A comparative study of remote sensing and gene expression programming for estimation of evapotranspiration in four distinctive climates

摘要

An accurate estimation of Evapotranspiration (ET) is an important issue in hydrology, water resources management and irrigation scheduling. There are a wide range of methods for estimation of ET, among which, machine learning techniques and remote sensing-based approaches demonstrated more reasonable results. Accordingly, this study attempts to compare the capability of two developed models of Gene-Expression Programming (GEP) and Surface Energy Balance Algorithm for Land (SEBAL) in estimation of ET, at four different climate types of Temperate-Warm (T-W), Wet-Warm (W-W), Arid-Cold (A-C), and Arid-Warm (A-W). In this way, a-two year of daily records as weather variables (i.e., maximum and minimum temperature, dew-point temperature, vapor pressure, saturated vapor pressure, relative humidity, 24-h rainfall, sunshine hours, and wind speed) were considered as input variables, whereas ET values were computed as output variable (observed ET) by using FAO Penman-Monteith-56 method. After development of two predictive models, the statistical results were compared with well-known Hargreaves-Samani method. The results showed that while Hargreaves-Samani equation could not yield remarkable results in any of the climates, GEP and SEBAL demonstrated accurate predictions. In this way, GEP was the superior model in T-W (R-2 = 0.902 and RMSE = 0.713 mm/day) and A-W (R-2 = 0.951 and RMSE = 0.634 mm/day) climates but it dropped a bit in two other climates. However, SEBAL not only had the best performance in both climates of W-W (R-2 = 0.967 and RMSE = 0.515 mm/day) and A-C (R-2 = 0.990 and RMSE = 0.720 mm/day), but also demonstrated good predictions in T-W and A-W climates. Therefore, SEBAL is recommended as the best model for estimation of ET in all climate types.