Evaluation of the Operational Simplified Surface Energy Balance Model for Pastureland EvapotranspirationMapping and Drought Monitoring in North Central Kentucky

College of Agriculture, Communities, and the Environment, Kentucky State University, 400 East Main Street, Frankfort, KY 40601, USA United States Geological Survey, Earth Resources Observation and Science Center, North Central Climate Adaptation Science Center, Boulder, CO 80303, USA College of Business and Computational Sciences, Kentucky State University, 400 East Main Street, Frankfort, KY 40601, USA


Introduction
Evapotranspiration (ET), the turbulent transfer of water from the ground (evaporation; E) and plant surfaces (transpiration; T), is a key component of the hydrologic cycle, which is responsible for returning 60-90% of precipitation (P) to the atmosphere [1][2][3]. Previous studies (e.g., [4][5][6]) showed that ET exhibits considerable variation in space and time and, as a result, obtaining reliable and accurate measurements continues to present challenges [7]. ET is the second largest component in the terrestrial water balance (ET = P-R-S-I, where P = precipitation, R = runoff, S = soil storage, and I = infiltration) and is the single most important predictor of seasonal crop water consumptive use drought prevalence. ET estimation and drought monitoring are determined by the availability and proliferation of remotely sensed data. Furthermore, the need for improved methods for monitoring and modeling the water cycle has fueled interest in the rapid, widespread use of remote-based ET data.
Traditional land-based ET measurements have largely relied on in situ ground observation methods, such as lysimtery, energy balance, Bowen ratio (FR), eddy covariance (EC), scintillometry, and soil water balance [8][9][10][11][12]. e surface-based EC method for small areas of coverage has proven invaluable in providing direct measurement of turbulent water vapor fluxes between the ground and the atmosphere. For more than two decades, a network of systems [11] has provided the high-quality data needed to validate remotely sensed ET estimates. e EC method directly measures ET based on statistical covariance between vertical water-vapor flux and scalar concentration fluctuations [13]. Scaling up observational ET information is a major challenge of this method because it involves a complex and multidimensional process that varies with landscape, vegetation type, soil moisture, and crop management. It is also affected by underlying environmental factors [14,15]. In this regard, space-borne ET estimates [16][17][18] are ideal to circumvent some of the stringent requirements (e.g., sensor placement and calibration) and limitations (smaller footprint) of land-based ET observations [1,7].
For the past 18 years, simplified surface energy balance (SSEBop) has generated and issued ET maps for the continental United States on a monthly time interval at a 1 km spatial resolution scale. It also provides an efficient and timely data delivery system that is critical for operational decision-making by water resource planners and managers and hydraulic engineers. e model's use and application have successfully reported ET on different scales, i.e., watershed, basin, United States, and global scale [19,20]. In addition, the model has been used to generate ET a and anomaly maps over very large extents (the coterminous United States) using low spatial resolution (1 km) MODIS data.
One key attribute of satellite-based ET estimate models is model validation, which is defined as the process of determining the degree to which a model is an accurate representation of the real world from the perspective of the intended uses of the model [21]. SSEBop at basin and global scales is a good example of the successful application of satellite-based ET drought monitoring [22]. SSEBop increases our understanding of the global water cycle by mapping seasonal and year-to-year changes in ET across the landscape and providing complete spatial ET coverage, supplying input data aiding simulating irrigation requirements, and assessing trends of ET. Monthly maps of actual and anomaly ET and products are available publicly via the Famine Early Warning System (https://app.climateengine. org/climateEngine).
Few studies have evaluated SSEBop across wide geographical areas (see noted exceptions [19]) and less frequently on ecosystem scales. erefore, the goal of this study was to evaluate the performance of SSEBop in estimating ET from a pasture agroecosystem under the climate and soil conditions of north central Kentucky. Although the SSEBop model has been in use for 18 years, validation studies on its use over varied ecosystems are limited. e objectives of this study were (i) to evaluate SSEBop against EC data and its comparison to reference evapotranspiration (ET ref ) as its upper boundary and (ii) to assess the application of SSEBop for drought monitoring and identifying drought years/ seasons using rainfall as a surrogate variable.

Site Microclimate.
e study site is located at the Harold R. Benson Research and Demonstration Farm, 12 km southwest of the main campus of Kentucky State University, Frankfort, Kentucky (38.11°N, 84.88°W; 242 m above mean sea level). e study period had marked differences in total precipitation among the years, ranging from the lowest (1012 mm·yr −1 ) in 2016 to 1855 mm·yr −1 in 2018. 2018 was a 150-year record, which gave 697 mm (60%) more rain than the historical (1980-2010) average. Based on the 30-year average of 1159 mm for the region, the observation years can be classified as dry in 2016 (1012 mm), average in 2017 (1219 mm), and wet in 2018 (1855 mm). Air temperature (T a ) from 2016 through 2018 at the site showed no major departure relative to the thirty-year (1981-2010) average of 12.8°C. Averaged over the three years, monthly T a varied between −1°C in January and 24.2°C in July, with annual averages of 13.5, 13.8, and 13.3°C in 2016, 2017, and 2018, respectively. e only exception was January 2018, which was the coldest month (−5.0°C; Figure 1). e soil at the site is in the well-drained McAfee silt loam soil series, commonly found in a range of 6 to 12 percent slopes, slightly acidic in pH, and moderately drained with high permeability [23].

Pasture Management.
Kentucky's pasture agroecosystem is extensive and largely rain-fed. Approximately 25% (2.5 million acres) of the total 13.4 million land area is used for pasture agroecosystems.
e study area at the Harold R. Benson Research and Demonstration Farm (pasture) was seeded in 2005 and consisted of mixed grasses, predominantly fescue (Festuca arundinacea), as well as Johnson grass (Sorghum halepense), green foxtail (Setaria viridis), hemlock (Conium maculatum), common chicory (Cichorium intybus), and red clover (Trifolium pratense). During the observation years, the same mix of grass species was used and meat goats rotationally and intermittently grazed the pasture for a maximum of two weeks beginning in late June through late summer (August). Grazing was allowed until the stub height reached 0.1 m.

Latent and Sensible Heat Measurements.
Vertical exchange of H 2 O vapor, sensible heat, and latent heat were measured using the EC technique. e EC flux tower is located on a pasture with a sufficiently wide and horizontal fetch of at least 200-300 m in the major wind (westerly and southwesterly) direction. e EC tower was instrumented with a fast-response 3D sonic anemometer (WindMaster Pro; Gill Instruments, Lymington, Hampshire, UK) that measures wind speed (m s 1 ) and sonic air temperature (T s, °C ). e wind sensor is located 2.82 m above the ground level. Water vapor concentration was measured in situ using a closed path infrared CO 2 /H 2 O gas analyzer (LI-7200; LI-COR Inc., Lincoln, NE, USA). e two sensors are installed such that they are within the mixed surface layer, i.e., where z h is the measurement height (m), z m is the surface roughness length (m), and d is the zero plane displacement height (m; the theoretical height above the ground where the logarithmic wind profile vanishes to zero). e measurement height z h was nearly 9 times the average z h of the grass canopy height (0.30 m at peak developmental stage). e turbulent sensible (H) heat flux was computed as follows: where ρ a is dry air density (kg m −3 ) and C p is the specific heat capacity of the air at constant pressure (1004 J kg −1 K −1 ). e overbar indicates the averaging period, in this case 30 min. Similarly, latent heat (LE, W −2 ) was calculated as where λ is the latent heat of vaporization (2.45 MJ·kg −1 ) and q ′ is the fluctuation about the mean of density of water vapor (kg m −3 ), ρ w , in air (1000 kg m −3 ), and w ′ and q ' are the turbulent vertical wind speed (m/s) and water vapor density (kg m −3 ), respectively. Overbar indicates the averaging period, which in this case was 30 min. Daily ET (mm/day) was estimated using the following relationship: Daily LE totals were calculated by integrating the 30minute observations. e energy balance at a surface can be thought of as accounting of energy gains and losses within a specified time interval (e.g., 30 min, daily, monthly, seasonally, and annually). We assessed the relative energy balance by plotting the sum of H and LE against the available energy (i.e., R n -G) and the mathematical expression for a complete energy closure assumes where a and y are the slope and intercept of the linear regression, respectively, and G is ground heat flux. Unit for R n , G, LE, and H terms is W·m −2 . e last term, ε, is the error term, which includes both instruments and random errors. G was directly measured using a soil heat plate at a 10-cm depth.

Penman-Monteith (P-M) Method.
Reference evapotranspiration (ET ref , mm/d) measures the rate at which water vapor is released from actively growing vegetation (typically, grass height of vegetation is assumed to be 0.12 m) unlimited by water with a fixed crop surface resistance of 70 s·m −1 and an albedo of 0.23 [25,26]. e FAO-56 (Food and Agriculture Organization) Penman-Monteith (P-M) method is currently considered as the standardized method of computing reference ET from an assumed uniform grass reference surface worldwide [26]. e method computes reference evapotranspiration as daily totals (mm day −1 ) using the following equations: where ∆ is the slope of the saturation vapor pressure function (kPa/°C), R n is the daily average net radiation (MJ m −2 d −1 ), G is the daily average ground heat flux (MJ m −2 d −1 ), c is the psychometric constant (kPa/°C), T is the daily average temperature (°C), u 2 is horizontal daily average wind speed (m s −1 ), e s is the vapor pressure of the air at saturation (kPa), and e a is the actual vapor pressure (kPa).

Operational Simplified Surface Energy (SSEBop) Model.
A SSEBop model was used in this study as described by Senay et al. e model uses two data streams acquired from MODIS and Global Data Assimilation Systems (GDAS) which include elevation (z), temperature correction coefficient (c), land surface (T s ) and air (T a ) temperatures, temperature difference (dT), clear sky net radiation (R n ), reference ET (ET o ), aerodynamic resistance (r ah ), albedo (α), and Normalized Difference Vegetation Index (NDVI) as input variables [22]. e SSEBop directly solves for ET without explicitly resolving the other energy terms. A noteworthy feature of the model is that, for any given pixel being analyzed, the model uses two boundary surface temperature conditions (the hot and cold pixel approach) to scale LST values linearly between 0 and max ET fraction. Such codification would then permit the use of the linearization assumption by taking into account the fact that latent heat (LE) varies according to a linear function of the "hot" and "cold" pixels. Hence, the variable dT is assumed to remain constant from year to year and is typically calculated under clear sky conditions for each pixel and is derived as where R n is the daily cloud-free net radiation (J m −2 s −1 � W·m −2 ), r ah is aerodynamic resistance to sensible heat transfer from bare and dry surface with constant value of 110 s m −1 [22], and p a and C p are density of air (kg m −3 ) and specific heat of air at constant pressure (1.013 kJ·kg −1 ·k −1 ), respectively. In this case, cold temperature pixels (T c � C * T max ) are those from which there is little or no sensible heat flux, while hot temperature pixels (T h � T cold + dT) are described as those with little or no latent heat flux (T c ). e SSEBop does not solve all energy balance components; rather it sets limited boundary conditions based on clear-sky net radiation balance principles [20,22] and the only data required are surface temperature (T s , K), air temperature (T a , K), and potential ET. With this simplification, the model computes actual evapotranspiration (ET a ) as a fraction of the ET o ; that is, where k is a scaling coefficient applied for ET o to achieve a theoretical maximum set for an optimally grown alfalfa. e model uses inputs to produce anomaly maps, calculated as the arithmetic median ET deviation of a target year for measuring drought occurrence, which has a great deal of interest for geographical areas prone to drying. For a detailed and full description of the model, its application, and performance in estimating ET, please see [22,27].

Monthly SSEBop Anomaly.
To develop ET-modeled anomaly maps for the site and surrounding regions for each area considered in the study, the following steps were followed: (i) monthly averages were calculated; (ii) individual monthly anomaly of respective years was then calculated as percent difference of each year from the previous 18 years' average value using where ET m is the respective monthly percent anomaly, ET a is the monthly climatology parameter, and ET L is the longterm ET average.

Statistical
values, and b is an intercept, was used to assess model fitness. e monthly and yearly estimates of SSEBop ET modeled values were quantitatively evaluated by percent bias (PBIAS), root mean square (RMSE), and coefficient of determination (R 2 ), which were defined as where N is the number of observations and ET o and ET m are observed (measured) and modeled ET, respectively.   Figure 2). Measured monthly rates Estimates of monthly satellite-derived ET by SSEBop ranged from 0 to 131 mm·month −1 (Table 1)

Assessment of SSEBop Performance.
To evaluate the accuracy of the modeled ET estimates at monthly scale, performance indices including coefficient of correlation (R 2 ), root mean square error (RMES) and mean absolute error (MAE), and percent bias (PBIAS) were used (Table 2). Broadly, SSEBop showed a tendency to overestimate monthly ET flux, exhibiting BIAS, RMSE, MAPE, MRE, and R 2 of 5.47, 21.5, 30.9, 10.5, and 0.87, respectively.
Although the model (SSEBop) consistently tracked measured ET m , with higher than measured flux values, moderate mismatches existed on monthly scales (Figure 3). Differences between the modeled and measured values were greatly pronounced (under-or overestimated) at two critical phases of the ET cycle during the peak ET summer months and winter months but there was a satisfactory agreement during fall and spring months. As expected, both measured and modeled values followed the natural seasonal pattern of ET, consistent with the cooling (winter, low ET) and warming (summer, high ET). Obviously, the model exhibited a level of bias, but it did track the expected ET trend as a whole. e RMSE test set was within 150.64, 100, and 100 mm of the measured ET for 2016, 2017, and 2018, respectively. Substantial variations existed among months (within a year) and among years between P-M and flux: BIAS � 49.5, RMES � 56.7, MAPE � 95.5, MRE � 94.9, R 2 , and CV � 0.39. As shown in Figure 3, SSEBop tended to overestimate ET particularly during the summer months.
As expected, the performance of the P-M physical model at this pasture site was not as good as the performance of the SSEBop model at estimating monthly ET (Table 2). Figure 4 shows the scatter plots between SSEBop and measured ET values with the 95% confidence intervals (R 2 � 0.90). Regression analysis indicated that the slope of the regression line did not statistically differ (P < 0.0001) from 0, suggesting good performances of the remote SSEBop model in estimating ET. Underpredictions suggested that the SSEBop model provided relatively conservative (lower ranges) estimates during the winter.
is consistent pattern (i.e., overestimates at the high ET end and underestimates at low ET) may be associated with the inability of the model to fully capture ET fluxes from slowly growing grass during the winter months. Although the underestimation in winter could be offset by overestimation in other seasons (particularly during the summer growing season), the annual ET models values tended to be clustered around 673 mm·year −1 versus the measured value of 609 mm·year −1 . Figure 5 presents annual ET (estimated) monthly anomaly maps for the period between 2016

Discussion
In this study, we evaluated SSEBop's ET monthly estimates against observed data (from an eddy covariance flux station) over a pastureland site in north central Kentucky. Overall, the SSEBop monthly estimates were highly correlated and tracked well with seasonal data. We found month-to-month bias (∼5 mm/month) that led to a pattern of overestimation in the warmer months or underestimation in the cooler months. Such bias could likely impact the magnitude of the estimated ET. On the other hand, a high correlation (e.g., low R 2 ) between modeled and true concentrations is indicative of a high degree of association. We found that while the model captures the temporal variability of ET, month-tomonth differences were evident. Notable was the fact that SSEBop quantitative ET estimates tend to underestimate ET during cooler months. is probably could be attributed to strong cloud attenuation effects of the main winter rainy months. Such underestimation of ET a occurs when the ET fraction (ET f � ET a /kET o ) is small [19]. Noted differences between the observed and estimated values during the wet and dry months may stem from the key assumption the model makes in that differences in land surface temperature (LST) over a homogeneous landscape are primarily due to differences in vegetation and functional differences in water use.
e model also ignores the contribution of albedo (canopy reflectance) and ground heat flux as part of the ET estimation.
e close correspondence between SSEBop estimates demonstrates the promise of satellite-sensed ET approaches to remotely monitor crop water use and drought conditions at plot scales in contrasting precipitation years. e SSEBop model ET estimates have previously been validated using networked eddy covariance data flux station at watershed and regional scales, mainly in western United States [19] and to a lesser degree in the southeastern US.
Whereas space-based estimates, such as the MODIS and SSEBop, are the most promising with regard to spatial and temporal resolution of ET products, retrieval quality is strongly tied to the algorithm's ability to detect and screen cloud interference, reduce the attenuation and scattering effects arising from cloud shadow [27], and hence improve the values of the interpolated pixels [28]. ese and the next generation of future advances in retrieval procedures will enhance space-based realistic descriptions of the spatial distribution of ET.
is contrasting anomaly indicates the dependency of ET on precipitation and on soil moisture availability. In the past, SSEBop anomaly maps showed and clearly captured two major drought events that engulfed much of the southeastern United States during the 2007 and 2012 extreme droughts. Drought, as a consequence of soil moisture deficiency, causes an increase in ET anomalies by reducing evaporative (latent heat) fluxes. One of the advantages of estimating ET remotely is to be able to provide anomaly maps for decision-makers in response to emergencies, such

Conclusion
For this study, we examined and evaluated the performance of SSEBop model using direct, in situ ET eddy flux measurements at a managed pastureland in north central Kentucky between 2016 and 2018. e model was chosen because it provides data products at spatial and temporal scales suitable for comparison with ET data. SSEBop overand underestimated ET with a mean difference of 5 mm·month −1 averaged over the three years (2016-2018). e comparison between direct flux ET measurement and modeled ET allowed seasonal characterization of EC trends at this managed pasture site. Seasonality was captured well; however, the model underestimated ET during winter months and overestimated it during summer months. ese data showed the following: (i) SSEBop anomaly maps provide relevant drought information that can be used by various practitioners, and it offers tools for managing and mitigating drought, which helps guide mitigation strategies. (ii) e model's poor performance may not necessarily indicate the model's inaccuracy. One possible explanation for this is mismatched data on spatial scales on 1 point (100 m fetch area) versus 1000 m fetch area. (iii) e SSEBop at the pasture site demonstrated the model's efficacy, which enforces its value as a reliable proxy for precipitation and a valuable indicator of ET anomalies on a month-to-month scale; however, the model's validation in areas characterized by contrasting seasonal and year-to-year precipitation requires further testing and additional research.
is study supports the model's use as proxy variable for precipitation and as a viable method for monitoring and assessing ground moisture status (i.e., drought). e SSEBop model is less complex than other remote-based ET products (e.g., METRIC and MODIS) and it provides sufficient information for reliably assessing and monitoring droughts on shorter time scales (e.g., daily to monthly). [29].  Figure 6: Monthly ETanomaly (defined here as P-ET) during 2016-2018, where P is precipitation. e monthly P-ET variation is controlled by precipitation exhibiting positive (P > ET) and deficit (P < ET). Precipitation data were obtained from the nearby (<2 miles) Kentucky Mesonet weather station.

Data Availability
e data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest
e authors declare that there are no conflicts of interest regarding the publication of this paper.