Академический Документы
Профессиональный Документы
Культура Документы
www.elsevier.com/locate/agee
Abstract
Because of the interest in the value of medium-term weather forecasts to UK agriculture, we attempt to quantify this value for N fertiliser
use with arable crops. Model systems such as SUNDIAL provide arable N fertiliser advice by modelling the N supply from soil, but poor
knowledge of future weather reduces accuracy. Aweather generator was used to produce sets of simulated weather of a range of accuracies and
durations, for 10 regions in England and Wales. In a series of computer simulations, we tested the effect of prior knowledge of weather
following the date of N fertiliser application on the efficiency of use of applications using SUNDIAL. The changes in N leaching,
denitrification and crop N uptake due to the forecast quality were calculated. Yield and gross profit changes were estimated from N uptake, for
the arable industry in England and Wales.
Changes in losses were small. With a perfect forecast, there was a small decrease in leaching (approximately 1 kg N ha1), and still less
change in denitrification. The increase in crop uptake due to a perfect 27-week weather forecast was 6 kg N ha1, and the increase in farm
profit in England and Wales amounted to £68M per annum.
With more accurate forecasts, the system can reduce the risk of under-application of N. A perfect 3-week forecast would increase uptake
by an average of 2 kg N ha1, and increase profit nationally by £23M per annum but have negligible impact on losses. These improvements
appear to be systematic and could be expected to be achieved with any recommendation system that makes explicit use of post-application
weather.
# 2006 Elsevier B.V. All rights reserved.
Keywords: Nitrogen management; Nitrogen losses; Economic value; Arable farming; Computer model; Optimised fertiliser advice
0167-8809/$ – see front matter # 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.agee.2006.03.004
A.G. Dailey et al. / Agriculture, Ecosystems and Environment 117 (2006) 22–28 23
been shown to agree well with RB209 (MAFF, 2000), the The parameters describing rainfall were derived from 95
current Defra-sponsored recommendation system for N weather stations in England and Wales, each having at least
(ADAS, 2005). Predictions from SUNDIAL, however, are 20 years’ of data. The parameters describing ET and T were
dependent on the estimates of the weather post-application derived from a subset of 34 sites reporting ET and T for 20 or
and in general the estimated and actual weather will differ. more years (Dailey et al., 2005).
Thus the fertiliser advice may turn out to be less than Two sets of weather data were generated. The first series
optimal. If too much N is applied, losses to the environment represents expected weather, which is the weather used to
may increase. Under application may result in a shortfall in optimise SUNDIAL at the time a decision on fertiliser
yield and profit. application needs to be made; the second series, which
There is potential for medium-term weather forecasts to deviates from the first, represents realised weather. These
be made available to the farming community and it seems terms, expected and realised weather will be used throughout.
likely that such information might improve the accuracy The extent of the deviation of the realised series from the
with which farmers can obtain site and season-specific expected is specified by a parameter, w, which has a value
fertiliser advice using fertiliser recommendation systems between 0 and 1. A value of 1 signifies no deviation of the
such as SUNDIAL. In the UK, an exercise known as realised weather from the expected; in other words the
Foresight (Anon, 2001) has tried to suggest the value of expected weather is a ‘‘perfect’’ forecast. A lower value of w
research into new areas that could benefit the economy. One signifies that the expected weather, used in SUNDIAL, is a
aspect of this was whether or not it could be worthwhile forecast of correspondingly lower accuracy. By increasing w
developing a capability for medium-term weather forecasts over the range of 0–1, prior knowledge of the weather is
of two or three months (Anon, 2001). The objective of this increased from zero to complete knowledge. Reliability of
article is to quantify the potential of medium-term forecasts prior knowledge is one issue, however. The other is the
to improve the accuracy and benefit of nitrogen fertiliser duration of the reliability of that prior knowledge. We
recommendations using SUNDIAL as an example. The examined the effect of having prior knowledge of a given
analysis carried out in this project reflects the avoided loss of accuracy (w = 0–1) for durations of 1, 3, 5, 7, 9 and 27 weeks.
yield or reduction in loss of N (denitrification or leaching) In order to assess the benefits of prior knowledge of weather,
that can come about with increasing prior knowledge of we generated 80 different sequences of weather correspond-
weather if SUNDIAL is used to provide fertiliser advice. It is ing to 10 different weather stations in England and Wales.
with this standpoint in mind that we examined the potential
of medium-term weather forecasts to reduce nitrogen 2.2. Generation of rainfall
pollution and to contribute to the farming economy in
England and Wales. The weather generator for a particular location is based
on a cumulative density function (CDF) of weekly amounts
of rain for each of the 13 4-week periods of the year. This
2. Materials and methods generator (Dailey et al., 2005) is of a kind sometimes known
as a Richardson type (Richardson, 1981). A rain amount is
SUNDIAL, in common with other N recommendation generated by first producing a random number (a uniform
systems, suggests a level of N application on the basis of deviate in the range 0–1), and transforming to rainfall
assumptions about the weather between application and amount using the CDF. Without further modification the
harvest. Unlike most N recommendation systems, however, transformed series would be expected rain. The random
SUNDIAL does this explicitly with a sequence of expected series, Rr, used for realised rain is given by
weather. This expected weather is nonetheless an assump-
tion, so by repeating the simulations to harvest with Rr ¼ Re þ Ri ð1 ’Þ (1)
sequences of weather that differed from the expected
weather in a predetermined manner (see below), we could where Re is the random series used for expected rain and Ri is a
make the expected weather represent a range of forecasts of different, independent random series. Rr is constrained to the
different accuracies and durations. range 0–1. Persistence is implemented as a first order Markov
chain (Gabriel and Neumann, 1962) based on the observed
2.1. The weather generator probability of 0.57 across the UK that two consecutive weeks
are both dry or both wet (Dailey et al., 2005). The resultant
A weather generator (Dailey et al., 2005) was used to random numbers are transformed to rain amounts using the
produce series of weekly values for rain, evapotranspiration CDF of rainfall at the weather station of interest.
(ET) and the weekly mean of daily mean temperature (T)
required by SUNDIAL. The generator is controlled by 2.3. Generation of ET and T
characteristic parameters that describe the weather for a
particular location, such that the data generated is The long-term weekly amounts and standard deviations
statistically similar to observed weather at that location. of evapotranspiration (ET) and weekly means and standard
24 A.G. Dailey et al. / Agriculture, Ecosystems and Environment 117 (2006) 22–28
deviations of temperature (T) at a particular location are each Three soil types, that are the default soils in the
parameterised as a single-term Fourier series. For tempera- SUNDIAL model, were used in this project: sand, loam
ture this is and clay containing 10, 23 and 40% clay, respectively. Ten
weather stations having at least 20 years of records were
Tt ¼ a sinðbtÞ (2)
chosen for the simulations to represent wet, moderate
where Tt is the seasonal mean temperature at time t, and a and b rainfall and dry regions and warm, moderate or cool
are constants. A process that generates a series of residuals for conditions (Table 1). These ten stations represent the range
ET and T was suggested by Matalas (1967), and was modified of conditions found in England and Wales and are
in our weekly weather generator (Dailey et al., 2005) to geographically spread throughout the countries.
provide a description of the correlation between, and the We took the nine Government Office Regions of England
persistence of, temperature and evaporation. There is, for (e.g. Defra, 2002), and separate data for Wales. However, we
instance, an increased probability that a warmer than average combined the London and the South East regions, giving
week will coincide with above average evapotranspiration, nine in all. One, or in the case of the South West two,
and be followed by another warm week. The expected weather appropriate weather stations were allocated to represent
is first generated by adding these residuals to the Fourier series each region. The regional results were then combined
(2) as described by Richardson (1981). To generate the according to climate into five general categories (Table 1).
realised weather, a new series of residuals is generated that Results were weighted by the extent of cropping in each of
tracks the original series, but deviates from it by a random the regions.
amount with an average magnitude dependent on w. The Four crops were investigated to represent major classes of
realised ET and T are calculated by adding the new series of arable crops grown in England and Wales: (i) a winter cereal,
residuals to (2). Full details and testing of the weather gen- winter wheat; (ii) a spring sown cereal, spring barley; (iii) a
erator and of its evaluation can be found in Dailey et al. (2005). high nitrogen residue crop winter oil seed rape (osr); (iv) a
shallow-rooting crop, potatoes. The potential yields and dates
2.4. SUNDIAL of establishment and harvest of the crops used in the
simulations are shown in Table 2. Winter oilseed rape had a
The SUNDIAL Fertiliser Recommendation System 30 kg dressing of fertiliser nitrogen at drilling. For all
(SUNDIAL-FRS, Smith et al., 1996) uses a computer simulations, the crop preceding the crop of interest was winter
simulation model (Bradbury et al., 1993) of N turnover in wheat. The total application of fertiliser nitrogen to the
soil to estimate the supply and loss of N to a number of preceding crop was 160 kg for sand, 220 kg for loam and for
arable crops. For this work, SUNDIAL used potential yield clay in wet or moderate rainfall locations, and 180 kg for clay
and weather data for the location to simulate up to the date at soils in dry locations. These recommendations were taken
which fertiliser is to be applied. Then using the expected from current fertiliser advice, RB209 (MAFF, 2000), and thus
weather for the period up to harvest, the system optimised represent the actions likely to be taken by local farmers.
the amount of fertiliser that should be applied for the SUNDIAL does not predict yield, but if the potential uptake
expected conditions. A second series of simulations was is reduced for any reason, yield is reduced. This loss of yield
then performed using this same fertiliser recommendation. and losses to the environment by leaching and denitrification
This time, however, realised weather for the period are the yardsticks by which we assess the value of access to
following fertiliser application was used with different medium-term weather forecasts. As a context in which to
degrees of prior knowledge (achieved by setting w to 0.1, 0.2, interpret these losses, Table 3 gives the results from the
0.3, 0.5, 0.7) and different durations of that prior knowledge baseline simulations using the expected weather.
of 0, 1, 3, 7, 9 or 27 weeks. For the remainder of the
simulation until harvest, realised weather data that 2.5. Estimation of economic benefits
represents no prior knowledge was used. For each
simulation, results for leaching, denitrification and crop The benefits to the farmer in terms of avoided loss of
uptake of nitrogen were obtained. yield were calculated using the following sources of
Table 1
Climate regions used in the simulations
Climate Temperature Rainfall Corresponding Area cropped to the crops
(mean annual 8C) (annual mm) Defra regions used in this study (ha)
Cool wet 8.8 1034 Wales and North West 126498
Cool dry 9.1 597 North East, Yorkshire and Humberside 573253
Moderate 9.3 656 East and West Midlands 885167
Warm wet 10.2 851 South West 371634
Warm dry 10.0 584 East and South East and London 1248747
Data from two weather stations representative of each region was used.
A.G. Dailey et al. / Agriculture, Ecosystems and Environment 117 (2006) 22–28 25
Table 2
Potential yield (tonnes ha1) and dates of establishment and harvest of the crops used in the simulations
Crop Potential yield Dates Area of crop in England
Sand Loam or clay Establish Harvest and Wales (ha)
information. Product prices for April 2002 were obtained 3.2. Crop N offtake
from Farmers Weekly (5 April 2002); nitrogen and dry
matter content for oilseed rape from A.J. Macdonald (RRes, Crop N offtakes increased with forecast quality relative to
personal communication); crude protein and dry matter offtakes obtained with no prior knowledge (Fig. 1). With
contents for other crops from Agro Business Consultants knowledge of the weather up to 3 weeks in advance that is
(1998); crop areas for England from Defra (2002); crop 60% accurate, an increase of 1 kg N ha1 on average can be
areas for Wales, year 2001 from John Bleasdale, Agricul- expected in crop uptake in England and Wales. A forecast
tural Statistics, Welsh Assembly Government (personal that is 100% accurate 3 weeks into the future could push that
communication). gain up to 2 kg N ha1. A 7 weeks forecast that is 70 or 80%
Results are expressed, in general, as the benefit from reliable could increase N offtake (or rather, avoided shortfall
having prior knowledge of a given accuracy of the weather in N offtake) by up to 2 kg N ha1. These amounts are not
forecast. The benefit has been quantified as avoided loss of large but are averaged, systematic effects over the 80
crop offtake, avoided leaching or denitrification
(kg N ha1), or as the avoided loss of gross profit to the
Table 4
arable industry in England and Wales (pounds sterling for Mean decrease in leaching due to perfect knowledge, compared with no
the total cropped area). knowledge of future weather between fertiliser application and harvest
Climate region Mean decrease in leaching (kg N ha1)
Cool wet 1.5 *
3. Results and discussion Cool dry 1.2 *
Moderate 0.7
3.1. Losses of N Warm wet 3.1 *
Warm dry 0.8 *
Increased prior knowledge had a small direct environ- Weighted mean 1.1
*
mental benefit by decreasing the amount of N leached and of Significant at the 95% confidence level.
that denitrified before harvest. We expected these changes to
be small as only a small proportion of these losses occurs Table 5
ordinarily during the summer. Mean decrease (kg N ha1) in leaching due to perfect knowledge, compared
The reduction in losses by denitrification was not with no knowledge of future weather between fertiliser application and
statistically significant at any location, the biggest reduction harvest
being 0.2 kg (data not shown). With perfect knowledge of Crop Soil type
the weather for three or more weeks into the future, Sand Loam Clay
denitrification was reduced by an average of 0.1 kg N ha1 Winter wheat 1.4 2.0 *
1.9 *
only. Spring barley 0.5 0.6 0.6
The mean decrease in leaching of nitrogen due to the Winter oilseed rape 1.2 1.1 1.4 *
maximum level of prior knowledge was 1.2 kg N ha1 Potatoes 0.5 0.7 0.5
*
calculated over all soils, sites and crops. The wetter areas Significant at the 95% confidence level.
26 A.G. Dailey et al. / Agriculture, Ecosystems and Environment 117 (2006) 22–28
simulation years and represent a consistent benefit. There is and 25 kg N ha1 are roughly equivalent to a loss of half or
variability between the simulation years so that in some years one tonne ha1 of cereal yield respectively. For consistency
the benefit was more (see next section). With a virtually these same amounts of N were used as cut-off values in
perfect forecast, we found little difference in benefit between looking at oil seed rape and potatoes even though such
soil types (range 5.8–6.6 kg ha1) but more between crops amounts of N correspond to different amounts of yield in
(spring barley 4.7, winter wheat 7.2 kg N ha1, data not these crops. The larger number and magnitude of shortfalls
shown). Perhaps surprisingly there was little difference in N offtake rather than gains in the results (Fig. 2) arises
ascribable to the region and its weather (Table 6). This may be from the fact that advice from SUNDIAL is geared towards
because our analysis compares differences in prior knowledge supplying fertiliser N below an upper bound (related to the
within regions of the same weather. expected yield) in order to minimise the losses to the
Table 6
Mean benefit in crop offtake using a 27 week forecast at 90% accuracy
Climate region Mean benefit (kg N ha1)
Cool Wet 6.6*
Cool Dry 5.7*
Fig. 3. Frequency with which weather forecasts for the specified duration
Moderate 5.9*
and accuracy (w) would increase crop N offtake by 12.5 kg N ha1
Warm Wet 7.0*
(0.5 tonnes cereal yield) relative to SUNDIAL’s advice and with no prior
Warm Dry 5.9*
knowledge of the weather (w = 0). Averaged over all crops and weather
*
Significant at the 95% confidence level. stations.
A.G. Dailey et al. / Agriculture, Ecosystems and Environment 117 (2006) 22–28 27
might be worth more than £54M per annum. The benefits Dailey, G.J., Smith, J.U., Whitmore, A.P., 2005. Weekly weather generation
presented in this article represent the extra value of yield in for a nitrogen turnover model: getting representative results from fewer
simulations. Nutr. Cycl. Agroecosyst. 73, 257–266.
relation to nitrogen fertiliser only and make no attempt to Defra, 2002. Agricultural and Horticultural Census: 5 June 2002 England.
assess the value of better pest and disease predictions or of Defra, UK.
the timing and scheduling of operations. Defra, 2003. Agriculture in the United Kingdom. The Stationery Office,
London.
Farmers Weekly, 5 April 2002, Reed Business Information, Sutton, Surrey.
Gabriel, K.R., Neumann, J., 1962. A Markov chain model for daily rainfall
Acknowledgements occurrence at Tel Aviv. Quart. J. R. Met. Soc. 88, 90–95.
IPCC, 2001. Climate Change: The Scientific Basis. Cambridge University
Rothamsted Research is grant aided by the Biotechnol- Press, Cambridge, UK.
ogy and Biological Sciences Research Council. We grate- MAFF, 2000. Fertiliser Recommendations for Agricultural and Horticul-
fully acknowledge support from the UK Government tural Crops. RB209, 7th ed. The Stationery Office, London.
Matalas, N.C., 1967. Mathematical assessment of synthetic hydrology.
Department of the Environment, Food and Rural Affairs Water Resour. Res. 4, 937–945.
(Defra) and useful discussions with Margaret Glendining. Pretty, J., Brett, C., Gee, D., Hine, R.E., Mason, C.F., Morison, J.I.L., Raven,
H., Rayment, M., van der Bijl, G., Dobbs, T., 2001. Policy challenges
and priorities for internalising the externalities of agriculture. J. Environ.
References Plan. Manage. 44, 263–283.
Richardson, C.W., 1981. Stochastic simulation of daily precipitation,
temperature, and solar radiation. Water Resour. Res. 17, 182–190.
Agro Business Consultants, 1998. Agricultural Budgeting and Costing Smith, J.U., Bradbury, N.J., Addiscott, T.M., 1996. SUNDIAL: a PC-based
Book No. 47. Agro Business Consultants Ltd., Leicestershire, UK. system for simulating nitrogen dynamics in arable land. Agron. J. 88,
ADAS, 2005. http://www.planet4farmers.co.uk/welcome/index.html.
38–43.
Anon, 2000. Directive 2000/60/EC of the European Parliament and of the Sylvester-Bradley, R., Addiscott, T.M., Vaidyanathan, L.V., Murray,
Council of 23 October 2000 establishing a framework for community A.W.A., Whitmore, A.P., 1987. Nitrogen advice for cereals: present
action in the field of water policy. Off. J. L 327, 1–73. realities and future possibilities. In: Proceedings of the Fertilizer
Anon, 2001. Seasonal weather forecasting for the food chain. Final report.
Society, vol. 263, London, pp. 1–36.
Department of Trade and Industry, London, UK. Available online at Whitmore, A.P., Van Noordwijk, M., 1995. Bridging the gap between
http://www.foresight.gov.uk. environmentally acceptable and agronomically desirable nutrient sup-
Bradbury, N.J., Whitmore, A.P., Hart, P.B.S., Jenkinson, D.S., 1993. Mod- ply. In: Glen, D.M., Greaves, M.P., Anderson, H.M. (Eds.), Ecology
elling the fate of nitrogen in crop and soil in the years following
and Integrated Farming Systems. John Wiley and Sons, Chichester, UK,
application of 15N-labelled fertilizer to winter wheat. J. Agric. Sci. pp. 271–288.
121, 363–379.