A simple method to directly retrieve reference evapotranspiration from geostationary satellite images

Application of FAO-56 methodology for the assessment of reference evapotranspiration, ET0, is challenging in areas of the world with sparse meteorological network stations. For this reason alternative procedures using remotely observed data have been proposed in the literature. In this work, a simplified version of the Makkink approach [J. Inst. Wat. Eng. 11: 277–288, 1957] was tested in a typical Mediterranean environment (Sicily, Italy). The implemented Makkink approach MAK) uses remotely estimated solar radiation derived from Meteosat Second Generation (MSG) satellite data and in situ observations of air temperature to assess ET0 at daily time scale. Alternatively, taking advantage of well-defined relationships that exist between seasonality, elevation and air temperature, a deterministic procedure for estimating air temperature inputs used in the MAK approach (named RS) was also tested. This approach allows the assessment of daily ET0 without the need of auxiliary air temperature ground observations. A comparison between the FAO-56 and MAK approaches was performed for 45 sites in Sicily over the period 2007–2010. Assuming FA0-56 as the benchmark, the average accuracy of the MAK methodology was 0.4 mm/d, with a relative error of 12%. Similar to other applications of the same procedure, the MAK approach showed a slightly underestimation of ET0 high values; however, an average regression slope of 0.96 (and negligible intercept) suggests a satisfactory agreement with the FAO-56 modeled values. Air temperature observations acquired during 2002–2006 were used to calibrate the deterministic relation between air temperature, seasonality (as a function of the DOY) and orography (as a function of elevation). For the period 2007–2010, the RS approach performs similarly to MAK, with an average difference of less than 0.05 mm/d. Analysis of monthly, seasonal and yearly ET0 maps shows a slight decrease in RS performance during June and July; nevertheless, the differences between MAK and RS approaches are negligible at all analyzed temporal scales.