone payments. Spoil-disposal licences cap the volume that can be deposited at a given site. Both problems are invisible to a site visit unless someone is counting lorry loads around the clock. A satellite revisit every few days is not around the clock, but it is systematic, timestamped, and difficult to dispute.
What tri-stereo photogrammetry actually measures, and where it fails
Tri-stereo acquisition, as implemented on Pléiades Neo, collects three images of the same area in a single overpass from forward, nadir, and aft viewing angles. Because all three are captured within seconds of each other, atmospheric conditions are essentially identical across the triplet. That matters: conventional stereo from two separate passes introduces differential atmospheric delay that can add decimetric noise to the resulting digital surface model. The three-image geometry also improves the density and reliability of matched point clouds compared with a simple stereo pair, particularly on the irregular, sloped surfaces of a spoil heap.
Volumetric change is computed by differencing two DSMs: a baseline acquired before significant deposition and a monitoring DSM acquired at a later date. The difference raster, integrated over the heap footprint, gives the deposited volume directly. Vertical accuracy on well-textured, coarse-grained spoil is typically 0.3–0.5 m RMSE for Pléiades Neo. On fine-grained, uniformly coloured material such as chalk or clay-rich overburden, photogrammetric matching degrades because the surface offers few distinctive features for the correlation algorithm. In those cases, errors can reach 1 m or more, which translates to meaningful volume uncertainty on a small heap. Ground control points, where accessible, constrain the error budget considerably.
Cloud cover is the operational constraint that does not yield to better algorithms. A single overcast acquisition is a wasted tasking slot. In high-cloud-frequency environments, planning for multiple tasking attempts per monitoring epoch is not optional.
SAR amplitude and radargrammetry: a cloud-independent alternative with its own caveats
TerraSAR-X in Spotlight mode provides sub-metre resolution SAR imagery regardless of cloud or darkness. Two Spotlight acquisitions from slightly different incidence angles can be processed as a radargrammetric stereo pair to extract a DSM, analogous to optical stereo but using radar backscatter intensity rather than pixel brightness. Published studies using TerraSAR-X radargrammetry report vertical accuracy in the range of 1–3 m RMSE over terrain with moderate relief, which is adequate for tracking a heap growing by several metres per month but insufficient for detecting small weekly increments.
Interferometric SAR (InSAR) is theoretically more precise, but loose spoil is a poor interferometric target. The surface shifts and settles between acquisitions, destroying phase coherence. Decorrelation is especially severe on freshly deposited, fine-grained material. Radargrammetry, which uses amplitude rather than phase, is therefore the more reliable SAR approach for active spoil heaps. The trade-off is coarser vertical precision than InSAR would provide over stable ground.
X-band wavelengths (around 3.1 cm for TerraSAR-X) are sensitive to surface moisture. A wet heap surface scatters differently from a dry one, which can introduce apparent height changes that are actually moisture-state changes. Acquiring pairs under similar meteorological conditions, or correcting for known rainfall events, is part of responsible processing.
Translating volume estimates into excavation advance
Converting heap volume to tunnel advance requires knowing the tunnel cross-sectional area and the bulking factor for the excavated material. Both are available from the contract documents or geotechnical baseline report. The cross-section is fixed by design. The bulking factor has a documented range for each rock class: massive granite typically bulks at 1.3–1.5, soft clays at 1.1–1.2, mixed ground somewhere between. Using the midpoint of the published range and propagating the uncertainty into the advance estimate gives an honest confidence interval rather than a spuriously precise single number.
A worked example illustrates the logic. A heap grows by 15,000 m³ over a 30-day period. Assuming a bulking factor of 1.35 for a sandstone formation and a 90 m² tunnel cross-section, the implied advance is approximately 123 m. The contract claimed 140 m. The discrepancy is outside the uncertainty range of the satellite measurement and warrants a formal query. That is the kind of specific, defensible output that project finance teams and planning authorities can act on.
Compliance monitoring and the limits of what a satellite can prove
Spoil-disposal licences issued by planning authorities typically specify a maximum deposited volume, a permitted footprint, and sometimes a maximum heap height. DSM differencing can check all three. Footprint expansion is visible in any high-resolution optical image. Height is directly measured by the DSM. Volume is the integral of the two. A monitoring cadence of once or twice per month is generally sufficient to detect licence exceedance before it becomes severe, provided cloud cover does not block every acquisition.
Satellite evidence has evidential weight, but it is not a court-ready measurement without supporting documentation. Positional accuracy depends on ground control or the satellite's direct geolocation accuracy, which for Pléiades Neo is better than 0.3 m CE90 with ground control and around 3–4 m CE90 without. Volume estimates carry the uncertainties described above. A responsible compliance report states those uncertainties explicitly rather than presenting a single number as definitive. That candour, counterintuitively, makes the evidence more credible to a planning officer or judge, not less.
Satellize runs this type of DSM-differencing workflow on commercially tasked Pléiades Neo acquisitions, delivering georeferenced volume-change rasters and summary tables suitable for inclusion in planning compliance reports. The approach is the same class of photogrammetric method used in the Tonga crop-estimation programme, adapted from agricultural canopy height to industrial spoil geometry. Clients wanting to discuss a specific tunnelling project can request a scoping call with the analytics team.
Practical decisions before the first tasking order
A baseline DSM acquired before significant spoil deposition begins is essential. Without it, the reference surface is unknown and volume estimates rest on assumptions about pre-existing terrain that introduce large errors. Commissioning a baseline acquisition at contract award, before portal excavation starts, costs one tasking slot and saves considerable argument later.
Sensor choice depends on cloud climatology and required precision. In persistently cloudy regions, a SAR-primary strategy with optical confirmation when skies clear is more reliable than waiting for cloud-free Pléiades windows. In arid or semi-arid environments, optical tri-stereo is usually the better choice given its superior vertical accuracy. Heap size matters too: for heaps smaller than roughly 50,000 m³, the volume uncertainty from SAR radargrammetry may exceed 20%, which limits its usefulness for payment verification. Optical stereo is preferable at that scale.
Typical figures
| Best-case vertical accuracy (optical tri-stereo) | 0.3–0.5 m RMSE (Pléiades Neo, coarse-textured surface, ground control present) |
| Typical vertical accuracy (SAR radargrammetry) | 1–3 m RMSE (TerraSAR-X Spotlight stereo pair) |
| Spatial resolution (optical) | 30 cm pan (Pléiades Neo); 1.5 m pan (SPOT 6/7) |
| Spatial resolution (SAR) | 0.25–1 m (TerraSAR-X Spotlight); 5×20 m (Sentinel-1 IW) |
| Revisit / tasking cadence | Pléiades Neo: 1-day revisit capacity; TerraSAR-X: 11-day repeat, off-track tasking to ~2.5 days; Sentinel-1: 6-day free repeat |
| Minimum detectable volume change | Approximately 500–2,000 m³ (optical tri-stereo); 5,000–15,000 m³ (SAR radargrammetry), depending on heap extent and surface texture |
| Cloud sensitivity | Optical methods fully blocked by cloud; SAR unaffected |
| Archive depth | Pléiades 1A/1B from 2012; TerraSAR-X from 2007; Sentinel-1 from 2014 |
| Delivery formats | GeoTIFF DSM differencing rasters, volume-change CSV tables, georeferenced heap-footprint shapefiles |
| Typical processing latency | 3–7 days from acquisition to delivered report under standard workflow |
Analytics Satellize can run
| Baseline digital surface model | Tri-stereo photogrammetric point cloud generation and rasterisation (Pléiades Neo or SPOT 6/7) | GeoTIFF DSM at 0.5–1 m post spacing, delivered before excavation begins |
| Periodic volume-change raster | DSM differencing between baseline and monitoring epoch; integration over heap footprint polygon | GeoTIFF difference raster plus CSV table of net volume added per epoch |
| Implied excavation advance estimate | Volume-to-advance conversion using contract cross-section and published bulking factors, with uncertainty propagation | PDF summary table with confidence intervals, formatted for project-finance reporting |
| Heap footprint compliance check | Automated polygon comparison of measured heap boundary against licensed footprint limit | GIS shapefile with flagged exceedance zones and percentage overage metric |
| SAR-based height change monitoring | TerraSAR-X radargrammetric stereo DSM generation and differencing for cloud-persistent sites | GeoTIFF height-change raster with stated RMSE, delivered within 5 days of acquisition |
| Multi-epoch growth rate time series | Sequential DSM differencing across all available epochs, fitted with linear or piecewise-linear advance model | Time-series chart and CSV of cumulative volume versus date, suitable for lender progress reports |
| Licence exceedance alert | Threshold comparison of computed volume or height against permit limits, triggered on each processed epoch | Email alert with supporting imagery and measurement summary, issued within 24 hours of threshold breach detection |
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.