A GENERAL SYSTEMS APPROACH TO APPLYlNG SEASONAL CLIMATE

FORECASTS

GRAEME HAMMER

Agricultural Production Systems Research Unit (APSRU)

Queensland Departments Primary Industries and Natural Resources and
CSIRO Tropical Agriculture
PO Box 102, Toowoomba, Qld 4350, Australia

Abstract

Agricultural and natural ecosystems and their associated business and government
systems are diverse and varied, ranging from farming systems to water resource systems
to species population systems to marketing and government policy systems, among
others. These systems are dynamic and responsive to fluctuations in c1imate.
Production, profit, conservation, and policy issues provide the major focus for
intervention in these systems. Risk, or the chance of incurring a fmancial or
environmentalloss, is a key factor pervading decision-making. Skill in seasonal c1imate
forecasting offers considerable opportunities to managers via its potential to realise
system improvements (ie. increased profits and/or reduced risks). Realising these
opportunities, however, is not straightforward as the forecasting skill is imperfect and
approaches to applying the existing skill to management issues have not been developed
and tested extensively.

The concepts associated with a systems approach to management are presented as a

suitable means to apply seasonal climate forecasting within a decision-making context.
An effective application of a seasonal c1imate forecast is defined as use of forecast
information leading to a change in a decision that generates improved outcomes in the
system of interest. A simple example of tactical management of row configuration in a
cotton crop on the Darling Downs, Queensland, is presented to demonstrate an effective
application. The profit outcomes with a fixed (aH years the same) decision, or a tactical
decision based on the SOI phase, or perfect knowledge of the season were compared
using a crop simulation study. Over the complete historical c1imate record, the tactical
approach gave an average profit increase of 11 %. Adopting a tactical approach,
however, did not give increased profit in every year. In 80% ofyears adopting a tactical
approach was as good as or better than not adopting it, but in 20% of years the manager
would have been worse off. It is suggested that effective imp1ementation requires
understanding of these risks and highlights the point that a1though tactical response to a
forecast may pay off on average over a period of years, there can be no guarantees for
the ensuing season.
51
G.L. Hammer et al. (etis.), The Australian Experience, 51-65.
© 2000 Kluwer Academic Publishers.
52

The systems approach to applying climate forecasts in decision-making across the range
of agricultural and natural ecosystems can be generalised to -
• understand the system and its management
• understand the impact of climate variability
• determine opportunities for tactical management in response to seasonal forecasts
• evaluate worth of tactical decision options
• participative implementation and evaluation
• feedback to climate forecasting
The nature of the interdisciplinary approach needed to pursue this systems approach to
applying seasonal climate forecasts is discussed.

1. Introduction

Agricultural and natural ecosystems are diverse and varied, ranging from farming
systems to water resource systems to species population systems, among others. These
ecosystems are dynamic and responsive to fluctuations in climate. Production and
conservation issues in these systems provide the major focus for management
intervention. While management focus will largely depend on the specific objectives
for intervening in any particular system, in general, a short term outlook concemed with
production and profitability issues needs to be combined with a longer term outlook
concemed with resource conservation and biodiversity issues. Climate variability
generates risks for management decision-making on both short and long time horizons
because outcomes of decisions cannot be predicted with any surety, be they decisions
on crop management, stocking rate, water allocation, or fish or insect population
management. Risk, or the chance of making a financial or environmentalloss, is a key
factor pervading decision-making in management of agricultural and natural ecosystems
(Hardaker et al., 1997).

Beyond the biophysical ecosystems, there is a range of other systems that effect, or are
affected by, management of those ecosystems. These are the business and govemment
systems that produce inputs required for management, market outputs, and determine
policies influencing ecosystem management. These business and govemment systems,
although operating at a different scale, are also dynamic and responsive to fluctuations
in climate. Climate variability generates risks for decision-making in these systems in
much the same way, because again, outcomes of decisions cannot be predicted with any
surety, be they decisions on inventories, marketing strategies, or drought or taxation
policy.

The introduction over the last decade of seasonal climate forecasts based on the El Nifio
- Southem Oscillation (ENSO) phenomenon and other research on global climate
forcing factors (Cane, 2000), has provided the basis to consider taking advantage of
climate variability, rather than passively accepting the risks it generates (Hammer and
Nicholls, 1996). The possibility of adjusting management to what the next season is
predicted to be offers considerable opportunities to managers of agricultural and natural
 53
ecosystems and the associated business and govemment systems. Realising these
opportunities, however, is not straightforward as the forecasting skill is imperfect and
approaches to applying the existing skill to management issues have not been deve10ped
extensively.

Hence, the aim of this paper is to

• present the concepts of a systems approach to management issues in agricultural and
natural ecosystems, and their associated business and govemment systems,
• consider what is an effective application of seasonal climate forecasting in these
systems, and
• via use of a simple example of an application using the systems framework, set out a
general framework for approaching applications

2. What is a Systems Approach?

A system can be defmed as a network of interacting e1ements receiving certain inputs

and producing certain outputs. The dynamics of a system is implicit in this defmition,
which can be applied not only to the full range of agricultural and natural ecosystems
and their associated business and govemment systems, but also to climate systems or
economic systems. It is important to consider boundaries cleady when defming a
system. The scale of the system and what is internal and external to the system are key
issues to be clarified in the systems approach. These issues influence the perspective
adopted when considering system function and management. A systems approach seeks
and utilises understanding of system composition and dynamics derived from relevant
research to enable prediction of the responses or behaviour of a system. In many cases,
this capacity to predict utilises system models, which are a simplified representation of
the system, often expressed mathematically. These concepts originated in the 1960's in
industrial systems (eg. Forrester, 1961) and were adapted to agricultural and natural
systems shortly after (eg. Duncan el al., 1967; Patten, 1971)

The range of systems relevant to agricultural and natural ecosystems and their
associated business and govemment systems includes
• a crop field and its management
• a farm and its management
• the national wheat crop and its marketing
• the national drought policy and its development
• a catchment and its management
• a species population and its management

For example, a crop field and its management consists of the crops, the soil and its
physical and chemical attributes, and the manager undertaking a range of management
practices, such as planting, cultivating, spraying and harvesting (Figure 1). Inputs to
this system include daily weather, fuel, fertiliser, and pesticide. Outputs include
harvested products, runoff, and soilloss. One could also look at balances for elemental
C and N in compounds entering and leaving the system. The dynamics involve crop
54

growth and development processes and soil chemical and physical processes given the
specific conditions of the system. The other systems listed could be similarly described
in terms of their components, inputs, outputs and dynamics.

Crop production
Transpiration

t
, I ' , I I
Evaporation I l ' I

t 1r~
6
Soil management
114 11

Runoff, erosion,
chemicals

T M E

Figure 1. Schematic ofa cropping system showing transitions ofthe system state through time in response to
management activities. Some system inputs and outputs are indicated.

Management of a system is the manipulation of that system to achieve specific

objectives. Hence, managers must be considered as an element of the system, Of at least
their actions considered as inputs to the system. The consequences of management
manipulations may be predicted using system models, although the adequacy of such
models for this task must be examined critically.

This systems approach can be applied to the full range of management and policy
decisions associated with agricultural and natural ecosystems and their associated
business and government systems. It is necessary, however, to place the systems
approach within a problem solving context to realise its potential in application to
decision-making. This concept is not new and stems from the rational approaches to
problem solving derived from operations research methods developed in industry in the
1960' s (Robertshaw et al., 1978). It has formed the basis for the development of
decis ion analysis procedures in agriculture (Dent and Anderson, 1971; Anderson and
White, 1991).

A general approach to problem solving takes the following steps

• de fine the issue or problem in the system of interest
• determine objectives and criteria for measuring them
• gather information and knowledge about the system of interest
 55
• identify and evaluate decis ion options
• implement preferred decis ion
• monitor the outcome and feedback new knowledge

Clearly defining the system of interest and the relevant issue and objective are critical
components in problem solving. Credible system models facilitate problem solving in
that they provide the means to evaluate altematives via simulation analyses.
Considerable scientific effort has been put to developing a range of system models. It
has been argued that this effort has been most effective where connection between
scientific rigour and predictive capability has been at the fore in driving model
development (Hammer, 1998).

A systems approach in a problem-solving context requires on-going connections

between decision-makers, advisors, modellers and researchers for effective outcomes.
This integration of skills is required to achieve the balance needed between
practicalities of system management, needs of decision-makers, and development and
use of system simulation or expert knowledge to evaluate options. Checkland (1983)
noted that the failure to be more aware of the human factor had resulted in a general
failure of "hard" systems analysis approaches to influence what practitioners do. The
failure of the "hard" methods to adequately address the human factor has lead to the
emergence of different methodologies for dealing with the "soft" people-related issues,
such as leaming processes (eg. Bawden and Packham, 1991). McCown et al (1994)
argued that the best prospects for developing betier policy and management strategies
lie with skilful use of "hard" system tools within a "soft" systems philosophical
framework. An interdisciplinary and participative approach facilitates this combination
and has been found effective in farming systems activities (McCown et al., 1994).

3. What is an Application of a Seasonal Climate Forecast?

An effective application of a seasonal climate forecast is the use of forecast information

that leads to the change in a decision that generates improved outcomes. A seasonal
forecast has no value if it does not generate changed decisions. But to be effective the
decision changes must produce positive changes in value by improving the relevant
aspect of system performance targeted. In agricultural and natural ecosystems this most
often relates to management decisions in those systems pertaining to their profitability,
use, and conservation. In the associated business and govemment systems changed
decisions may relate to profitability of the business and policy positions that are
designed to influence ecosystem management indirect1y.

An effective application differs considerably from the impact of a seasonal forecast.

Impact measures the extent of effect, whereas application is the management response
to change the anticipated impact. There have been many reports of significant impacts
of predictions of certain season types (White, 2000; Nicholls, 1986; Rimmington and
Nicholls, 1993; Meinke and Hammer, 1997) and while this is of interest in establishing
the relevance of seasonal forecasting, it is of no value if nothing can be done to change
56

from passive acceptance of impact to active response. Some studies have pursued
effective applications by exam:ining tactical changes in decisions associated with a
seasonal forecast (Mjelde et al., 1988; Hammer et al., 1991; Keating et al., 1993;
Hammer et al., 1996; MarshaH et al., 1996; Meinke and Stone, 1997).

3.1 EXAMPLE APPLICAnON - COTTON CROP MANAGEMENT

A simple example of potential tactical management of row configuration in a cotton

crop on the Darling Downs, Queensland, demonstrates an effective application of
seasonal climate forecasting. Dryland cotton is planted as a row crop and can be
configured with aH rows planted (solid), every third row missed (single skip), or two
rows planted and two rows missed (double skip). The rationale for manipulating row
configuration is that although yield potential in good seasons is lost as rows are deleted,
the slower rate of water use can give better yield in low rainfaH seasons and, thus,
reduce production variability risks associated with climatic variability. In addition,
pesticide spray and some other management costs are reduced as rows are missed. With
the solid row configuration, aerial spraying of insecticides is required, whereas more
frequent use of less costly ground spraying technologies is feasible with the skip row
configurations.

Is it possible to improve profitability by tacticaHy manipulating row configuration in

dryland cotton in response to a seasonal climate forecast? A simulation analysis using
the cotton crop model, OZCOT (CSIRO Division of Plant Industry), in the agricultural
production system simulator, APSIM (McCown et al., 1996) was used to compare the
management options over aH years of the historical climate record (1887 - 1993) for
Dalby, Queensland. Options were then compared for profitability over either aH years
or over sets of analogue years associated with a seasonal climate forecast. The
historical analogue sets were defmed using the SOI phase analysis system of Stone et
al. (1996). By taking the SOI phase immediately preceded sowing (August-September),
the climate of the subsequent growing season (October-April) could be aHocated to one
of five types for each year of the historical record. The best tactical management
system was determined by selecting the row configuration giving maximum profit in
each of the five season types associated with the SOI phases. This was compared with
the row configuration giving maximum profit over aH years to determine the degree of
profit improvement associated with responding to the climate forecast. Both the fixed
and tactical management approaches were compared to the "perfect knowledge" case,
where the management system giving highest profit each year was implemented.

The scenario simulated was for a Siokra-type cotton planted with row configuration
either solid, single skip, or double skip on 1 October on a Brigalow soil type with
260mm plant available water capacity to 1.8m depth and with 100mm of stored water in
the soil profile at planting. These starting conditions were reset every year of the
simulation so that the outcome reflected only variability associated with row
configuration and seasonal climate conditions. Standard management practices for the
region were assumed so that variable costs ofproduction were $850, $790, and $750 per
hectare for solid, single, and double skip, respectively (DPI Crop Management Notes).
 57

Retums were assumed to be $400 per bale of cotlon produced after alIowing for retums
from lint and seed and costs of ginning. Gross margin was calculated as the product of
retum per bale and bales produced per hectare less costs of production per hectare.

The simulated yield time series for the solid planting configuration (Figure 2) showed
extreme variability in annual production, which emphasises the riskiness associated
with growing dryland cotlon. The variability associated with single or double skip row
configurations was less but remained high (data not shown). Average simulated
production over alI years was 3.4, 3.2, and 2.8 bales per hectare for solid, single, and
double skip configurations, respective1y (Figure 3).

Average production varied among sets of years defmed by SOI phases (Figure 3).
When the SOI phase in August-September was consistent1y positive, average yield was
greater. In contrast, if the SOI phase was rapid1y falIing, average yield was lower. In
general, average yield was greatest with solid planting (Figure 3). It was only in years
with consistent1y negative Of rapidly falling SOI phase in August-September that
average yield for single skip exceeded solid planting. However, the production
averages do not reflect the economic outcome as costs differ among row configurations.
They also contain no information about the risk associated with the high year-to-year
variability (Figure 2).

-~
as
10
8
~ 6

-.-
as
.a
4
"C
a; 2
>
O
O o O O o
,.... oO)
O) ..- M LO
co O) O) O) O)
..-
O)
..-
..- ..- ..- ..-

Year
Figure 2. Simulated cotton yield versus year of planting from 1887 to 1992 for crops planted at Dalby,
Queensland, on 1 October each year.
58

-ca
~CI>
4

3 liI!I double

-"
(ij
.o
Q)
2

1
ll8 single

o solid

>=
O
neg pos fali rlse zero AII years
SOI phase
Figure 3. Average simulated cotton yield for three row configurations - solid, single skip row, and double
skip row. Averages are given for the entire period 1887-1992 (ali years) or for subsets of years within that
period derived from the SOI phase preceding planting (August-September). The five phase types are
consistent1y negative (neg), consistent1y positive (pos), rapidly falling (fali), rapidly rising (rise), and near
zero (zero).

Average gross margin over an years was greatest with solid planting ($495/ha), but this
was not consistent across groups of years associated with SOI phases (Figure 4).
Assuming the simple criterion of maximising gross margin or profit, the best tactic al
decision making use of the forecast information would be to plant single skip, solid,
double skip, solid, or single skip when the SOI phase preceding planting was
consistent1y negative, consistent1y positive, rapidly falling, rapidly rising, or near zero.
Ignoring the forecast information, the profit maximising strategy would be to plant solid
in all years. In years when the August-September phase was consistent1y positive,
average gross margin was much higher ($919/ha for solid planting). While this is a
substantial impact on retums, associated with identifying high yield years, it had no
consequence on the decision in those years, which remained as solid planting. This
contrasted with the situation when the August-September SOI phase was either
consistent1y negative or rapidly falling. In these season types, the application of the
forecast was effective as it ;ndicated benefit from a change of dec;s;on, as single and
double skip options (average gross margins $427/ha and $206/ha) were more profitable
than the solid planting configuration in those years ($340/ha and $ 129/ha, respectively).
This was associated with the likely lower rainfall in those season types, which caused
more favourable yield due to the slower water use of the single and double skip
systems, combined with the cost savings of the single and double skip systems.

Comparing the tactical and fixed management approaches over the complete historical
climate record, gave an average gross margin increase of $28/ha for the tactical
approach (gross margins $523/ha and $495/ha, respectively). This is an increase of
about 6% in gross margin or 11% in profit (calculated by deducting fixed costs). The
"perfect knowledge" case gave an average gross margin of $819/ha, which is an
 59

increase of 65% in gross margin or 130% in profit over the fixed management
approach. Hence, the seasonal climate forecasting system used yielded about 10% of
the value contained in a perfect forecast. These percentage increases for value of the
forecast information and its relationship to perfect knowledge are similar to that
reported for wheat in NE Australia (Hammer et al., 1996; Marshall et al., 1996).
Although the "perfect knowledge" situation is an unrealistic target, this indicates that
considerable scope remains for improving forecasts. This methodology provides a
general means to evaluate forecasts using the extent of the shift in value from no use of
forecasts to the perfect knowledge case.

-ca
1000

-...
~
.5C)
800

600 IiIIdouble
f:ŞJ s ingle
ca 400
osolid
==ti>ti>
...O
(!J
200

O
neg pos fali rise zero AII years

SOI phase
Figure 4. Average simulated gross margin for cotton grown using three row configurations - solid, single skip
row, and double skip row. Averages are given for the entire period 1887-1992 (all years) or for subsets of
years within that period derived from the SOI phase preced ing planting (August-September) as defined in
Figure 3. The solid line represents the level offixed costs ($2S0Iha).

In this example application, a number of simplifications were maintained for ease of

communication. For example, the risk, or chance ofmaking a loss, associated with each
management option was not considered. The simple criterion of maximising gross
margin or profit, independent of risk, was used. The trade-off between profit and risk,
however, is a vital component in decision analysis in this type of situation (Hardaker et
al., 1997) and has been an important aspect ofprevious studies (Hammer et al., 1996).
In this example, if risk is defined as the chance of making a loss (ie. not recovering
fixed costs) in any year, then there was a shift in risk with season type (Figure 4). There
was a considerably increased chance of making a loss in years with rapidly falling SOI
(as the average gross margin was less than the fixed costs) and there was a considerably
decreased chance of making a loss in years with SOI consistent1y positive. These shifts
in risk can have implications on tactical management responses if the risk differs among
management options within season types. This will not be examined in this case study
as it is dealt with in greater detail in subsequent chapters (eg Carberry et al., 2000), but,
in general, it should form part of the analysis procedure.
60
A further simplification is that this example related only to a specific situation - a
specific combination of soil condition, location and planting date. The outcome will
vary with changes in these conditions and cannot be generalised without further
analysis. To implement an application ofthis nature would require such further analysis
and direct interaction with decision-makers. The analysis presented provides only a
basis for discussion as a starting point in this process.

One of the key issues in discussing such analyses with decision-makers concems the
chance of coming out ahead in any one year if the forecast information was used to
adjust decisions as suggested. This average increase found for the tactic al approach is
derived from a distribution of differences between the two approaches (ie. using or not
using the forecast information) on a year-to-year basis. Adopting a tactical approach
did not give increased profit in every year. In some years it was substantially better,
whereas in others it was substantially worse (Figure 5). Overall, there were about 40%
of years in which the tactical approach gave higher profit, 40% of years with no
difference, and 20% of years with lower profit. This distribution is what the manager
faces each year and the risks must be understood for effective implementation.
Statistical testing of the time series of the benefit of the tactical over the fixed decision
strategy showed no significant pattems or trends in these data over the last century (data
not shown), indicating that outcomes from this distribution are equally likely in any
year. While it is clear that, on average, outcomes can be improved in the longer term,
there can be no guarantees for the ensuing season. The question a manager faces in
deciding whether or not to apply a seasonal climate forecast is - am I content with an
80% chance of do ing as well as Of better than I would do without applying the forecast?
This percentage is the targeted piece of information needed and cannot be derived by
simple examination of the forecast. Dealing with probabilistic information remains an
important consideration in implementing such applications. This issue is addressed in
subsequent chapters (Meinke and Hochman, 2000; Nicholls, 2000).

100U
800

., .-
60O
40O .
..
•
20O 1.. •'1 1...
'.
..1 1. 1_ 1 III
O :1 I '1 ~
1- ~-
-20
"
-40'U ~
-60'U
~ ~
~ -
~ ~

-80IV-
Year
Figure 5. Difference in gross margin between tactical (responsive to seasonal climate forecast) and fixed
(non-responsive) row configuration management strategies for each year ofthe cotlon simulation study.
 61

4. Using the Systems Approach in Applying Seasonal Climate Forecasts

To generalise from the cotton management example, the emphasis in a systems

approach is to develop targeted information for inf1uencing the most relevant decisions
in the system of interest. This concept is relevant across the range of scale and issues
associated with agricultural and natural ecosystems and their associated business and
government systems (Figure 6). Whether the key issue is short-term profitability or
long-term resource conservation, the emphasis needs to be on the analysis required to
target the forecast information to the issue and the decision-maker. Generalised
seasonal forecasts, which have information relevant across alI systems, are likely to
have little value if their targeting is not considered. The relevant decision-maker at each
scale must be included as part of the systems approach to ensure clear problem
definition and understanding of relevant decisions and information needs. The systems
approach will often involve systems modelling as a means to move from general to
targeted information. A number of subsequent chapters will pursue this systems
approach for this diverse range of systems.

Farmer
Business & Resource Managers
[ Government

t t t t t
-g Crop Pasture Property Land & Water Marketing
4i Herd Policy
~ Management
CII
II) 1-

~
c
.2
m
E
....
o
c_
- I!
G>
C
c!
Paddock Farm Catchment Reglon Natlon

Scale Axis
Figure 6 (B). The relationships between scale, information content, and decision-makers in defining relevant
systems and the systems approach to applying seasonal climate forecasts in agricultural and natural
ecosystems. (This figure is reproduced in the colour section al the end of the book as plate B)

The systems approach to applying climate forecasts to decision-making in agricultural

and natural ecosystems can be generali sed to a sequence of steps that are relevant to the
wide range of types of decis ion in these systems. The steps can be summarised as-
62
1. Understand the system and its management - Whether the system is a paddock, a
farm, or the national wheat crop it is essential to understand its dynamics and
management and the points at which significant decisions influence desired system
performance. Such understanding can only be gained by close interaction with system
managers or their advisers. Developing predictive system models is an excellent means
to leam about the system as well as generate a suitable tool for subsequent use.

2. Understand the impact of climate variability - It is essential to know just where in the
system climate variability is an issue. Only at those points will a seasonal forecast be
worth pursuing.

3. Determine opportunities for tactical management in response to seasonal forecasts -

If a seasonal forecast was available, what possible options are there at relevant decision
points? That is, how might decisions be changed in response to a seasonal forecast?
What nature of forecast would be most useful? What lead-time is required for
managementresponse?

4. Evaluate worth of tactical decision options - Given a range of potential tactic al

management responses to a seasonal forecast, an evaluation of their expected outcomes
and risks is required as a basis for discussion with decision-makers. Simulation
analyses using system mode1s provide a means to evaluate tactic al options.

5. Participative implementation and evaluation - Working with managers/decision-

makers generates valuable insights and leaming throughout the entire process. An
evaluation of outcomes with respect to changed decisions or processes highlights issues
for subsequent focus.

6. Feedback to climate forecasting - Rather than just accept a given climate forecast,
consider what specific improvements would be of greatest value in the system. This
can provide some direction for the style of delivery of forecasts and for climate
research.

5. Discussion

Connecting forecasts into applications remains an area where we have a lot to leam.
This paper argues that the systems and operational research approach to applying
seasonal climate forecasts provides an appropriate and general methodology for
integrating forecasts into decision-making. Implementing this approach, however,
imposes requirements on design of forecasts. Using agricultural and natural system
models to analyse and evaluate decision options, usually requires input streams of daily
climate data that sample the seasonal variability likely to be experienced. Many
seasonal climate forecasts provide only general predictions and, hence, are difficult or
impossible to use in the systems approach to applications. A forecasting scheme that
identifies historical analogue seasons or years, as in the cotlon example given above, is
perhaps the most useful concept to date for facilitating applications. The use of this
 63

style of forecasting scheme in many of the case studies to follow in this book is
testament to this. If seasonal forecasts are intended for targeted use as distinct from
general information (Figure 6) then they must be designed to accommodate this
requirement.

The analogue seasons or years concept does not, however, need to be confmed to
partitioning the historical record as in the cotton case study. It may be possible to
derive synthetic analogues in some way via stochastic weather generators, for example.
This would facilitate connection of forecasts to applications for forecasting systems that
do not generate historical analogues, such as forecasts derived from dynamic ocean
and/or atmosphere models. Developments in weather generators would also facilitate
inclusion of effects of potential climate change, which may underlie trends in historical
data that need to be considered when using historical analogues. The point remains,
however, that development of appropriate methods to derive analogue seasons or years
in order to connect forecasts and applications needs to be viewed as an essential
component of forecasting research and development. It is an area where increased
effort is needed.

Quantifying the unexplained variability in a seasonal climate forecast is as important in

applications as the predicted shift in mean climate. This unexplained variability is what
drives decision-making risks and opportunities. It needs to be communicated in a useful
way, not hidden in an average, median, or simple probability statement. Generating an
average forecast from some ensemble may simplify forecast delivery, but it removes the
unexplained variability from the forecast. The issuing of forecasts as simple probability
statements is better, but provides on1y general inforrnation. Defming analogue seasons
or years captures this variability in a way that enables riskiness of alternative decisions
to be evaluated by examining each year in the analogue set separately (eg. Figure 3).

Separate analysis of each analogue year is essential, as is careful consideration of how

to condense the resulting output to the most useful form. A verages are often a far less
meaningful statistic of the probability distribution of outcomes than some consideration
of the likelihood of exceeding (or not exceeding) some critical system state, be it profit
or land condition. Given this approach, outcomes of decision options can then be
assessed in terms of both their expected outcome and the associated risk. Both aspects
are important to the decision-maker when having to decide among competing options.
One of the most important features of a useful forecast may be the extent to which it
reduces risk ofmaking a loss (Hammer et al, 1996).

Understanding targeted information about effects of seasonal climate forecasts on likely

outcomes of decision-making options is the key to successful application of forecasts.
The inforrnation is probabilistic in nature and is relevant to managing risks over a
number of years. In any one year, responding to a forecast may or may not pay-off.
Over a number of years, however, responding to a forecast results in significant
advantages (eg. Figure 5). The notions of a "win" or a "loss" or "getting it right" or
"getting it wrong" on a single occasion are not really appropriate. The shift in outcome
probabilities associated with application of a forecast is valuable in risk management if
64

used with such understanding. This requires close interaction with decision-makers as
part of the development process. The participative approach facilitates leaming by the
decision-maker in relation to the nature ofthe information and how to use it effectively,
while facilitating leaming by the researcher in relevant decisions to target and in
communication process. This approach provides case studies that are useful for broader
educative activities with decision-makers and researchers. The difficulties associated
with using this type of information are explored in some depth in subsequent chapters
(Meinke and Hochman, 2000; NichoBs, 2000).

Converting general information on seasonallikelihood to targeted information on likely

outcomes of decision options requires robust and reliable models of the systems in
question. This is not a trivial task and should not be taken for granted. The models
used in the cotlon example have taken many years of intensive research, development,
and testing to get to their current state. There are many possibilities for improvement in
models (eg. Hammer, 1998) making it important to maintain on-going effort in this
area.

The systems approach to applying seasonal climate forecasts offers considerable

opportunities to improve decision-making in our agricultural and natural ecosystems. It
does, however, require interdisciplinary action. ContinuaI interaction of forecast
application R&D, forecast development R&D, and decision makers managing the target
systems is essential to ensure that forecasts are produced in a way that maximises their
utility and relevance. Such interaction also provides opportunities to generate targets
for improvement of forecasts as weB as a means to capture quickly any innovations in
forecasting.

Acknowledgments

The use of the OZCOT cotlon model developed by CSIRO Division of Plant Industry,
Cotlon Research Unit, Narrabri and assistance from Dr P Carberry and Mr D Holzworth
(APSRU) in conducting the simulation analysis are gratefuBy acknowledged.

References

Anderson, J.R. and White, D.H. (1991) Systems thinking as a perspective for the management of dryland
farming, in V. Squires and P.G. Tow (eds.), Dryland Farming - A Systems Approach. Sydney
University Press, Sydney. pp.16-23.
Bawden, RJ. and Packham, R.G. (1991) Improving agriculture through systemic action research, in V.
Squires and P.G. Tow (eds.), Dryland Farming - A Systems Approach. Sydney University Press,
Sydney. pp.261-270.
Cane, M.A. (2000) Understanding and predicting the wor1d's climate system, in G.L. Hammer, N. Nicholls,
and C. Mitchell (eds.), Applications of Seasonal Climate Forecasting in Agricultural and Natural
Ecosystems - The Australian Experience. Kluwer Academic, The Netherlands. (this volume)
Carberry, P.S., Hammer, G.L., Meinke, H. and Bange, M. (2000) The potential value of seasonal climate
forecasting in managing cropping systems, in G.L. Hammer, N. Nicholls, and C. Mitchell (eds.),
Applications of Seasonal Climate Forecasting in Agricultural and Natural Ecosystems - The
Australian Experience. Kluwer Academic, The Netherlands. (this volume)
 65

Checkland, P.B. (1983) OR and the systems movement: Mappings and conf1icts. J. Operational Research
Society 34,661-675.
Dent, lB. and Anderson, J.R. (1971) Systems analysis in agricultural management. John Wiley, Sydney.
pp.394.
Duncan, W.G., Loomis, R.S., Williams, W.A. and Hanau, R. (1967) A model for simulating photosynthesis in
plant communities. Hilgardia 38, 181-205.
Forrester, J.W. (1961) Industrial dynamics. Massachusetts Institute of Technology Press, Cambridge,
Massachusetts, pp.464.
Hammer, G.L. and Nicholls, N. (1996) Managing for climate variability - The role of seasonal climate
forecasting in improving agricultural systems. Proc. Second Australian Conference on Agricultural
Meteorology. Bureau of Meteorology, Commonwealth of Australia, Melboume. pp.19-27.
Hammer, G.L. (1998) Crop modelling: Current status and opportunities to advance. Acta Horticulturae 456,
27-36.
Hammer, G.L., McKeon, G.M., Clewett, J.F. and Woodruff, DR (1991) Usefulness of seasonal climate
forecasts in crop and pasture management. Proc. First Australian Conference on Agricultural
Meteorology. Bureau of Meteorology, Commonwealth of Australia, Melboume. pp.15-23.
Hammer, G.L., Holzworth, D.P. and Stone, R. (1996) The value of skill in seasonal climate forecasting to
wheat crop management in a region with high c1imatic variability. Aust. J. Agric. Res. 47,717-737.
Hardaker, lB., Huime, R.B.M. and Anderson, lR. (1997) Coping with risk in agriculture. CAB International.
pp.274.
Keating, B.A., McCown, R.L. and Wafula, B.M. (1993) Adjustment of nitrogen inputs in response to a
seasonal forecast in a region ofhigh climatic risk, in F.W.T. Penning de Vries et al. (eds.), Systems
Approachesfor Agricultural Development. Kluwer, The Netherlands. pp.233-252.
Marshall, G.R., Parton, K.A. and Harnmer, G.L. (1996) Risk attitude, planting conditions and the value of
seasonal forecasts to a dryland wheat grower. Aust. J. Agric. Econ. 40,211-233.
McCown, R.L., Cox, P.G., Keating, B.A., Hammer, G.L., Carberry, P.S., Probert, M.E. and Freebairn, D.M.
(1994) The development of strategies for improved agricultural systems and land-use management,
in P. Goldsworthy and F.W.T. Penning de Vries (eds.), Opportunities, Use, and Transfer of Systems
Research Methods in Agriculture to Developing Countries. Kluwer, The Netherlands. pp.81-96.
McCown, R.L., Hammer, G.L., Hargreaves, J.N.G., Holzworth, D.P. and Freebairn, D.M. (1996) APSIM: A
novel software system for model development, model testing, and simulation in agricultural research.
Agric. Sys. 50, 255-271.
Meinke, H. and Hammer, G.L. (1997) Forecasting regional crop production using SOI phases: a case study for
the Australian peanut industry. Aust. J. Agric. Res. 48, 231-240.
Meinke, H. and Stone, R.C. (1997) On tactical crop management using seasonal climate forecasts and
simulation modelling - a case study for wheat. Sei. Agric., Piracicaba 54,121-129.
Meinke, H. and Hochman, Z. (2000) Using seasonal c1imate forecasts to manage dryland crops in northern
Australia, in G.L. Hammer, N. Nicholls, and C. Mitchell (eds.), Applications of Seasonal Climate
Forecasting in Agricultural and Natural Ecosystems - The Australian Experience. Kluwer
Academic, The Netherlands. (this volume)
Mjelde, J.W., Sonka, S.T., Dixon, B.L., and Lamb, P.J. (1988) Valuing forecast characteristics in a dynamic
agricultural production system. American Joumal of Agricultural Economics 70, 674-684.
Nicholls, N. (1986). Use of the southern oscillation to predict Australian sorghum yield. Agric. For.
Meteorol. 38, 9-15.
Nicholls, N. (2000) Impediments to the use of climate predictions, in G.L. Hammer, N. Nicholls, and C.
Mitchell (eds.), Applications of Seasonal Climate Forecasting in Agricultural and Natural
Ecosystems - The Australian Experience. Kluwer Academic, The Netherlands. (this volume)
Patten, B.C. (1971) Systems Analysis and Simulation in Ecology. Academic Press, New York.
Rimmington, G.M. and Nicholls, N. (1993) Forecasting wheat yields in Australia with the southern oscillation
index. Aust. J. Agric. Res. 44, 625-632.
Robertshaw, J.E., Mecca, S.J. and Rerick, M.N. (1978) Problem Solving: A Systems Approach. Petrocelli
Books Inc., New York. pp.272.
Stone, R.C., Hammer, G.L. and Marcussen, T. (1996) Prediction of global rainfall probabilities using phases of the
Southern Oscil1ation Index. Nature 384, 252-256.
White, B. (2000) The importance ofclimate variability and seasonal forecasting to the Australian economy, in
G.L. Hammer, N. Nicholls, and C. Mitchell (eds.), Applications of Seasonal C/imate Forecasting in
Agricultural and Natural Ecosystems - The Australian Experience. Kluwer Academic, The
Netherlands. (this volume)