Actual evapotranspiration and crop water use mapping
Surface energy balance models applied to satellite thermal imagery can estimate actual crop water use at field scale, exposing irrigation inefficiency invisible to ground sensors. The method is powerful and genuinely difficult to get right.
Sensors
- Landsat 8/9 TIRS: Thermal Infrared Sensor bands 10 and 11 at 100 m native resolution (resampled to 30 m in products). The workhorse for field-scale energy balance: fine enough to resolve individual irrigated parcels. Revisit 16 days per satellite; combined Landsat 8+9 constellation gives roughly 8-day repeat at the equator, less at higher latitudes under clear skies.
- Sentinel-3 SLSTR: Sea and Land Surface Temperature Radiometer delivers LST at 1 km resolution with a ~1-2 day revisit. Too coarse to resolve individual fields in fragmented landscapes, but valuable for regional water-use accounting and for filling temporal gaps between Landsat overpasses.
- MODIS Terra/Aqua: Thermal bands at 1 km, twice-daily overpass. Longest continuous archive (2000 to present) makes it the standard for multi-year seasonal water-use trend analysis at basin or district scale. Field-scale application is not credible below roughly 50 ha.
- ECMWF ERA5 reanalysis: Hourly gridded meteorological fields at ~31 km: air temperature, wind speed, humidity, solar radiation. Used to drive the reference evapotranspiration (ET0) calculation and to supply boundary conditions for energy balance models on days without satellite overpass.
What a thermal band actually measures, and why it matters
Land surface temperature (LST) is the physical quantity at the heart of every satellite-based evapotranspiration estimate. A wet, actively transpiring canopy stays cool because latent heat flux carries energy away as water vapour. A dry or stressed field cannot sustain that cooling, so its surface temperature rises. On a clear summer afternoon, the LST difference between a well-irrigated maize field and an adjacent fallow plot can exceed 15 °C. That contrast is the signal.
Surface energy balance models such as SEBAL (Surface Energy Balance Algorithm for Land) and METRIC (Mapping EvapoTranspiration at high Resolution with Internalised Calibration) partition the net radiation at the surface into latent heat (ET), sensible heat (warming the air), soil heat flux, and a small storage term. Latent heat flux is not measured directly. It is the residual after the other terms are estimated from satellite-derived albedo, NDVI, and LST combined with meteorological inputs. That residual structure is both the method's elegance and its principal vulnerability.
From instantaneous flux to seasonal water budget
A satellite overpass gives you one snapshot, typically around 10:00 to 10:30 local time for Landsat. Converting that instantaneous latent heat flux to a daily ET total relies on the evaporative fraction, the ratio of latent to available energy, which empirical studies have shown to be relatively stable through the day under clear-sky conditions. Daily ET is then extrapolated using reference ET (ET0) calculated from ERA5 or from ground station data. Seasonal totals are built by interpolating between clear-sky overpass dates, with ET0-weighted gap-filling on cloudy days.
At Landsat resolution this produces per-field estimates of seasonal actual ET, typically expressed in millimetres. Published validation studies against eddy-covariance flux towers report root-mean-square errors in the range of 15 to 30 mm per month under good conditions, though errors can be larger in heterogeneous or windy landscapes. That accuracy is sufficient to rank fields by irrigation efficiency, identify chronic over-irrigation, and support water-rights auditing, but it is not precise enough to drive day-to-day irrigation scheduling on its own.
Where the method struggles, honestly stated
Cloud cover is the most obvious constraint. Landsat's 8-day revisit under clear skies becomes far sparser in humid or monsoon climates: in parts of South and South-East Asia, entire growing seasons can yield only two or three usable Landsat thermal scenes. Gap-filling with MODIS or Sentinel-3 introduces a resolution mismatch that inflates uncertainty in fragmented field systems.
Atmospheric correction of thermal data is a second, less-discussed problem. TIRS band 10 on Landsat 8 suffered a stray-light artefact that biased LST retrievals by up to 2 °C in some scenes before a correction was published in 2017. Even with current corrections, residual errors in atmospheric water vapour estimation propagate directly into LST and then into the energy balance. A 1 °C LST error can shift the sensible heat flux estimate by 20 to 50 W m², which translates to a meaningful ET bias over a season.
The 'hot pixel' and 'cold pixel' anchor points used in METRIC to internally calibrate the sensible heat flux require careful selection. Automated selection algorithms exist, but they can fail in landscapes where truly bare dry soil or full-cover well-watered crops are absent within the scene. Results should always be cross-checked against ET0 and against any available ground truth before being used for water-rights or payment decisions.
Combining sensors to patch the temporal gap
The practical answer to Landsat's revisit limitation is data fusion. The STARFM (Spatial and Temporal Adaptive Reflectance Fusion Model) family of algorithms blends Landsat's spatial detail with MODIS's daily frequency to generate synthetic daily imagery at 30 m. Applied to ET estimation, this approach can produce plausible daily field-scale ET surfaces through short cloudy periods, though the synthetic pixels carry the uncertainty of both source datasets and should not be treated as equivalent to a real Landsat acquisition.
Sentinel-3 SLSTR adds independent LST observations at 1 km, useful for validating basin-scale totals and for detecting gross anomalies. The combination of Landsat thermal for spatial precision and Sentinel-3 for temporal density is now a standard architecture in operational irrigation district monitoring programmes, including work reported under the FAO's WaPOR (Water Productivity through Open-access of Remotely sensed derived data) initiative.
Operational uses: who actually buys this analysis
Irrigation district managers use seasonal ET maps to compare actual water use against water allocations, identify fields drawing more than their licensed entitlement, and prioritise on-farm audits. The analysis is objective in a way that meter readings are not: it covers every field in the district simultaneously, not just those with functioning flow meters.
National water ministries in water-stressed countries use basin-scale ET accounting, typically at MODIS or Sentinel-3 resolution, to close the water balance and verify reported agricultural abstractions against satellite-derived consumption. The gap between reported and satellite-estimated use is sometimes striking.
Agricultural development banks and food security agencies use multi-year ET time series to assess whether irrigation infrastructure is being used productively, and to detect gradual degradation in irrigated area or crop performance. Satellize runs satellite-derived crop analytics for clients including the Kingdom of Tonga crop-estimation programme; the same energy balance infrastructure underpins water-use work in comparable smallholder and peri-urban irrigation contexts.
What good output looks like, and what to do with it
A well-executed seasonal ET product for an irrigation district delivers a raster at 30 m resolution, a per-field summary table (field ID, area, total ET in mm, implied water volume in cubic metres, ratio to district median), and a ranked list of statistical outliers. The outlier list is where the operational value sits: a field consuming 900 mm of ET against a district median of 550 mm is either growing a high-value perennial crop, has a broken infrastructure leak, or is being illegally over-irrigated. The satellite cannot distinguish between those explanations, but it identifies which fields deserve a site visit.
Delivery formats depend on the client's GIS environment. Standard outputs are GeoTIFF seasonal ET rasters, field-boundary-clipped vector summaries in GeoPackage or Shapefile, and a PDF technical report documenting the scene selection, atmospheric correction approach, anchor pixel quality, and validation statistics. Clients with existing SCADA or irrigation management systems can receive field-summary data as CSV or via a simple API feed keyed to field identifiers.
Typical figures
| Spatial resolution (thermal) | 100 m native / 30 m resampled (Landsat TIRS); 1 km (Sentinel-3 SLSTR, MODIS) |
| Revisit frequency | ~8 days clear-sky (Landsat 8+9 combined); 1-2 days (Sentinel-3); daily (MODIS) |
| Minimum field size reliably resolved | ~1 ha at Landsat resolution; ~50 ha at MODIS/Sentinel-3 resolution |
| Thermal bands used | Landsat TIRS Band 10 (10.6-11.2 µm), Band 11 (11.5-12.5 µm); SLSTR S8 (10.85 µm), S9 (12.0 µm) |
| Meteorological forcing | ECMWF ERA5 hourly reanalysis at ~31 km, or in-situ station data where available |
| Typical monthly ET accuracy (vs flux tower) | RMSE 15-30 mm per month under clear-sky conditions; higher in humid or heterogeneous landscapes |
| Satellite archive depth | Landsat from 1982 (TM); MODIS from 2000; Sentinel-3 from 2016 |
| Cloud cover constraint | Energy balance requires cloud-free thermal scenes; cloudy-day ET gap-filled using ET0 scaling or data fusion |
| Delivery formats | GeoTIFF seasonal ET rasters, GeoPackage/Shapefile field summaries, CSV tabular outputs, PDF technical report |
Analytics Satellize can run
| Seasonal actual ET map at field scale | SEBAL or METRIC surface energy balance applied to Landsat 8/9 TIRS + OLI, with ERA5 meteorological forcing | 30 m GeoTIFF raster of seasonal ET (mm) clipped to area of interest, with per-field summary table |
| Basin-scale ET accounting | MODIS or Sentinel-3 LST-based energy balance, aggregated to sub-basin polygons | Monthly ET volume estimates (million m³) per sub-basin, delivered as GeoPackage and PDF report |
| Irrigation efficiency ranking | Per-field ET normalised by crop coefficient (Kc) and growing degree days to isolate management signal from crop-type variation | Ranked field list with efficiency score, flagging statistical outliers for ground audit |
| Multi-year ET trend analysis | Time-series regression on annual seasonal ET totals derived from Landsat archive (2013 to present for TIRS) | Trend map (mm per year) and district-level summary chart, PDF report with interpretation |
| Synthetic daily ET surface (gap-filled) | STARFM-class Landsat-MODIS fusion to generate daily 30 m ET estimates through short cloudy periods | Daily GeoTIFF stack for the growing season, with data-quality flag layer distinguishing real from synthetic acquisitions |
| Water-use anomaly alert | Comparison of current-season ET accumulation against 5-year median for each field polygon; threshold-based flagging | Seasonal alert report identifying fields with ET more than 25% above district median, updated at each clear-sky overpass |
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.