Crop water productivity benchmarking across irrigation schemes
Crop water productivity (yield per cubic metre of evapotranspiration) can be computed at field scale by combining satellite-derived ET from surface energy balance models with vegetation-index yield proxies. The method works across farms, districts and borders, but ET uncertainty of 10–20% propagates directly into the ratio.
Sensors
- ECOSTRESS (ISS-mounted): Thermal infrared at 70 m spatial resolution, five TIR bands centred near 8–12 µm. Revisit is irregular, roughly 1–5 days depending on ISS orbit precession, which gives opportunistic high-resolution land surface temperature for surface energy balance at field scale.
- Landsat 8/9 TIRS + OLI: Thermal bands at 100 m (resampled to 30 m product) with a 16-day revisit per satellite; two satellites together give 8-day repeat. OLI reflectance bands supply the albedo and vegetation index inputs that SEBAL and METRIC require alongside surface temperature.
- Sentinel-2 MSI: 10 m visible and near-infrared, 20 m red-edge and shortwave infrared, 5-day revisit at mid-latitudes with both satellites. No thermal band, so it contributes reflectance-based inputs (NDVI, LAI, albedo) rather than temperature directly. Cloud cover remains the main operational constraint.
- MODIS MOD16 ET product: Global daily ET at 500 m, derived from the Penman-Monteith equation driven by MODIS land cover and meteorology. Useful for scheme-level or national benchmarking where field-scale resolution is not required, and for gap-filling cloudy Landsat periods.
What crop water productivity actually measures, and why it is hard
Crop water productivity (CWP) is yield divided by the volume of water consumed through evapotranspiration (ET), expressed in kilograms per cubic metre. It is not the same as irrigation efficiency, which counts conveyance losses. CWP counts only the water that actually left the soil and canopy as vapour, and asks how much grain or biomass that vapour purchase bought.
The difficulty is that ET is invisible. You cannot meter it from a satellite the way you can read a flow gauge. What satellites measure is land surface temperature, reflectance and, indirectly, the energy fluxes that drive evaporation. Surface energy balance models convert those observations into latent heat flux, which is ET expressed in energy units. The conversion is physically grounded but introduces uncertainty at every step: emissivity assumptions, aerodynamic resistance parameterisation, the choice of hot and cold anchor pixels in SEBAL or METRIC. Published inter-comparison studies put typical ET uncertainty at 10–20% at field scale, and that error travels unchanged into the CWP ratio. Buyers should know this before commissioning field-level rankings.
The physical chain from photons to productivity ratios
The computation runs in three linked stages. First, surface energy balance. A model such as SEBAL (Surface Energy Balance Algorithm for Land), METRIC (Mapping EvapoTranspiration at high Resolution with Internalized Calibration), or the simpler SSEBop (Operational Simplified Surface Energy Balance) ingests land surface temperature from Landsat TIRS or ECOSTRESS alongside shortwave albedo and vegetation indices from OLI or Sentinel-2 MSI. It solves the energy balance residual: net radiation minus soil heat flux minus sensible heat flux equals latent heat flux, which is actual ET. ECOSTRESS at 70 m is particularly useful for resolving within-field gradients that 100 m Landsat TIRS blurs.
Second, yield proxies. Direct yield data from satellites does not exist at field scale. The practical substitute is a vegetation index integral: the area under a seasonal NDVI or NDWI curve correlates with above-ground biomass accumulation, and calibrated against ground-truth harvest data it becomes a relative yield proxy. Sentinel-2 at 10 m and 5-day revisit is the current workhorse for this, provided cloud cover does not destroy the seasonal profile. In persistently cloudy climates, Sentinel-1 SAR backscatter can partially substitute.
Third, the ratio. Seasonal ET (in cubic metres per hectare) is divided into the yield proxy (in kilograms per hectare) to produce CWP in kg/m³. The ratio can then be mapped at field scale, aggregated to irrigation district level, and compared across schemes or across years. The comparison is most defensible when the same sensor, model and calibration approach is applied consistently, because absolute ET errors partially cancel in relative rankings.
What the numbers look like in practice
Published benchmarks from peer-reviewed literature give a sense of expected ranges. Wheat in well-managed irrigated systems in South Asia and the Middle East typically achieves CWP of 0.8–1.2 kg/m³. Maize under centre-pivot irrigation in semi-arid regions reaches 1.0–1.8 kg/m³. Cotton, a notoriously thirsty crop, often falls below 0.5 kg/m³. Rice paddy, because of percolation and open-water evaporation, frequently sits at 0.3–0.6 kg/m³ even where yields are reasonable. These figures come from ground-based lysimeter and flux-tower studies; satellite-derived estimates track the same order of magnitude but carry the 10–20% ET uncertainty on top.
The value of satellite benchmarking is not the absolute number for any single field. It is the spatial pattern across hundreds or thousands of fields computed consistently. A field at the tail-end of a canal that shows 30% lower CWP than its neighbours is a signal worth investigating, regardless of whether the absolute ET estimate is off by 15%.
Where the method breaks down
Cloud cover is the primary operational obstacle. SEBAL and METRIC require cloud-free thermal imagery on days representative of the growing season. In humid tropics or monsoon-dominated systems, a full season can pass with fewer than four usable Landsat overpasses. MODIS MOD16 at 500 m fills temporal gaps but loses field-scale resolution entirely. ECOSTRESS helps because its irregular ISS-driven revisit sometimes catches clear windows that Landsat misses on its fixed 8-day cycle, though it cannot be relied upon for systematic coverage.
Mixed pixels are a persistent problem at 100 m Landsat thermal resolution. A field boundary running through a pixel contaminates the temperature retrieval. ECOSTRESS at 70 m improves this marginally; true 10–30 m thermal from a dedicated satellite remains an unmet need in the open-data ecosystem.
The anchor-pixel calibration in SEBAL and METRIC requires identifying a reliably dry bare-soil pixel and a fully evapotranspiring wet pixel within the scene. In landscapes where everything is irrigated, or where bare soil is rare, the calibration degrades. SSEBop avoids this by using a pre-defined cold boundary condition from climatology, which makes it more automatable but less locally tuned.
Finally, yield proxies from NDVI integrals conflate biomass with harvestable yield. A crop that lodges late in the season may show high seasonal NDVI but low grain yield. Ground-truth calibration at the scheme level is essential if CWP is to be used for anything beyond relative ranking.
From scheme benchmarking to water-policy decisions
The practical use cases for CWP benchmarking fall into three broad categories, though each requires a different level of analytical rigour. Irrigation district management is the most straightforward: a water authority wants to know which farmers or zones are converting water into yield most efficiently, so it can target extension services or adjust water allocations. Here, relative ranking across a consistent dataset is sufficient, and the 10–20% ET uncertainty is acceptable.
Cross-border or national benchmarking, say comparing productivity across irrigation schemes in the Nile basin or the Indus system, is more demanding. Differences in crop mix, climate, soil and management all confound the comparison. The satellite data must be accompanied by crop-type maps and meteorological correction before the numbers mean anything politically defensible.
The third use case is temporal trend analysis: is a given scheme improving or degrading its water productivity over a decade? Landsat's archive back to 1984 (with consistent thermal calibration from Landsat 5 TM onward, though TIRS quality improved markedly with Landsat 8 in 2013) makes long-run trend detection feasible. Satellize runs this kind of multi-year ET and productivity analysis on open constellations, including Sentinel and Landsat, and has applied similar vegetation-index productivity methods in the Kingdom of Tonga crop-estimation programme.
Building a defensible benchmarking dataset
A credible CWP benchmarking exercise needs four things to hold up to scrutiny. A consistent ET model applied across the full domain, not a patchwork of methods. A cloud-gap-filling strategy, whether MODIS MOD16 interpolation, SAR-assisted phenology reconstruction, or simple temporal interpolation between clear-sky images. A yield proxy calibrated against at least a sample of ground-truth harvest records. And a stated uncertainty budget that travels with every output map.
The last point is the one most often omitted. An irrigation authority that commissions a CWP map and receives it without uncertainty bounds has no way to distinguish a real 20% productivity gap between two districts from a modelling artefact. Honest uncertainty communication is not a caveat; it is what makes the analysis actionable.
Typical figures
| Thermal spatial resolution | 70 m (ECOSTRESS); 100 m native / 30 m resampled product (Landsat 8/9 TIRS); 1 km (MODIS) |
| Reflectance spatial resolution | 10–20 m (Sentinel-2 MSI); 30 m (Landsat OLI) |
| Revisit cadence | ~8 days combined (Landsat 8+9); 5 days at mid-latitudes (Sentinel-2 A+B); 1–5 days irregular (ECOSTRESS/ISS); daily (MODIS) |
| ET model output uncertainty | 10–20% at field scale (published range for SEBAL, METRIC, SSEBop) |
| Minimum field size for reliable ET retrieval | Approximately 1 ha with ECOSTRESS or Landsat; 25 ha practical minimum with MODIS MOD16 |
| Spectral bands used | Thermal infrared (8–12 µm for LST); visible, NIR, SWIR for albedo and vegetation indices |
| Archive depth | Landsat thermal from 1984 (consistent TIRS quality from 2013); Sentinel-2 from 2015; ECOSTRESS from 2018 |
| Cloud sensitivity | High: SEBAL/METRIC require cloud-free thermal scenes; MODIS MOD16 provides gap-fill at coarser resolution |
| CWP output units | kg of yield equivalent per m³ of ET, or relative index normalised within scheme |
| Delivery formats | GeoTIFF per season or per dekad; vector aggregates by field or district; tabular CSV for time-series comparison |
Analytics Satellize can run
| Seasonal actual ET map | SEBAL or METRIC surface energy balance driven by Landsat TIRS + OLI; ECOSTRESS for higher-resolution spot checks | GeoTIFF raster (30 m or 70 m) of cumulative seasonal ET in mm, with per-pixel uncertainty band |
| Yield proxy map from NDVI integral | Seasonal NDVI time series from Sentinel-2 MSI; trapezoidal integration over growing season; calibrated against available ground-truth harvest records | GeoTIFF of relative biomass accumulation index, per field or per pixel |
| Field-scale CWP ratio map | Pixel-wise division of yield proxy by seasonal ET; aggregated to field boundaries using cadastral or auto-delineated field polygons | GIS layer (GeoPackage or Shapefile) with CWP value and uncertainty flag per field polygon |
| Scheme-level benchmarking report | Statistical aggregation of field-scale CWP across irrigation district boundaries; percentile ranking; year-on-year trend if archive depth permits | PDF report with ranked district table, spatial maps and stated uncertainty budget; CSV of underlying statistics |
| Multi-year productivity trend analysis | Consistent METRIC or SSEBop applied to Landsat 8/9 archive (2013 onward); linear trend fitted to annual CWP per district | Time-series chart and GIS layer showing trend direction and significance per zone |
| Cloud-gap-filled ET time series | MODIS MOD16 500 m product used to interpolate between cloud-free Landsat scenes; temporal disaggregation using STARFM or similar data-fusion approach | Dekadal ET raster stack covering full growing season, flagged by data source per pixel |
| Low-productivity alert layer | Threshold detection: fields below the 20th percentile of scheme CWP for two consecutive seasons flagged for field investigation | Spatial alert layer updated each season; optional tabular export for extension service targeting |
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.