Performance of METRIC in estimating hourly and daily evapotranspiration fluxes over an irrigated field in Saudi Arabia

Accurate estimation of evapotranspiration (ET) is essential for hydrological modelling and efficient crop water management in hyper-arid climates, like the one in the Eastern Region of Saudi Arabia. Therefore, a study was designed to 10 apply the METRIC algorithm on Landsat-8 images, acquired from June to October 2013, for the development of ET maps for a 50-ha center pivot irrigated alfalfa field. The METRIC estimated energy balance components and ET were evaluated against the data provided by an Eddy Covariance (EC) flux tower installed in the field. Results indicated that the METRIC algorithm provided accurate ET estimates over the study area, with RMSE values of 0.09 mm h-1 and 0.38 mm d-1. The METRIC algorithm was observed to perform a relatively better in full canopy conditions compared to that in partial canopy 15 conditions. On the average, the METRIC algorithm overestimated the hourly ET by 6.6% in comparison to the EC measurements; however, the daily ET was underestimated by 4.2%.


Introduction
In the Kingdom of Saudi Arabia (KSA), which suffers from severe water shortage problems, the agricultural sector was 20 estimated to consume about 85% of the total freshwater used in 2008 (Al-Kahtani and Ismaiel, 2010), which increased to 90% by 2012 (Elnesr and Alazba, 2013).With the limited water resources and to fulfill the needs of the increasing population in the KSA, efficient use of water for crop production is essential (Hussain et al., 2010;Praveen et al., 2012).
Various studies (Kassem and Al-Moshileh, 2008;Atta et al., 2011;Al-Ghobari et al., 2013;Mohammad et al., 2013) recommended advanced irrigation systems for improving agricultural water use efficiency in the Kingdom.As an example, a 25 reduction in the irrigation water of 30-40% was attained through sprinkler irrigation compared to traditional methods, and an additional saving of 10-25% can be reached with drip irrigation systems (Rizaiza and Al-Osaimy, 1996).Furthermore, implementation of recent innovative precision irrigation technologies, in conjunction with the accurate estimation of crop water requirements through remotely sensed data, can significantly enhance the efficient use of irrigation water in the agricultural sector.Evapotranspiration (ET) measurements are recognized as an essential prerequisite for various water management applications such as irrigation scheduling, hydrologic modelling, nutrient cycles and drought monitoring (Bastiaanssen et al., 2000;Allen et al., 2005;Chavez et al., 2005;Senay et al., 2008;Santos et al., 2010;Anderson et al., 2012;Mkhwanazi and Chavez, 2012;Gowda et al.,2013;Lagos et al., 2013;Moorhead et al., 2013;Yee et al., 2014).In semiarid climates, ET can significantly influence the hydrologic cycle through the crop water use.Therefore, an accurate determination of ET is 5 considered as one of the crucial and essential factors influencing the optimal management of crop water use through different irrigation systems (Hoedjes et al., 2008).On the other hand, the accurate determination of energy balance elements such as sensible heat (H), soil heat (G) and latent heat (LE) fluxes is of great importance for water management practices, especially in arid and semi-arid regions (Zeweldi et al., 2010).Estimation of ET has been performed through different direct/indirect methods including the lysimeter, water balance, Bowen Ratio (BR), Eddy Covariance (EC), scintillometer and Remote 10 Sensing (RS) techniques (Allen et al., 2011a;Rana and Katerji, 2000).Among all the methods used for the estimation/measurement of ET, only lysimeter and EC methods can provide direct measurements of LE or ET; while, other methods use certain equations and assumptions to estimate heat fluxes using measurable parameters (FAO, 1977;Drexler et al., 2004;Chavez et al., 2009a).
Recently, EC systems have gained a lot of popularity in the determination of ET.Several studies have been 15 conducted to investigate the effectiveness of the EC systems in estimating accurate values of H, LE and ET.As an example, Chavez et al. (2009b) conducted a study to investigate the accuracy of the EC measured energy balance components (H and LE) compared to those measured by the large weighing lysimeter on an irrigated cotton field in the USA.Their results showed that the mean values of the EC-measured H and LE were underestimated by about 30% compared to those measured by the lysimeter method.However, adjusting the EC data, through the techniques of energy closure and BR, contributed 20 significantly to improving the accuracy of the ET values measured by the EC system.Ding et al. (2010) compared the EC measured ET (ETEC) with the lysimetric data (ETL) for maize in 2009.Their results showed a non-closure ratio of 0.84 of the energy balance for the daytime fluxes.However, the forced energy balance closure with BR data improved the accuracy from 79.2% to 95.2%.Therefore, the ET values measured by the EC system need to be adjusted, through an appropriate method, to improve their accuracy.25 In-situ point measurements provide accurate ET values; however, these methods are limited to small areas only.An alternative method is to use numerical models, which are successfully used in assessing the spatial distribution of ET on larger scales.However, the process of obtaining accurate estimates of ET, using these computational models, requires the access to high-quality meteorological data.Based on the fact that most of the methods employed to accurately estimate ET provide only point measurements, and therefore, not suitable for applications in large areas, the development of new 30 techniques for continuous determination of ET over large agricultural areas are of great importance.RS systems can provide a sound basis for the development of such technologies (Bala et al., 2013).RS techniques have significantly contributed to the provision of different methods used for the estimation of the spatial ET over large areas, which is hard to achieve by conventional methods (Farah et al., 2004;Colaizzi et al., 2006).
Different methods of RS systems have been successfully implemented to estimate the spatially distributed ET values over agricultural fields (Kalma et al., 2008;Gowda et al., 2008).One of the RS models, most commonly used for the determination of ET values, is the Surface Energy Balance Algorithm for Land (SEBAL) with an estimation accuracy ranging from 67% to 97%.Recently, several SEBAL based models have been developed and successfully implemented in different water management practices (Paul et al., 2013).Examples of such models include the Mapping Evapotranspiration 5 at high Resolution and with Internalized Calibration (METRIC), Modified SEBAL (M-SEBAL), Simplified Surface Energy Balance (SSEB), Remote Sensing of Evapotranspiration (ReSET), Surface Energy Balance System (SEBS) and Surface Energy Balance with Topography Algorithm (SEBTA).The METRIC model, which was developed to determine the quantity and spatial distribution of ET over large areas, is considered as one of the most appropriate models for the continuous estimation of ET over crops during the growing season (Allen et al., 2007;Patil et al., 2015).10 Accurate estimation of ET is critical for crop water management, especially under hyper-arid climates with scarce freshwater resources, as in the KSA.Long-term spatial and temporal ET measurements are not yet available in the Kingdom.
Therefore, the main goal of this study was to employ the Eddy Covariance (EC) system for the assessment of the performance of the METRIC model, which is based on satellite data, in estimating the hourly-and daily-based ET over a center pivot irrigated alfalfa field.15 2 Materials and Methods

Study area
The study was carried out on a 50-ha alfalfa field, which was one of the 48 agricultural fields of Todhia Arable Farm (TAF) located about 250 km Southeast of Riyadh, the capital city of Saudi Arabia, at coordinates of 24° 11' 00'' E and 48° 56' 14.6'' N (Figure 1).The farm was under an arid climate with hot summers (40 ± 2 °C) and cold to moderate winters (15 ± 3 °C) 20 and a mean air temperature of 35°C.The annual rainfall was about 90 mm, most of which occurred in the period from November to February.Due to the high crop water demand combined with the highly erratic rainfall, irrigation is entirely provided using underground water delivered by center pivot irrigation systems.The major crops cultivated in the study area were alfalfa, Rhodes grass and corn.

Eddy Covariance (EC) system 25
Eddy Covariance system (LI-COR, Lincoln, NE, USA) was installed, on May 27th, 2013, over the selected alfalfa field.The EC flux tower was equipped with response sensors, including an open-path gas analyzer (Licor, LI-7500), a 3-axis ultrasonic anemometer (GILL), a net radiometer (CNR-4), soil heat flux plates, pyranometer and a quantum sensor.The EC measurements were recorded at the height of 3.67 m above the soil surface.
The recorded EC data, for the period from June to October 2013, was used for this study.Flux computations were 30 performed using Eddypro software program (ver.5.0).During the process, the collected 30-minute data sets were subjected Hydrol. Earth Syst. Sci. Discuss., doi:10.5194/hess-2017-206, 2017 Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 29 May 2017 c Author(s) 2017.CC-BY 3.0 License.
to raw data processing, such as wind speed measurement offsets, axis rotation for tilt correction, de-trending turbulent fluctuations, time lag optimization settings, spectral corrections and footprint estimation following the procedures described in the Eddypro software instruction manual (LI-COR).As the sonic anemometer was tilted 243º towards the North, an angleof-attack correction for wind components was also performed.During the filtering process, EC data was excluded whenever there was a condensation, a covariance with missing values (-9999) or periods with incorrect sonic temperatures (i.e.>50 5 °C).The missed data was filled with the standard methods as described in the manual (LI-COR).Subsequently, the processed EC data was used in the computation of the ET (i.e.ETEC).

Satellite images
Eight cloud free Landsat-8 (OLI/TIRS) satellite images (Path 165, Row 43) were downloaded for the study period (June to October 2013) from the United States Geological Survey (USGS) Earth Explorer site (http://earthexplorer.usgs.gov),Table 10 1.The downloaded data was georeferenced to the Universal Transverse Mercator (UTM) Map projection using the World Geodetic System 84 (WGS84) datum.Subsequently, Fast Line-of-sight Atmospheric Analysis of Hypercubes (FLAASH) algorithm in ENVI software was used to convert Landsat-8 digital numbers to Top-Of-Atmosphere (TOA) spectral reflectance.Also, land surface temperature (LST) layers were generated using split window algorithm (USGS).The obtained spectral reflectance and the land surface temperature values were utilized as inputs for the ET estimation employing the 15 METRIC algorithm.Image analysis and the execution of the METRIC algorithm were performed using ENVI software program (ver.5.1).

METRIC algorithm and ET estimation
The METRIC algorithm has been developed exclusively for the estimation of ET from Landsat data (Allen et al., 2005).A total of eight Landsat-8 images were processed and used for the estimation of ET (ETMETRIC) over the experimental alfalfa 20 field.During the ETMETRIC computation, surface characteristics such as surface albedo, vegetation indices, surface emissivity and surface temperature were estimated as intermediate products.Anchor pixels (hot and cold) were selected, and the energy components such as the net radiation (Rn), the soil heat flux (G) and the sensible heat flux (H) were estimated as well.
Finally, the latent heat flux (LE) was predicted as a residual of the land surface balance (Allen et al., 2005(Allen et al., , 2007)), Equation (1).Consequently, the instantaneous ET (ETinst) for each pixel was calculated.25 The first step in the METRIC model was to compute the net radiation (Rn) using the surface radiation balance, Equation (2).
The Rn estimation was accomplished in a series of steps by summing up the net short-wave radiation and net long-wave 30 radiation (Hipps, 1989;Brunsell and Gillies, 2002;Allen et al., 2007).
Where, Rs is the incoming shortwave radiation (W m -2 ),  is the broadband surface albedo (dimensionless), RL and RL are the inbound and outgoing longwave radiation (W m -2 ), respectively.o is the broad-band surface thermal emissivity (dimensionless), The (1-o) RL term represents the fraction of incoming longwave radiation reflected from the surface.5 Spectral reflectance of bands 1-7 was used to calculate surface albedo () as described in Allen et al. (2007).
The surface emissivity correction was performed using an empirical Equation (3) by Tasumi (2003) based on soil and vegetative thermal emissivities.Leaf area index (LAI) is computed as per Equation (4) as proposed by Bastiaanssen (1998).During the correction, the typical bare soil and the fully vegetated surface values were set as 0.93 and 0.98, respectively.where, the soil adjusted vegetation index, SAVI was calculated based on TOA reflectance of bands 4 and 5 10 (Huete, 1988).Subsequently, surface temperature (Ts) was computed by converting the TIRS band data from spectral radiance to brightness temperature using the provided thermal constants in the metadata file of Landsat-8 images (USGS).For the estimation of soil heat flux (G), various empirical equations can be found in the literature (Bastiaanssen, 1998;Singh et al., 2008;Gowda et al., 2011).However, this study adopted the empirical model described by Bastiaanssen (2000) representing values near midday for the prediction of Landsat-8 G (i.e.GMETRIC), as a ratio G/Rn based on the NDVI, 20 Equation (5).
Where Ts is the surface temperature (K), and  is the surface albedo.Subsequently, the GMETRIC was obtained by multiplying 25 G/Rn with Rn.
The sensible heat (HMETRIC), however, was estimated from an aerodynamic function as expressed in Equation ( 6).In the calculation of rah, wind speed measurements were used.Where ρ is the air density (kg m -3 ), Cp is the specific heat capacity of the air (J kg -1 °K-1 ), ΔT is the near surface air temperature and rah is the aerodynamic resistance for heat transfer (s m -1 ) between two near-surface heights (i.e. at alfalfa canopy height and the EC measurement height.For further detail steps of HMETRIC computation refer Allen et al. (2007).

Calculation of ET 5
Following the establishment of Rn, G and H from the Landsat-8 processing, LE was calculated as a residual of the energy balance equation.The obtained LE is equivalent to the ETinst at the time of Landsat-8 overpass, (Equation (7).
= 3,600 (7) Where ETinst is the instantaneous ET (mm h -1 ); 3,600 converts from seconds to hours, ρω is the density of water (~1,000 kg m -3 ) and  is latent heat of vaporization (J kg -1 ) representing the heat absorbed when a kilogram of water evaporates.The  component was computed as per Equation ( 8).
= 2.501 − 0.00236 ( − 273.15) × 10 (8) 15 Finally, as presented in Equation ( 9), the reference ET fraction (ETrF) was calculated as the ratio of the computed ETinst from each pixel to the reference ET (ETr), which was computed from the weather data.

= (9) 20
Where ETinst is from Equation (7) and ETr is for the standardized 0.5 m tall alfalfa at the time of the image.The EC system recorded weather parameters were used to calculate ETr as described in ASCE-EWRI (2005).The obtained ETrF was subsequently extrapolated to daily values.In the processes, ET24 was computed by assuming that the instantaneous ETrF computed at satellite overpass is the same as the average ETrF over the 24 h average (Allen et al., 2007), Equation (10).25 Where ET24 is the cumulative ET (mm day -1 ) for the day of the image overpass; Crad is the correction term used in sloping terrain to correct for variation in 24 h versus instantaneous energy availability.ETrF is from Equation (9) and ETr24 is the 30 cumulative reference ET for the day of the image (Van Niel et al., 2012;Cammalleri et al., 2014).

Data analysis
The spatially estimated energy flux components, derived from the METRIC algorithm and the EC flux tower, were subjected to heat flux correction and energy balance and footprint (FTP) analysis, as described in Schmid (1994), Twine et al. (2000) and Kustas et al. (2005).In the FTP analysis, the utilized wind parameters such as wind speed, wind direction and yaw angle at the time of Landsat-8 overpass were provided in Table 2. On the other hand, different statistical performance indicators 5 (RMSE, MBE and Nash-Sutcliff coefficient) were utilized for the assessment of the METRIC performance in estimating ET against the EC system.
3 Results and discussion

Footprint analysis (FTP)
The simple arithmetic averages of weighted/integrated heat fluxes over the fetch areas are the FTPs (Chavez et al., 2005).10 FTPs are widely used in validating the spatially distributed fluxes obtained from remote sensing (RS) with EC measured fluxes.The fetch area of the FTP was classified into six classes, based on the cumulative contribution of the eddies, as provided in Figure 2. Based on the FTP analysis, about 90% of the EC system observed eddies originated from the 122 to 144 m in the upwind direction.However, the peak (vertical) fetch was within 42 to 51 m.It was also observed that 10% of the fluxes were registered between 6 and 17.5 m from the flux tower.Whereas, about 30%, 50% and 70% of the fluxes were 15 recorded between 35 to 63 m, 52 to 97 m and 81 to 136 m, respectively.On the other hand, energy fluxes were analyzed as per Chavez et al. (2005) by integrating Landsat-8 image acquisition (i.e.overpass), which showed the 90% of the FTP overlap with METRIC model (Kustas et al., 2005).Hence, the obtained leaf area index (LAIG) and the reference ET (ETr) over the alfalfa field were used as inputs to the METRIC-based Landsat-8 ET prediction (Table 3).The ground-based LAI (LAIG) of alfalfa was recorded on the day of satellite overpass using plant canopy analyzer (LAI-2200).20

Surface temperature (T)
As illustrated in Figure 3, linear regression analysis of Landsat-8 derived surface temperature (TLST) using the METRIC algorithm against the EC flux tower measured temperature (TEC) showed a good correlation with an R 2 value of 0.71 (P = 0.0084).However, the METRIC algorithm was observed to underestimate the TLST compared to the TEC as evidenced by the RMSE and MBE performance indicators of 4.23°C (-12.82%) and -3.40° C (-9.34%), respectively.The recorded errors (4.2 25 % to 19.6%) in the METRIC estimated surface temperature were slightly higher than the values reported in previous studies.
For example, compared to lysimeters, Chavez et al. (2009a) reported METRIC errors of 11.1% and 1.9% in estimating surface temperatures for corn and sorghum fields, respectively.According to Omran (2012), LAI affected the amount of heat stored in the soil surface through transpiration process, and this was evident with the Landsat-8 derived TLST with an error of about 20%, where the alfalfa crop was in full foliage cover (i.e.LAI~6).Thus, the temporal variation in LAI would result in 3 °C) values were recorded in the experimental alfalfa field.However, there is a discrepancy at higher temperatures when the crop density of alfalfa was low (i.e.bare soil was visible to the radiometer) as evidenced by the low LAI values over the footprint and the recorded TEC and TLST (Chavez et al., 2005).Hence, the cooling effect of vegetation on the TLST accelerated the error by 1 to 20% during the low and high LAI cases, respectively.5

Energy balance (EB) components
The mean values of energy balance (EB) components obtained from both the EC system and METRIC algorithm are provided in Table 4. While, the performance indicators (RMSE, MBE and Nash-Sutcliff coefficient) are given in Table 5.

Net radiation (Rn)
The temporal trend of EC (CNR-4) measured Rn (RnEC) was analyzed for discrepancies.About 9% to 16% of the collected 10 data exhibited infrequent errors; and hence, the abnormal data has been discarded, and a gap-filling process was performed using the Eddy Pro software program (version 5.0).A scatter plot (Figure 4), representing the relationship between RnEC and METRIC estimated Rn (RnRS), was generated and analyzed for performance accuracy.Moderate linear correlation (R 2 = 0.54 and P>F = 0.0381), low RMSE (18.32 W m -2 ; 3.8%) and very low MBE (8.66 W m -2 ; 1.76%) indicated that the METRIC model accurately (96%) estimated the Rn.The obtained RMSE and MBE values are in agreement with those reported by 15 earlier studies of Mkhwanazi and Chavez (2012) where the RMSE value was 4.1% and MBE value was 3.3%, and Chavez et al. (2007) where the RMSE value was 9.8% for METRIC model estimated Rn using Landsat-8 TM image.

Soil heat flux (G)
Figure 5 illustrates a scatter plot between the EC measured (GEC) and the METRIC estimated (GRS) soil heat flux values.
The relationship was relatively fair with an R 2 value of 0.67 and a Nash-Sutcliffe coefficient of 0.59.The RMSE and MBE, 20 however, were determined at 28.46 W m -2 (37.33%) and 12.42 W m -2 (16.29%), respectively.These recorded errors were relatively higher than the values reported in similar previous studies.Mkhwanazi and Chavez (2012) reported an RMSE value of 14.2 W m -2 (27.6%) and an MBE value of -3.0 W m -2 (-5.8%) in estimating G with the METRIC algorithm for irrigated alfalfa.The average observed values of G were similar to the majority of values recorded by Ham et al. (1991).
Furthermore, the ratio of Rn to the G (G/Rn) derived from METRIC model against the LAI inferred from the Canopy 25 analyser (PCA 2200, Li-COR, USA) is presented in Figure 6.The G/Rn is one of the essential components in the analysis of the accuracy of Bowen ratio (Allen et al., 2011b).Correlating the LAI with the G/Rn produced a moderate polynomial (3 rd order) relationship for both the METRIC (R 2 = 0.288 and P>F = 0.23) and the EC (R 2 = 0.313 and P>F = 0.15) methods.
However, the results suggested that beyond a certain value of LAI (up to 4.2), the relationship between the G/Rn and LAI was constant.The scatter in the G/Rn where the values for LAI > 4 is not surprising since the values of G/Rn for full canopy 30 cover are ranging between 0.05 and 0.15 as showed in Waters et al. (2002).

Sensible heat flux (H)
Comparison between the METRIC estimated H (HRS) and the EC measured H (HEC), after the adjustments of EB closure, showed values of mean estimation error ranging from 10.89 W m -2 (low LAI conditions) to 38.91 W m -2 (full canopy condition) with an MBE value of 15.72 W m -2 (13.87%).The high RMSE value of 72.01 W m -2 (63.54%) for the HRS might 5 be due to the advection and variability in the canopy density.Hence, most of the Rn has been partitioned into LE than into H, as introduced by the near surface air temperature difference (ΔT) and the aerodynamic resistance of heat transfer (rah), i.e.
propagation errors.This was evident in the linear regression analysis (Figure 7), where a good correlation between the HRS and HEC (R 2 = 0.61) was observed; however, it was not significant (P>F = 0.022), and it was also confirmed with the RMSE of 63.5%.In contrast, Carrasco-Benavides et al. (2013) reported that the METRIC algorithm overestimated the H component 10 by 39 W m -2 with a mean absolute error of 10%.

Latent heat flux (LE)
A scatter plot was established between the METRIC estimated (LERS) and the EC determined LE (LEEC) values as illustrated in Figure 8.The correlation was relatively good (R 2 = 0.66); however, it was not statistically significant (P>F = 0.14).This was attributed to the fact that the LERS was a residual component of the energy balance equation that mainly depends on the 15 calibration accuracy of the HRS, which was purely based on the quality of the selected anchor pixels (Weaver, 1990;Field et al., 1994).
Although the regression analysis showed a non-significant correlation, the performance indicators such as the MBE (2.45 W m -2 ; 0.73%) and the RMSE (115.04W m -2 ; 34.33%) are in accordance with the earlier reported values indicating that the METRIC algorithm was useful in the accurate determination of LE over large areas.However, Carrasco- Benavides 20 et al. (2013) stated that the METRIC algorithm overestimated the LE by 14% (RMSE).The obtained average absolute error of the LERS for the eight images was 35%. Figure 9 illustrates the Energy Balance (EB) components, where the LE and the G are presented as the major and the minor EB components, respectively.

Evapotranspiration (ET)
The correlation between the ET values obtained from both the METRIC algorithm and the EC system was found to be highly 25 significant (R 2 = 0.93 and P>F = 0.0001) as illustrated in Figure 10.Further assessment of the accuracy of the METRIC algorithm in estimating the ET was performed using the RMSE and MBE indicators.Comparing the hourly ET calculated using the METRIC algorithm to that using the EC system resulted in an RMSE value of 0.09 mm h -1 (14.33%) and an MBE value of 0.04 mm h -1 (6.46%).For the daily mean ET, the RMSE and MBE values were 2.25 mm d -1 (25.93%) and 0.38 mm d -1 (4.40%), respectively.These results are in agreement with the results reported by Mkhwanazi and Chavez (2012), where 30 the performance errors in estimating the ET using the METRIC model were determined at 0.14 mm h -1 (17.6%) and -0.08 Hydrol.Earth Syst.Sci. Discuss., doi:10.5194/hess-2017-206, 2017 Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 29 May 2017 c Author(s) 2017.CC-BY 3.0 License.mm h -1 (-10.3%) for RMSE and MBE, respectively.Similarly, Chavez et al. (2007) reported a METRIC error of 0.7 mm d -1 (7.4 %) in predicting the ET compared to lysimeter data.

Energy balance (EB)
Eddy Covariance (EC) system facilitates the process of accurate measurements of energy balance (EB) components.Hence, 5 using the measured EB components, the surface energy budget can be estimated (Equation 1) along with the EB closure.The four flux components (Rn, H, G and LE) measured by the EC system were assessed for the EB closure (Kustas et al., 2005).
The EB closure rate for the EC system ranged between 71% and 99%, with a mean value of 87% (Table 6).Energy balance closure was significantly influenced by the amount of the unaccounted energy lost (i.e.lack of closure in the energy balance equation due to advection and measurement errors in the field) in the partition of the net radiation into latent and sensible 10 heat fluxes and soil heat flux.This was attributed to the inherent errors in measuring the EB components by the ground systems, which was reported to be 15 -20%, 5 -10% and 20 -30% according to Weaver (1990), Field et al. (1994) and Twine et al. (2000), respectively.

Conclusions 15
The METRIC algorithm was applied on Landsat-8 images for the ET mapping, and its performance was evaluated against the EC flux tower measured EB components and ET at hourly and daily intervals over irrigated alfalfa field in the Eastern Region of Saudi Arabia.The following are the specific conclusions of this study:  The METRIC algorithm was successfully utilized to estimate the ET with an average error of 6.6% (hourly ET) and 4.2% (daily ET).The performance of the METRIC algorithm was found to be more accurate in estimating the 20 hourly ET (R 2 = 0.93) than the daily ET (R 2 = 0.57) compared to the ETEC.
 Due to the cooling effect of vegetation, it was found that the METRIC algorithm underestimated the surface temperature (TLST) by 1% and 20% for partial and full canopy conditions, respectively.However, the METRIC algorithm produced accurate LE estimates which were indicated by the low mean value of MBE of 2.45 W m -2 (0.74%). 25  Compared with the EC data, the RS estimated Rn component was found to be highly accurate (an accuracy of more than 95%).In addition, the sensible heat flux obtained from satellite data was associated with errors ranging between 10.89 W m -2 and 38.91 W m -2 at low (< 1.2) and high (>5.0)LAI values, respectively.

Figure 7 .
Figure 7.Comparison between the measured (HEC) and METRIC derived (HRS) sensible heat flux.

Figure 9 .Figure 10 .
Figure 9. Relationship between EB components (W m -2 ) measured by the METRIC algorithm and the EC system.5

Table 1 .
Details of Landsat-8 data used in the study.

Table 2 .
Details of wind parameters at the time of satellite overpass

Table 4 .
EC measured and METRIC estimated Energy Balance (EB) components over an alfalfa field.