SEA Working Paper 99/02

The economics of monitoring crops at the micro level:
Precision weed management

David J. Pannell and Anne L. Bennett

Agricultural and Resource Economics, The University of W.A., Nedlands, 6907

Abstract

Precision farming technologies for weed control offer the prospect of reduced herbicide use and greater profits. However evaluating their overall impact on profit can be very difficult. It is shown in this paper that simple evaluations based only on savings in herbicide can be very misleading. There are a number of important complexities which influence the level of benefits gained and unfortunately the complexities tend to reduce the level of benefits. In a case study of the WASP technology for detection of weeds during pre-crop treatment, it is found that the realistic level of benefits is very low in most scenarios - too low to cover likely costs. This is not primarily because of limitations in the technology. Even if these limitations could be overcome, the benefits would still be low because of the nature of the problem. This finding is reinforced by an assessment of an hypothetical in-crop precision sprayer. It is found that the benefits are likely to be even lower for the in-crop situation because of the greater importance of weed detection errors in this case.

Introduction

There has been a rapid growth of interest in precision weed management. This interest springs both from concerns about possible impacts of herbicides on farmers, consumers and the environment (e.g. Combellack 1989; Hoar et al. 1986; Kovach et al. 1992; Nielsen and Lee 1987; Pimentel 1983) and the potential for financial savings to farmers.

The benefits of precision weed management arise from use of information about the spatial variability of weeds (e.g. Wiles, Wilkerson and Gold 1992a). The nature and biological implications of this spatial variability have been widely studied (e.g. Auld and Tisdell 1988; Brain and Cousens 1990; Hughes 1989; Wiles et al. 1992b, 1992c), and there has been some work on the related issue of spatial distribution of applied herbicides (e.g. Dorr and Pannell 1992).

Although the aim of precision weed management technologies is to reduce herbicide usage, a full evaluation of its benefits should not be based solely on the reduction achieved. As outlined by Bennett and Pannell (1998), the economic benefits are affected by a number of complexities, including the following.

(a). The benefits of the technology depend on the spatial distribution of weeds. The more patchy the distribution, the greater the potential for precision weed management.

(b). The benefits of reducing herbicide usage are partly offset by the costs of increased weed competition in patches where the weed density is greater than zero, but too low to activate control by the precision weed controller;

(c). For various reasons, including inherent limitations of the technologies, the information collected by the precision weed controller will always be imperfect. Some areas of the paddock will be sprayed despite a low weed density, while some patches with many weeds will be missed.

(d). The impact of spraying or not spraying a weed is felt over subsequent years due to the level of carry over of weed seeds.

(e). From a social perspective, the net benefits may depend on the external impacts of herbicide usage (e.g., to consumers’ health, the environment, other farmers).

Although there have been various evaluations of precision weed management technologies in the past, most considered only a sub-set of the relevant issues (e.g. Ahrens 1994; Audsley 1993; Thompson, Stafford and Miller 1991) or have dealt only with hypothetical low-precision technologies (e.g. Oriade et al. 1996). In this paper we discuss the factors that influence the level of benefits from precision weed management and outline a framework within which the benefits can validly be estimated. We apply this framework to one existing method of precision weed management known as the Weed Activated Spray Process (WASP) (Felton et al. 1987, 1991) as well as to an hypothetical WASP-like technology which could be used for in-crop application of selective herbicides. The paper is an extension of a study by Bennett and Pannell (1998) which focussed only on WASP.

The technologies evaluated

WASP employs an electronic eye to detect the presence or absence of green matter beneath the spray nozzle of a boom sprayer, allowing reductions in herbicide usage by automatically preventing the application of herbicides to areas of a paddock which have an insufficient density of weeds to warrant the expense (Felton et al. 1987). Because it cannot distinguish between green matter of different plant species and treat them differently, WASP is only of use at times when there is no crop.

The WASP technology is fitted to a conventional boom sprayer. Each nozzle of the boom sprayer is fitted with its own sensor which has a field of view the same width as that covered by the spray from the nozzle. The user is able to specify the minimum proportion of this field of view which must be taken up by green matter before the spray nozzle is activated. With current commercially available technology this threshold field of view cannot be set lower than three percent.

There is no mechanism in WASP to control the dosage that is sprayed onto each patch of weeds, so a patch with a weed density just above the threshold is sprayed with the same dosage as a heavily infested patch. This threshold refers to the weed density above which spraying by the WASP is triggered. It does not necessarily coincide with the economic threshold, defined as the weed density above which application of a fixed recommended herbicide dose is economically justified (Pannell 1990).

Like any technology, the WASP system is liable to occasional mistakes, spraying bare ground or failing to spray dense patches. We lack evidence on the frequency of these mistakes and so have assessed their significance using sensitivity analysis.

There have been a number of field evaluations of WASP in Australia, focusing on savings in herbicides. For example, Felton et al. (1991) measured a 90 percent reduction in herbicide use over 2,681 hectares on 33 farms in New South Wales and Queensland. WASP reduced herbicide use on summer weeds by 67 to 87 percent in a trial in Western Australia (Martin 1992). In the USA, Ahrens (1994) measured reductions in spray volume of 47 to 88 percent using WASP in two North Dakota fallow sites.

The purchase cost of WASP technology is currently substantial, estimated as A$38,000 for a complete modification to an average sized boom sprayer in Western Australia. Using optimistic assumptions about interest rates, the useful life of the technology and its salvage value at the end of its life, the sprayer's equivalent annual cost (annuity) would be at least $4,000 per year. In addition there would be costs in maintaining the equipment and replacing sensors.

As noted above, WASP is not suitable for use in-crop. Our first set of modelling results is for use of the technology for pre-emergence application of a broad-spectrum herbicide. Another possible use is for treatment of summer weeds in Mediterranean environments, but we do not evaluate this option here. However we do present results for an hypothetical technology which can be used in the crop. This might conceivable be a more sophisticated WASP-like technology which can detect and distinguish between different plant species, or it might be based on stored maps of weed patches. We assume that the in-crop spraying is conducted over the whole paddock. Another possibility not evaluated here is the constraining of spraying to the inter-rows of row crops. Further details of numerical assumptions about both of the analysed technologies are given below.

The Model

Overall framework

This framework is based on that presented by Bennett and Pannell (1998). Consider a field in which weed density prior to spraying, y, is spatially distributed within the field according to the probability density function f(y). Suppose that it is possible to subdivide the field into n smaller regions or patches in which the assumption of a spatially uniform weed distribution is reasonable. In patch i, representing a proportion pr(i ) of the total area, initial weed density is y_i. If herbicide is applied, a proportion k is killed, where in general k < 1. Weed density after spraying, W, is given by

(1) W = y (1 - hk ),

where h is a binary variable taking values 1 if herbicide is applied or zero if not. Crop yield is reduced by competition with those weeds which survive or avoid herbicide application. Let D(W) represent proportional yield loss. Empirical evidence (Cousens 1985) indicates that a suitable functional form is

(2) D(W) = 1 - a/[1 + a/(bW)]

where parameter a can be interpreted as the asymptotic yield loss as W ® ¥. Crops typically give some positive yield even at very high weed densities, so a is normally less than one. The parameter b is the proportional yield loss per weed as W ® 0.

Crop yield in the absence of weeds would be Y0, assumed for convenience to be spatially uniform. In patch i, final crop yield after allowing for weed competition, Y, is given by

(3) Yi = Y0 [1 - D(Wi)]

Average crop production per unit area for the field is

(4) Y = Si=1..n Y0 [1 - D(yi {1 - hik})] pr(i)

Profit for patch i is

(5) pi = Py Y0 [1 - D(yi {1 - hik})] - hiPh - Pf,

where Py is output price, Ph is herbicide cost per unit area and Pf is other production costs, assumed to be fixed. Average profit per unit area for the field is

(6) p = Si=1..n {Py Y0 [1 - D(yi {1 - hik})] - hiPh - Pf}pr(i).

Using a similar model, Pannell (1990) showed that as initial weed density falls, the benefits of applying a fixed herbicide dose also fall. There exists an economic threshold weed density (t) below which the costs of purchasing and applying the herbicide outweigh the benefits. If the threshold is known, information about how the weed density varies over space can be used to increase the expected net profit from the crop. Information of this type about spatial variability is the source of the benefits from precision farming technology. If pw represents a stream of annual profits when WASP technology is used during the evaluation period, pt represents a stream of profits using traditional spray application technology, CW represents a stream of purchase and operating costs of the WASP technology and A signifies the annuity (the equivalent annual value), the problem is whether

(7) A(pw - pt) > A(-CW).

Because WASP is purchased as an add-on to a traditional boom sprayer, the cost of the boom sprayer is not included in equation 7; it must be borne whether or not WASP is purchased.

The value of information about weed density

The field can be divided into patches where the weed density exceeds the economic threshold for spraying (y > t) and patches where it does not. When WASP technology is used, the field can also be divided into patches where herbicide is applied (h = 1) and patches where it is not (h = 0). In practice, these two divisions of the field will not coincide exactly. Reasons for this may include: failure of the operator to set the sprayer's threshold at the true economic threshold; a divergence between the field of view of a sensor and the piece of ground sprayed by the corresponding nozzle; wind moving herbicide droplets before reaching their target; and intrinsic limitations of the technology in sensing weeds. This means that patches can be categorised into groups where, relative to traditional spray application technology, the use of WASP technology (a) makes no difference to profit, (b) increases profit, and (c) decreases profit. Table 1 shows how these categories apply to the different types of patch. It is clear from Table 1 that the level of profit improvement provided by the WASP depends on the accuracy of the technology in detecting patches with sufficiently low weed density to not spray. The model used here explicitly represents the four categories of land shown in Table 1.

Table 1. Impact of WASP technology on profit relative to traditional spray application technology (+ = profit improved; - = profit reduced; 0 = profit unchanged)

Legend: h = 1 means spray; h = 0 means do not spray; y is the initial weed density; t is the economic threshold weed density.

Assumptions for the Numerical Model

Parameters for the evaluation are specified to represent a wheat producer in Western Australia with 1,000 hectares of crop per year. The analysis is based on a broad-spectrum herbicide such as glyphosate or paraquat/diquat applied with WASP prior to emergence of the crop. Additional weed control is assumed to be conducted in the crop with a traditional boom sprayer. In the first set of results the precision sprayer is only used prior to seeding, as WASP would be.

Different weed densities are examined with the standard assumption corresponding to an average of 700 weeds m^-2 as observed in a set of field measurements after weed germination in May 1994 on a farm at Cunderdin, in Western Australia’s central wheatbelt. The alternative assumptions explored in the modelling were for this density to be doubled or quartered. The distribution of weed densities within a paddock is important to the analysis. We used as a standard the distribution measured at Cunderdin, as shown in Figure 1. For higher and lower density scenarios, this distribution was scaled proportionately to the change in mean density.

Figure 1. Cumulative probability density function for weed density within the paddock.

Crop yield losses from competition of the weeds with the crop were based on the equation estimated by Pannell (1995) for ryegrass; parameters a and b from equation 2 are set at 0.75 and 0.002 respectively. We do not specify exactly what mixture of weed species is present. Implicitly, the competitiveness of a given population of ryegrass is assumed to be a good approximation of the competitiveness of a mixed population of weeds of the same density. This assumption is justified on the bases that (a) ryegrass is the most widespread and economically important weed in the region and will be present in most weed populations, (b) there are few other weeds for which the information is available, and (c) even if it is a little inaccurate, the sensitivity of model results to weed competitiveness is amongst the lowest of any of the model’s parameters.

The 100 sample patches underlying Figure 1 are assumed to be representative of the whole area of crop. Yield loss is calculated separately for each of the patches and scaled up to an average value per hectare. Allowance is made for the probability that the patch will be sprayed, based on its weed density, the threshold field of view and the probability of a WASP error.

Not all weeds germinate and emerge from the soil in time to be sprayed by the WASP, which must be used prior to crop emergence because of WASP’s inability to distinguish crop from weeds. A wide range of emergence assumptions is tested, reflecting the range which occurs in practice.

Allowance is made for the cost in future years of allowing weeds to survive and set seed in the current year. This is achieved by assigning a shadow cost of $0.10 per plant based on results from a modified version of the dynamic model of Gorddard et al. (1995).

Procedure for sensitivity analysis

Many of the parameters of the model are subject to uncertainty or to change over time and space. For this reason, results are subjected to a sensitivity analysis. The general approach is consistent with Pannell's (1997) Strategy A for sensitivity analysis. It proceeds by (a) identifying those parameters most subject to change or uncertainty, (b) selecting minimum, maximum and standard or most-likely values for each of these parameters, (c) assessing the sensitivity of results to parameter changes within the ranges selected in (b), and (d) for a subset of the most sensitive parameters, conducting a complete factorial experiment. Steps (a), (b) and (c) are undertaken assuming that the threshold field of view for WASP is set at its current minimum value of three percent. Step (d) is repeated four times: for pre-crop and in-crop use of the precision sprayer and for two different settings of the threshold (three percent, and the optimal value for the scenario being considered). Results for several different output variables are presented: the impact of the WASP technology on per-hectare profits from cropping, the optimal threshold level of weeds in the field of view, and the proportion of the paddock area which is sprayed by the WASP technology.

The selection of parameter values for the sensitivity analysis was based on previous studies (Pannell 1995; Gorddard et al. 1995), discussions with the developers of WASP and with weed scientists at the government agency Agriculture Western Australia. The parameters used are shown in Table 2

Results and Discussion

Table 2 shows the minimum, standard and maximum values for each parameter used in the sensitivity analysis. The parameters are ranked according to the absolute value of a "sensitivity index", defined as follows.

(8) I = (Bmax - Bmin)/Bst

where Bmax is the benefit of WASP when the parameter in question is set at its maximum value, Bmin is the benefit given the minimum parameter value and Bst is the benefit for the standard parameter value. This index is almost the same as one proposed by Hoffman and Gardner (1983) (who used Bmax in place of Bst).

Table 2. Values and sensitivity index results for parameters of the model

* Sensitivity to parameter changes of the impact of WASP sprayer on profits from cropping, based on equation (8)

The purpose of this ranking is to select the most important parameters for more detailed analysis. Five of the six top-ranked parameters were subjected to a complete factorial experiment using all three parameter levels in Table 2, giving 35 = 243 solutions. Herbicide cost was not varied in this experiment as farmers normally have good knowledge of the herbicide cost. For those parameters which were not varied, the standard values in Table 2 were used.

Table 3 shows results in which these 243 scenarios are assumed to constitute a joint probability distribution, which is used to calculate the expected value of benefits from WASP used pre-crop. For each of the random variables, the standard value is assigned a probability of 0.5 and the extreme values have probabilities of 0.25. All random variables are assumed to be statistically independent. Results are presented for the case where the threshold field of view is fixed at 0.03 (the maximum sensitivity of WASP) and the hypothetical case where the threshold is set at the economically optimal level, regardless of how low that is.

Table 3. Expected value of benefits of WASP used as pre-crop treatment over 243 scenarios

Note that in Table 3 and all other results presented below, purchase and maintenance costs have not been deducted. The results represent gross benefits; net benefits would be lower by the annualised cost of purchase and maintenance.

The expected value of benefits of WASP is calculated in a variety of ways. The first line of results in Table 3 shows the expected value of herbicide savings relative to a conventional boom sprayer. These are calculated at A$11 to A$12 per hectare for this case where herbicide costs A$15 per hectare. Figure 2 illustrates the probability density function for the optimal threshold case. The level of herbicide saving varies widely within the paddock, but is substantial overall. This is the most common method by which WASP has been evaluated in previous studies. However, the other results in Table 3 reveal that this simple partial criterion for evaluating WASP is highly misleading. As realistic complexities are added to the calculation, the estimated benefits fall steadily.

Figure 2. Cumulative probability density function for the value of herbicide saving from WASP used pre-crop, assuming threshold field of view is adjusted to optimal value for each scenario.

The first realistic complexity is the fact that if the WASP is used, there will be some regions of the crop which face increased competition from weeds as a result of not being sprayed. It is true that these may be patches for which spraying is not economically justified, but nevertheless the increased competition must be set against the herbicide savings to obtain a valid estimate of the overall benefits. The second line of results in Table 3 shows that this makes a substantial difference to the level of benefits.

Line 3 introduces the possibility that the WASP may make errors in its detection of weeds, either by spraying low density patches or failing to spray high density patches. With the assumed probability of these errors set at five percent, the impact on the estimated benefits is relatively minor.

Results in lines 4, 5 and 6 all include the additional consideration that the correct comparison may not necessarily be with the traditional boom sprayer, but instead with no herbicide application. In many of the scenarios in which the WASP performs much better than the traditional sprayer, it is because the circumstances do not favour herbicide application in general. In these cases, zero herbicide may be little worse, or even better, than the WASP. Table 3 shows that this is the single most important factor to get right when attempting to evaluate WASP. It brings the estimated benefits down to very low levels, or even to negative values in the realistic case where the threshold is fixed at 0.03. Figure 3 shows the probability density function for the result in line 6 for the fixed threshold case. In this graph it is striking that the probability that WASP actually reduces returns is approximately 75 percent, even before purchase and maintenance costs are considered.

Figure 3. Cumulative probability density function for the gross benefits WASP used pre-crop, assuming threshold field of view is fixed at 0.03.

The differences between the results in lines 4 to 6 are due to different assumptions about when WASP would be used, as shown in the table. However these differences are much less important determinants of the value than the factors of increased weed competition and the possible need to compare with zero herbicide.

Table 4 shows a selection of individual results for particular scenarios. To save space the results are only given for high and low values of the parameters. The two columns of estimated benefits correspond to lines 6 and 5 in Table 3. There seems to be no clear trend in the results apart from the tendency for benefits to be lower if the level of subsequent in-crop weed control (by conventional means) is higher. The results here reflect the low benefits presented in Table 3. The second column of benefits includes a number of negative results, indicating that in these scenarios, weed control with WASP is less profitable than with the better out of zero herbicide application or the conventional sprayer.

Table 4. Results for WASP for individual scenarios (assuming optimal thresholds and including weed detection errors)

^AThis is from a separate subsequent application of selective herbicide in the crop, not from the WASP, which is used pre-crop.
^BProportion of weeds which have emerged at time of WASP use
^CMean weed density in paddock in the absence of any herbicide application
^DSale price net of transport and other selling costs
^EBenefit without deducting purchase and use costs, based on assumption that WASP only used if it is better than both a conventional sprayer and zero herbicide use.
^FBenefit without deducting purchase and use costs, based on assumption that if herbicide is applied, the WASP is used
^GProportion of field of view of WASP which should be set as trigger for herbicide application
^HProportion of paddock sprayed if threshold field of view is set at optimum

The last column of Table 4 shows the proportion of the crop which would be sprayed in each of the scenarios. This helps to explain why the WASP does not perform better. The highest benefits are for cases where approximately 50 percent of the paddock is sprayed. However, in most of the scenarios, it is optimal to spray almost all of the crop, so that WASP is little different to a conventional sprayer, or very little of the crop, so that WASP is little different to not spraying at all.

From these results, the prospects for current WASP technology for pre-crop weed control appear bleak. The benefit in the single most favourable scenario in Table 4 ($2.95) would need to be generated every year over more than 1,300 hectares of crop to cover the annuity cost of purchasing the WASP technology even under the most favourable assumptions regarding discount rate and machinery life and ignoring maintenance and repair costs. Even then, the benefits in Table 4 are over-estimates since they are based on the user having complete flexibility in selecting the threshold weed cover which triggers herbicide application. In reality, the sensitivity of the current technology is substantially lower than would be desirable in most pre-crop scenarios examined (apart from those where no herbicide application is the preferred option). Further, if a fully dynamic model had been used, the effect on results would have been to reduce the value of WASP even further. This would occur because of the tendency to move toward a uniform weed density across the field with repeated use of WASP. The nearer the density moved towards uniformity, the lower would be the value of WASP.

Overall, these results underline the difficulties facing the developers of precision sprayers. Even if the technology can be developed to the extent that the threshold for weed spraying can be set at any level and all errors of weed detection are removed, the use of precision technology for pre-crop weed control is still unlikely to generate benefits in excess of costs for most farmers in extensive farming systems similar to the type modelled here.

Finally, consider the results for an hypothetical WASP-like technology for in-crop use (Table 5). For these model runs, the cost of herbicide is increased to $25 per ha, mean pre-crop weed control (by conventional means) is set at 35 percent and the threshold weed level for detection for the "fixed" case was set at the same density (but obviously a higher biomass) as for pre-crop use. The probability of detection errors was left unchanged at five percent, although this is probably overly optimistic for an in-crop situation. Nevertheless, the results in Table 5 are even more discouraging than for pre-crop treatment. One reason is that the consequence of detection errors in the in-crop situation are much worse than they are pre-crop. In pre-crop situations, the WASP would not be the main method of weed control. Errors can be compensated for to some extent in subsequent in-crop control. However when a precision sprayer is used to deliver the primary weed treatment, any detection errors can become very costly. For the "Fixed threshold" case, the threshold is clearly too high for in-crop use in many scenarios and the use of WASP can be especially unprofitable.

Table 5. Expected value of benefits of hypothetical precision sprayer used as in-crop treatment over 243 scenarios

Conclusion

If the results from this case study apply more generally, there appears little prospect for precision technologies to make a major economic contribution in weed control in Australian extensive agriculture. This negative outlook stems primarily not from limitations of the technology, but from the nature of the farm management problem which it is intended to address. The model developed here indicates that in most plausible scenarios, the proportion of a crop which should optimally receive a herbicide spray is either so high that WASP is little different to a conventional sprayer, or so low that WASP is little different to not spraying at all. Even in the minority of scenarios where WASP improves the profitability of cropping, the purchase cost of the technology would have to be reduced substantially from current levels for the net benefits to be positive. The prospects for an in-crop precision sprayer appear no better, particularly because of the greater importance of weed detection errors in the in-crop situation. The situation we have not investigated is treatment of summer weeds, which have a more pronounced patchiness and so be more likely to benefits from an automatic detection system. Also, because of the timing of treatment and the lower densities, herbicide savings are more likely to reasonably approximate the economic benefits for this case.

References

Ahrens, W.H. 1994,. ‘Relative costs of a weed-activated versus conventional sprayer in northern Great Plains fallow,’ Weed Technology, vol. 8, pp. 50-57.

Audsley, W.P. 1993, ‘Operational research of patch spraying,’ Crop Protection, vol. 12, pp. 111-119.

Auld, B.A. and Tisdell, C.A. 1988, ‘Influence of spatial distribution of weeds on crop yield loss,’ Plant Protection Quarterly, vol. 3, p. 81.

Bennett, A.L. and Pannell, D.J. 1998. ‘Economic evaluation of a weed-activated sprayer for herbicide application to patchy weed populations,’ Australian Journal of Agricultural and Resource Economics, 42(4): 389-408.

Brain, P. and Cousens, R. 1990, ‘The effect of weed distribution on prediction of yield loss,’ Journal of Applied Ecology, vol. 27, pp. 735-742.

Combellack, J.H. 1989, ‘The importance of weeds and the advantages and disadvantages of herbicide use,’ Plant Protection Quarterly, vol. 4, pp. 14-31.

Cousens, R. 1985, ‘An empirical model relating crop yield to weed and crop density and a statistical comparison with other models,’ Journal of Agricultural Science, Cambridge, vol. 105, pp. 513-521.

Dorr, G.J. and Pannell, D.J. 1992, ‘Economics of improved spatial distribution of herbicide for weed control in crops,’ Crop Protection, vol. 11, pp. 385-391.

Felton, W.L., McCloy, K.R., Doss, A.F. and Burger, A.E. 1987, ‘Evaluation of a weed detector,’ in Lemerle, D. and Leys, A.R. (eds.), Proceedings of the Eighth Australian Weeds Conference, Sydney, 21-25 September 1987, pp. 80-84.

Felton, W.L., Doss, A.F., Nash, P.G. and McCloy, K.R. 1991, ‘A microprocessor controlled technology to selectively spot spray weeds,’ in Proceedings of the 1991 Symposium of Automated Agriculture for the 21st Century, American Society of Agricultural Engineers, St Joseph, Michigan, pp. 427-432.

Gorddard, R.J., Pannell, D.J. and Hertzler, G.L. 1995, ‘An optimal control model of integrated weed management under herbicide resistance,’ Australian Journal of Agricultural Economics, vol. 39, pp. 71-87.

Hoar, S.K., Blair, A., Holmes, F.F., Boysen, C.D., Robel, R.J., Hoover, R. and Fraumesi, J.F. Jr. 1986, ‘Agricultural herbicide use and risk of lymphonoma and soft tissue sarcoma,’ Journal of the American Medical Association, vol. 256, pp. 1141-1147.

Hoffman, F.O. and Gardner, R.H. 1983, ‘Evaluation of uncertainties in environmental radiological assessment models,’ in: Till, J.E. and Meyer, H.R. (eds.), Radiological Assessments: A Textbook on Environmental Dose Assessment, US Nuclear Regulatory Commission, Washington D.C., Report no. NUREG/CR-3332, pp. 11.1-11.55.

Hughes, G. 1989, ‘Spatial heterogeneity in yield-weed relationships for crop loss assessment,’ Crop Research, vol. 29, pp. 87-94.

Kovach, J., Petzoldt, C., Degni, J. and Tette, J. 1992, ‘A method to measure the environmental impacts of pesticides,’ New York Food and Life Sciences Bulletin, New York State Agricultural Experimental Station, Geneva, Bulletin N139.

Martin, R.J. 1992, ‘Evaluation of remote sensing techniques for the detection of skeleton weed,’ Unpublished report to the Skeleton Weed Advisory Committee, 20 February 1992, Agriculture Protection Board of Western Australia, Perth.

Nielsen, E.G. and Lee, L.K. 1987, ‘The magnitude and costs of groundwater contamination from agricultural chemicals: a national perspective,’ Natural Economics Division, USDA-ERS Staff Report, AGES 870318.

Oriade, C.A., King, R.P., Forcella, F. And Gunsolus, J.L. 1996, ‘A bioeconomic analysis of site-specific management for weed control,’ Review of Agricultural Economics, vol. 18, pp. 523-535.

Pannell, D.J. 1990, ‘An economic response model of herbicide application for weed control,’ Australian Journal of Agricultural Economics, vol. 34, pp. 223-241.

Pannell, D.J. 1995, ‘Optimal herbicide strategies for weed control under risk aversion,’ Review of Agricultural Economics, vol. 17, pp. 337-350.

Pannell, D.J. 1997, ‘Sensitivity analysis of normative economic models: Theoretical framework and practical strategies,’ Agricultural Economics, vol. 16, pp. 139-152.

Pimentel, D. 1983, ‘Effects of pesticides on the environment,’ in 10th International Congress of Plant Protection 1983, British Crop Protection Council, Croyden, England, pp. 685-691.

Thompson, J.F., Stafford, J.V. and Miller, P.C.H. 1991, ‘Potential for automatic weed detection and selective herbicide application,’ Crop Protection, vol. 10, pp. 254-259.

Wiles, L.J., Wilkerson, G.G. and Gold, H.J. 1992a, ‘Value of information about weed distribution for improving postemergence control decisions,’ Crop Protection, vol. 11, pp. 547-554.

Wiles, L.J., Oliver, G.W., York, A.C., Gold, H.J. and Wilkerson, G.G. 1992b, ‘Spatial distribution of broadleaf weeds in North Carolina soybean (Glycine max) fields,’ Weed Science, vol. 40, pp. 554-557.

Wiles, L.J., Wilkerson, G.G., Gold, H.J. and Coble, H.D. 1992c, ‘Modeling weed distribution for improved post-emergence control decisions,’ Weed Science, vol. 40, pp. 546-553.

Citation: Pannell, D.J. and Bennett, A.L. (1999). Economic feasibility of precision weed management: Is it worth the investment? In: R.W. Medd and J.E. Pratley (eds), Precision Weed Management in Crops and Pastures, CRC for Weed Management Systems, Adelaide, pp. 138-148. (SEA Working paper 99/02, Agricultural and Resource Economics, University of Western Australia). http://www.general.uwa.edu.au/u/dpannell/dpap9903f.htm

SEA News issue #3

The SEA News index is at http://welcome.to/seanews

Copyright © David J. Pannell and Anne L. Bennett 1999
Last revised: May 21, 2003.