Profit and risk in dryland cropping: seasonal forecasts and fertiliser management
Peter C. McIntosh1, Senthold Asseng2, Enli Wang3
CSIRO Oceans & Atmosphere Flagship, GPO Box 1538, Hobart TAS 7001, Australia, [email protected]
Department of Agricultural & Biological Engineering, University of Florida, Gainesville, FL 32611, USA, [email protected]
3
CSIRO Agriculture Flagship, GPO Box 1666, Canberra ACT 2601, Australia, [email protected]
1
2
Abstract
Fertiliser management in dryland cropping involves a trade-off between risk and return. Increasing the
fertiliser application rate can increase profit in good years, but may lead to a loss in bad years. Farmers try
to avoid the potentially serious consequences of consecutive loss years by adopting a conservative strategy,
typically applying $1 of fertiliser only when the long-term net return would be $2. A forecast of seasonal
rainfall can help inform fertiliser management decisions; application rates can be increased in forecast good
years and decreased in forecast bad years. The value of a seasonal forecast of above/below median rainfall
was determined at a variety of sites in the Australian wheat belt. The long-term value of the forecast varied
from $16/ha to $80/ha, compared to a realistic conservative strategy without a forecast. This value was
achieved with little increased risk. The same value could be obtained by increasing fertiliser rates in every
year, but the risk was higher. Seasonal forecasts give probabilistic information about the seasonal bias and
are therefore valuable in the long-run rather than in any given year. At the sites studied, and at an 80% level
of surety, seasonal forecast information was of value over time periods ranging from three to eight years.
Key words
Risk-averse nitrogen management, variable climate
Introduction
The application of nitrogen fertiliser in dryland cropping requires a balance between risk and return. The
optimum fertiliser rate will vary from year to year depending on seasonal rainfall. Too much fertiliser in a
dry year can lose money, while too little fertiliser in a wet year is a lost opportunity. Forecasts of seasonal
rainfall are now sufficiently skilful that they can be of considerable value when benefits are averaged over
a number of years. In the southern part of the Western Australian wheat belt, it has been shown that even
moderately skilful seasonal forecasts of rainfall can have a value exceeding A$50 /ha when averaged over
27 years (Asseng et al., 2012). This value exceeds that of previous studies (e.g. Ash et al., 2007; McIntosh et
al., 2007; Moeller et al., 2008; Wang et al., 2008) because of the use of a realistic and conservative baseline
management strategy in the absence of a forecast.
Seasonal forecasts are, by necessity, probabilistic rather than deterministic, regardless of the method used.
It is not possible to give definitive predictions at seasonal timescales because of the chaotic behavior of
the climate system. However, it is possible to make probabilistic statements about the chances of rainfall
in the coming season. The simplest format is to specify whether the rainfall is likely to be above or below
the long-term median value. The use of probabilistic information is not familiar to many people, and there
can be a tendency to reject a forecast system if the most likely outcome does not eventuate when it is first
used. However, the value of a probabilistic forecast system lies in the longer term, and it is important to
communicate this fact clearly.
This work had two goals. The first was to demonstrate that the excellent results achieved by Asseng et al.
(2012) in Western Australia are likely to be achievable in the eastern wheat-belt of Australia as well. The
second goal was to start to explore the number of years a forecast should be used in order to be reasonably
certain (at some level of probability) that it adds value.
Methods
The methods used here closely follow those used by Asseng et al. (2012) in Western Australia. The
experiments were designed to explore the value of climate forecasts in the absence of other factors, such
© 2015 “Building Productive, Diverse and Sustainable Landscapes “
Proceedings of the 17th ASA Conference, 20 – 24 September 2015, Hobart, Australia. Web site www.agronomy2015.com.au
as carryover from year-to-year of nitrogen and stored soil water, disease, pest, frost, heat stress and crop
rotation. It will be desirable in future studies to extend the realism of the simulation and explore non-linear
interactions by incorporating some or all of these additional factors.
There were four study sites in New South Wales (NSW) spanning low to moderate rainfall regimes for the
May-Nov growing season: Griffith (245 mm), Ardlethan (292 mm), Temora (329 mm) and Young (421
mm). A continuous wheat system on red Chromosol soil was simulated using APSIM Wheat version 7.5
(Holzworth
et al.,
2014).
cultivar
was sown
according
tomoderate
a variablerainfall
sowing
rule: when
There
were four
study
sitesThe
in New
South“Scout”
Wales (NSW)
spanning
low to
regimes
for the30 mm
total
rain
fell
in
10
consecutive
days
between
15
April
and
10
July,
or
on
10
July
if
there
was
insufficient
May-Nov growing season: Griffith (245 mm), Ardlethan (292 mm), Temora (329 mm) and Young (421
rain. Soil
water was
resetsystem
to the on
lower
and nitrogen
to 50 using
kg N/ha
on 15Wheat
April.version 7.5
mm).
A continuous
wheat
red limit
Chromosol
soil wasreset
simulated
APSIM
(Holzworth et al., 2014). The cultivar “Scout” was sown according to a variable sowing rule: when 30 mm
total
rain fell
in 10 consecutive
15 April by
anda10
July,oforprofit
on 10ratios
July iffrom
there$0was
Nitrogen
fertiliser
was applieddays
at abetween
rate determined
range
to insufficient
$3 per $1 of
rain.
Soil
water
was
reset
to
the
lower
limit
and
nitrogen
reset
to
50
kg
N/ha
on
15
April.
fertiliser applied. A profit ratio of $0 means that the fertiliser rate is determined from the maximum of the
gross margin (GM) versus N curve from past years. A profit ratio of $3 is determined from where the GM
Nitrogen fertiliser was applied at a rate determined by a range of profit ratios from $0 to $3 per $1 of
versus N curve has a slope of 3:1, and will generally result in a much smaller application of N. Farmers in
fertiliser applied. A profit ratio of $0 means that the fertiliser rate is determined from the maximum of the
Western
Australia
that they
a profit
of $2-3 (Asseng
et al.,
gross
margin
(GM) indicated
versus N curve
fromgenerally
past years.operated
A profitatratio
of $3ratio
is determined
from where
the2012)
GM so as to
reduce
the
risk
of
losing
money
in
a
dry
year.
versus N curve has a slope of 3:1, and will generally result in a much smaller application of N. Farmers in
Western Australia indicated that they generally operated at a profit ratio of $2-3 (Asseng et al., 2012) so as to
reduce
the risk
of losing
money inrainfall
a dry year.
Forecasts
of May
to November
made on 1 May were obtained from the Australian Bureau of
Meteorology’s Predictive Ocean-Atmosphere Model for Australia (POAMA). This model is a dynamical
Forecasts
of Mayocean-atmosphere-land
to November rainfall made
on 1 May
wereinitialized
obtained from
theforecast
Australian
Bureau
of a vast
global coupled
computer
model
at each
start
time from
Meteorology’s
Predictive
Ocean-Atmosphere
Model
for
Australia
(POAMA).
This
model
is
a
dynamical
network of atmosphere and ocean observations (e.g. Hudson et al., 2013; Zhao et al., 2014). The model
global coupled ocean-atmosphere-land computer model initialized at each forecast start time from a vast
provides retrospective forecasts extending back to 1981. While POAMA provides multiple forecasts to
network of atmosphere and ocean observations (e.g. Hudson et al., 2013; Zhao et al., 2014). The model
explore the
spread of forecasts
possible extending
outcomes,back
we simply
thePOAMA
mean rainfall
at each
site forecasts
and compared
this to
provides
retrospective
to 1981.used
While
provides
multiple
to
the
model’s
long-term
mean
rainfall
there.
At
the
four
sites
in
NSW,
POAMA
correctly
predicted
whether
explore the spread of possible outcomes, we simply used the mean rainfall at each site and compared this to
rainfall
would
be above
or rainfall
below the
median
74%
and 81%
of thecorrectly
time. predicted whether
the
model’s
long-term
mean
there.
At thebetween
four sites
in NSW,
POAMA
rainfall would be above or below the median between 74% and 81% of the time.
If the forecast in a particular year was for above median rainfall, then the GM versus N curve from which
Ifthe
thefertiliser
forecastrate
in a was
particular
year was
forconstructed
above median
rainfall,
then theyears
GM versus
curve
fromwas
which
the
determined
was
from
all historical
whereNthe
rainfall
observed
fertiliser
rate
was
determined
was
constructed
from
all
historical
years
where
the
rainfall
was
observed
to
be
to be above the median. The below median case was treated similarly. In experiments where no forecast was
above the median. The below median case was treated similarly. In experiments where no forecast was
required, all historical years back to 1981 were used. Example curves for Nyabing are shown in Figure 1; the
required, all historical years back to 1981 were used. Example curves for Nyabing are shown in Figure 1; the
othersites
sitesare
arequalitatively
qualitativelysimilar.
similar.
See
Asseng
et al.
(2012)
further
details
about
the method.
other
See
Asseng
et al.
(2012)
for for
further
details
about
the method.
Figure 1.
application
forfor
all all
years
(black),
above
median
rainfall
yearsyears
(blue)(blue)
Figure
1. Gross
Grossmargin
marginversus
versusNNfertiliser
fertiliser
application
years
(black),
above
median
rainfall
and
years
(red)
at at
Nyabing,
WA.
TheThe
N rate
appropriate
to a to
profit
ratioratio
of $2of
for$2each
andbelow
belowmedian
medianrainfall
rainfall
years
(red)
Nyabing,
WA.
N rate
appropriate
a profit
for $1
each $1
applied N is indicated by the green arrows. The other sites are qualitatively similar.
applied N is indicated by the green arrows. The other sites are qualitatively similar.
The “break-even time” is the number of years a farmer would need to grow a crop in order to be, say, 95%
© 2015
“Building
Productive,
Sustainable
“ time can then be compared between experiments
sure
that
the farm
is not Diverse
losing and
money.
The Landscapes
break-even
Proceedings of the 17th ASA Conference, 20 – 24 September 2015, Hobart, Australia. Web site www.agronomy2015.com.au
using different profit ratios, or between experiments with and without a forecast. Figure 2 shows the 5th and
© 2015 "Building Productive, Diverse and Sustainable Landscapes "
2
The “break-even time” is the number of years a farmer would need to grow a crop in order to be, say, 95%
sure that the farm is not losing money. The break-even time can then be compared between experiments
using different profit ratios, or between experiments with and without a forecast. Figure 2 shows the 5th
th
95 percentiles
of gross margin for all combinations of modelled years taken n at a time, where n is the
th
and
95
percentiles
margin
for all combinations
modelled
taken
n atmodelled,
a time, where
is the
number of years along of
thegross
x-axis.
For example,
at Nyabing,of
where
there years
were 27
years
therenare
number
of yearsofalong
the10x-axis.
Nyabing, where
27 yearsmeans
modelled,
therethe
are
many
subsamples
length
years.For
Theexample,
statisticalatdistribution
of thethere
set ofwere
subsample
provides
many
subsamples
of
length
10
years.
The
statistical
distribution
of
the
set
of
subsample
means
provides
the
percentiles when n=10.
percentiles when n=10.
Figure 2. Gross margin 5th and 95th percentiles over varying numbers of years (subsample sizes) for a profit
th
Figure
margin
and ratio
95th percentiles
over results
varyingare
numbers
of years
(subsample
sizes) for
a profit
ratio 2.
of Gross
$2 (green)
and5profit
$0 (red). These
for Nyabing,
WA.
The break-even
time
is indicated
ratio
of $2the
(green)
and profit
ratio $0positive.
(red). These results are for Nyabing, WA. The break-even time is indicated
where
5th
percentile
becomes
where the 5th percentile becomes positive.
Results
We compared the original Western Australian site (Nyabing) with preliminary modeling at the four eastern
Results
Australian sites. Further refinement of starting soil moisture and the sowing rule will be necessary to fully
Weaccount
compared
the differences
original Western
Australian
sitein(Nyabing)
withwest.
preliminary
modeling
at the
four easternhave
for the
between
conditions
the east and
However,
these initial
experiments
Australian sites. Further refinement of starting soil moisture and the sowing rule will be necessary to fully
shown encouraging gains from a seasonal forecast. At a $2 profit ratio, using a forecast increased the longaccount for the differences between conditions in the east and west. However, these initial experiments have
term profit at all sites between $16 /ha (Griffith) and $80 /ha (Ardlethan). The original value at Nyabing (on a
shown encouraging gains from a seasonal forecast. At a $2 profit ratio, using a forecast increased the longclay
soil)atwas
It is clear
the good results
obtained
in WA byThe
Asseng
et al.value
(2012)
will be feasible
term
profit
all $65
sitesha.
between
$16 that
/ha (Griffith)
and $80
/ha (Ardlethan).
original
at Nyabing
(on
in eastern
Australia.
a clay
soil) was
$65 ha. It is clear that the good results obtained in WA by Asseng et al. (2012) will be
feasible in eastern Australia.
There are various possible risk metrics for dryland cropping, such as the percentage of loss years, the mean
long-term
loss, possible
the chance
ofmetrics
two lossforyears
in a cropping,
row, and the
number
years to break
a
There
are various
risk
dryland
such
as the of
percentage
of losseven.
years,Using
the mean
forecast
increased
all
these
risks
at
Nyabing,
and
increased
the
mean
loss
at
Temora.
All
other
risks
remained
long-term loss, the chance of two loss years in a row, and the number of years to break even. Using a
the same
or decreased.
must
be compared
with the alternative
strategy
for obtaining
therisks
sameremained
profit;
forecast
increased
all theseThis
risks
at Nyabing,
and increased
the mean loss
at Temora.
All other
decreasing
the
profit
ratio
and
effectively
applying
more
N
in
each
year
regardless
of
the
forecast.
This
the same or decreased. This must be compared with the alternative strategy for obtaining the same profit;
amounts the
to moving
the green
arrow to the
right along
in Figure 1ofuntil
profit This
equals
decreasing
profit ratio
and effectively
applying
more the
N inblack
eachcurve
year regardless
the the
forecast.
that obtained
using
forecast.
In this
case,
outalong
of thethefour
risks
mentioned
and1 over
sites,
the that
risk
amounts
to moving
thea green
arrow
to the
right
black
curve
in Figure
until the
the five
profit
equals
obtained
using
a forecast.
In this
of the four
risks amentioned
andotherwise
over the five
sites, the
increased
in 15
out of these
20case,
casesout
compared
to using
forecast, and
remained
therisk
same.
increased in 15 out of these 20 cases compared to using a forecast, and otherwise remained the same.
A forecast adds value when the 5th percentile using a forecast exceeds the 5th percentile when a forecast is not
used. It may take a number of years for this to occur. This is a generalisation of comparing break-even times.
The number of years taken for a forecast to be of value (at the 80% level) was estimated this way at each site.
© 2015 “Building Productive, Diverse and Sustainable Landscapes “
Proceedings
ofProductive,
the 17th ASA
Conference,
20 – 24 September
2015,
© 2015
"Building
Diverse
and Sustainable
Landscapes
" Hobart, Australia. Web site www.agronomy2015.com.au
Proceedings of the 17th ASA Conference, 20 – 24 September 2015, Hobart, Australia. Web site www.agronomy2015.com.au
3
Future calculations will use a more robust bootstrapping method. A forecast was found to be of value after
between three and eight years at all sites. This calculation assumes the climate in each year is uncorrelated
with neighbouring years. If correlation between years is allowed for, the calculation is only approximate due
to the short time sequence available, but the indication is that this time increases by one to three years.
Conclusions
There is a trade-off between risk and return in dryland cropping because of Australia’s variable climate.
Applying more fertiliser each year increased the long-term return, but at an increased risk. By comparison,
strategic application of more fertiliser based on a seasonal climate forecast increased the return at reduced
risk. A seasonal forecast was found to be of value after three to eight years at the sites studied, and the return
was between $16 /ha and $80 /ha, indicating the great potential value of seasonal forecasts. Extension of
this work in future studies might take a whole-of-system approach and incorporate additional factors such as
knowledge of stored soil moisture at planting.
Acknowledgement
This work was funded in part by the Managing Climate Variability Program of the Grains Research and
Development Corporation.
References
Ash, A., McIntosh, P., Cullen, B., Carberry, P. and Smith, M.S., 2007. Constraints and opportunities in
applying seasonal climate forecasts in agriculture. Crop and Pasture Science, 58(10): 952-965.
Asseng, S., McIntosh, P.C., Wang, G. and Khimashia, N., 2012. Optimal N fertiliser management based on a
seasonal forecast. European Journal of Agronomy, 38: 66-73.
Holzworth, D.P. et al., 2014. APSIM–evolution towards a new generation of agricultural systems simulation.
Environmental Modelling & Software, 62: 327-350.
Hudson, D., Marshall, A.G., Yin, Y., Alves, O. and Hendon, H.H., 2013. Improving intraseasonal prediction
with a new ensemble generation strategy. Monthly Weather Review, 141(12): 4429-4449.
McIntosh, P.C., Pook, M.J., Risbey, J.S., Lisson, S.N. and Rebbeck, M., 2007. Seasonal climate forecasts for
agriculture: Towards better understanding and value. Field Crops Research, 104(0378-4290, 0378-4290):
130-138.
Moeller, C., Smith, I., Asseng, S., Ludwig, F. and Telcik, N., 2008. The potential value of seasonal forecasts
of rainfall categories—Case studies from the wheatbelt in Western Australia’s Mediterranean region.
Agricultural and Forest Meteorology, 148(4): 606-618.
Wang, E., Xu, J.H. and Smith, C.J., 2008. Value of historical climate knowledge, SOI-based seasonal climate
forecasting and stored soil moisture at sowing in crop nitrogen management in south eastern Australia.
Agricultural and forest meteorology, 148(11): 1743-1753.
Zhao, M., Hendon, H.H., Alves, O. and Yin, Y., 2014. Impact of improved assimilation of temperature and
salinity for coupled model seasonal forecasts. Climate Dynamics, 42(9-10): 2565-2583.
© 2015 “Building Productive, Diverse and Sustainable Landscapes “
Proceedings of the 17th ASA Conference, 20 – 24 September 2015, Hobart, Australia. Web site www.agronomy2015.com.au