Basin-scale evapotranspiration and water-balance closure
Actual evapotranspiration is the largest unmeasured flux in most catchments. Combining thermal infrared land-surface temperature, vegetation indices, GPM rainfall and GRACE-FO gravity change lets analysts close the water balance and expose unaccounted abstractions.
Sensors
- Landsat 8/9 TIRS: Thermal Infrared Sensor bands 10 and 11 (10.6–12.5 µm) provide land-surface temperature at 100 m native resolution (resampled to 30 m in products). Revisit is 8 days combined for Landsat 8 and 9. Clear-sky only; cloud cover renders scenes unusable.
- MODIS MOD11 LST: Terra and Aqua MODIS deliver daytime and night-time LST at 1 km resolution with daily to twice-daily revisit globally. Lower spatial detail than TIRS but far more frequent, making it the backbone of operational global ET products including SSEBop and GLEAM inputs.
- GPM IMERG: Global Precipitation Measurement IMERG provides half-hourly precipitation at 0.1° (~10 km) globally between 60°N and 60°S, with a Final Run latency of roughly 3.5 months and an Early Run available within hours. The primary precipitation input for satellite water-balance closure.
- GRACE-FO (GRACE Follow-On): Gravity Recovery and Climate Experiment Follow-On measures changes in terrestrial water storage at basin scales, with a spatial resolution floor of roughly 300–400 km. Monthly solutions allow detection of storage anomalies of a few centimetres of equivalent water height. The residual term in water-balance closure.
- Sentinel-2 MSI: 10 m multispectral bands deliver NDVI and fractional vegetation cover at high resolution, used to disaggregate coarser LST-based ET estimates and to parameterise the vegetation resistance term in Penman-Monteith frameworks. 5-day revisit at the equator.
Why evapotranspiration is the hardest flux to measure
In most river basins, evapotranspiration returns between 60 and 70 percent of annual precipitation to the atmosphere. Yet it is the one major flux that cannot be metered at the catchment outlet. Streamflow can be gauged. Precipitation can be raingauged or estimated from radar. Storage change in soil and groundwater is slow enough to approximate. ET, by contrast, happens everywhere simultaneously, varies with crop type, soil moisture, wind, humidity and radiation, and leaves no physical residue to count.
This invisibility has practical consequences. Water-resource models that estimate ET poorly will misattribute abstractions, misforecast low flows, and produce water accounts that do not close. In basins where irrigation withdrawals are politically contested or legally regulated, a credible independent ET estimate is not a technical luxury.
From skin temperature to latent heat: the energy-balance chain
The dominant satellite ET methods rest on the surface energy balance: net radiation minus soil heat flux equals sensible heat plus latent heat. Latent heat flux is ET expressed in energy units. The key satellite observable is land-surface temperature (LST), which constrains sensible heat flux through the temperature gradient between surface and air. Once sensible heat is estimated, latent heat is the residual.
SEBAL, METRIC and SSEBop are the three most widely applied implementations of this logic. SSEBop, developed by the USGS, simplifies the approach by using pre-defined hot and cold boundary conditions derived from climatological LST and air temperature, which makes it operationally scalable across large domains. GLEAM takes a different route, using a Priestley-Taylor formulation driven by microwave-derived soil moisture and vegetation optical depth rather than LST, which gives it an advantage under cloud cover but removes the spatial sharpness that thermal data provides.
Penman-Monteith methods, the FAO standard for reference ET, require surface resistance terms that are not directly observable from space. In practice, satellite-derived leaf area index and fractional cover from MODIS or Sentinel-2 are used to parameterise canopy resistance, and reanalysis products such as ERA5 supply the meteorological forcing. The result is a hybrid: physically grounded but dependent on the quality of the reanalysis winds and humidity, which can be poor in complex terrain.
Closing the water balance: what the residual reveals
The water-balance identity is simple: precipitation minus ET minus runoff equals change in storage. Satellite observations can now supply estimates of all four terms independently. GPM IMERG provides precipitation. Streamflow gauges, or satellite-derived discharge estimates, supply runoff. GRACE-FO measures the storage change. ET from SSEBop or GLEAM completes the set.
When the four terms do not sum to zero within their combined uncertainty bounds, something is wrong. The discrepancy might reflect gauge undercatch in snowfall, IMERG bias over complex topography, GRACE signal leakage across basin boundaries, or, most interestingly, unaccounted abstractions. Groundwater pumping that does not appear in official statistics shows up as a storage decline that is larger than the modelled ET and runoff can explain. This is one of the few remote-sensing methods that can flag illegal or unregistered extraction at basin scale without requiring field access.
The honest caveat is that the uncertainty budget is large. GRACE-FO storage anomalies carry errors of roughly 1–2 cm equivalent water height per month at the 400 km scale. IMERG biases in arid and mountainous regions can exceed 30 percent. ET products disagree with each other by 10–30 percent in many climates. Closure analysis is most useful as a consistency check and an anomaly detector, not as a precise audit instrument.
Where energy-balance ET goes wrong: the advection problem
In arid and semi-arid irrigated landscapes, energy-balance ET methods carry a well-documented positive bias. The cause is advection: dry, hot air from surrounding unirrigated land blows across irrigated fields, adding sensible heat that drives ET above what local net radiation alone would support. The energy-balance residual approach interprets this additional energy as latent heat, producing ET estimates that can exceed potential ET calculated from meteorological data alone. This is physically possible over a well-watered surface in an advective environment, but the magnitude is frequently overestimated.
The practical implication is that SSEBop and similar products should be used with caution in dryland irrigation schemes, particularly at field scale. Aggregating to sub-basin or basin scale reduces the bias somewhat because the advective energy is redistributed across the full domain. Cross-checking against Penman-Monteith reference ET multiplied by a crop coefficient provides a useful sanity bound.
GLEAM and SSEBop: what the public products actually deliver
GLEAM (Global Land Evaporation Amsterdam Model) produces daily actual ET, transpiration, bare-soil evaporation, interception loss and soil moisture at 0.25° resolution globally, with a record extending back to 1980. It is freely available and updated with a short lag. Its strength is temporal depth and physical decomposition of ET components. Its weakness is coarse spatial resolution, which makes it unsuitable for field-level or small-catchment analysis.
SSEBop, maintained by the USGS, produces monthly actual ET at 1 km resolution globally, driven by MODIS LST and a simplified energy-balance framework. The USGS publishes it through the Earth Resources Observation and Science Center. For basins larger than a few hundred square kilometres, SSEBop provides a defensible operational estimate with a documented validation record across diverse climates.
Neither product is a black box that can be applied without local calibration checks. Both should be validated against flux-tower measurements where available, and against independent water-balance estimates from gauged sub-catchments. Satellize applies this cross-validation step as standard when running basin water accounts, drawing on the same open-constellation inputs that underpin our crop-estimation work in Tonga.
Resolution limits and the disaggregation trade-off
The fundamental tension in satellite ET is between temporal frequency and spatial resolution. MODIS delivers daily LST at 1 km. Landsat TIRS delivers 100 m LST every 8 days under clear skies. In a humid climate with frequent cloud cover, a Landsat-based ET product may produce only four or five valid scenes per growing season, leaving large temporal gaps that must be filled by interpolation or fusion with MODIS.
The STARFM and ESTARFM spatial-temporal fusion algorithms combine Landsat and MODIS to produce synthetic daily 30 m LST estimates. These are useful but introduce their own errors, particularly when land cover changes between the Landsat anchor dates. For basins where high-resolution ET is operationally important, such as irrigation districts with metered allocation, the honest answer is that current open-constellation revisit is a binding constraint. Commercial thermal satellites with higher revisit, or dense flux-tower networks, are the only routes around it.
Typical figures
| LST spatial resolution (Landsat TIRS) | 100 m native, 30 m resampled |
| LST spatial resolution (MODIS MOD11) | 1 km |
| Landsat 8+9 combined revisit | 8 days (clear sky required) |
| MODIS LST revisit | Daily to twice-daily globally |
| GPM IMERG resolution | 0.1° (~10 km), half-hourly; Final Run latency ~3.5 months |
| GRACE-FO spatial resolution floor | ~300–400 km equivalent; monthly solutions |
| GLEAM ET product resolution | 0.25°, daily, record from 1980 |
| SSEBop ET product resolution | 1 km, monthly |
| Typical ET product uncertainty | 10–30% depending on climate zone and method |
| GRACE-FO storage anomaly error | ~1–2 cm equivalent water height per month at basin scale |
Analytics Satellize can run
| Monthly actual ET map | SSEBop energy-balance framework applied to MODIS MOD11 LST with ERA5 temperature boundary conditions | GeoTIFF raster layer at 1 km, delivered monthly with uncertainty band |
| High-resolution growing-season ET | METRIC/SEBAL applied to Landsat 8/9 TIRS scenes, gap-filled by STARFM fusion with MODIS | 30 m GeoTIFF per available clear-sky scene, aggregated to seasonal totals |
| Basin water-balance closure report | Four-term balance: GPM IMERG precipitation, satellite ET, gauged or estimated runoff, GRACE-FO storage change | Monthly PDF report with closure residual, uncertainty decomposition and anomaly flags |
| Unaccounted abstraction signal | Statistical comparison of GRACE-FO storage trend against modelled ET plus gauged outflow; persistent negative residuals flagged | Time-series chart and alert when residual exceeds two standard deviations of historical variability |
| ET component decomposition | GLEAM outputs disaggregated into transpiration, bare-soil evaporation and interception loss using Sentinel-2 fractional cover | Tabular CSV by sub-catchment and land-cover class, quarterly |
| Penman-Monteith reference ET surface | FAO-56 PM equation parameterised with ERA5 reanalysis meteorology and Sentinel-2 LAI for canopy resistance | Weekly 0.1° raster for basin domain, used as upper-bound sanity check on energy-balance ET |
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.