Evapotranspiration model comparison and an estimate of field scale Miscanthus canopy precipitation interception

The bioenergy crop Miscanthus × giganteus has a high water demand to quickly increase biomass with rapid canopy closure and effective rainfall interception, traits that are likely to impact on hydrology in land use change. Evapotranspiration (ET, the combination of plant and ground surface transpiration and evaporation) forms an important part of the water balance, and few ET models have been tested with Miscanthus. Therefore, this study uses field measurements to determine the most accurate ET model and to establish the interception of precipitation by the canopy (Ci). Daily ET estimates from 2012 to 2016 using the Hargreaves–Samani, Priestley–Taylor, Granger–Gray, and Penman–Monteith (short grass) models were calculated using data from a weather station situated in a 6 ha Miscanthus crop. Results from these models were compared to data from on‐site eddy covariance (EC) instrumentation to determine accuracy and calculate the crop coefficient (Kc) model parameter. Ci was measured from June 2016 to March 2017 using stem‐flow and through‐flow gauges within the crop and rain gauges outside the crop. The closest estimated ET to the EC data was the Penman‐Monteith (short grass) model. The Kc values proposed are 0.63 for the early season (March and April), 0.85 for the main growing season (May to September), 1.57 for the late growing season (October and November), and 1.12 over the winter (December to February). These more accurate Kc values will enable better ET estimates with the use of the Penman‐Monteith (short grass) model improving estimates of potential yields and hydrological impacts of land use change. Ci was 24% and remained high during the autumn and winter thereby sustaining significant levels of canopy evaporation and suggesting benefits for winter flood mitigation.


Introduction
The planting of perennial bioenergy crops is expected to grow following an increased focus on renewable energy generation in order to meet global greenhouse gas Correspondence: Amanda J. Holder, tel. +44 1970823197, fax +44 1970 emission targets (IPCC, 2014;Energy Technologies Institute, 2015). Adaptations to changes in climate are also being considered as it is now anticipated that some of the predicted impacts of climate change are unavoidable (IPCC, 2007(IPCC, , 2014. In the UK, repeated flooding events have stimulated interest in identifying mitigation strategies and have highlighted the potential role for farmland and upland areas for buffering against high rainfall (Marshall et al., 2009;Christen & Dalgaard, 2013;Wynne-Jones, 2016). This need is leading to an interest in finding commercially viable climate change resilient crops (Environment Agency, 2015) that can be located within these landscapes to provide wide-ranging environmental benefits. Miscanthus 9 giganteus Greef et Deu (Greef & Deuter, 1993) is a low input biomass feedstock that, beyond simply burning in power stations, is also marketable in the biorefining industry (producing liquid fuels and chemicals) and as animal bedding (Brosse et al., 2012;Van Weyenberg et al., 2015).
The current commercial clone, Miscanthus 9 giganteus (hereafter Miscanthus), is a tall-growing (up to~3 m) sterile perennial grass hybrid with an efficient C 4 photosynthetic pathway. Requiring few agricultural inputs, it has the potential to grow on poorer soils (Lewandowski et al., 2000;Hastings et al., 2008;Lovett et al., 2009;Cadoux et al., 2012). Miscanthus has limited stomatal control, a high water demand used to quickly increase biomass, and rapid canopy closure with a large leaf area index providing effective rainfall interception (Clifton-Brown et al., 2002;Joo et al., 2017). The site-specific impacts of land use change to Miscanthus on water balances vary depending on factors including altitude, climate, and stage of crop maturity (Dunkerley, 2000;Stephens et al., 2001a). Increased planting of Miscanthus could potentially increase evapotranspiration (ET) and affect ecosystem water dynamics through impacts on boundary layer temperatures, humidity, and solar radiation to the ground (Hickman et al., 2010;Milner et al., 2016). However, these traits may also reduce flooding, soil erosion, and nutrient run-off. Information regarding these potential impacts is vital for accurate modelling of land use change scenarios to fully inform policymakers.
ET is mainly estimated using models due to the cost of equipment and time-consuming nature of field studies. A number of models can be used to calculate estimates of actual ET (ET a , evaporation from all surfaces under natural conditions), potential ET (ET p , the ET rate where there is no shortfall in soil water for vegetation use), and reference crop ET (ET o , ET p from a specific reference crop type (e.g. short grass) with no water shortage) (Allen et al., 1998;Xu & Chen, 2005;McMahon et al., 2013). Different models require varying levels of data and have different approaches to the basis of the calculations, and the impacts of these differences for the prediction of ET rates for a novel crop like Miscanthus are not clear. The Hargreaves-Samani (HS, Hargreaves & Samani, 1985) model is based on air temperature, Priestley-Taylor (PT, Priestley & Taylor, 1972) on solar radiation, and the Granger-Gray (GG, Granger & Gray, 1989) model uses a complementary relationship where land and atmosphere feedbacks lead to a mutual dependency between ET a and ET p (Bouchet, 1963;Morton, 1965). The simplified Penman-Monteith model (PMgrass, Allen et al., 1998) uses net incoming radiation and atmospheric and surface resistance terms to provide an estimate of ET o for a reference short green crop. PMgrass results can be further adapted to provide estimates of ET for a specific crop type (ET c. ) with the use of a crop coefficient value (K c ) (Allen et al., 1998).
To our knowledge, there are no published studies comparing different ET models with a Miscanthus crop. The PMgrass model in conjunction with K c values has been used for Miscanthus plants by Beale et al. (1999) in a water use efficiency study, and by Triana et al. (2015) and Liu et al. (2014) in water balance studies. K c values reported for Miscanthus range from 0.31 to 1.20 (Beale et al., 1999;Stephens et al., 2001b;Triana et al., 2015), based on data obtained from locations with different climates, and do not always include the full Miscanthus growing season. Hydrology models incorporating ET have also been used to model land use change to Miscanthus: Stephens et al. (2001b) and Borek et al. (2010) used the WaSim model (calculating ET using PMgrass with the option of K c values); Finch et al. (2004) the Met. Office Surface Energy Scheme (MOSES) model; Vanloocke et al. (2010) the Agro-IBIS model; and Cibin et al. (2015) the Soil & Water Assessment Tool (SWAT) model. The SWAT model can calculate ET p via the Penman-Monteith equation (Monteith, 1965), PT, or HS methods (SWAT, n.d.). Only Stephens et al. (2001a,b) and Finch et al. (2004) model hydrology for Miscanthus in a UK climate type. Simulations by Stephens et al. (2001a) show reductions in run-off and groundwater recharge under Miscanthus compared to grass, whereas simulations by Finch et al. (2004) show Miscanthus having lower water use than grass, whilst pointing out that measurements over a full year are required to confirm this. More crop-specific measurements for energy grasses are required to provide accurate estimates of ET and validate model predictions (Stephens et al., 2001a;Finch et al., 2004;Vanloocke et al., 2010;McCalmont et al., 2017a). Of the few studies that have measured ET for Miscanthus, Finch et al. (2004) recorded growing season highs of~5 mm day À1 with eddy covariance (EC) equipment, Hickman et al. (2010) measured highs of ~7 mm day À1 using a residual energy balance approach and Triana et al. (2015) report a maximum 11 mm day À1 using lysimeters.
Knowledge of the accuracy of commonly used ET formulae is not only of use in modelling the hydrological impacts of land use change but will also be of benefit in the modelling of potential yields and other environmental impacts such as greenhouse gas emissions where models require ET p as an input (Richter et al., 2008;Hastings et al., 2009;Dondini et al., 2016).
In addition to ET, canopy precipitation interception (C i ) is an important metric in understanding winter evaporation and soil moisture recharge. To date, there have been few studies relating to tall grass energy crops and interception, with only one UK Miscanthus study. Finch & Riche (2010) reported measured Miscanthus C i of 24%. However, measurements took place in small trial plots and the effect cannot be assumed to be the same at field scale as surface resistance becomes a smaller factor in water vapour diffusion to the atmosphere with increasing canopy cover forming a uniform layer (Monteith & Unsworth, 2008;Finch & Riche, 2010).
This study aims to: • Determine the most accurate ET model compared to EC ET data (ET EC ) for use with Miscanthus.
• Establish C i in a commercial-scale Miscanthus plantation under the UK climate conditions.
To achieve this, four base ET models, with further adjustments taking account of soil moisture status, were used to compare to ET EC at a commercial-scale mature Miscanthus plantation in Wales, UK, where in situ weather station and EC equipment have been recording since land use conversion from grassland in 2012. A field study was set up in the plantation to record C i from June 2016 to March 2017.

Site description
Field experiments, EC measurements, and weather data collection took place at a 6 ha plantation of Miscanthus located in Aberystwyth, Wales (52°25 0 17″ N 4°04 0 14″ W) (Fig. 1). The site elevation is~110 m a.s.l. with coastal cliffs~0.5 miles west of the field boundary. It is predominantly flat with a slight slope (7°) to the south. The soil, a mixture of clay loam and sandy/ silty clay loam, is formed over Denbigh series bedrock. The field capacity is 0.38 m 3 m À3 , as shown in Saxton & Rawls (2006) and confirmed from in situ soil moisture probes (29 CS616 Campbell Scientific (CSI), Logan, UT, USA, soil water content reflectometer installed at 25 cm depth). Permanent wilting point is 0.22 m 3 m À3 (Saxton & Rawls, 2006). The field was converted from semi-improved grass pasture to Miscanthus in April 2012.

Meteorological data
EC data were recorded by two open-path systems (EC150/ CSAT3A OPEC system, CSI, Logan, UT, USA) located at two towers ( Fig. 1) covering the central and most level 3.9 ha portion of the cropped area. Sensors were raised during the growing season to maintain a height of 2 m above the canopy. The systems included a sonic anemometer (CSAT-3A, CSI), infrared gas analyser (EC150, CSI), and air temperature (Ta,°C) and relative humidity (Rh, %) probes (HMP155A, CSI) recording to Fig. 1 Map showing the outline of the 6 ha (approx.) Miscanthus field with the cropped area, sampling points, and meteorological and atmospheric measuring equipment locations marked. data loggers (CR3000, CSI) at 20 Hz and processed to 30 min averages using EddyPro software (EddyPro version 4.2.0, LI-COR bioscience, Lincoln, NE, USA). Data were quality controlled and gap-filled as described in McCalmont et al. (2017b). Latent heat flux (LE) values surrounding gap-filled values were further checked for abnormally high figures caused by wet instrumentation and were replaced using averages of nearby nongap-filled values. ET figures were determined from LE using Eqn (1) and were converted to mm day À1 .
where ET EC is the ET flux (mm h À1 ), LE is the 30 min latent heat flux after corrections and gap filling (Wm 2 ), and k is the latent heat of vaporization constant. The value used for the hourly rate constant was 690.42 Wm 2 (2.4855 MJ m À2 ), as determined by the EddyPro software. Excepting Rh and Ta (measured at each eddy covariance tower) meteorological data were collected from a station located in the centre of the field (Fig. 1) and logged in 30 minute intervals using a CR1000 (CSI) data logger. Precipitation (mm) was recorded using a tipping bucket rain gauge (52203, R.M. Young, Ann Arbor, MI, USA). Photosynthetic photon flux density (lmol m À2 s À1 ) was measured with a SKP215 Photosynthetically Active Radiation (PAR) Quantum sensor (Skye systems, Llandrindod Wells, UK). Wind speed (ms À1 ) and direction (from north) were collected using a 05013 wind monitor (R.M. Young). Small gaps in the weather data (<1% overall) were filled from a nearby weather station.
where GG is the Granger-Gray ET model (mm day À1 ), Gg is based on Eqns (3) and (4), G is the soil heat flux (MJ m À2 day À1 ), c the psychrometric constant (kPa°C À1 ), R n the net daily radiation (MJ m À2 day À1 ), k the latent heat of vaporization (MJ Kg À1 ), and Ea the drying power of the air calculated from Eqn (5).
where Ea is calculated using Eqn (5).
where f(u) is the wind function shown in Eqn (6), v*a the daily saturation vapour pressure (kPa) and va the mean daily actual vapour pressure (kPa).
where PT is the Priestley-Taylor ET model (mm day À1 ), aPT is a constant of 1.26 for advection-free saturated surfaces, Δ is the slope of vapour pressure curve (kPa°C À1 ), c the psychrometric constant (kPa°C À1 ), R n the net daily radiation (MJ m À2 day À1 ), k the latent heat of vaporization (MJ Kg À1 ), and G the soil heat flux (MJ m À2 day À1 ). The Hargreaves-Samani (HS) formula (McMahon et al., 2013) calculates reference ET for a short grass crop with no water shortage (Eqn 8).
where HS is the Hargreaves-Samani ET model (mm day À1 ), C HS is a coefficient based on Eqn (9), R a is extraterrestrial radiation (MJ m À2 day À1 ), k the latent heat of vaporization (MJ Kg À1 ), T max and T min the maximum and minimum daily temperatures (°C), and T a the average daily temperature (°C).
where C HS is the Hargreaves-Samani coefficient and T max and T min are the maximum and minimum daily temperatures (°C). The Penman-Monteith (PMgrass) formula (Allen et al., 1998) calculates reference ET for a short grass crop with no water shortage (Eqn 10) where Δ is the slope of the vapour pressure curve (kPa°C À1 ), R n the net radiation (MJ m À2 day À1 ), G the soil heat flux (MJ m À2 day À1 ), c the psychrometric constant (kPa°C À1 ), T a the mean daily air temperature (°C), u 2 the average daily wind speed (at 2 m) (m s À1 ), v*a the daily saturation vapour pressure (kPa), and va the mean daily actual vapour pressure (kPa). The inputs required for the models along with the values used for the constants are shown in Table 1. Global radiation, also known as solar radiation (Rs) (Allen et al., 1998), was calculated as 29 PAR (Monteith & Unsworth, 2008) and converted to MJ m 2 day À1 .

Adjustment from ET p to ET a
The PT and HS models were adjusted daily to provide a prediction of ET a via the use of a soil moisture function (Mintz & Walker, 1993;Dingman, 2002;Xu & Chen, 2005) which reduces ET estimates as soil water becomes depleted to critical levels. The relationship between ET p , precipitation, the soil moisture function (F), and ET a is as follows: if ET p > precipitation then ET a = ET p 9 F if ET p = precipitation then ET a = ET p if ET p < precipitation then ET a = ET p The soil moisture function is calculated from a basic soil water balance using Eqns (11-13).
where F is the soil moisture function restricted to between 0 and 1, W the soil moisture estimated from Eqn (12), and W* the soil storage capacity calculated from Eqn (13).
where W t is the soil moisture (mm) restricted to between 0 and the field capacity (96 mm, from Eqn 13), W tÀ1 the soil moisture (mm) from the previous day, P t the precipitation (mm), and ET pt the calculated ET p (mm).
where W* is the site-specific soil moisture storage capacity (mm), 1000 the conversion to mm, 0.38 the site-specific field capacity (m 3 m À2 ), 0.22 the site-specific wilting point (m 3 m À2 ), and 0.60 the site-specific approximate soil/rooting depth (m). Following the method in Allen et al. (1998), the PMgrass results were adjusted with a water stress coefficient (K s ) and a crop coefficient (K c ) to provide an estimate of ET a , as shown in Eqn (14). The K c values for sugarcane, also a C 4 plant with tall stems and a large leaf area index, were used. Sugarcane published K c values are 0.40 for the early growth stage, 1.25 for the main growing season, and 0.75 for the late season (Allen et al., 1998). 0.75 was also used for the winter season.
where ET a is the PMgrass results adjusted for the soil moisture depletion and crop type, K s the water stress coefficient calculated from Eqns (15-17), K c the crop-specific coefficient, and ET o the PMgrass result.
where K s is the water stress coefficient (between 0 and 1), TAW the total available water (mm) calculated in the same way as W* (Eqn 13), Dr the root zone moisture depletion calculated from Eqn (16), and RAW the readily available water (mm) calculated from Eqn (17).
where Dr is the root zone depletion (mm), Dr tÀ1 the water content in the root zone on the previous day (mm), P t the precipitation (mm), and ET ct the crop evapotranspiration (mm).
where RAW is the readily available water (mm), TAW the total available water (mm) calculated in Eqn (13), and p the fraction of TAW that the plant can extract without suffering water stress applied on a seasonal basis [values of p used were 0.76 for the early and late season, 0.67 for the main season, and 0.77 for the winterbased on the values and adjustments given for sugarcane in Allen et al. (1998)].

Miscanthus crop coefficient (K c )
To calculate the Miscanthus-specific K c , ET EC and PMgrass daily ET rates were divided to approximately correspond to the relevant stages of plant growth ( Table 2). The K c value for each season was calculated using Eqn (18), and the value multiplied by the results of PMgrass to provide the Penman-Monteith K c (PMK c ) estimated ET.
where K c is the crop coefficient, ET EC the mean daily EC calculated evapotranspiration for the season, and PM the mean daily evapotranspiration calculated by the Penman-Monteith (short grass) model.

Canopy precipitation interception
Measurements took place from 23rd June 2016 until 13th March 2017 using methods similar to those used by Riche & Christian (2001). Eight sampling locations (2 m 2 ) within the cropped area ( Fig. 1) were selected by stratified random sampling using a preconversion topsoil moisture map to take account of wetter and drier areas. Three stem-flow and three through-fall gauges (Fig. 2a-c) were randomly placed within each sampling location. Two further sampling areas to collect gross precipitation were located outside the crop canopyone to the north and the other in a clearing along the centre track (Fig. 1). A monthly count of the number of mature stems in 1 m 2 along with the average stem thickness was carried out in an area immediately adjoining the sampling locations. Gauges were checked approximately twice weekly with measurements taken in dry weather when water levels were high enough in the gauges for accurate measurement with the use of a graduated cylinder. After the first few weeks of data collection, an error level of less than or equal to 4.75% was calculated from the sums of squares and coefficient of variation using the means of the eight zones within the crop (Raghunath, 2006).
The C i was taken to be the difference between the gross precipitation recorded outside the crop and the net precipitation recorded within the crop (Eqn 19).
where C i is the interception (mm), GP the measured gross precipitation (mm), TF the measured through-fall (mm), and SF the measured stem-flow (mm). For each recording event, the amount of precipitation collected in the through-fall bottles was converted into a depth measurement based on the area of the funnel. Gross rainfall was collected and converted to a depth measurement in the same way as the through-fall using the four gauges located in each of the two locations outside the crop. For each recording event, stem-flow amounts were adjusted for the average size of the stem and reduced by the amount collected by the closest through-fall bottle to account for through-fall that would also have been collected by the funnel (Eqn 20). Total stem-flow was then calculated as a mean depth measurement (Eqn 21). During measurement, 19 samples of a total of 2856 (2.62%) were rejected as a result of broken stems or damage to the collecting system.
where SFA is the stem-flow amount (ml), SFC the amount collected in the stem-flow bottle (ml), TFC the amount collected in the closest through-fall bottle (ml), and SP the percentage of the funnel/overflow bottle area taken up by the stem (%).
where SFD is the total stem-flow depth (mm), SFA the mean stem-flow amount (calculated from the mean stem-flow amount in each sampling area) (ml), S the mean number of stems in 1 m 2 , 1000 the conversion to mm 3 , and SA the surface area of the stem count (mm 2 ).

Statistics
Statistics were carried out using R version 3.2.3 (R Core Team, 2015). Model residual plots were checked for the appropriateness of linear regression, and the linear model function was used to obtain the R 2 values (with ET EC as the independent variable). The seasonal daily means, standard deviation, and standard error of the mean were calculated for all the daily ET results. The HydroGof (Zambrano-Bigiarini, 2017) R package was used to calculate the mean absolute error (MAE), Root Mean Square Error (RMSE), modified Index of Agreement (md) (return of between 0 and 1 where 1 = a perfect match), and the modified Nash Sutcliffe Efficiency (mNSE) (return of between -infinity and 1 where 1 = a perfect match and (a) (b) (c) Fig. 2 (a) Through-fall within the crop and precipitation outside the crop canopy was measured using 500 ml plastic bottles with 95 mm diameter funnels. The funnel and bottle were attached to a garden stake and secured with an elastic band and tent peg. (b) Stem-flow was measured using 750 ml plastic bottles (of the same height as the 500 ml bottles) with a 95 mm diameter funnel adapted to fit around the stem and sealed with silicon sealant. (c) As a precaution against overflowing the stem-flow bottle was placed inside a plastic container. 0 = predictions as accurate as the mean of the observed data) as described in Legates & McCabe (2005).

Experimental data
Meteorological data from the weather station and eddy covariance instrumentation are shown in Fig. 3. Wind direction at the site is predominantly from the west with mean wind speeds and annual precipitation of 2.45 ms À1 and 871 mm for the period 2012 to 2016. Over the C i study period (23rd June 2016 to 13th March 2017), the total precipitation was 776 mm. Conditions at the site during the C i sampling period were generally within the five year average with the exception of short periods of high wind speeds due to seasonal storms, and particularly high rainfall during the summer of 2016 caused by shifts in the gulf stream (  (Fig. 3b). Rh was mostly above 80% for all of the 5 years (Fig. 3d). Soil moisture only dropped below the wilting point from 24th July 2014 to 20th October 2014 (Fig. 3e). Rs levels and LE and sensible heat (H) fluxes were comparable across each of the 5 years ( Fig. 3f-h).

ET results
The mean annual ET rates ( (Fig. 4) were similar over the five year study period with 2014 and 2015 showing the highest summer peaks, and the winters of 2012/2013 and 2014/2015 showing the lowest drops. ET EC was higher in the winter than predicted by all the models. There was no drop in ET EC during the period when the soil moisture was below wilting point, although there was a drop over the following late and winter seasons. GG, PMgrass, and PT.adj correspond well to the summer peak of 2012 which was the conversion year, but all the potential ET models overestimate the summer peaks. Adjustments for soil moisture with HS.adj and PT.adj reduce the main growing season levels too much compared to ET EC , but PMsugarcane.adj overestimates them. Whilst HS results are considerably higher in the summer, the model performs better over the winter. PMgrass and PMsugarcane results are also close to ET EC over the winter, although the late growing season higher values are not captured by any of the models.
Statistics carried out for the early season (Table 3) show low R 2 values for all the models compared with ET EC . The seasonal daily mean of the GG results is the closest to ET EC and is followed by PT.adj. All the model predictions overestimate with the exception of PMsugarcane. HS is shown to be the worst model for the early season with the most unfavourable outcomes of all the statistical tests performed compared to the other models. Adjustments for soil moisture during the early season improved PT and HS results (mean F values for the early season for PT and HS were 0.89 and 0.84, respectively) but made no difference to PMsugarcane (mean K s for the early season was 1). PT.adj, GG, PMsugarcane, and PMsugarcane.adj show a moderate fit using the modified Index of Agreement (md); however, the modified Nash Sutcliffe Efficiency (mNSE) test results in below zero values for all the models, with PT.adj being closest to it at À0.04. Overall for the early season, PT.adj performs the best, closely followed by GG. Comparing the potential ET models shows the PT results to be closest to PMgrass.
The mean values and statistics for the main season (Table 4) show PMgrass to be the best model compared to ET EC , followed by GG. All the model means (except HS.adj and PT.adj) show an overestimation for the season, but GG and PMsugarcane.adj show the smallest difference to ET EC . However, PMsugarcane.adj has a high MAE and low mNSE compared with the other models. PMgrass is the only model to have a mNSE value above zero (0.08). The impact of soil moisture across the adjusted models is not the same with the mean F values for the season for HS and PT as 0.25 and 0.42, respectively, whereas the seasonal mean K s value is 0.74. Comparing the potential ET models to PMgrass again shows the PT results to be closest. The model with the worst fit to ET EC for the main season is the HS model.
During the late season, the means for all the models underestimate ET EC , including the potential ET formulae (Table 5). Only the Penman-Monteith-derived models have mediocre R 2 values, whereas the values for the other models are low. Results of the md test for all the models are in a similar range, although PMgrass shows the best fit at 0.49. All of the results of the mNSE test are below zero, although PMgrass and HS are slightly better than the other models with values of À0.06 and À0.09, respectively. Of the potential ET models, PMgrass performs better than HS and PT, but HS is closest to the PMgrass results. The means of the models adjusted for soil moisture were further away from the mean ET EC than their unadjusted potential ET base models. Overall for the late season, PMgrass shows the best fit, followed by HS. GG is the worst fit for the season.
During the winter season, as in the late season, all the models' means were less than the ET EC mean (Table 6). PMgrass was closest mean to ET EC and also had the most favourable md result of 0.51. PMsugarcane and PMsugarcane.adj were similar to PMgrass with md values of 0.48. PMgrass was the only model with a mNSE result above zero (0.10). Both the PMsugarcane models mNSE results were zero. Adjustments for moisture were minimal for this season with only HS being adjusted (the winter seasonal mean F value for HS was 0.97 and for PT was 1, and the mean K s value for adjusting  The models are as follows: GG, Granger-Gray; PMsugarcane.adj, PMgrass adjusted with a water stress coefficient and the crop coefficient for sugarcane; PMsugarcane, PMgrass adjusted with the crop coefficient for sugarcane; PMgrass, Penman-Monteith (short grass) model; HS, Hargreaves-Samani; HS.adj, HS adjusted with a soil moisture function; PT, Priestley-Taylor; PT.adj, PT adjusted with a soil moisture function. Model results are compared to eddy covariance (EC).
PMsugarcane was also 1). Overall for the winter season, PMgrass showed the most favourable fit of the models tested, followed by PMsugarcane. The worst fit for the season was GG.

Miscanthus K c value
In the early and main growing seasons, there is a difference in the Miscanthus K c values (calculated from the eddy covariance data and the PMgrass results) when data are used from the whole 5 year period compared to just 2013 to 2016, but values are almost the same for the late and winter seasons ( Table 7). The early season in 2012 represents an atypical period being the time of land conversion to Miscanthus with a dominance of bare soil during the crop's initial establishment. Figure 5 shows ET EC results in comparison with PMgrass adjusted with the calculated K c values (PMK c ).

Canopy interception
Fifty-one recording events took place over the sampling period June 2016 to March 2017. Data were only removed from one of these occasions due to the high winds in November 2016 causing damage to the gauges. Measured C i was 24% for the period. The total gross precipitation (outside of the crop) was 776 mm, and the net precipitation (a combination of stem-flow and through-fall) was 588 mm. The net precipitation was made up of 133 mm stem-flow and 455 mm throughfall. Gross precipitation was related to net precipitation with an R 2 value of 0.9 (Fig. 6a). Interception is highest from July to September when the canopy is mature (Fig. 6b). The highest level of interception for a measuring occasion was 52% recorded during the period 15th -18th July, and the highest mean monthly level of interception was 34% recorded The models are as follows: GG, Granger-Gray; PMsugarcane.adj, PMgrass adjusted with a water stress coefficient and the crop coefficient for sugarcane; PMsugarcane, PMgrass adjusted with the crop coefficient for sugarcane; PMgrass, Penman-Monteith (short grass) model; HS, Hargreaves-Samani; HS.adj, HS adjusted with a soil moisture function; PT, Priestley-Taylor; PT.adj, PT adjusted with a soil moisture function. Model results are compared to eddy covariance (EC). for the month of August. Whilst the interception levels drop over the autumn and winter as leaves are dropped in senescence, the remaining canes continue to intercept rainfall until the harvest at the end of March. There are four instances where there was a higher net than gross precipitation (Fig. 6b). Examination of the data suggests these are related to occasions when wind direction may have caused gauges to record higher levels in the within-crop sampling due to the canopy intercepting rain being blown horizontally by the wind.

ET models
The mean ET EC of 483 mm yr À1 was over half of the mean annual rainfall demonstrating the importance of obtaining accurate estimates of ET in hydrological modelling. The maximum measured ET EC value of 4.65 mm day À1 was considerably lower than the highs of 7 and 11 mm day À1 found in the USA by Hickman et al. (2010) and in Italy by Triana et al. (2015). This is as expected for the very different climatic conditions of the studies. However, it was similar to the ET EC of around 5 mm day À1 obtained in Hereford, UK, and within the range of the MOSES model predictions both shown in the study carried out by Finch et al. (2004). The eddy covariance technique is a recognized method for obtaining field estimates of ET and is regarded as having a good level of accuracyprovided careful data processing and gap-filling strategies are employed (Aubinet et al., 2012;Gebler et al., 2015;Wagle et al., 2016). The use of daily ET results has provided a detailed insight into the performance of the models within each season. Although none of the ET models provide a good fit compared to ET EC , the highest modified Index of Agreement (md) results for each season were generally in the medium range (early 0.49, main 0.60, late 0.49 and winter 0.51).
A combination of factors in this study has allowed for reasonable comparisons of reference and potential ET models to ET EC in this study. Whilst reference and potential ET models calculate ET on the basis of no crop water shortage, this was the case at the field site for the majority of the study period, with only a short time when the soil moisture status was below wilting point. Adjustments to the HS and PT base models to account for soil moisture stress generally resulted in ET rates less than ET EC (Fig. 4). This Miscanthus genotype has also been shown to have a slower initial response to drought, with limited stomatal control (Clifton-Brown et al., 2002;Joo et al., 2017) and the ability to exploit the maximum soil depth and hence available water (Neukirchen et al., 1999) enabling the maintenance of high ET rates compared to other crops. However, prolonged water stress is likely to reduce Miscanthus ET rates (Joo et al., 2017).
PMgrass performed the best in all the seasons with the exception of the early season. This model had the highest md result for the main, late, and winter seasons and was the only model to achieve a mNSE score of above zero (main 0.08 and winter 0.10). Miscanthus emerges later than the start of the grass pasture growing season and can continue transpiring to the end of October (in favourable years). These are likely to be factors in differing early and late season ET rates of Miscanthus compared with grass.
GG was the second best model for both the early and main seasons (early: MAE 0.43, md 0.48, mNSE À0.12; main: MAE 0.54, md 0.58, mNSE À0.03), but it was the worst performing model for the late and winter seasons (late: MAE 0.85, md 0.34, mNSE À0.88; winter: MAE 0.53, md 0.35, mNSE À0.71). For the early season, GG Both GG and PMgrass require wind speed data as an input, whereas this is not required by PT and HS. PT and HS models can also be used within the SWAT hydrology model to calculate ET in the absence of wind speed data (Arnold et al., 2012) making them suitable for sites with more limited instrumentation. Comparing PT and HS to ET EC has shown that PT performs better than HS over the early (PT, md 0.41; HS, md 0.29) and main growing seasons (PT, md 0.48; HS, md 0.30) but that over the late (PT, md 0.37; HS, md 0.41) and winter (PT, MAE 0.51, md 0.35, mNSE À0.64; HS, MAE 0.36, md 0.34, mNSE 0.36) seasons HS out performs PT. HS is more commonly used for warmer climates (Tabari, 2010) so was least suited to the UK climate type.
Winter ET EC values were higher than all of the model predictionsan important point to consider when modelling the impacts on water balance and potential flood mitigation benefits. Winter precipitation interception by stalks and dead leaves in the field is not taken into account in PMgrass. Interception is an important factor in ET rates where differences of 30% between ET calculated with and without adjustment for the impact of C i have been observed (Robinson et al., 2017). The field site's coastal proximity and localized weather systems could also be impacting on lower model results compared with ET EC . ET a may be higher at times on site due to advection of sensible heat energy either from the sea or the presence of nearby hilly terrain causing localized wetter and drier air systems creating greater mixing in boundary layers (Van Dijk et al., 2015).
Whilst the use of the more complex Penman-Monteith formulae (Monteith, 1965) may provide better results, the detailed data input requirements are not always available, and the simplified short grass equation (PMgrass) in conjunction with crop-specific K c values has been used (Stephens et al., 2001b;Borek et al., 2010;Triana et al., 2015). The use of K c values for sugarcane did not perform as well as using the PMgrass base model (Fig. 4). Based on the data in this study, the following Miscanthus-specific K c values are suggested: early season 0.63; main season 0.85; late season 1.57; and 1.12 over winter. The main growing season K c value is the same as the 0.85 proposed by Beale et al. (1999) and within the wide range of 0.31 to 1.93 found by Triana et al. (2015). However, it is lower than the 1.20 suggested by Stephens et al. (2001b) and the 1.15 for maize and 1.25 for sugarcane given by Allen et al. (1998). Clearly, these measurements will to a degree be site specific and would benefit from testing at a wider number of sites; however, they do represent an improvement in our knowledge especially for the nongrowing season (Hay & Irmak, 2009).

Canopy precipitation interception
This study has shown that the Miscanthus crop is having a greater impact than short grass pasture on precipitation reaching the ground surface from the months of June (with the growth of leaves) through to the spring harvest date. High interception over July to September reflects the time when the canopy is at its fullest. However, it remains high into the autumn when the crop continues to intercept moisture after senescence due to stem density and some dead leaves remaining attached to stems until the end of January.
The measured interception of 24% from June to March is similar in value to the annual interception estimated for a mixed deciduous forest of 25% (Herbst et al., 2008), 21% for short rotation coppice (SRC) poplar (Hall & Allen, 1982), and the model prediction for SRC willow of 20% (Stephens et al., 2001a), suggesting benefits for flood alleviation by reducing soil moisture recharge (Marshall et al., 2009). However, in contrast to forestry, the Miscanthus crop has a period after harvest each year when there is no, or very little, interception with only short stubble left in the field before spring regrowth. Nonetheless, interception by the Miscanthus canopy will play a role in reducing soil moisture, particularly in the late autumn and early winter when higher rainfalls can occur.
Data collected in this study compare well to the measured results in plots of Miscanthus found by Riche & Christian (2001) of 25% in 1997/1998, and 24% in 1998. There was a longer period of interception in this study due to the late harvest date in March as opposed to the more typical harvest time of early February. When the interception is calculated over a shorter timescale of June to January, as in the study by Riche & Christian (2001), the result is slightly higher at 26%. The use of the Gash interception model by Finch & Riche (2010) suggested that interception might be reduced by as much as 6% in larger scale plantations, but the results of this study do not support this suggestion. This may be due to an estimated value for field scale wet canopy evaporation used in the Gash model and obtained from the full Penman-Monteith equation (Monteith, 1965). This component has a large influence in the result (Gash et al., 1995(Gash et al., , 1999, and therefore, the accuracy of the estimated evaporation rate will impact on the predicted interception. Higher measured interception than obtained via the Penman-Monteith equation has been noted before (Van Dijk et al., 2015) and shows the importance of this field estimate for accurate hydrological modelling. Another possible reason for this higher interception (and therefore wet leaf evaporation) than modelled is the lower albedo of 0.21 (Miller et al., 2016) for Miscanthus during October and November compared with 0.23 for grass (Allen et al., 1998). This means the crop is reflecting less solar energy and retaining more heat energy.
This study shows the potential benefits for flood mitigation of Miscanthus compared to a short grass pasture with similar levels of interception to forestry and SRC, which are coupled with the crop's high water use and conversion efficiency and higher winter ET rates. The most accurate of the formulae considered to predict ET rates was the simplified Penman-Monteith (short grass) equation. The Miscanthus-specific K c values suggested would benefit from being tested against other commercial-scale plantations where ET EC or other field measurements of ET are available. However, information from this study can be used to increase accuracy of yield models and in determining suitable areas for planting.