Illegal waste site and open-dump detection in urban fringes
Open dumps and unauthorised tipping grounds leave spectral fingerprints in SWIR and visible bands that change over time. Satellite analysis can flag candidate sites across entire city peripheries in days, though field verification before enforcement remains essential.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and NIR bands; 20 m in six SWIR bands (1610 nm and 2190 nm) that respond to plastic films and disturbed mineral soil. Five-day revisit at the equator with two satellites, enabling reliable monthly composites for temporal change detection. Free and globally archived from 2015.
- Landsat 8/9 OLI: 30 m multispectral including SWIR-1 and SWIR-2 bands. Sixteen-day single-satellite revisit, but the combined Landsat 8/9 pair reduces this to eight days. Archive depth to 1972 (Landsat 1) makes it the only sensor capable of establishing multi-decade baselines for long-standing dump sites.
- Planet Dove: 3 m resolution in four bands (blue, green, red, NIR). Daily revisit over most land areas. Useful for confirming spatial extent and detecting fresh tipping activity between Sentinel-2 acquisitions. No SWIR, so spectral discrimination of waste type is limited; Planet is primarily a change-detection and geometry tool here.
- Maxar WorldView-3: 0.3 m panchromatic, 1.24 m multispectral, and eight SWIR bands at 3.7 m covering 1195 nm to 2365 nm. The SWIR suite is the most capable commercially available for separating plastic, asphalt, organic decomposition and bare soil. Tasked on demand; not a systematic monitoring sensor. Cost and tasking lead time make it appropriate for site confirmation rather than area-wide screening.
What a dump looks like from 786 kilometres up
Open waste sites have three spectral characteristics that distinguish them from surrounding land. First, mixed-waste surfaces produce anomalously low NIR reflectance relative to visible bands compared with bare agricultural soil or construction spoil, because organic decomposition products and dark plastic films absorb strongly in the 700–900 nm range. Second, SWIR reflectance at 2190 nm is elevated by plastic polymers and certain leachate-stained soils relative to clean mineral substrates. Third, the surface is spatially heterogeneous at fine scales: a texture metric computed from a high-resolution image shows high local variance, because plastic sheeting, exposed soil patches and organic waste produce sharp reflectance contrasts within a few metres of each other.
None of these signals is individually diagnostic. Taken together in a spectral mixture model, and tracked through time as a site grows or changes composition, they become a reliable flag. The practical workflow is a two-stage funnel: Sentinel-2 composites screen large areas cheaply, and WorldView-3 SWIR imagery confirms candidates before any enforcement referral is made.
The construction-spoil problem
The single largest source of false positives is construction spoil and demolition rubble. Both produce bare-soil spectral signatures with elevated SWIR reflectance and low vegetation cover. Both appear suddenly in time-series analysis. Both are common on urban fringes, often adjacent to genuine waste dumps.
Separating them requires combining spectral evidence with contextual clues. Construction spoil tends to be geometrically regular, deposited in linear or fan-shaped forms consistent with truck tipping, and associated with nearby building activity detectable in Planet or WorldView imagery. Waste dumps are typically irregular in plan, show smoke plumes or thermal anomalies in Landsat thermal band (Band 10), and accumulate over months rather than arriving in a single episode. Even with all these indicators, ambiguity persists at sites that mix demolition rubble with general waste, which is common in rapidly urbanising areas. Field verification is not optional; it is the point at which satellite analysis ends and enforcement begins.
Temporal change is the most reliable signal
A single image rarely proves anything. A time series does. Sentinel-2's five-day revisit, combined with cloud-masking and median compositing over rolling 30-day windows, produces a monthly surface-reflectance record that reveals when a site first appeared, how fast it is growing, and whether tipping is ongoing or has ceased. Growth rates matter for enforcement prioritisation: a site expanding by more than a few hundred square metres per month is actively receiving waste, whereas a stable site may already have been remediated informally.
Landsat's archive extends this logic back to the 1980s for large sites. Several published studies have used Landsat time series to document the multi-decade accumulation of informal dumps in West African and South Asian cities, establishing both the spatial footprint and the approximate volume trajectory. Sentinel-2 then takes over for current monitoring from 2015 onward. The two archives are spectrally compatible enough for continuity analysis in the SWIR bands, though cross-calibration adjustments are advisable.
Cloud cover is the main operational constraint. In humid tropical cities, monthly compositing may still leave gaps in the wet season. In those environments, SAR data from Sentinel-1 can detect the roughness contrast between waste surfaces and surrounding land, though SAR cannot resolve spectral waste type at all.
Spectral indices worth knowing
Several published indices are applicable here. The Normalised Difference Vegetation Index (NDVI) confirms absence of vegetation cover; active dump surfaces typically show NDVI below 0.1. The Normalised Difference Built-up Index (NDBI), which uses SWIR-1 and NIR, responds to both built surfaces and waste. More specific is the Bare Soil Index (BSI), computed from SWIR-2, red, NIR and blue bands, which isolates exposed mineral and disturbed soil surfaces. None of these was designed for waste detection specifically, but their combination in a supervised classifier trained on confirmed dump sites produces reasonable detection rates in published literature.
WorldView-3's eight SWIR bands allow a finer analysis. Polyethylene and polypropylene films have absorption features near 1730 nm and 2310 nm that are resolvable at 3.7 m resolution. This level of discrimination is not possible with Sentinel-2's two SWIR bands. For governments that need to characterise waste composition for remediation planning, WorldView-3 SWIR tasking over confirmed sites adds real value. For initial screening across hundreds of square kilometres, it is impractical.
What the analysis can and cannot deliver
A well-designed detection pipeline over a city of one million people, with a 20-kilometre peri-urban buffer, can typically flag candidate sites down to roughly 1,000 square metres using Sentinel-2, and down to perhaps 500 square metres with Planet Dove where daily imagery is cloud-free. Below those thresholds, informal tipping at road verges or drainage channels falls below reliable detection. Volume estimation from optical imagery alone is not possible; depth requires either field survey or, for larger sites, photogrammetric DSM differencing from stereo VHR imagery.
Satellize runs this type of spectral-change screening as part of its urban analytics work, applying the same compositing and anomaly-detection pipeline used in the Tonga crop-estimation programme to urban land-cover problems. The output for a waste-detection engagement is a GIS layer of candidate sites with confidence scores, a monthly change log, and a site-priority ranking for field teams. Enforcement decisions remain with the client authority.
One honest caveat: detection performance degrades sharply in cities where construction activity is dense and ongoing, because the false-positive rate from spoil rises faster than the true-positive rate from waste. In those environments, the classifier needs frequent retraining against locally verified ground truth to stay calibrated.
From flag to enforcement: the verification chain
Satellite analysis produces a list of sites to investigate, ranked by evidence strength and growth rate. It does not produce evidence admissible in court or sufficient for formal enforcement without field confirmation. That distinction matters for how the output is framed to client agencies.
A practical workflow runs in three stages. Sentinel-2 composites generate a monthly candidate list for the whole peri-urban zone. Planet Dove imagery, pulled for the highest-ranked candidates, confirms recent activity and provides a more precise boundary. WorldView-3 is tasked only for the subset of sites that field teams have confirmed as waste rather than spoil, providing the spatial precision and spectral detail needed for remediation scoping. This staged approach keeps costs proportionate to the evidence required at each decision point.
Typical figures
| Screening spatial resolution | 10–20 m (Sentinel-2 MSI); 30 m (Landsat 8/9 OLI) |
| Confirmation spatial resolution | 3 m (Planet Dove); 1.24 m multispectral / 3.7 m SWIR (WorldView-3) |
| Revisit for change detection | 5 days (Sentinel-2, two satellites); ~8 days (Landsat 8+9 combined); daily (Planet Dove) |
| Key spectral bands | SWIR-1 (1610 nm), SWIR-2 (2190 nm), NIR (842 nm), Red (665 nm); WorldView-3 adds eight SWIR bands from 1195–2365 nm |
| Minimum detectable site area | ~1,000 m² with Sentinel-2; ~500 m² with Planet Dove (cloud-free conditions) |
| Archive depth | 2015–present (Sentinel-2); 1972–present (Landsat); Planet Dove from ~2016 |
| Cloud-cover constraint | 30-day compositing mitigates cloud in temperate climates; wet-season tropical gaps may persist |
| Delivery formats | GeoTIFF change layers, GeoJSON candidate-site polygons, CSV priority rankings, PDF monthly report |
| Latency (screening product) | Typically 3–5 days after end of compositing period, depending on cloud cover and processing queue |
Analytics Satellize can run
| Peri-urban waste-site candidate map | Supervised spectral classification using BSI, NDBI and SWIR band ratios on Sentinel-2 monthly composites | GeoJSON polygon layer with confidence score per candidate site, updated monthly |
| Site growth rate time series | Temporal change detection on rolling 30-day Sentinel-2 median composites; area differencing between periods | CSV table of site area by month; flagged alert when growth exceeds configurable threshold |
| Construction-spoil vs. waste discrimination score | Multi-feature classifier combining spectral indices, texture variance (GLCM), geometric regularity and proximity to active construction detected in Planet imagery | Attribute field in candidate-site GeoJSON; sites below discrimination threshold flagged for mandatory field check |
| Historical dump baseline (pre-2015) | Landsat 4–9 archive time-series analysis using SWIR band composites and NDVI suppression masks | Decade-by-decade site-boundary reconstruction as GeoTIFF stack; summary PDF |
| Waste composition proxy (confirmed sites only) | WorldView-3 SWIR spectral unmixing using published plastic and organic endmembers across eight SWIR bands | Per-site composition fraction map (plastic, organic, bare soil, other) as GeoTIFF; remediation-scoping annex |
| Enforcement priority ranking | Multi-criteria scoring combining site area, growth rate, proximity to watercourses, residential density and confidence score | Ranked site list as PDF and GIS layer, refreshed monthly, for field-team tasking |
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.