Impervious surface expansion and stormwater runoff risk from construction

As construction sites compact and pave, peak runoff rises faster than most site managers expect. Multispectral time-series from Sentinel-2 and Planet SuperDove can track impervious surface fraction week by week, giving regulators and developers early warning before watercourses breach compliance thresholds.

Sensors

Sentinel-2 MSI (ESA): 10 m resolution in visible and near-infrared bands (B2–B8A), 20 m in the two shortwave-infrared bands critical for distinguishing moist bare soil from dry concrete. Five-day revisit at the equator with both satellites. Free and open archive from 2015.
Planet SuperDove: 3 m native resolution, eight spectral bands including red-edge and two SWIR-adjacent bands, daily revisit over most land surfaces. Commercial licence required. Resolves access roads and small hardstanding pads that Sentinel-2 pixels straddle ambiguously.
Landsat 8/9 OLI (USGS/NASA): 30 m multispectral resolution, 16-day revisit per satellite (8-day combined). SWIR bands 6 and 7 on OLI are well-calibrated for impervious surface indices. Useful for historical baselines stretching back to 1984 via the Landsat archive.
Sentinel-1 SAR C-band (ESA): Not a spectral classifier, but cloud-penetrating C-band backscatter distinguishes smooth concrete and asphalt (low return) from rough disturbed ground (high return) and can bridge cloud-contaminated optical gaps during wet construction seasons. 5–20 m resolution, 6–12 day revisit.

What spectral physics separates concrete from grass

Impervious surfaces and permeable ones reflect sunlight differently in ways that persist across seasons. Concrete and asphalt absorb strongly in the near-infrared (NIR) compared with healthy vegetation, which reflects it vigorously. The Normalised Difference Vegetation Index (NDVI) captures this contrast directly: vegetated pixels typically score above 0.4, while hardstanding and bare compacted ground fall below 0.2. The Modified Normalised Difference Water Index (MNDWI) and the Index-based Built-up Index (IBI), which combines SWIR, NIR and SWIR-to-visible ratios, sharpen the separation further.

Bare compacted earth is the awkward middle case. Freshly graded soil can resemble low-density hardstanding in NDVI alone. The shortwave-infrared bands resolve this: dry concrete reflects SWIR strongly, while moist compacted soil absorbs it. Sentinel-2's Band 11 (1610 nm) and Band 12 (2190 nm) at 20 m resolution are the primary discriminators. Planet SuperDove's spectral configuration is slightly different but its 3 m pixels allow the analyst to see individual poured slabs rather than a mixed pixel spanning both slab and adjacent soil.

From classified pixels to runoff coefficients

The rational method, the most widely used first-order stormwater model in regulatory practice, expresses peak runoff as a product of a runoff coefficient (C), rainfall intensity and catchment area. Runoff coefficients for impervious surfaces typically range from 0.70 to 0.95; for short-grass or cultivated land they fall between 0.10 and 0.40. When satellite classification maps the impervious fraction of a catchment at successive dates, the area-weighted composite C can be recalculated for each epoch.

A site that starts as agricultural land with C around 0.25 and reaches 60 percent hardstanding mid-construction may see its composite C climb above 0.60. That shift, applied to a design storm, can more than double the peak discharge entering a downstream culvert or watercourse. The satellite time-series does not replace hydraulic modelling, but it provides the input that is most often missing or estimated crudely: the actual impervious fraction at any given date. Regulatory frameworks in the UK (the Schedule 3 SuDS approval regime in Wales, and the emerging mandatory SuDS provisions in England) increasingly require developers to demonstrate that runoff rates do not exceed pre-development greenfield rates. Satellite-derived impervious fraction gives that comparison an objective, dated evidence base.

The 10-metre pixel problem and how to manage it

A Sentinel-2 pixel at 10 m covers 100 square metres. A standard temporary site access road is 4–6 m wide. That road occupies at most 60 percent of a pixel it crosses, so the pixel's spectral signature is a mixture of road surface and adjacent ground. Sub-pixel unmixing can estimate the impervious fraction within a mixed pixel, but the uncertainty on any individual pixel is substantial. Narrow linear features, small hardstanding pads and individual drainage swales are genuinely below the reliable detection threshold of Sentinel-2 alone.

Planet SuperDove at 3 m resolves most access roads cleanly. The practical workflow pairs the two: Sentinel-2 for site-wide area accounting at five-day intervals, Planet for targeted verification of specific features where regulatory scrutiny is highest. Neither sensor is infallible. Planet's radiometric calibration across its large constellation has improved markedly since the SuperDove generation, but inter-sensor consistency still requires normalisation before multi-date comparison.

Cloud cover during wet construction seasons: the real operational constraint

Construction in temperate and tropical climates proceeds through seasons when cloud cover can obscure optical sensors for weeks at a stretch. In the UK, winter cloud cover renders many Sentinel-2 passes unusable. In monsoon-affected regions, the gap can extend to months. This is not a minor caveat; it is the dominant operational risk for optical-only monitoring programmes.

Three mitigations exist. First, Sentinel-1 SAR C-band imagery is cloud-independent. Backscatter change between passes can flag new hardstanding even when optical data is unavailable, though it cannot produce the same classification precision. Second, Planet's daily revisit increases the probability of at least one clear acquisition per week, even in cloudy climates. Third, temporal gap-filling using interpolated spectral indices (carrying forward the last valid observation with flagged uncertainty) can maintain a continuous impervious fraction estimate with honest confidence intervals attached. No approach eliminates cloud entirely. A monitoring programme should specify, upfront, the maximum acceptable gap in the evidence record and the SAR-based fallback that fills it.

Sedimentation as a co-indicator

Impervious surface expansion increases not just runoff volume but sediment load in the runoff that does flow. Bare disturbed ground on an active construction site is among the highest sediment-yield land cover types. Satellite imagery cannot directly measure suspended sediment concentration in a narrow drainage ditch, but it can monitor the area of exposed bare earth upstream, which is the proximate cause.

In watercourses wide enough to be resolved (roughly 10 m or more for Sentinel-2, 3 m or more for Planet), turbidity can be estimated from the red and NIR band ratio, a method validated extensively in estuarine and coastal settings. For most construction drainage channels this is below the resolution floor. The practical output is therefore a bare-earth exposure index for the site, updated at each cloud-free acquisition, which correlates with sedimentation risk and satisfies the evidence requirements of many environmental permit conditions without claiming to measure the ditch directly.

Fitting this into a regulatory evidence record

Environment agencies and local planning authorities increasingly accept satellite-derived land cover change as supporting evidence in environmental impact assessments, permit compliance reports and post-consent monitoring. The key requirements are reproducibility, traceability to calibrated sensor data, and clear documentation of classification method and uncertainty. Sentinel-2 and Landsat data are freely archived and radiometrically calibrated to surface reflectance through the Copernicus and USGS processing chains, which satisfies traceability. Classification outputs should be delivered as georeferenced GIS layers with per-pixel confidence scores, not just summary statistics.

Satellize has applied similar multispectral classification workflows to agricultural land cover in the Kingdom of Tonga crop-estimation programme, where the spectral separability challenges between bare soil and low-canopy crops are directly analogous to the hardstanding-versus-compacted-earth problem on construction sites. The next concrete step for a developer or environmental consultant is a site-specific feasibility assessment: cloud statistics for the site location, available archive depth, and a sample classification run on a historical Sentinel-2 scene to demonstrate separability before committing to a monitoring contract.

Typical figures

Spatial resolution (Sentinel-2 MSI)	10 m (visible/NIR), 20 m (SWIR bands used for impervious classification)
Spatial resolution (Planet SuperDove)	3 m native; resolves access roads ≥ 4 m wide
Spatial resolution (Landsat 8/9 OLI)	30 m multispectral; useful for historical baseline, not narrow-feature detection
Revisit frequency	Sentinel-2: 5 days (dual satellite); Planet: daily; Landsat 8+9 combined: ~8 days
Cloud penetration	Optical sensors (Sentinel-2, Planet, Landsat) blocked by cloud; Sentinel-1 SAR provides cloud-independent fallback
Key spectral bands for impervious classification	NIR (~835 nm), SWIR-1 (~1610 nm), SWIR-2 (~2190 nm); IBI and MNDWI indices derived from these
Minimum reliably detectable hardstanding patch	~400 m² with Sentinel-2 (2×2 pixels); ~20 m² with Planet SuperDove (2×2 pixels at 3 m)
Archive depth	Sentinel-2: from 2015; Landsat: from 1984; Planet SuperDove: from ~2021
Typical classification latency	1–3 days after cloud-free acquisition, depending on processing pipeline
Delivery formats	GeoTIFF impervious fraction raster, GeoPackage/Shapefile polygon layer, per-epoch summary CSV, PDF compliance report

Analytics Satellize can run

Impervious surface fraction map	Supervised spectral classification using IBI, MNDWI and NDVI indices derived from Sentinel-2 or Planet surface reflectance; random forest or support vector machine classifier trained on site-specific samples	GeoTIFF raster and polygon GIS layer, per acquisition date, with per-pixel confidence score
Impervious area change time-series	Multi-date classification stack differenced to produce epoch-by-epoch hardstanding growth; change pixels flagged above a minimum mapping unit threshold	CSV time-series of total impervious area (m²) and rate of change (m²/day), with cloud-gap flags
Area-weighted runoff coefficient estimate	Classified impervious fraction combined with published rational-method C values for each land cover class; composite C calculated per sub-catchment polygon	Per-epoch runoff coefficient table keyed to catchment zones, formatted for input to hydraulic models
Bare-earth exposure index	Spectral unmixing of disturbed bare soil fraction from Sentinel-2 SWIR and NIR bands; index correlated with sediment mobilisation risk per published erosion-risk frameworks	Weekly raster layer and site-level risk score, suitable for environmental permit compliance reporting
SAR-optical fusion gap-fill product	Sentinel-1 C-band backscatter change detection used to flag new hardstanding during cloud-obscured optical periods; change detections merged with nearest cloud-free optical classification	Continuous impervious fraction estimate with uncertainty bounds, covering cloud-affected periods
Regulatory compliance summary report	Automated comparison of satellite-derived impervious fraction against permitted development footprint and pre-development greenfield baseline; exceedance flagged against threshold	Dated PDF report with georeferenced evidence, traceable to calibrated Sentinel-2 or Landsat surface reflectance products

Sources

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.