Evaluating the performance of SEBAL and EEFLux algorithms in estimating daily evapotranspiration using Landsat 8 satellite data

Evapotranspiration (ET) is one of the key components of the Earth's hydrological cycle and its accurate estimation is very important in the water resources management and planning in agricultural usages. ET as a main hydrological flux, links energy, carbon and water cycles, and has an important role in meteorology, hydrology and water resource management, especially in Agricultural Water Management (AWM). Precise and accurate estimation of ET is an essential for the Integrated Water Resources Management (IWRM). Since ET is considered as water consumption over the agricultural areas, knowledge about ET (water consumption), as one of the most important features of the hydrologic cycle, plays an important role in irrigation planning and agricultural management. Quantification of water consumption in agricultural areas can be carried out using climatic and environmental variables, e.g. reference evapotranspiration (ETo), crop coefficient (Kc) and crop evapotranspiration (ETc).

Over the past few decades, satellite imagery based methods have come to the attention of researchers, which they have developed varieties of remote sensing methods to estimate evapotranspiration. In the present study, the Surface Energy Balance algorithm for Land (SEBAL) and Mapping Evapotranspiration with Internalized Calibration (METRIC)/Earth Engine Evapotranspiration Flux (EEFLux) algorithms were used to estimate ET for every pixel of the Landsat 8 at the Arak maize farm, Iran. Various observed data was used in this study. The meteorological data were first obtained from Markazi Regional Water Authority (sunny hours and solar radiation at selected dates) and Arak synoptic station to calculate actual ET and the selected satellite data and ET models were then evaluated. Moreover, wind speed, dew point temperature and daily temperature at the corresponding dates of the Landsat 8 satellite overpass time were gathered from Iran Meteorological Organization. In order to obtain the Leaf Area Index (LAI) at the overpass time of the Landsat 8 satellite over the study area, field measurements were performed. For this purpose, at each date of field operation (which was at the same date as the overpass time of the Landsat 8 satellite) and based on the Landsat pixel size, a 30*30 m plot was designed for field sampling in the selected maize farmland, and a 1*1 m square sub-plot was then designed in the center of each plot. In order to compare the used models, some quantitative criteria are required to measure model performance. In this study, daily ET from SEBAL and EEFlux were used to compare the relative performance of the algorithms for the eight Landsat images during the growth period. Based on commonly used statistical metrics, percent bias error (PBIAS), root mean squared error (RMSE), Nash-Sutcliffe coefficient of efficiency (NSE) and coefficient of determination (R2) criteria were used to evaluate the models. Taylor diagram was used to provide a visual framework and graphically summarizes how closely a set of patterns matches the observed data. In this study, Taylor diagram was used for visual comparing Evapotranspiration derived from SEBAL and EEFLux algorithms, based on the ET obtained from Lysimeter (as reference data).

The results of estimating daily Evapotranspiration from Surface Energy Balance algorithm for Land (SEBAL) and Earth Engine Evapotranspiration Flux (EEFLux) algorithms indicated that daily ET were low at the beginning of the growing season and then came up until middle of the growing season, and then ET values have been decreased due to decreasing temperature as well as changes in maize cover in the study area. Comparison of SEBAL and EEFLux algorithms showed that SEBAL algorithm has estimated about 7.71% of daily ET more that EEFLux algorithm at Arak maize farm. The results of performance evaluation showed that RMSE, NSE, PBIAS and R2 were obtained 0.711, 0.807, 7.398 and 0.885, respectively based on SEBAL algorithm, and for EEFLux algorithm were equal to 1.046, 0.582, 15.080 and 0.793, respectively. The Taylor diagrams showed that the SEBAL model had a lower RMSE and higher correlation than the EEFLux model. Comparing the standard deviation of both models, it was found that the SEBAL model was more in agreement and closer to measured daily ET values than the EEFLux model. This can also be seen in Fig. 6, where SEBAL has a close standard deviation to the measured data, but EEFLux model has a lower standard deviation than the measured data. This indicates that the EEFLux model cannot predict the daily ET the same as SEBAL model. According to observed data (lysimeter data), evaluating the performance of SEBAL and EEFLux algorithms showed that SEBAL algorithm had higher correlation and less standard deviation than EEFLux algorithm. Therefore, SEBAL algorithm had better estimation than the EEFLux algorithm.