Evapotranspiration estimation using Surface Energy Balance Model and medium resolution satellite data: An operational...

Monitoring spatial and temporal trends of water use is of utmost importance to ensure water and food security in river basins that are challenged by water scarcity and climate change induced abnormal weather patterns. To quantify water consumption by the agriculture sector, continuous monitoring is required over different spatial scales ranging from field (< 1 ha) to basin. The demand driven requirement of covering large areas yet providing spatially distributed information makes the use of in-situ measurement devices unfeasible. Earth observation satellites and remote sensing techniques offer an effective alternative in estimating the consumptive use of water (Actual EvapoTranspiration (ET_a) fluxes) by using periodic observations from the visible and infrared spectral region. Optical satellite data, however, is often hindered by noises due to cloud cover, cloud shadow, aerosols and other satellite related issues such as Scan Line Corrector (SLC) failure in Landsat 7 breaking the continuity of temporal observations. These gaps have to be statistically filled in order to compute aggregated seasonal and annual estimates of ET_a. In this paper, we introduce an approach to develop a gap-filled multi-year monthly ET_a maps at medium spatial resolution of 30 m. The method includes two major steps: (i) estimation of ET_a using the python based implementation of surface energy balance model called PySEBAL and (ii) temporal interpolation using Locally Weighted Regression (LWR) model followed by spline based spatial interpolation to fill the gaps over time and space. The approach is applied to a large endorheic Lake Urmia Basin (LUB) basin with a surface area of ~ 52,970 km² in Iran for the years 2013-2015 using Landsat 7 and 8 satellite data. The results show that the implemented gap filling approach could reconstruct the monthly ET_a dynamics over different agriculture land use types, while retaining the high spatial variability. A comparison with a similar dataset from FAO WaPOR reported a very high correlation with R² of 0.93. The study demonstrates the applicability of this approach to a larger basin which is extendible and reproducible to other geographical areas.