Accurate and spatially distributed estimates of evapotranspiration (ET) are increasingly important with the growing global
population and economy putting strain on fresh water resources and food supplies. The utility of ET maps has been
demonstrated in a variety of applications, ranging from water rights management, through drought and food shortage
monitoring, to more efficient use of land and water in agriculture and crop stress assessment. Those applications are
directly relevant for reaching at least two of the UN Sustainable Development Goals (SDGs): goal 2 – zero hunger,
and goal 6 – clean water and sanitation; and could also prove useful in trying to achieve other SDGs (e.g. goal 15 – life on land).

The most efficient methods for deriving such ET maps at local (field), regional and global scales make use of remote
sensing observations, particularly from satellite-based sensors since they provide a synoptic view at regular time
intervals. Majority of remote sensing models that estimate actual (as opposed to potential) ET do so through the
estimation of land surface energy fluxes. This requires both knowledge of the state of vegetation (e.g. leaf area
index or fraction of green vegetation) and of land surface temperature (LST) which acts as one of the model boundaries.
Therefore, remote sensing observations using both visible/near-infrared and thermal infra-red sensors serve as an
important source of input data for such models.

There are existing ET datasets utilising remote sensing observations but none of them fully satisfies the need for an
accurate, operational, and field-scale flux estimates. One of the widely available datasets is the MODIS Global
Evapotranspiration Project (MOD16) which while being suitable for regional and/or scientific applications (with
an 8-day temporal and 1-km spatial resolution) is not fulfilling the need of field-scale operational applications. Another
ET dataset is provided by USDA-ARS using a range of satellite sensors ranging from geostationary GOES with a spatial
resolution of 5 km, through MODIS at 1 km to Landsat at 60 m. However, this dataset is currently only produced
over the continental United States. In addition, the two mentioned datasets rely on the MODIS sensors on board Terra
and Aqua satellites which are already working past their designed lifetime. Another freely available ET dataset is provided
by LSA SAF and uses geostationary observations (without using LST) from Meteosat Second Generation (MSG) to derive
30 minute and daily ET observations over the whole MSG disc. However, the spatial resolution of this product (3-
6 km) is also insufficient for field-scale applications. Finally, there are various advanced modelling approaches which
indirectly incorporate remote sensing observations, such as ECMWF or GLDAS models, but the spatial
resolution of their outputs is even coarser.

The recent launches of the Sentinel-2A , Sentinel-2B and Sentinel-3A satellites (and upcoming launch of Sentinel-3B)
present new opportunities for accurate and operational modelling of actual ET with the use of remote sensing data. The
Sentinel satellites were designed with the aim of serving operational needs of diverse user communities and fields of
application. In the field of agriculture this aim is already partially realized with ESA projects such as “Sentinel-2 for
Agriculture” providing a system which utilises high spatial and temporal resolution optical data from Sentinel-2 satellites
(and Landsat-8) for estimating growing area extent for main crop types as well as crop status from temporal evolution of
Leaf Area Index (LAI) over a growing season. Time-series of Sentinel-1 Synthetic Aperture Radar observations
have also been used to operationally map rice crop evolution or to discriminate between different crop types.
Sentinel-3 satellite has added the capacity of acquiring thermal observations to the existing suite of Sentinels’ capabilities.
This should allow for operational modelling of ET and of other land surface energy fluxes. However, for many agricultural
applications the resolution of the modelled fluxes should be smaller than the typical field size. In the European Union
agricultural context, where the Agricultural Census 2010 reported that an average farm size was 14.2 hectares and that 6
million farms were smaller than 2 hectares, this requires model outputs with a pixel size on the order of tens of meters. This is not possible when using Sentinel-3 alone (Sentinel-3 thermal sensor SLSTR has a spatial resolution of
1000 m) but might be achievable if synergies between Sentinel-3 and Sentinel-2 are exploited. In particular the Multi
Spectral Instrument (MSI) on board Sentinel-2, with its high spatial resolution (up to 10 m) and red-edge spectral bands,
could allow for disaggregation of ET model inputs and/or outputs to higher spatial resolution. By utilising the synergies
between the sensors of those two satellites it might be possible to derive accurate flux estimates at high spatial resolution.