Daily Actual Evapotranspiration Estimation in a Mediterranean Ecosystem from Landsat Observations Using SEBAL Approach

Quantifying actual evapotranspiration (ETa) over natural vegetation is crucial in evaluating
the water status of ecosystems and the water-use patterns in local or regional hydrological basins.
Remote sensing-based surface energy balance models have been used extensively for estimating
ETa in agro-environments; however, the application of these models to natural ecosystems is
still limited. The surface energy balance algorithm for land (SEBAL) physical-based surface energy
balance model was applied to estimate the actual evapotranspiration over a heterogeneous coverage
of Mediterranean maquis in a natural reserve in Sardinia, Italy. The model was applied on 19 Landsat
5 and 8 images from 2009 to 2014, and the results were compared to the data of a micrometeorological
station with eddy covariance flux measurements. Comparing the SEBAL-based evaporative
fraction (ΛS) to the corresponding tower-derived evaporative fractions (ΛT) showed good flux estimations
in the Landsat overpass time (Coefficient of determination R2 = 0.77, root mean square
error RMSE = 0.05 and mean absolute error MAE = 0.076). Three methods were evaluated for
upscaling instantaneous latent heat flux (λE) to daily actual evapotranspiration (ETa,D). The upscaling
methods use the evaporative fraction (Λ), the reference evapotranspiration fraction (EFr) and
the ratio of daily to instantaneous incoming shortwave radiation (Rs24/Rsi) as upscaling factors under
the hypothesis of diurnal self-preservation. A preliminary analysis performed using only insitu
measured data demonstrated that the three factors were relatively self-preserved during the
daytime, and can yield good ETa,D estimations, particularly when obtained at near the Landsat scene
acquisition time (≈10:00 UTC). The upscaling factors obtained from SEBAL retrieved instantaneous
fluxes, and some ancillary measured meteorological data were used to upscale SEBAL-estimated
instantaneous actual λ to daily ET. The Λ EFr and Rs24/Rsi methods on average overestimated the
measured ETa,D by nearly 20, 61 and 18%, respectively. The performance of the Λ and Rs24/Rsi methods
was considered satisfactory, bearing in mind the high variable ground cover and the inherent
variability of the biome composition, which cannot be properly represented in the Landsat moderate
spatial resolution. In this study, we tested the potential of the SEBAL model application in a
complex natural ecosystem. This modeling approach will be used to represent the spatial dynamics
of ET, which will be integrated into further environmental and hydrological applications.