Metrics of biomass, live-weight gain and nitrogen loss of ryegrass sheep pasture in the 21st century.

This study argues that several metrics are necessary to build up a picture of yield gain and nitrogen losses for ryegrass sheep pastures. Metrics of resource use efficiency, nitrous oxide emission factor, leached and emitted nitrogen per unit product are used to encompass yield gain and losses relating to nitrogen. These metrics are calculated from field system simulations using the DAYCENT model, validated from field sensor measurements and observations relating to crop yield, fertilizer applied, ammonium in soil and nitrate in soil and water, nitrous oxide and soil moisture. Three ryegrass pastures with traditional management for sheep grazing and silage are studied. As expected, the metrics between long-term ryegrass swards in this study are not very dissimilar. Slight differences between simulations of different field systems likely result from varying soil bulk density, as revealed by a sensitivity analysis applied to DAYCENT. The field with the highest resource use efficiency was also the field with the lowest leached inorganic nitrogen per unit product, and vice versa. Field system simulation using climate projections indicates an increase in nitrogen loss to water and air, with a corresponding increase in biomass. If we simulate both nitrogen loss by leaching and by gaseous emission, we obtain a fuller picture. Under climate projections, the field with the lowest determined nitrous oxide emissions factor, had a relatively high leached nitrogen per product amongst the three fields. When management differences were investigated, the amount of nitrous oxide per unit biomass was found to be significantly higher for an annual management of grazing only, than a silage harvest plus grazing, likely relating to the increased period of livestock on pasture. This work emphasizes how several metrics validated by auto-sampled data provide a measure of nitrogen loss, efficiency and best management practise.


Food Production and Sustainable Management
Agricultural production needs to increase to feed an increasing global population under a changing climate. Strategies that promote long-term sustainability and yields, rather than purely peak quantity, should be introduced (Heinemann et al., 2013). Unsustainable farming practises run the risk of environmental pollution due to nutrient run-off, soil degradation and the loss of biodiversity through inappropriate management (Tilman et al., 2002;Hayarti et al, 2010). Nitrogen (N) fertilizer increases crop production, but a large proportion of agricultural N is leached to the environment in chemical forms that have caused contamination of drinking water and eutrophication of water bodies (Diaz andRosenberg, 2008, EPA Science Advisory Board, 2011) and its gaseous emission is the form of nitrous oxide participates in photochemical reactions in the upper atmosphere (the stratosphere) that destroy ozone (Crutzen 1970).
Improving one aspect of the field system, does not always have a beneficial effect on other environmental features. A test of beneficial and harmful effects, or gains against losses, can be viewed by using metrics to compare management methods, and innovations could be compared to a baseline of traditional agronomy to compare benefits and offsets. Many agricultural metrics exist, however there is no consensus on a correct or most suitable one. Hayati et al. (2010) advise to construct metrics which are location specific and within the context of the situation. Our interest in this study is to view how several related metrics can improve agronomic information.
1.2 Comparative resource use and productivity

A C C E P T E D M A N U S C R I P T
N use efficiency (NUE) can be measured in different ways, as crop N offtake per unit of N applied, or as defined by Moll et al. (1982) as grain production per unit of N available in the soil, thereby translating it to a measure of biomass per unit N applied. Resource use efficiency (RUE) is a ratio of productivity per unit of resource (Sheriff et al., 1995) where the resource can be any limiting factor to growth. If the resource is soil N, the definition of RUE overlaps that of Moll's definition of NUE, and these metrics on traditional management can act as a benchmark from which future improvements can be assessed. Low values usually indicate inefficient use of the added N whereas very high values usually indicate the mining of soil N (Norton et al., 2015). NUE is not necessarily a direct quantitative estimate of N loss from the system, because N not removed in the harvest might remain in the soil. Over the long term, however, changes in soil N stocks are usually low relative to inputs and outputs, and therefore, low NUE values over multiple years are reasonably reliable indirect indicators of probable significant N loss to the environment (Norton et al., 2015).
RUEs relating to productivity are important agronomic indicators focussing on production as the aim rather than efficient use of the N. An advantage of RUE relating to productivity and fertilizer is that the biomass and fertilizer data are generally available at the field level. In this study RUE is used, and termed f-RUE (fertilizer RUE).

Leached N
Plant available N loss depends upon a balance of the timing and rate of N application and the demand for N by the crop, or by microbial uptake. If uptake has a lower rate than application, or heavy rains follow application, excess NO 3 and NH 4 + is susceptible to water transport.
NO 3 flows through soil pores more rapidly than NH 4 + which is held back by chemical bonding (Mekala and Nambi, 2016). Nitrate is a common risk in leached runoff to water bodies, due to the tendency of eutrophication to result in reduced oxygen in water,

A C C E P T E D M A N U S C R I P T
detrimental to aquatic and human life. In this study, we refer to leached inorganic N, predominantly NO 3 -, as leached N because dissolved organic N cannot be automatically sensed in runoff like inorganic N, due to the need to digest the sample prior to analysis which is not possible to automate under field conditions (ASA Analytics, 2017). Studies in agriculture have generally shown less leaching from dissolved organic N than inorganic N (Siemens and Kaupenjohann, 2002;Lehmann et al., 2003), but we accept that the lack of measured dissolved organic N measurement is a gap in the system.

N 2 O emissions
Agriculture practices of N amendments cultivation, excess soil water, can increase N 2 O production and emissions (Del Grosso, 2006). Mineral N supply, plant N demand, and abiotic soil conditions interact to control N 2 O emissions from soils. Agricultural practices also increase NO 3 leaching, which enters aquatic systems or is transported to a non-farm plantsoil system, and undergoes denitrification which results in indirect N 2 O emissions.
To reduce leaching losses, best practise field management tries to minimize the amount of excess nitrate (NO 3 -) present in the soil at any given time, timing the application of fertilizer to smaller and more frequent applications. However, the fact that pores hold back ammonium (NH 4 + ), allows it to be in contact with microbial matter longer (Mekala and Nambi, 2016).
This increases the risk of conversion from NH 4 + to nitrite and then to NO 3 by nitrifying bacteria in aerobic conditions, and then conversion to N 2 O by heterotrophic bacteria in anaerobic conditions. Both aerobic and anaerobic processes result in the production of N 2 O, a potent greenhouse gas and precursor of stratospheric ozone loss. These processes occur simultaneously and in proximity in grassland soils (Abbasi & Adams, 1998).
1.5 Agronomic modelling as a precursor to metric calculation

A C C E P T E D M A N U S C R I P T
Data collected manually, or by sensor, is not available every day for every year, and weather cannot be measured for future climate projections. The only way to obtain consistent multiannual production and N loss data is to simulate the data using a calibrated model. We have chosen the DAYCENT model (Parton et al., 1998) for its applicability to our study. This is a field-scale model concerning soil emissions, leaching and crop production, which calculates the grazed offtake of biomass, from which we determine the live-weight gain efficiency of livestock.

Aim of Study
Our aim is to view how collating several related metrics, related to the gains or losses of nitrogen in traditional agronomic practises, can build up information on the agronomic system. This is carried out across three neighbouring sheep pastures under a similar soil type, and the same historic climate and projected late 21 st century climate, where the main difference is the seasonal pasture management.
We will use measured variables from manual soil sampling and air and water quantity and quality sensors of the North Wyke Farm Platform (NWFP) site to calibrate and validate the DAYCENT agricultural systems model. The calibrated model will provide information to calculate the metrics concerned with field-scale gains in production against losses of N. Three fields will be compared under two types of traditional annual management, grazed use only and a silage crop followed by grazing. A null hypothesis is that the two types of management result in the same yield gain to nitrogen loss.

Site description
The NWFP (Orr et al., 2016) (Parton et al., 1998) is an agricultural system model simulating crop growth and biogeochemical cycling between the soil-water-crop-atmosphere. Plant production is a function of genetic potential, phenology, nutrient availability, water/temperature stress, and solar radiation. The model includes soil organic matter decomposition pools (active, slow and passive) with different decomposition rates, above and belowground litter pools and a surface microbial pool. Soil NO 3 and NH 4 + , labile soil carbon, water content and temperature determine N 2 O production (Parton, 1998). DAYCENT was used because it has been globally validated against forage production (Henderson et al., 2015). It was chosen because of its flexibility, many field management techniques are simulated and linked to the system, and it incorporates forage removal and nitrogen return by ruminants. It was also chosen for ease of use and applicability to our study; it has a daily time-step and a scheduling file which controls the simulation, bringing together all input files and process modules.  (Defra, 2010).
A moderate grazing regime was selected in the model's grazing management options, which simulated a linear decrease in production through the growing season, involving the offtake of 40% of biomass as live shoots and 10% of biomass as leaf litter. In the case of sheep grazing, the management option was set to return 90% of N in offtake to the soil, and proportion 34% of excreted N into faeces, the rest in urine. It is advised (Eblex, 2016) that the percentage of live biomass grazed is normally 50% or above, hence this was increased by 50% but this resulted in no difference in model output of N in soil, leached or emitted.
DAYCENT allows the creation of new management options, to create tailored effects of each type of management. We created new grazing options to switch grazing on and off for the exact dates when livestock were in the field. The option to switch on grazing simulation requires fraction of biomass grazed and fraction of N returned, so these were set to zero and

A C C E P T E D M A N U S C R I P T
this module listed in the model's scheduling file to switch off the grazing. New fertilizer options have been created, each new fertilizer type and application rate requires a value containing specific rates of N applied, calculated from different formulations and spreading rates using RB209 recommendations. Each fertilizer module has a different name and each one is listed as required in the scheduling file to set up an application.
Two schedule files were created: a spin-up file of grazed grassland without inorganic fertilizer for years 1 -1900 to balance biogeochemical cycling, leading on to the main schedule file for years 1901 -2016. Using a spin-up output for initialization of the main simulation, is commonly utilized with DAYCENT to represent the historic land use and management of the site and initialize soil organic matter pools before current practices are simulated. Our focus is the period 2011-2015, for which we have very detailed field management operational data records for type, rate and dates of fertilizer application, number of days of sheep grazing, dates of harvest, from which we constructed the annual summary in Table 2, and the fertilizer schedule (Table 3). We did not use grazing numbers of livestock, the DAYCENT model does not use livestock numbers, it assumes a fraction of live and dead dry matter biomass removed". The NWFP attempts to maintain a constant rate of grazing.

The DAYCENT model calibration
The DAYCENT model obtained from the USA had been calibrated for use in that country, so required calibration for a precipitation-heavy UK agriculture. The type of growth module used was changed, from one relating carbon allocation to rainfall, to a module for the UK using a growth based on degree-day accumulation. The climate record for the site, common to all three neighbouring fields, was analysed to determine degree-day parameters Table S1). The soil parameters were similar, only bulk density and pH were modified for each field (Table 1). Field management for the three grazed fields in the study was unique to each field, for each year (Table 2 and 3).
The calibration was carried out on biomass, followed by soil moisture, soil nutrients, and finally gaseous emission, as advised in the DAYCENT 4.5 INSTRUCTIONS (NREL online, accessed Mar 8, 2019).
There is very little harvest yield data available on the three fields studied. Silage crop fields (for which there is harvest data) surrounding the three used for study and had the same management regime in the same year, the same soil type and climate, and therefore mean field parameters were used to obtain simulated yield. Therefore, the harvest yield of L.
perenne grass cut for silage was collected from the mean yield data of 10 neighbouring fields to the three study fields, and was compared against the modelled yield for calibration. The For all other parameters than biomass (soil moisture and runoff, soil N and leached N, and N 2 O emission), simulated output using data from the Longlands South field were compared against measured field data to calibrate the model. During the process of calibration using data of Longlands South, parameters were modified and seven calibration versions of the model formed, until the seventh version obtained a balance between the fit of simulated- After model calibration, it is good practise to do a sensitivity analysis, as model inputs such as soil parameters from field averages contain uncertainty (Wu & Shepherd, 2011;Jørgensen, 1995). A sensitivity analysis was conducted on the DAYCENT model using inputs for the Longlands South field. Changes were made with respect to fertilizer (for a change in soil N), pH, precipitation (for a change in soil moisture) and bulk density; these have previously been the inputs found most likely to influence the N 2 O (Fitton et al., 2014b), as a proxy for the effect on general N cycling. Each of the four input parameters was separately modified by an increase and a decrease in 5% and 10% from for the site value, holding the remaining inputs at the field values.

A C C E P T E D M A N U S C R I P T
To automate the process, two shell scripts were written in R to run the DAYCENT model in

Calculation of resource use, product and N loss metrics
The f-RUE is an indicator of N use for productivity, whereas the N 2 O emission factor (EF) and the emission or leaching per product are indications of the loss of N to the field system.

A C C E P T E D M A N U S C R I P T
f-RUE was calculated annually, from grams m -2 harvest product / grams m -2 nitrogen, from fertiliser applied and livestock excreted (Moll et al., 1982;Sheriff et al., 1982). The harvest product is annual aboveground biomass, or alternatively the live-weight gain of lamb that the biomass would support in these sheep fields. This is based on the average feed conversion of 8 kg dry matter biomass to 1 kg live-weight gain (Eblex sheep BRP manual 5, 2014).
N 2 O emission EFs for grassland are calculated annually from simulations using fertilizer and grazing returns of N, and control simulations with zero N applied. EF = g N 2 O-N m -2 (fertilizer and grazed return to the soil)g N 2 O-N m -2 (zero N) / g m -2 total N applied. The total N applied is fertilizer N plus N from excreta of grazing animals applied annually (g N m -2 y -1 ) (following Rafique et al, 2011;Barton et al, 2008).
N loss metrics were calculated annually as g N 2 O-N m -2 (or g N leached m -2 ) / g harvest product m -2 , where harvest product is either g aboveground sward biomass per m -2 , or the live-weight gain of lamb that the biomass would support.
DAYCENT outputs most variables in units of g m -2 and have been reported as such, as the metrics are ratios of the same units. The exception is EFs which have been reported as kg N 2 O-N per kg fertilizer applied (for the same area), for comparison against EFs from literature.
All metrics for the three fields were calculated annually using output from the validated DAYCENT model, for 2011-2015 when precise management records were available, and by 30-year mean for the climate scenarios.
The annual simulations comprising 3 fields, for 5 years (2011)(2012)(2013)(2014)(2015) or 30-year climate scenarios, contain managements for grazing only (i.e. no silage harvest) or one silage harvest plus grazing. To test for differences between grazing and silage-grazing managements, a one- sample t-test for one variate with group factor was carried out on each of the metrics produced.

Model validation
Measured yield data was very limited. The site is a working farm and documented yield data had been measured when contractors cut grass for silage and weighed grass from a collected harvest from all fields together. So for the calibration of yield, the simulation was based on average field conditions. The simulation of harvest yield is compared against measured yield values, we cannot properly compare these datasets statistically, because stocking rates will vary and some years will include a silage cut, however as a general guidance they give confidence that simulated grazed offtake values are not unreasonable, and being heavily reliant on simulated biomass, by proxy this serves as an extra check that biomass values are not unreasonable.
Soil N was measured over 4 separate dates and compared to a continuous profile of simulated soil N. Simulated soil NO 3 and NH 4 + follow the pattern ( Fig. 3a and b, respectively) and value of mean observations quite closely but there is a time lag of about 14 days, the simulations having a more rapid rate of decay than the observations. For both NO 3 and NH 4 + there is a large variation in observations for the first date of field measurement, and therefore a large variation in the observed rate of decay between the first and second date of observations.
It is easier to assess the simulated vs. observed soil moisture when using sensors rather than sparse manually sampled data, because sensors provide a continuous profile to match the simulations. Simulation-observation compared favourably overall with no time lag (Fig. 3c).
The correlation coefficient between DAYCENT simulated moisture and sensor data was 0.98, modelling efficiency was 0.94 and coefficient of determination was 0.85, which indicated a high positive degree of association. Fig. 3c shows a discrepancy in DAYCENT and sensor soil moisture for the replenishment of soil water after a dry summer. This is because most agricultural models' do not simulate cracking clay soils. On the study fields, the

A C C E P T E D M A N U S C R I P T
Figs. 4a, 4d and 4e show a comparison of simulated-observed variables for the N system over the three swards studied, adding in Dairy North and Golden Rove. Figs. 4b and 4c show a comparison of simulated-observed soil moisture and runoff, reflecting the hydrological processes of the three swards which impact emission and leaching of nitrogen. Field measurements vary spatially whereas a model simulates a field average, so a greater number of outliers can be expected from field observations.
Golden Rove data has, in part, been taken from Horrocks et al. (2014), but over the same growing season as the other two fields. Golden Rove has a more variable slope across the field, which explains the greater variability in observed soil water runoff compared to other fields.
For most simulated-observed pairs in Fig. 4, the dense area of points on the plots falls near the 1:1 line. Soil inorganic N simulation generally appears to be lower than observed data ( Fig. 4a), but Fig. 3a and b suggest the cause is a faster rate of soil N assimilation in the DAYCENT model than measured, however measured soil inorganic N data is in limited supply and also variable.
For the North Wyke site with high rainfall and heavy clays, the N 2 O emissions have been described as higher than most sites (Fitton et al., 2014a). If we alter DAYCENT calibration parameters relating to nitrification and denitrification to match a high measured rate of gaseous emission, we also speed up the depletion of soil inorganic N. Smaller estimation of soil N by DAYCENT compared to measurement is known and has been described in literature (Senapati et al., 2016).
Over several fields the occurrence of daily simulated and observed leached N is concurrent, generally simulations are higher than observations but data has a wide spread. Since daily values are so variable, leached data is accumulated annually for use in the metrics. Annual

A C C E P T E D M A N U S C R I P T
simulations-observations compare better than daily, by a simulated to measured ratio of 1.5:1, however the under-estimation inherent in this type of sensor system for leached inorganic N has been discussed earlier.
The simulated-observed daily N 2 O emissions are spread widely and evenly over the plot (Fig.4e)

A C C E P T E D M A N U S C R I P T
obtain a reasonable simulation of N 2 O, the calibration may inadvertently increase the rate of N turnover in the soil, shown by a faster rate of simulated soil NH 4 + and NO 3 decrease, although the uncertainty in soil observations is high because of the low sampling frequency. Senapati et al. (2016) have similarly commented on this relationship of soil N transformations and N 2 O emission. Our sensitivity analysis likely explains differences in simulated output between Golden Rove and the other two fields, with Golden Rove having the lowest bulk density of 0.9 for the top 10cm depth of its clay loam compared to 1.07. Although all soils are of the Halstow series, bulk density will vary with soil compaction and soil organic matter content, which are related to previous field management. Comparatively, Senapati et al. (2016) found DAYCENT to be most sensitive to field capacity and a decrease in bulk density, followed by pH, fertilizer-N and soil organic matter.

A C C E P T E D M A N U S C R I P T
Our model compares satisfactorily with the literature. For Irish grassland, Rafique et al. (2011) calculated EFs at 0.01 -0.031 and Hyde et al. (2006) at 0.007 -0.05. For Scottish grasslands Dobbie and Smith (2003) reported EFs at 0.01 -0.03. Cardenas et al. (2010) reported a N 2 O flux minus background flux of 3.9 kg N 2 O-N ha −1 yr −1 , for the west of England in a field close to this study, using a fertilizer application of 100 kg N ha -1 , resulting in an EF of 0.039.
All these studies have a higher limit than the 2006 IPCC EF of 0.01 for direct N 2 O emissions.
Deviations of observed N 2 O emissions from those calculated using the IPCC Tier 1 EF approach clearly shows that this methodology is too simplistic to reflect regional variations of biologically produced N 2 O emissions (Skiba et al., 2012). The Department for Environment,

Food and Rural Affairs and devolved UK governments funded the GHG Platform in order to
improve the UK's agricultural greenhouse gas emission inventories which should improve regional N 2 O EFs. Research since the adoption of the IPCC EF for grassland strongly suggests that weather and management modifies EFs. Smith et al. (1999) (Table 5b) is

A C C E P T E D M A N U S C R I P T
slightly lower than the 30-year record and also the mean of 2011-15 (Table 5a), and the baseline stochastic precipitation falls between the 30-year mean and the mean of 2011-15.
Baseline climate N 2 O EFs are higher than 2011-2015, partly because the differences in a 30year period to a 5-year period. This is also partly because we cannot re-create the same concurrence between baseline stochastic precipitation / temperature and the time of fertilizer application date or grazing period in the same way as it occurred 2011 -2015.
For UK climate projections, the baseline climate produces EFs of 0.056, 0.076 and 0.048 kg N 2 O-N per kg applied fertilizer for Longlands South, Dairy North and Golden Rove, respectively. The medium GHG emission climate projection (36.6% increase of mean temperature over baseline climate) increased EFs by a value of 0.03, 0.01 and 0.03 respectively, above baseline climate. The high GHG emission climate projection (48.4% increase of mean temperature) increased EFs by a value of 0.03, 0.03 and 0.03, respectively.
Simulated N 2 O EFs for Golden Rove are lower than the other fields, the sole exception being 2011. This is due to the lower top-soil bulk density, which is considered a key factor in reducing emission via the soil porosity and hence oxygen levels reducing microbial denitrification (Oenema et al., 1997). However, a factor to also bear in mind is the DAYCENT model's known sensitivity to bulk density.
The aim of a calibrated model is to obtain a reasonable agreement for the fit of all simulated output against measured data, and to do this generically for a crop and soil type, therefore we do not expect to obtain a perfect fit for all variables for all fields. A source of error in measured data is the spatial heterogeneity of the physical and biological factors in a grazed field that control the rate of N 2 O emissions. The limited area of static chambers covering the field means that it is possible that N 2 O emissions are under-or over-estimated (Chadwick et al., 2014). This is especially true for N hotspots created by urine patches (Cowan et al.,

2015)
, and it is difficult when setting up static chambers to know beforehand where these exist. The chambers are moved every few weeks to a different part of a field containing livestock, this results in hotspots from fresh urine being covered and therefore varying values of soil N as the chambers are moved around the field, whereas a model simulates the processes of nutrient cycling resulting from the average rate of N applied to the field.

N 2 O or leaching per unit product
Averaged over 2011 -2015, 0.002 g N 2 O-N m -2 was emitted annually per g m -2 of aboveground sward biomass (Table 6a) for all three fields, and 0.016 g N 2 O-N m -2 was emitted annually per g m -2 of grazing stock live-weight gain. N 2 O per product is shown to be consistent, both between fields and between years 2011 -2015. There appears little increase in these metrics under future climate projections (Table 6b) from the baseline values; but this metric hides the fact that with warmer projected temperatures there is a corresponding increase in biomass plus a proportional increase in annual N 2 O emissions.
In contrast to N 2 O, the inorganic N leached per unit product 2011-2015 was variable (Table   7a) Rove. Because fertilizer N inputs for the three fields were similar, the reason behind higher leaching of Longlands South is likely animal derived, the longer the total period of grazing over a year, the higher the risk of leaching (Cuttle et al., 1998).

A C C E P T E D M A N U S C R I P T
The climate projection under both the medium and high GHG scenarios (Table 7b) resulted in a small increase in leaching above the baseline scenario together with a small increase in biomass, resulting in a small increase in inorganic N leached per unit product.
Based on the same simulated units of g m -2 , metrics for leaching per unit product are smaller than N 2 O emissions per unit product. This agrees with other findings from DAYCENT which showed that fine textured soils emit more N 2 O, but with smaller leaching losses (Del Grosso et al., 2008).
In this study the Dairy North field with the highest f-RUE was also the field with the lowest leached N per unit product, and Longlands South field with the lowest f-RUE was also the field with the highest leaching. Norton et al. (2015) reported that improvements in N use efficiency from increased productivity coincide with reductions in N pollution of surface waters.

Significant differences between managements for grazing or silage crop with grazing
The annual simulations contain a mixture of field management types, either grazing only or one silage harvest plus grazing. There was a significant difference (p<0.05) in the simulated N 2 O emission per unit product, with values for the grazing only management found to be significantly higher than the silage plus grazing management. This disproves the null hypothesis that the two managements would result in the same yield gain to N loss. The higher N 2 O emission per unit product for grazing only management related to the total period livestock spent on pasture, which were 171 -277 days per year for grazing only, and 27 -152 days per year for silage plus grazing. Total annual inorganic fertilizer applied was not significantly different between the two managements. N 2 O emission in grazed pastures are known to be primarily associated with animal excreta and soil compaction from livestock (Saggar et al., 2004; rather than resulting indirectly from a reduction in the grass

A C C E P T E D M A N U S C R I P T
biomass (Zhang et al., 2015), and livestock numbers plus number of grazing days have shown an increase in N 2 O (Wang et al., 2012). Here however, the simulated grazing intensity is assumed constant, and the DAYCENT model does not directly simulate grazing intensity with livestock numbers. The metrics for leaching, fertilizer N use efficiency or EFs did not display a significant difference between the silage plus grazing management and the grazing management.

Conclusion
By applying the automated sensor data to model calibration, the simulations provided continuous data to create the metrics to build up a picture for the health of the field system in terms of gains in product offset by the losses in nitrogen.
Comparing the three field systems, there appeared to be no difference in absolute leached N,

Supplementary data
Supplementary data: S1. DAYCENT calibration parameters.     A C C E P T E D M A N U S C R I P T A C C E P T E D M A N U S C R I P T  *Fertilizer resource use efficiency in this study is the g m -2 of harvest product per g m -2 of nitrogen applied from fertiliser and excreta, where harvest product is defined as (a) annual aboveground biomass harvested and grazed (mixed annual management), or (b) **in square brackets, the live-weight gain of stock that the biomass would support. This is based on the average feed conversion ratio of 8 kg dry matter per kg live weight gain (Eblex sheep BRP manual 5, 2014). 44% of the sward's dry matter biomass is carbon, and is used to convert model output of carbon to biomass.

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T Table 6. Metrics between three grazed fields growing Lolium perenne for nitrous oxide (N 2 O-N) per unit product, (a) simulated for 2011-2015, where the product is grass biomass or, in square brackets the live-weight gain of lamb**, and (b) simulated 30-year mean results for climate scenarios. LS=Longlands South, DN=Dairy North, GR=Golden Rove.

A C C E P T E D M A N U S C R I P T
Highlights:  Provides yield-nitrogen metrics for ryegrass sheep pastures  Combination of climate, sensor and sampling data applied to agri-system modelling  Highest resource use efficiency coincides with lowest leaching per unit product  N 2 O per unit biomass significantly lower for silage & grazing than grazing only  Several related metrics reveal info about the system, hidden by using one metric ACCEPTED MANUSCRIPT