Unauthorised irrigation water abstraction detection
Unlicensed irrigation leaves a spectral fingerprint: anomalously high NDVI and elevated evapotranspiration in areas with no registered water rights. Fusing Sentinel-2, Landsat and FAO WaPOR data turns that fingerprint into enforcement evidence.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; 5-day revisit at the equator with both satellites. Provides NDVI (bands 4 and 8) and NDWI (bands 3 and 8) time series for detecting actively irrigated parcels. Cloud cover is the principal operational limit in humid climates.
- Landsat 8/9 OLI-TIRS: 30 m multispectral resolution with a 16-day revisit per satellite, reduced to roughly 8 days when both are combined. The thermal infrared band (Band 10, ~100 m resampled to 30 m) enables land-surface temperature estimation, a proxy for evaporative cooling that distinguishes irrigated from dryland parcels.
- Sentinel-1 SAR C-band: 10 m ground range detected imagery, 6-day revisit (IW mode, dual-polarisation). Cloud-independent. Detects new metallic infrastructure such as pump housings, pipes and centre-pivot frames through backscatter change analysis. Cannot directly measure water flow but identifies physical installation of abstraction equipment.
- MODIS Terra/Aqua: 250 m to 500 m resolution with daily revisit. Too coarse to resolve individual small parcels but useful for basin-scale evapotranspiration anomaly screening and for gap-filling during extended cloud periods over Sentinel-2 and Landsat acquisitions.
- FAO WaPOR (Copernicus-derived ET products): FAO's Water Productivity Open-access Portal provides actual evapotranspiration at 100 m (Level 2, dekadal) and 30 m (Level 3, dekadal) for Africa and the Near East, derived from Sentinel-2 and MODIS. Outside WaPOR coverage, SEBAL or METRIC algorithms applied to Landsat TIRS yield comparable ET estimates.
What irrigated fields reveal that paper records conceal
Water abstraction licensing is a bureaucratic process. Irrigation is a physical one. When an unlicensed operator pumps from a river or sinks an illegal borehole, the water goes somewhere: into soil, into plant tissue, into the atmosphere as vapour. Each of those pathways changes the spectral and thermal character of the land surface in ways that sensors in orbit can measure.
Actively irrigated crops in an arid or semi-arid setting will show NDVI values typically above 0.4 during the growing season, while surrounding dryland parcels remain below 0.2. Evapotranspiration over irrigated fields commonly runs two to four times higher than over rain-fed land in the same period. Neither figure is subtle. The challenge is not detection sensitivity; it is cross-referencing the spectral signal against the water-rights register to flag the parcels that should be brown but are green.
Building the evidence chain: NDVI, ET and the licence boundary
The analytical workflow has three stages. First, a dense NDVI and NDWI time series is constructed from Sentinel-2 imagery, typically at monthly composites over two or more growing seasons to distinguish persistent irrigation from a single wet year driven by rainfall. Parcels showing sustained high NDVI through dry months, when rain-fed agriculture would be dormant, are the primary candidates.
Second, actual evapotranspiration is estimated for those candidate parcels. FAO WaPOR dekadal ET products cover much of Africa and the Near East at 100 m resolution. For the Murray-Darling Basin or the Tagus and Guadalquivir systems in Spain, METRIC or SEBAL applied to Landsat 8/9 thermal data provides equivalent estimates. Published studies in the Murray-Darring Basin have demonstrated that this approach can attribute irrigation activity to individual cadastral parcels with spatial accuracy consistent with 30 m Landsat resolution.
Third, the irrigated-area map is overlaid against the water-rights register. Parcels outside the licensed perimeter, or parcels licensed for dryland use showing clear irrigation signatures, become the enforcement shortlist. The output is a spatial anomaly layer, not a volumetric abstraction figure. Volume requires flow-meter data or hydrological modelling; satellite observation supplies the where, not the how much.
SAR as the infrastructure witness
Spectral methods confirm that irrigation is happening. Sentinel-1 SAR adds a different kind of evidence: it can detect the physical infrastructure that makes abstraction possible. Centre-pivot irrigation systems produce distinctive circular backscatter patterns visible in C-band SAR at 10 m resolution. Pump stations, header tanks and above-ground pipework introduce localised increases in cross-polarisation backscatter relative to surrounding agricultural land.
Change detection between SAR acquisitions separated by weeks or months can identify new installations. A field that shows no metallic infrastructure in a January stack but clear backscatter anomalies by April, coinciding with the onset of an NDVI increase, builds a temporal narrative that is harder to dismiss than a single optical snapshot. SAR is cloud-independent, which matters in regions where optical coverage is interrupted precisely during the wet-to-dry transition when illegal abstraction often begins.
Where the method reaches its limits
Honesty about what this approach cannot do is part of making it legally useful. The method does not measure abstraction volume. A regulator who needs to quantify how many megalitres were taken cannot derive that figure from NDVI or ET estimates alone without additional hydrological assumptions. The satellite evidence establishes that irrigation occurred and approximately where, which is sufficient for initiating an inspection but not for calculating a penalty based on volume.
Small parcels below roughly 0.5 hectares are difficult to resolve cleanly in Sentinel-2 at 10 m, and individual smallholder plots may fall below the reliable detection threshold. Persistent cloud cover, common in tropical and subtropical regions during the wet season, can create gaps in the optical time series of several weeks. SAR partially compensates, but SAR alone cannot confirm crop greenness. Mixed land-cover parcels, where irrigated rows alternate with bare soil, will produce lower composite NDVI values and may be missed at coarser resolutions. Finally, the method depends entirely on the quality and completeness of the water-rights register it is compared against; if the register is itself incomplete or spatially imprecise, false positives increase.
Published precedent and the regulatory context
This is not a speculative method. Peer-reviewed studies published in Remote Sensing (MDPI) and related journals have applied NDVI and ET anomaly analysis to identify unlicensed irrigation in the Tagus basin in Spain, in Chilean river systems under the country's historically permissive water code, and across the Murray-Darling Basin in Australia, where the method has been discussed in the context of compliance monitoring by state water authorities. The scientific basis is settled; the operational gap is the institutional plumbing to move from anomaly map to enforcement action.
Satellize's crop-estimation work in the Kingdom of Tonga uses the same Sentinel-2 NDVI and phenological-signature pipeline that underpins this detection method, which gives the team direct operational familiarity with the edge cases: mixed cloud, small parcel sizes and the difficulty of separating irrigation signal from unusually high rainfall years. That experience informs how confidence thresholds are set before an anomaly is escalated to a regulator.
From anomaly map to inspection order
The practical output for a water authority is a ranked shortlist of parcels, each with a confidence score derived from the consistency of the NDVI anomaly across multiple seasons, the magnitude of the ET excess relative to regional baselines, and any SAR infrastructure signal. High-confidence cases, where all three signals align over two or more seasons, are the ones worth committing inspection resources to first.
Delivery as a GIS layer overlaid on cadastral boundaries lets field inspectors arrive at a site with coordinates already loaded. Archival depth is substantial: Sentinel-2 data runs back to 2015, Landsat to 1972 in lower resolution. That archive means a regulator can reconstruct when irrigation activity began, which is relevant when calculating the duration of an unlicensed abstraction for penalty purposes. The satellite record does not replace the inspector; it tells the inspector where to go.
Typical figures
| Optical spatial resolution | 10 m (Sentinel-2 VNIR), 30 m (Landsat 8/9 OLI/TIRS), 250–500 m (MODIS) |
| SAR spatial resolution | 10 m ground range detected, Sentinel-1 IW mode |
| Revisit cadence | 5 days (Sentinel-2 dual satellite); ~8 days (Landsat 8+9 combined); 6 days (Sentinel-1); daily (MODIS) |
| Evapotranspiration product resolution | 100 m dekadal (FAO WaPOR Level 2); 30 m dekadal (WaPOR Level 3, Africa/Near East); ~30 m via METRIC/SEBAL on Landsat TIRS elsewhere |
| Minimum detectable irrigated parcel | Approximately 0.5 ha at Sentinel-2 resolution; smaller parcels risk spectral mixing with adjacent bare soil |
| Spectral bands used | Red (B4), NIR (B8, B8A), SWIR (B11, B12) for Sentinel-2 NDVI/NDWI; Thermal IR Band 10 for Landsat ET; C-band VV/VH for Sentinel-1 SAR |
| Archive depth | Sentinel-2 from 2015; Landsat from 1972 (30 m from 1984); Sentinel-1 from 2014 |
| Cloud limitation | Optical methods require cloud-free compositing; multi-month gaps possible in humid tropics. SAR unaffected by cloud. |
| Latency (routine monitoring) | Sentinel-2 Level-2A products available within 3–5 hours of acquisition via Copernicus Data Space |
| Delivery formats | GeoTIFF anomaly raster, GeoPackage/Shapefile parcel shortlist, PDF inspection brief, optional WMS feed for GIS integration |
Analytics Satellize can run
| Seasonal NDVI anomaly map | Dense Sentinel-2 time-series compositing; NDVI computed from bands B4 and B8; anomaly scored against multi-year baseline for each parcel | GeoTIFF raster and parcel-level ranked shortlist, updated each growing season |
| Evapotranspiration excess layer | FAO WaPOR actual ET products (Africa/Near East) or METRIC/SEBAL applied to Landsat 8/9 TIRS; excess computed relative to regional dryland ET baseline | GeoTIFF ET anomaly layer at 30–100 m, overlaid on cadastral boundaries |
| Licence-boundary violation map | Spatial intersection of irrigated-area mask with water-rights register polygons; parcels outside licensed perimeter flagged | Shapefile of candidate violation parcels with confidence tier (high/medium/low) and season of first detection |
| New pump-infrastructure change detection | Sentinel-1 SAR backscatter change analysis (VV and VH polarisation) between baseline and monitoring stacks; metallic infrastructure identified by localised backscatter increase | Point or polygon layer of detected infrastructure changes with acquisition dates, delivered as GeoPackage |
| Multi-season irrigation onset timeline | Phenological analysis of Sentinel-2 NDVI time series using harmonic regression or breakpoint detection to identify season in which irrigation activity first appeared | Per-parcel table of estimated irrigation onset date, supporting duration-of-violation calculation for regulatory use |
| Basin-scale screening report | MODIS 250 m NDVI and ET anomaly screening to prioritise sub-basins for higher-resolution Sentinel-2 and Landsat analysis, reducing processing area and cost | PDF basin screening report with priority sub-basin map and recommended tasking areas |
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.