Vous êtes sur la page 1sur 9

Agricultural Systems 25 (1987) 219-227

Economic Thresholds and Response to Uncertainty in


Weed Control

Bruce A. Auld
Agricultural Research and Veterinary Centre,
Forest Rd., Orange, NSW 2800, Australia
&
Clem A. Tisdell
Economics Department, University of Newcastle,
NSW 2308, Australia

(Received 2 February 1987; revised version received 21 May 1987;


accepted 22 May 1987)

S UMMA R Y

The influence of uncerta&ty on economic thresholds~critical density models &


weed control is examined. Two principal sources of uncertainty, potential
weed density and the form of the crop loss function, are discussed. A range of
criteria including maximisation of net gain, minimax gain and mean-risk
analysis are considered in relation to the decision to control or not control
weeds in a erop. A notional example of wild oats in wheat illustrates how the
various decision criteria can also be considered as thresholds. I f risk aversion
is present, uncertainty about the weed loss function and weed densities
increases the likelihood that control of weeds is optimal and specOqc threshold
weed densities become less relevant

INTRODUCTION

Economic threshold or critical density models are used to determine


circumstances in which profit from controlling weeds exceeds the cost of
doing so. Their use in decision making in weed control has recently received
increasing attention (Marra & Carlson, 1983; Cousens et al., 1986; Doyle et
al., 1986). Eco nom i c thresholds are influenced by variation in control costs
219
Agricultural Systems 0308-521X/87/$03"50 Elsevier Applied Science Publishers Ltd,
England, 1987. Printed in Great Britain
220 Bruce A. Auld, Clem A. Tisdell

and prices received for crop products although these can sometimes be
assumed to be constant. However, if costs of weed control increase with
increasing weed density, it is possible that there are two thresholds (Auld
et al., 1987). In this case, control at weed densities greater than the higher
threshold density is uneconomic. It is also possible that control costs
increase in a stepwise fashion with weed density (e.g. Medd et al., 1987); this
gives rise to the possibility of multiple thresholds. In addition, agronomic
factors such as crop sowing time and rate, as well as weather, affect potential
yield and, consequently, economic thresholds.
Apart from these factors, many uncertainties can arise in weed control
which will also influence the precision with which an economic threshold can
be estimated. These include effectiveness of herbicides and phytotoxicity of
herbicides to crops. In this paper, we consider the uncertainties of the weed
population density and the crop loss function in relation to weed control. We
discuss possible decision criteria which can be used in response to these
uncertainties, given the two options: to control or not control a weed
population.
Several concepts of weed control thresholds exist (Cousens, 1987). This
paper extends the critical density/weed control model to allow for risks and
uncertainty and thereby links economic and 'safety' thresholds as described
by Cousens (op. cit.). In doing so it incorporates risk in an objective manner.
However, one should not assume a false sense of accuracy since the degree of
risk aversion will vary among farmers.

U N C E R T A I N T I E S IN WEED C O N T R O L

Weed population

A major factor in decision making about weed control is the size of the weed
population in current and future years. Marra & Carlson (1983) suggest that
future benefits from carryover effects of controlling weeds in one year to
succeeding years may be so uncertain that it is best to ignore them.
Nevertheless, increased uncertainty about the future may be taken into
account by applying larger discounts to future costs and benefits, thereby
putting a reduced weight on these in decision making.
In the current cropping year uncertainty about the potential weed
population is relevant. Particularly where pre-emergence herbicides are to
be used, a farmer may be uncertain about the level of weed infestation
expected in the absence of treatment; loss to be anticipated and therefore
gains from treatment are uncertain.
Even for emerged weeds where weed density at the economic threshold is
Economic thresholds and uncertainty in weed control 221

low, it may be difficult to obtain an accurate estimate of the weed population


because of non-random distribution of weeds and difficulties in sampling
low density populations in the field.

Yield response function

The effect of weed density on crop yield is generally to cause increasing yield
loss, but at a decreasing rate (Dew, 1972; Chisaka, 1977). Although Zimdahl
(1980) has suggested a sigmoidal relationship there appears to be no
empirical evidence for it and it has been refuted on theoretical grounds
(Cousens et al., 1984). Marra & Carlson (op. cit.) suggest that a linear loss
function is a reasonable approximation and this may be so in the region of
the economic threshold. However, while the general form of the crop loss
function for a specific weed/crop system may be known, its precise shape
may vary with location and agronomic factors. Thus, the economic
threshold derived from the value of increased yield from weed control will
vary accordingly. In Fig. 1 possible revenues from weed control in relation to
weed density are represented by a range of values bounded by two curves
(OBD and OB'D'). The economic threshold determined by the intersection
of these curves and the cost of control curve (CE) (assumed a constant for all
weed densities) is represented by a range W 1 to W2. If the maximum yield
loss function occurs W 1 is the economic theshold. If the minimum yield loss
function occurs W2 is the economic threshold.
Given that functions for the value of increased yield from treatment are
positively sloped throughout, it follows that if the weed density is less than
W1 it is not economic to treat the weed, but is is always economic to treat it if

d
c

0
ii
W, w2
Weed density (W)
Fig. 1. The economic threshold is represented by the range of weed densities W 1 to W 2
where there is uncertainty about the value of increased yield function between the bounds
O B D and O B ' D ' and C E represents cost of treatment.
222 Bruce A. Auld, Clem A. Tisdell

the density exceeds W 2. In cases where W 1 < W < W 2 the attitude of the
farmer towards uncertainty will influence the decision about whether it is
optimal or not to treat the weed.

RESPONSE TO U N C E R T A I N T Y

Weed population uncertainty

Figure 2 depicts the net gain function (J(W)) in relation to weed density (W)
after allowing for control costs. If yield loss decreases, at a decreasing rate in
the usual way, then J'(W)> 0 and J"(W)<0. That is, the net gain function
is an increasing strictly concave function of weed density. W2 is the threshold
weed density; the expected value E(J) = 0, gains from treatment of weeds just
cover costs.
Let us assume there is some uncertainty about the size of the weed
population such that there is an 0"5 probability of it being W 1 and an 0"5
probability of it being W 3 (where W2 = 0-5 WI + 0"5 W3). (Alternatively, we
could express the assumption as an equal probability of the weed density
being anywhere between W~ and W3.) The expected value of J is J~ and is
negative. Weed control in this case (Fig. 2) would be uneconomic on average.
Thus, if a farmer wishes to maximise his expected net gain (where there is
uncertainty of the type shown in Fig. 2) he would not control weeds. Where
the net gain function (Fig. 2) can be represented by a quadratic, E(J) declines
linearly with an increase in the variance of W(and, with two observations,

~ J(W)

)y
J' "" -~- - - ' ~ l " r weW~d density (W)

Fig. 2. The net gain function from treatment of weeds is indicated by J(W). Where there is
uncertainty about the weed density and it may occur anywhere between W1 and W2 with
equal probability the expected gain at the economic threshold weed density ( W2) is Jl and
negative.
Economic thresholds and uncertain O' in weed control 223

W 1 and W3,with an increase in the difference between them). If gain function


J(W) were linear, then uncertainty would not change the expected value of J.
This response to uncertainty, showing no risk aversion, does not appear to
be what generally occurs in practice. Reichelderfer (1980), for instance, notes
that risk perception and aversion may be a stronger motivation than profit
maximisation among farmers. It is also conceivable that a farmer may wish
to maximise utility (Luce & Raiffa, 1966) (satisfaction) from profit rather
than maximise profit. Utility may increase with profit but at a decreasing
rate. Consequently, since profit in the absence of weed control tends to
decline with increased weed density, utility can also be expected to decline
with greater weed density. Thus in this case, the farmer's utility curve as a
function of weed density becomes strictly concave. In these circumstances
greater uncertainty about weed density tends to favour weed control as weed
densities become more uncertain.
An extremely risk-averse strategy the minimax gain criterion (Luce &
Raiffa, op. cit.), is one which attempts to maximise minimum possible gains.
As the degree of uncertainty about weed density increases, the likelihood
that the density may exceed the economic threshold value rises and the
minimax strategy is treatment. In contrast, the maximax criterion (Luce &
Raiffa, op. cit.) is one which assumes that the most favourable conditions will
occur. In this case if there is any chance that the actual weed density will be
less than the threshold density, the weed is not treated. Greater uncertainty
reduces the likelihood of weed treatment by this criterion.
These criteria are extreme. The expected gain criterion is risk neutral, the
minimax gain criterion is ultra-pessimistic whereas the maximax is ultra-
optimistic. In most cases, there may be some risk-aversion but not of the
extreme type assumed by the minimax criterion. This can be captured to
some extent by E-V or mean-risk analysis (Anderson et. al., 1977). Each
individual is assumed to have a preference function for mean gains and risks
(variance in expected gains). If risk aversion is present the acceptable level of
risk increases with increasing expected gains.

Crop loss function uncertainty

We now consider the impact of uncertainty or increased uncertainty about


the crop loss function assuming that its expected value in relation to each
possible weed density remains constant. In the case of the minimax and
maximax criteria and E-V risk aversion, the consequences of greater
uncertainty about the crop loss function are similar to that where the weed
density is uncertain. However, in the case of the maximisation of expected
gain criterion there is a difference. Increased uncertainty about the crop loss
function does not alter the decision about whether to treat weeds provided
224 Bruce A. Auld, Clem A. Tisdell

that with this increased uncertainty the expected value of the function for
each weed density remains constant.

AN E X A M P L E - - W I L D OATS IN WHEAT

Figure 3 illustrates a range of (single year) crop loss functions for wheat
infested by wild oats (Arena fatua) in eastern Australia; the resulting
thresholds are within the wider range which may be encountered in this
region (Martin, 1986). For the purposes of this example, we shall consider
that L1 is more likely and that the expected (arithmetic average) loss
function is E(L). Control cost, by herbicide including application cost, is
A$40 ha-1 (2 litres ha-1 Diclofop methyl).

,1
t
601
0 2 0 ~

5 10 20
Wild oat density plants / m2
Fig. 3. Variation in thresholds for a range of possible loss functions between curves L I and
L 2 for wild oats in wheat in eastern Australia in one year. Cost of control is A$40 and E(L)
represents the expected (arithmetic average) loss function (i.e. L 1 is more likely than Lz).

If expected gain is to be maximised, the threshold density for wild oats is


eight plants per square metre. IfE-V analysis is applied and there is some risk
aversion, there will be a tendency for the likelihood of treatment being
optimal to rise as uncertainty about the loss function rises. This is because
although the mean gain associated with the non-treatment strategy remains
unaltered, the variance of gain rises, whereas after treatment gain is assumed
to be certain.
If the minimax gain criterion is applied to the weed treatment decision,
greater uncertainty about the weed loss function increases the probability
that treatment is optimal. It becomes increasingly likely that, for a possible
weed loss function, the actual weed density exceeds the threshold value. In
Economic thresholds and uncertainty in weed control 225

Fig. 3 if the actual wild oat density is six plants per square metre non-
treatment is optimal if the loss function is definitely E(L). However, if it is
possible that it is L1 or L 2, the minimax criterion requires treatment of
weeds. This is so because if the loss function L 1 holds a greater loss than
A$40 would occur.
In the case of the maximax criterion, greater uncertainty about the weed
loss function reduces the likelihood that treatment is optimal. This is
because the possibility of the actual weed density being below the critical
density rises. For example, in Fig. 3, if the actual weed density is twelve plants
per square metre, treatment is optimal if the weed loss function is definitely
E[L], However, if L 1 and L 2 each become probable, there is a possibility of
the threshold weed density being twenty plants per square metre and a user
of the optimistic maximax criterion would not treat the weed.
A further risk decision criterion can be considered in relation to
uncertainty about the crop loss function. It is the minimax regret (or risk)
criterion (Luce & Raiffa, op. cir.). It involves minimising the maximum
possible regret or opportunity cost in deciding on a weed control strategy.
This is illustrated graphically in Fig. 3. Ten plants per square metre is the
weed density at which AB = BC; at this weed density the minimum of the
maximum possible opportunity cost or regret possible from non-treatment
of weeds equals that from treatment. Under the minimax regret criterion,
where W > ten plants per square metre, the weeds are treated. The greater
the possible range of the critical weed densities, the more likely the minimax
regret criterion indicates treatment of weeds.
If a farmer assumed L 2 (Fig. 3) were the loss function for his particular
circumstances, uncertainty about wild oat density would be unlikely to
influence his decision about whether to control or not if, expected profit
maximisation were his criterion for action (because the function is almost
linear in the threshold region). However, if he anticipated loss functions E(L)
or L~, greater uncertainty about wild oat density increases the likelihood
that non-control maximises expected gains, on average. This is because of
the strict concavity of the loss functions in the threshold region.
Given the original hypothesis that E(L) is the expected loss function but
that possible loss functions range between L~ and L 2, the various criteria can
each be considered as 'thresholds'. The minimax gain threshold is five wild
oat plants per square metre, the maximax gain threshold is 20, the
minimax regret threshold is 10 and the maximisation of net gain threshold
is 8.

CONCLUSION

Increased uncertainty about either the weed density or the weed loss
function increases the likelihood that weed treatment is optimal if the
226 Bruce A. Auld, Clem A. Tisdell

minimax gain criterion is adopted but reduces this likelihood if the maximax
criterion is adopted. In the case of the expected gain criterion and given a
strictly concave weed loss function, there is a tendency for greater
uncertainty about weed densities to increase the expected gain from non-
treatment. F r o m the total community point of view, adoption of the
maximisation of net gain criterion is the more desirable strategy; the
minimax gain criterion would tend to encourage over-use of herbicides.
Where risk aversion is present, increased uncertainty about weed densities or
the crop loss function tends to increase the likelihood that treatment of
weeds is optimal. In practice, this may be the most c o m m o n consequence.
The importance of the latter finding is that where risk aversion occurs,
specific economic threshold weed densities become less relevant to farmers'
decision making in weed control.

ACKNOWLEDGEMENT

The authors wish to thank an anonymous referee for his comments.

REFERENCES

Anderson, J. R., Dillon, J. L. & Hardaker, J. B. (1977). Agricultural decision analysis.


Iowa St. Univ. Press, Ames, Iowa, 90 6.
Auld, B. A., Menz, K. M. & Tisdell, C. A. (1987). Weed control economics. Applied
Botany and Crop Science Series. Academic Press, London, 55-6.
Chisaka, H. (1977). Weed damage to crops: Yield loss due to weed competition. In:
Integrated control of weeds (Fryer, J. D. & Matsunaka, S. (Eds)), Japan Scientific
Societies Press, Tokyo, 1 t6.
Cousens, R. (1987). Theory and reality of weed control thresholds. Plant Protection
Quarterly, 2, 13-20.
Cousens, R., Doyle, C. J., Wilson, B. J. & Cussans, G. W. (1986). Modelling the
economics of controlling Avenuafatua in winter wheat. Pestic. Sci., 17, 1-12.
Cousens, R., Peters, N. C. B. & Marshall, C. J. (1984). Models of yield loss-weed
density relationships. European Weed Research Society Proceedings 7th
Internat. Symp. on Weed Biology, Ecology and Systematics. Paris, 367-74.
Dew, D. A. (1972). An index of competition for estimatimg crop loss due to weeds.
Can. J. Plant. Sci., 52, 921 7.
Doyle, C. J., Cousens, R. & Moss, S. R. (1986). A model of the economics of
controlling AIopecurus myosuroides Huds. in winter wheat. Crop Protection, 5,
143-50.
Luce, R. D. & Raiffa, H. (1966). Games and decisions: Introduction and critical survey.
Wiley, New York, 275 87.
Marra, M. C. & Carlson, G. A. (1983). An economic threshold model fbr weeds in
soybeans (Glycine max). Weed Sci., 31,604 9.
Economic thresholds and uncertainty in weed control 227

Martin, R. J. (1986). Using herbicides to control wild oats in crops. In: Weed control
in winter crops (Gammie, R. L. & Dellow, J. J. (Eds)). Dept. Agriculture, NSW,
26-7.
Medd, R. W., Kemp, D. R. & Auld, B. A. (1987). Management of weeds in perennial
pastures. In: Temperate pastures: Their production, use, management. (Wheeler,
J., Pearson, C. J. & Robards, G. E. (Eds)). Australian Wool Corporation,
Melbourne. (In press.)
Reichelderfer, K. H. (1980). Economics of integrated pest management: Discussion.
American Journal of Agricultural Economics, 62, 1012-13.
Zimdahl, R. L. (1980. Weed crop competition. A review. International Plant
Protection Centre, Oregon St. Univ., Corvallis, Oregon, 29-31.

Vous aimerez peut-être aussi