Assessing Daily Evapotranspiration Methodologies from One-Time-of-Day sUAS and EC Information in the GRAPEX Project

Daily evapotranspiration (ETd) plays a key role in irrigation water management and is particularly important in drought-stricken areas, such as California and high-value crops. Remote sensing allows for the cost-effective estimation of spatial evapotranspiration (ET), and the advent of small unmanned aerial systems (sUAS) technology has made it possible to estimate instantaneous high-resolution ET at the plant, row, and subfield scales. sUAS estimates ET using “instantaneous” remote sensing measurements with half-hourly/hourly forcing micrometeorological data, yielding hourly fluxes in W/m2 that are then translated to a daily scale (mm/day) under two assumptions: (a) relative rates, such as the ratios of ET-to-net radiation (Rn) or ET-to-solar radiation (Rs), are assumed to be constant rather than absolute, and (b) nighttime evaporation (E) and transpiration (T) contributions are negligible. While assumption (a) may be reasonable for unstressed, full cover crops (no exposed soil), the E and T rates may significantly vary over the course of the day for partially vegetated cover conditions due to diurnal variations of soil and crop temperatures and interactions between soil and vegetation elements in agricultural environments, such as vineyards and orchards. In this study, five existing extrapolation approaches that compute the daily ET from the “instantaneous” remotely sensed sUAS ET estimates and the eddy covariance (EC) flux tower measurements were evaluated under different weather, grapevine variety, and trellis designs. Per assumption (b), the nighttime ET contribution was ignored. Each extrapolation technique (evaporative fraction (EF), solar radiation (Rs), net radiation-to-solar radiation (Rn/Rs) ratio, Gaussian (GA), and Sine) makes use of clear skies and quasi-sinusoidal diurnal variations of hourly ET and other meteorological parameters. The sUAS ET estimates and EC ET measurements were collected over multiple years and times from different vineyard sites in California as part of the USDA Agricultural Research Service Grape Remote Sensing Atmospheric Profile and Evapotranspiration eXperiment (GRAPEX). Optical and thermal sUAS imagery data at 10 cm and 60 cm, respectively, were collected by the Utah State University AggieAir sUAS Program and used in the Two-Source Energy Balance (TSEB) model to estimate the instantaneous or hourly sUAS ET at overpass time. The hourly ET from the EC measurements was also used to validate the extrapolation techniques. Overall, the analysis using EC measurements indicates that the Rs, EF, and GA approaches presented the best goodness-of-fit statistics for a window of time between 1030 and 1330 PST (Pacific Standard Time), with the Rs approach yielding better agreement with the EC measurements. Similar results were found using TSEB and sUAS data. The 1030–1330 time window also provided the greatest agreement between the actual daily EC ET and the extrapolated TSEB daily ET, with the Rs approach again yielding better agreement with the ground measurements. The expected accuracy of the upscaled TSEB daily ET estimates across all vineyard sites in California is below 0.5 mm/day, (EC extrapolation accuracy was found to be 0.34 mm/day), making the daily scale results from TSEB reliable and suitable for day-to-day water management applications.


Introduction
Evapotranspiration (ET) is a key component in the hydro-ecological process, which couples water and energy budgets, links the land surface and the atmosphere [1], and represents water consumption for biomass production [2]. Routine monitoring of actual ET is important for a variety of applications, including water resource management, drought monitoring, climate change, and the efficiency of crop irrigation [3][4][5][6]. Numerous methods have been used over the past decades to measure ET, including lysimeters, Bowen ratio, and eddy covariance (EC) flux towers. However, these methods represent limited sampling areas [7], and the measurements are best interpreted for homogeneous surfaces [8]. Spatial techniques are needed to accurately quantify ET for improved irrigation scheduling and water management decision support, particularly in complex canopies such as vineyards, which have non-uniform and complex vertical canopy structure, wide and variable row efforts are being directed towards daily ET estimation using surface energy balance models, among them the Two-Source Energy Balance (TSEB) model. However, the TSEB model provides hourly surface energy fluxes, which requires a scaling/extrapolation approach for generating daily ET information. Several studies have compared different daily ET methods with an assumption that the ratio of latent heat flux (LE) to one energy balance term is constant throughout the day, yet no universal approach has been identified as suitable for all types of land surfaces. Previous studies have indicated that the accuracy of that approach (upscaling daily ET) is a function of land surface type. For example, the evaporative fraction (EF) approach produced the best agreement in bare soil [28] and soybean [19], while the incoming solar radiation (R s ) approach was deemed to be more efficient in estimating daily ET in grassland and woody savanna [29]. Another crucial issue for precise daily ET estimation is the proper selection of the time-of-day window. In the study conducted by Colaizzi et al. [28], the best time window for extrapolating the hourly ET to a daily scale was shown to be within 1 or 2 h of solar noon. This conclusion was also supported by Jackson et al. [30], who identified the time-of-day window for acquiring the ET for daily ET estimation as within 2 h of solar noon. Therefore, some concerns, such as actual and potential satellite overpass times and cloudiness vs. time of day, should be identified clearly to avoid any error propagation in the daily ET estimation.
The need for accurate daily ET (ET d ) estimates raises two fundamental questions: (1) which daily ET extrapolation approach at grapevine row scales can provide reliable values under a variety of crop and environmental conditions and thermal-based ET models like TSEB? and (2) what time window for acquiring a remotely-sensed ET provides the most reliable daily ET using an extrapolation approach? Multiple efforts have been made to estimate ET d for different crops; however, computing ET d for complex canopies, such as vineyards and grapevine row scales, has not been adequately addressed. In this study, different extrapolation approaches from the literature were assessed for estimating daily ET from instantaneous sUAS ET estimates for several vineyard sites across California. Specifically, this paper (a) assessed the performance of several daily ET extrapolation approaches using EC observations and sUAS information, and (b) determined an optimal time window for ET upscaling from a single to a daily estimate.
1.1.1. Evaporative Fraction (EF) Approach-One of the most common schemes to extrapolate instantaneous evapotranspiration to a daily value is the evaporative fraction (EF) [34]. EF is defined as the ratio of latent heat flux (LE) to the available energy (the difference between net radiation, R n , and soil heat flux, G), assumed to be constant throughout daytime hours. The EF approach is presented in Equation (1), as follows: where ET d is the daily ET (mm/day), LE is the instantaneous latent heat flux (W/m 2 ), R n is the instantaneous net radiation (W/m 2 ), G is the instantaneous soil heat flux (W/m 2 ), ρ w is the water density (kg/m 3 ), λ is the latent heat of vaporization for water (MJ/kg), (R n -G) d is the total daily available energy (MJ/m 2 /day), and c is a factor equal to 1000 to convert meters to millimeters.
Numerous studies have considered the tendency of the EF to be nearly constant during the daytime [35]; however, the combination of soil moisture, weather conditions, topography, and biophysical conditions has an impact on the conservation (or variability) of the EF in the daytime [31]. According to Hoedjes et al. [36], self-preservation of the EF approach is applicable under dry conditions, while under wet conditions, the EF is no longer valid.
Nonetheless, a previous study by Crago [32], which used Bowen ratio stations over natural grassland, indicated that, for clear days, the midday EF is a good indicator of the daytime average value of the EF compared with cloudy days, but the values are still underestimated from the daytime average EF due to the concave-up shape of the diurnal variation of the EF. This finding is also supported by Li et al. [37], who found that the EF is relatively close to the daily average EF in the 1000 to 1500 timeframe, and could be used to guide vineyard irrigation practices in arid regions. However, the study by Zhang and Lemeur [38], which used 12 surface network stations called Système Automatique de Mesure de l'Evaporation Rèelle (SAMER) over an area composed of forest (40%) and mixed agriculture (60%), indicated that the EF varies during the daytime and could not be used as a guide for ET d estimates due to factors such as available energy, soil moisture, and other environmental variables. According to the study by Gentine et al. [39], which examined the influence of environmental factors (incoming solar radiation, wind speed, air temperature, soil water content, and leaf area index) on the diurnal behavior of the EF over wheat and olive, indicated that EF is strongly linked to soil moisture availability and canopy cover. As such, the EF increases with increasing the soil moisture and/or fractional cover. On the other hand, they found that the phase difference between net radiation (R n ) and the soil heat flux (G) must be well-characterized in application models that invoke the EF daytime self-preservation.

Solar
Radiation (R s ) Approach-Another approach for extrapolating ET i to ET d is the R s approach, which is similar to the EF but replaces the available energy ((R n − G), instantaneous or daily) term with the incoming solar radiation (R s ) as a reference variable. This approach, developed by Jackson et al. [30], assumes that the diurnal ET variation is similar to the solar radiation (ET~R s ), that is, the ET is highly correlated and proportional to the R s . Equation (2) demonstrates the expressions for calculating ET d using the R s approach. 2) where R sd is the daily solar radiation (MJ/m 2 /day), and R s is the instantaneous solar radiation (W/m 2 ). Other parameters are similar to the EF approach.
According to Van Neil et al. [40], the R s approach is robust when upscaling ET i to multiple timeframes (e.g., daily, 8-day, and monthly). Moreover, many studies have indicated that solar radiation (R s ) is the most robust scalar approach that explains the ratio between the ET d and ET i [41].

Ratio of Net
Radiation-to-Solar Radiation (R n /R s ) Approach-The R n /R s approach is another approach to scale up ET i to ET d using the evaporative fraction (EF) and the ratio of net radiation-to-solar radiation (R n /R s ) [42]. The R n /R s approach is presented in Equation (3).
The parameters of this approach are explained in the EF and R s approaches.

Sine Approach-The
Sine approach, developed by Jackson et al. [30], showed that the generic trend of the ET i during the daylight period is similar to the solar irradiance and could be approximated by a Sine function, where the maximum irradiance occurs at solar noon (~12 p.m.). For cloudy days, the daily ET estimates using the Sine approach are less reliable or may be invalid. This implies that the ET i responds strongly to solar radiation [38]. The approach has been investigated by Zhang and Lemeur [38], who found the Sine approach to be preferable to others for upscaling instantaneous ET values.
1.1.5. Gaussian (GA) Approach-The Gaussian (GA) approach has been used recently by Liu et al. [43] to retrieve the ET d from remotely sensed instantaneous ET. The study used ET i observations from an EC system and found that the ET diurnal variation follows a Gaussian-fitting curve. When comparing this approach to the Sine and EF approaches, results from the study of Liu et al. [43] indicated that GA is more accurate using the eddy covariance (EC) system.
where w is the width that equals 2δ, δ is the standard deviation of ET i values, t i is the time of the instantaneous ET (ET i ), and t c is the time when ET i arrives at maximum value in the diurnal variation.

Two-Source Energy Balance (TSEB) Model
The TSEB model was developed by Norman et al. [44] to explicitly accommodate the difference between radiometric and aerodynamic surface temperatures that affect the energy exchange between soil and canopy systems and the lower atmosphere at instantaneous time scales. In the TSEB model, turbulent energy fluxes are partitioned between canopy and soil, with different versions applied to separate between those components. These versions include the TSEB-PT (Priestly-Taylor), the TSEB-DTD (Dual Time Difference), TSEB-2T-DMS (Data-Mining Sharpening of temperature), and TSEB-2T (Dual Temperature). The TSEB-PT version assumes a composite radiometric temperature (T rad ) that contains temperature contributions from the soil/substrate and canopy and is decomposed based on the vegetation fractional cover (f c ). The TSEB-DTD version, developed by Norman et al. [45], uses two observations of T rad : the first observation obtained 1.5 h after the sunrise (T rad,0 ), and the second one during the daytime (T rad,1 ). The TSEB-DTD version uses the same approach as TSEB-PT to divide the composite T rad between the soil/substrate and canopy temperatures. Using TSEB-DTD could reduce the error in flux estimations when uncertainty exists in local air temperature observations and absolute T rad [46]. TSEB-2T-DMS uses a data-mining fusion algorithm to sharpen the land surface temperature (LST), which allows better discrimination between the soil/substrate and canopy temperatures [47]. The TSEB-2T approach was originally developed by Kustas and Norman [48] and was further refined and tested by Nieto et al. [49]. The main concept underpinning the TSEB-2T approach is to estimate the T s and T c from composite LST imagery using the relationship between the vegetation index (VI) and the LST to extract the T s and T c within a spatial domain. An early attempt at estimating vineyard water use at a field scale using aerial imagery with TSEB and a simple thermal-based contextual scheme suggests the TSEB is a robust approach for vineyard ET estimation [50]. In this study, the TSEB model was used to calculate the instantaneous ET at the time of the sUAS overpass, and the various schemes were used to extrapolate this one-time-of-day ET to a daily value. The TSEB-2T model was used for the Sierra Loma vineyard analysis, while the TSEB-PT was used for Ripperdan and Barrelli due to limitations in applying the TSEB-2T model to those two sites. The average value of the LAI was used for these sites, but the TSEB-2T requires the LAI spatial information to identify the threshold values of NDVI of soil, which is based on the empirical relationship between the NDVI and LAI. More details about the TSEB-2T can be found in Nieto et al. [49]. Applying the energy conservation and balance principles, the energy budget in the TSEB model can be described in the following equations: R ns = H S + LE S + G, (11) where R n is the net radiation, and G is the soil heat flux. H and LE are heat fluxes, where H is the sensible heat flux and LE is the latent heat flux. All flux units are expressed in W/m 2 . Subscripts of c and s represent the canopy and soil components, respectively. To estimate the sensible heat flux for soil and canopy, Norman et al. [44] proposed a series of soil vegetation resistive schemes (following an analogy with Ohm's law), as illustrated in Figure 1. 12) where ρ air is the air density (kg/m 3 ), C p is the heat capacity of the air at constant pressure (J/kg/K), T A is the air temperature (Kelvins), T c and T s are the canopy and soil temperatures (Kelvins), respectively, and T AC is the temperature of the canopy air space (Kelvins), which is calculated with Equation (13). 13) where R A is the aerodynamic resistance to heat transport from the soil/canopy system, R x is the boundary layer resistance of the canopy leaves, and R s is the aerodynamic resistance to heat transport in the boundary layer close to the soil surface. All resistances are expressed in (s/m). The mathematical expressions used to compute the resistance network are detailed in Equations (14)- (16).
Nassar et al. 16) where u * is the friction velocity, calculated as the following: In Equation (17), z u and z T are the measurement heights for wind speed (u) and air temperature (T A ), respectively, d 0 is the zero-plane displacement height, and z 0M is the roughness length for momentum. The unit of z 0M is expressed in m. In the TSEB model versions, the roughness length of momentum (z 0M ) is assumed to equal the roughness length for heat transport (z 0H ), as the aerodynamic resistance of the canopy elements (R x ) already takes into account the different efficiencies between momentum and heat transport. κ′ represents the von Karman's constant, which is equal to 0.4. Ψ h and Ψ m are the adiabatic correction factors for heat and momentum, respectively. The details of these two factors are described in Brutsaert [51]. In Equation (15), C′ is assumed to be 90 s 1/2 /m and l w represents the average width of leaf (m). The coefficients (b and c) in Equation (16) [11,49,52,53].

Study Area
The experiment was conducted within three different climate regions located in California, as shown in Figure 2. All of these sites are part of the Grape Remote Sensing Atmospheric Profile and Evapotranspiration eXperiment (GRAPEX) project [54], led by the USDA ARS in collaboration with E&J Gallo Winery, University of California in Davis, Utah State University, NASA, and others. The overall objective of the GRAPEX project is to provide the vineyard manager and grower with spatially distributed, remotely sensed ET information for improving irrigation water use efficiency and detecting crop stress in multiple vineyard blocks. This would facilitate water conservation efforts in California's Central Valley, which has been experiencing frequent and severe drought conditions. The project began in 2013 at two pinot noir blocks located within the Sierra Loma Vineyard near Lodi, California (38.29°N, 121.12°W) in Sacramento County (see Figure 2) [7]. The two vineyard blocks, north and south, differed in maturity and age, having been implemented in 2009 and 2011, respectively. The configuration of the trellising system in both fields is the same, with vine trellises 3.35 m apart and an east-west orientation. In 2017, the GRAPEX project extended the observations to include two additional vineyards: Barrelli  (IOPs) in these sites were conducted in the veraison period (from mid-July to early August), when the crop evaporative demand increases. Figure 3 illustrates the procedure used for this study. First, available eddy covariance (EC) flux tower data was filtered to select cloud-free days only. Then, five different ET d approaches were applied to the LE fluxes from the EC measurements for upscaling the ET to the daily timescale. The analysis was performed using EC observations at different vine phenological stages (April-May, June-August, and September-October).

Procedure
Finally, daily sUAS ET information, produced using the TSEB model, and results from the five approaches for upscaling/extrapolating the daily ET were compared against the measured ET d from the EC tower data. Two time windows were selected for the daily ET estimation: the first was near solar noon (1030-1330), and the second was in the afternoon (1430-1630). The reasons for these selections were (a) satellite overpass time, (b) sUAS flexibility, which allows for flights at different hours, including mid to late afternoon, and (c) an opportunity to assess the suitability of using later (2+ hours after solar noon) sUAS flights for the estimation of daily ET.

sUAS Data Processing-The AggieAir sUAS Program at Utah State University
(https://uwrl.usu.edu/aggieair/ accessed on 10 December 2020) [55] acquired highresolution imagery at 450 m above ground level (agl), resulting in visible and near-infrared data at a 0.10 m spatial resolution, and a thermal spatial resolution at 0.6 m. The spectral range of the visible and near-infrared data was similar to Landsat; however, the thermal band range was wider, with a bandwidth spanning from 7 to 14 μm. Thermal data was acquired using a radiometrically calibrated micro-bolometer camera. Table 1 lists the information concerning the different AggieAir sUAS flights. In this study, the obtained sUAS images were georectified using ground control points (GCPs). Details of the optical and thermal information are presented below.
Thermal Data: Changes in the transmissivity and atmospheric radiance can adversely affect the sUAS thermal data [56]. Details about thermal data calibration can be found in the work by Torres-Rua [56], while the work by Torres-Rua et al. [57] shows that the TSEB model is insensitive to surface emissivity. The AggieAir sUAS Program has a thermal protocol to use over 90% of overlap for thermal raw imagery collected after sUAS launching but before mission data collection upon internal lens temperature stabilization of the microbolometer Nassar et al. Page 10 camera. These two steps address potential vignetting as well as the temperature drifting effect observed in other sUAS applications.
Optical Data: Radiometric agreement between different remote sensing platforms is important for further integration. An internal evaluation of the optical data obtained from different sUAS flights was performed by aggregating the high-resolution imagery up to Landsat scale using a point-spread function (PSF). The resulting 30-m pixels were found to agree with Landsat reflectance information. This is due to the use of different sensors than the ones used by Hassan-Esfahani et al. [58].

Eddy Covariance (EC) Fluxes-Surface energy fluxes (LE and H)
were calculated from the EC measurements of the sonic temperature, water vapor, and vertical wind speed. In this study, the measurements obtained from the EC were averaged over a 60-min time interval to ensure appropriate averaging time for calculating the H and LE. The sensible heat flux was calculated from the product of the air density, the specific heat of air, and the covariance between the vertical wind speed and sonic temperature. The expression used to calculate H is shown in Equation (18).
where ρ a is the air density (kg/m 3 ), C p is the heat capacity of the air at constant pressure (J/kg/K), U z ′ is the vertical wind speed (m/s), and T s ′ is the sonic temperature (Kelvins).
The latent heat flux (LE) was calculated from the product of the latent heat of vaporization (λ) and the covariance between the vertical wind speed (U z ′) and the water vapor density (ρ v ′). The formula used to calculate the LE is illustrated in Equation (19). 19) where ρ v ′ is the water vapor density (kg/m 3 ). Table 2 describes the EC towers installed at the different vineyard sites to monitor ET. The EC measurements (April to October) obtained are the surface energy fluxes (R n , H, and LE) and micrometeorological data. More details about the in situ micrometeorological measurements can be found in the work by Nassar et al. [7].
In Sierra Loma, each EC tower monitors grapevines of different ages, while 4 flux towers in Ripperdan 720 measure different water management approaches at 4 different blocks. In this study, the footprint analysis of each EC tower was performed to validate the results obtained from the TSEB model. The Kljun et al. [59] model was used for describing the fetch of the EC contribution area for the hourly period encompassing the sUAS flight times. The shape and orientation of the EC footprint depend on multiple micro-meteorological conditions that are observed by the EC towers installed at the sites, which include the friction velocity, wind speed, wind direction, roughness length, standard deviation of the crosswind velocity, and Monin-Obukhov length as well as the EC tower height. In this study, the authors did not include any energy balance closure to the EC information to minimize biases.

Quantitative Statistics-
The performance indices to evaluate the daily ET approaches in this study involved comparisons of the modeled ET from the five different approaches against daily ET measurements from the EC towers. Computed statistical metrics included the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), Nash-Sutcliffe efficiency coefficient (NSE), and the coefficient of determination (R 2 ). The NSE coefficient checks the capability of the model to reproduce the following statistical components: correlation coefficient of (r), mean (μ), and variance (s). NSE values range between −∞ and 1, where 1 represents a perfect agreement, while a value of 0 means that the model results are not better than the average of the variable of interest, and values < 0 indicate unacceptable model performance [60].
where O i denotes the observed value, P i denotes the modeled value, O denotes the mean observed value, P denotes the mean modeled value, and N represents the number of observations.

Graphical
Representations-Different graphical representations were used to visualize and evaluate the datasets from the EC towers and the performance of the extrapolation techniques. Boxplots were created to describe the variance of surface energy fluxes (R n , H, LE, and G) at each hour in the dataset. Boxplots were also used to evaluate the performance of the five daily ET extrapolation schemes by presenting the distribution of relative error at each individual hour during the daytime, as shown in the Appendix. Moreover, scatterplots were used to compare the modeled fluxes from TSEB and the measurements from EC systems to evaluate model performance.

Diurnal Variation of Energy Fluxes from EC Measurements
An example of the diurnal variation of surface energy fluxes (R n , H, LE, and G) is shown in Figure 4 for the Sierra Loma vineyard. Diurnal variation plots for the other vineyard study sites (Ripperdan 760, Ripperdan 720, and Barrelli) are shown in Appendices B.1, C.1 and D.1 The boxplot at each individual hour represents the seasonal variation (April to October) of surface fluxes due to changes in the irrigation scheduling and variations in weather conditions (wind speed, air temperature, vapor pressure deficit, and soil moisture) [61]. Overall, the behavior of R n diurnal variation is similar among the different sites, as the solar radiation is relatively consistent. As shown in Figure 4, R n values are negative in the nighttime and late evening. In the daytime, R n values vary, with maximum values of nearly 700 W/m 2 at solar noon depending on the daily solar radiation. The diurnal pattern of R n is almost systematic with a peak value appearing during midday, around 1200 standard time. The diurnal distribution of both H and LE exhibits a typical concave-down shape, with minimums in the early morning and late afternoon. The peak value appears near solar noon, between 1030 and 1330. Overnight, the H is almost negative, while the LE is approximately equal to zero, as the incoming solar radiation (R s ) value is 0 at night.
Although this is not always the case, the approximation may be acceptable for night [62].
In summertime, the LE value overnight is very small and rarely exceeds 5-10% of the daily total [63]. The study by Shapland et al. [64], which was conducted to estimate the ET over vineyards in California, assumed that the turbulent fluxes are zero during the night to avoid the uncertainty associated with the flux measurement. Another study by Tolk et al. [65], which aimed to quantify the nighttime evapotranspiration ET N -to-24-h ET (ET 24 ) of irrigated and dryland cotton in a semiarid climate, indicated that the ratio of ET N -to-ET 24 ranged from an average of 3% for a dryland cotton crop to around 7% for irrigated alfalfa. The contribution of ET N -to-ET 24 was the result of a relatively high nighttime vapor pressure deficit (VPD) and wind speed.
Flux observations indicated that the LE values were higher than the H across the different vineyards, as shown in Figure 4 and Appendices B.1, C.1 and D.1. These results stem from the fact that the vineyards are drip irrigated and, during most of the growing season, the cover crop is senescent, so ET is largely controlled by the vine canopy and, hence, mainly affected by the vine leaf stomatal conductance. The diurnal variation of soil heat flux (G) does not follow symmetric behavior, having a right skewness. As demonstrated in Figure 4 and Appendices B.1, C.1 and D.1, the G value is much lower than other energy fluxes (R n , H, and LE), where the peak does not persist across different vineyard sites. For overnight and later evening, G is negative and could yield values around −100 W/m 2 , as shown in Figure 4d at Sierra Loma vineyard, with similar results obtained at the other vineyard sites included in this study. In the energy balance, usually, the G value is estimated as a portion of R n (~0.35 R n ) for remote sensing ET models. Meanwhile, the G value is highly affected by the LAI, canopy architecture, row direction, and trellis design, as well as the incoming solar radiation. Reducing the canopy fractional cover results in an increased daytime soil heat flux (G), while increasing the areal coverage of vegetation leads to decreased soil heat flux and greater above-canopy latent heat fluxes, as long as there is ample root zone soil moisture to meet the atmospheric demand. Figure 5 shows the LE diurnal variation at each individual EC tower included in this study.
The boxplot at every hour represents the seasonal variation from April to October due to weather changes and irrigation scheduling. Overall, the general temporal trend of the LE has a shape that resembles solar radiation at different vineyard sites, with a peak value near solar noon, between 1030 and 1330. In early morning and overnight, the LE values were close to zero. Comparing the diurnal variation of LE at different vineyards, the Barrelli site had the lowest LE values. The Barrelli vineyard is located near the Pacific Coast shoreline, which brings cool maritime air that cools the warm interior valleys. The cool and moist air over Barrelli is associated with a decrease in the vapor pressure deficit (VPD) and more cloudiness, which causes a decrease in ET demand. In Sierra Loma and Ripperdan, the VPD and air temperature were higher than Barrelli, as both sites are exposed to a warm Mediterranean climate, which is characterized by abundant sunshine and a large day-to-night temperature difference and, therefore, increases the ET demand [66].
To compare the contribution of the ET at different hours to the daily ET, additional statistics were included, such as the ratio

of hourly ET (ET h )-to-daily ET (ET h /ET d ) and the ratio of ET h -to-maximum hourly ET (ET h(max) ) (ET h /ET h(max) ). An example of the diurnal variation of both ratios (ET h /ET d and ET h /ET h(max) ) at different phenological vine
stages (bloom, April-May; veraison, June-August; and post-harvest, September-October) is shown in Figure 6 for

. The general trends of ET h /ET d and ET h /ET h(max) resemble a Gaussian behavior, with peak values at solar noon. The
results also indicate that the vine phenological stage could affect both ratios in terms of the variation at each individual hour during the daytime. In the veraison stage, low variation was observed in the ET h /ET d and ET h /ET h(max) compared with the bloom and post-harvest stages. In the early growing season (April), the inter-row cover crop was at peak greenness, which was senesced by early June as the vines' leaves were fully developed (see the phenocam data at different study sites showing the different vine phenological stages: https://hrsl.ba.ars.usda.gov/awhite/CAM/ accessed 10 December 2020). This transition resulted in the main source of transpiration from the inter-rows, where the turbulent exchange was relatively suppressed to the vines with high potential coupled with the atmosphere [67]. On the other hand, the high variability observed in ET h /ET d and ET h /ET h(max) ratios in the time period between September and October were due to vines senescence and stress in the post-harvest stage due to a lack of irrigation and low atmospheric demand, where the daily ET decreased significantly. Moreover, as shown in Figure 6, the results of the ET h /ET d indicate that the major contribution of the daily ET came from the midday time between 1030 to 1530, which represents at least 65% of the daily total. However, in early morning (~0630 to 0930) and evening (~1630 to 1930), the value of ET h /ET d was low, which together represents 25-35% of the daily ET.  .2, B.4, B.5, C.4, C.5, D.4 and D.5 The analysis also considered all months segregated into three vine stages/periods (April-May, June-August, and September-October) to investigate how vine phenology could affect the accuracy of estimated daily ET due to the timing of both water uptake and growth. In general, the results indicate that the performance of the methods had different utility in computing an accurate daily ET at different vine canopy development and grapevine phenological stages (bloom, veraison, and post-harvest). As shown in Table 3 The results indicate that three methods (R s , GA, and EF) among the five daily ET models have the best performance (low RMSE and MAPE values and a high NSE value). The R s showed better agreement with the ground measurements among the other extrapolation approaches and was less sensitive to LE variation due to seasonal and climate differences, and particularly when using the one-time-of-day ET in the time window between 1030 and 1330. Using the R s approach, RMSE values were less than 0.4 mm/day, while the NSE value was higher than 0.9 for all vine stages (season). These results are also supported by a previous study conducted by Cammalleri et al. [29], which compared several upscaling daily ET methods using observations from flux towers within the United States and were evaluated over multiple seasonal cycles. They reported that using solar radiation (R s ) for converting the instantaneous to a daily ET value is more robust. Comparing the less accurate daily ET extrapolation techniques, the Sine method marginally outperformed the R n /R s approach in terms of moderate to high error within the time window (1030-1330) in the bloom and veraison stages, while in the post-harvest stage, the R n /R s method gave better results than the Sine approach. Using these approaches increased the RMSE, which yielded values above 0.65 mm/day, while the MAPE values were greater than 20% in the time window between 1030 and 1330 for all vine stages (season). This implies that the Sine and R n /R s techniques do not work properly for a daily ET estimate in vineyards.

Assessing the Instantaneous TSEB ET versus EC Measurements
As a first step toward evaluating the performance of the TSEB model, a comparison between the field observations from the EC and modeled fluxes using the TSEB and the sUAS (Table 1) at four different study sites are presented in Figure 7. A more detailed model Nassar et al. Page 15 performance assessment for each energy flux term is shown in Table 4. Surface fluxes were estimated from the sUAS based on the TSEB model, averaged over the EC footprint, and then compared against the measured fluxes. As shown in Figure 7, the estimated fluxes derived from the TSEB model generally align along However, the Barrelli site results indicate that the RMSE of the LE was less than the H. Overall, the higher values of the RMSE obtained for the LE are attributed mainly to the TSEB method for calculating the LE, which is solved as the residual component of the surface energy balance, LE = R n -H-G. Therefore, the uncertainties associated with the calculation of energy fluxes (R n , H, and G) within the TSEB method can adversely affect the estimation of the LE. Another potential uncertainty could be related to the no use of flux closure in the eddy covariance (EC) data. According to previous studies (e.g., Neale et al. 2012) [68], heat fluxes (H and LE) are acceptable when the RMSE ranged between 20 W/m 2 and 60 W/m 2 . This implies that the results of the H and LE obtained from the TSEB model across different vineyards were within an acceptable range and similar to prior studies [50].
The results for G indicate poor performance across the different vineyard sites, except for Ripperdan 720 vineyard, which had a MAPE of less than 25%. Part of these discrepancies between the modeled and observed G can be attributed to the assumption used in this study for calculating G, which is that as a portion of the soil net radiation (R ns ), G = 0.35 R ns .
This value was obtained based on a proposed method by Nieto et al. [49], which takes into consideration the diurnal variation of the G/R ns and found high scattering/uncertainty in the relationship, with an average value of 0.35 near solar noon. In this study, most of the flights were between 1000 and 1500, and at these time intervals around solar noon, the G/R ns fraction remained rather constant at ~0.35 (see Figure 4 in Nieto et al. (2019)) [49].
Therefore, for the sake of simplicity, and considering that the sinusoidal approach might be site-dependent, the constant fraction at 0.35 was used. This value is also broadly applied over a wide range of crops and environments. Meanwhile, vineyards are characterized by strong heterogeneity, which causes spatial and temporal variability in G values. According to Kustas et al. [69], the simple remote sensing methods for estimating G as a portion of R n have significant uncertainty due to temporal variability in the G/R n ratio.

Assessment of the Daily ET Extrapolation Approaches Using TSEB sUAS Results
The accuracy of the daily high-resolution ET from the TSEB depends largely on an accurate instantaneous ET estimate at the time of acquisition of the sUAS imagery, as well as the reliability of the approach used to scale up the TSEB-derived ET to a daily value. The five daily ET methods (EF, R s , R n /R s , GA, and Sine) were applied using the modeled energy fluxes derived from the TSEB and compared against the EC-derived daily value, ET d , calculated by integrating the daytime LE fluxes measured by EC towers. Table 5 lists the goodness-of-fit statistics between the modeled daily ET using sUAS data sets and the ground-based EC daily measurements at two time windows during the day: 1030-1330 and 1430-1630. Figure 8 shows the relationship between the modeled and measured fluxes.  Jackson et al. [30] and Colaizzi et al. [28], where scaling instantaneous ET to daily values showed better agreement when the measurement was taken within about 1-2 h of solar noon.
Although the results indicate that three (GA, EF, and R s ) out of the five methods for daily ET upscaling agree reasonably well with the ground-based measurements, the R s technique yielded better agreement at all three sites (Sierra Loma, Ripperdan 720, and Barrelli). This approach generated a robust ET d when a single remote sensing-based ET estimate was taken within 1-2 h of solar noon and provided a close agreement with the ground truth ET measurement. This result also aligns with the EC ET d analysis, which indicates that the R s approach has better statistical performance (see Table 3). Using the R s approach for all vineyards, the RMSE values were 0.45 mm/day, and the MAPE was 10%, while the R 2 was 0.88 for the time window of 1030-1330 (see Table 5, All Vineyards section).
These results agree with a previous study conducted by Wandera et al. [41], which showed that the R s -based approach was better for upscaling compared with the EF method. That study was carried out over 41 FLUXNET validation sites for two different times of day, including 1100 and 1330. Furthermore, the found results are also supported by Cammalleri et al. [29], when comparing different daily extrapolation methods. Cammalleri et al. [29] found that the incoming solar radiation (R s ) was the most robust method with the least error when using EC data collected at different flux tower sites within the United States and over Nassar et al. Page 17 multiple seasons. The R s approach for ET upscaling is highly recommended in situations where obtaining the daily net radiation is not possible [19] or, in some cases, where the modeled R n is overestimated/underestimated, which will adversely affect the EF ratio. On the other hand, the G is more difficult to estimate than the R s and R n , which could limit the accuracy of the EF method. This might explain why the R s method has a slightly higher agreement than the EF. Comparing the approaches with the lowest performance, the Sine method demonstrated the worst performance, with the largest RMSE and MAPE values and the lowest NSE value in the time window between 1030 and 1330. However, between 1430 and 1630, the results indicate that Sine performed slightly better than R n /R s . Still, the RMSE and MAPE values were high and the NSE and R 2 values were very low. The hypothesis is that the heterogeneity in the field, due to vine biomass, cover crop, and bare soil, has a larger impact on the R n /R s and Sine approaches than other methods.

Conclusions
The objective of this study was to assess existing methodologies for upscaling ET from single time-of-day information to daily estimates over commercial vineyards in California's Central Valley using EC flux measurements and the TSEB model with sUAS imagery. The extrapolation approaches included the evaporative fraction (EF), solar radiation (R s ), net radiation to incoming solar radiation (R n /R s ), the Gaussian (GA), and Sine technique. First, analysis was performed using flux observations collected at eight EC towers located at three vineyards in California's Central Valley: Sierra Loma, Ripperdan, and Barrelli. These sites are characterized by different climates, soils, vine variety, and trellis designs. The analysis also considered months of the growing season to coincide with three vine phenological stages (April-May (rapid vine growth, bloom/berry establishment), June-August (berry development/veraison), and September-October (harvest/post-harvest/vine senescence)) to investigate how vine phenology could affect the accuracy of the modeled daily ET due to timing of both water uptake and growth.

Acknowledgments:
We would like to thank the Aggieair Service Center team (Ian Gowing, Mark Winkelaar, and Shannon Syrstad) for their extraordinary support in this research, whose cooperation greatly improved the data collection and data processing, and the staff of Viticulture, Chemistry and Enology Division of E&J Gallo Winery for the assistance in the collection and processing of field data during GRAPEX IOPs. This project was partially financially supported     Comparison of instantaneous TSEB sUAS energy fluxes against EC measurements (without flux closure). The presented subplots include the available sUAS imagery, as described in Table 1.   Description of EC towers in vineyards that were part of this study.