Case Study: Short-term forecasting of peak population density of the grain aphid Sitobion avenae on wheat (Entwistle and Dixon, 1986)
2.5.4 Geographic Information Systems The quantification of the spatial dynamics
of pest populations especially in relation to associated environmental changes, e.g. wind direction (Tucker et al., 1982) or rain- fall (Tucker and Pedgeley, 1983; Tucker, 1984a, b) is not new; regional and national density distribution maps of pest species constructed from trap catch records using computer graphics have been possible since the 1970s (Taylor, 1974; Campion et
al., 1977). However, the more recent
increase in computer power and sophisti- cation of software has made the analysis of geographic variables in relation to the spa- tial aspects of the dynamics of pest popula- tions more widely available. This type of analysis based on Geographic Information Systems (GIS) utilizes data referenced by spatial or geographic coordinates (Star and
Estes, 1990). Geographical information such as altitude, temperature, soil type and distribution of crop varieties can be stored as a spatial map layer. When a number of map layers are overlaid, valuable insights such as pest zones or areas prone to pest outbreaks may be identified. Information generated in this way can be fed into mod- els that use the maps as variables and analyse these to determine whether corre- lations exist and to construct forecast maps. GIS has been successfully used in pest management to forecast grasshopper incidence in Canada (Johnson and Worsbec, 1988), dispersal of rice insects in South Korea (Song et al., 1994) and to identify the risk posed by exotic insect pests in the UK (Baker, 1994; Head et al., 1998). It is likely that GIS analyses will fig- ure more commonly in the understanding Fig. 2.19. The peak mean number of Sitobion avenae per tiller (y) in relation to the forecast from
the multiple regression equation (see text) at the end of anthesis (GS 69). Symbols as Fig. 2.17, n results from 1984 for model validation. Numbers shown on log scale (after Entwistle and Dixon, 1986).
of pest populations and the ability to fore- cast susceptible outbreak regions in the future, as this powerful technology gains more widespread use.
2.6 Discussion
Sampling using relative measures of insect abundance tends to be popular among entomologists mainly because the various types of trap used provide quick and simple methods for acquiring a large amount of population data. However, the value of such data, for the purposes of monitoring and forecasting, may be lim- ited if the trap catch cannot be related to other appropriate population estimates. The type of monitoring system employed will be dependent on the user, the pest, the resources available and whether or not the monitoring is being carried out for research purposes or as a practical aid in insecticide application. More attention needs to be given at the start of a sam- pling programme to the eventual use of the data and whether or not a monitoring/forecasting programme could result from the work (Wall, 1990). If this is the case, then it is important to iden- tify the potential users, their needs and abilities so that an appropriate sampling and monitoring programme can be devised. The tendency to embark on long term monitoring programmes with the hope that ‘something will come out of it’ needs to stop and be replaced by a more rational approach where specific goals, time limits and available resources are identified. The type of sampling tech- niques used in a monitoring programme dictates the type of forecasting system that can be developed, hence both the monitoring programme and the perceived forecasting system must be appropriate for the situation and conditions to which it will be applied.
In the absence of any monitoring/fore- casting system a farmer would either apply an insecticide regardless of the level of pest attack (prophylaxis) or not apply an
insecticide at all. The no control option would mean that in non-outbreak years (or seasons) the farmer would maximize his profits with respect to that particular pest/crop system; however, in an outbreak year total yield and monetary loss may occur. The prophylaxis option in non-out- break years would reduce the farmer’s prof- its by the cost of the unnecessary insecticide application, whereas in out- break years the farmer would control the pest but the prophylaxis strategy may mean that the farmer uses too many (or too few) insecticide applications for effective pest control. An ineffective control due to pro- phylaxis could reduce the farmer’s profits. Hence, neither strategy will always provide the optimal solution in terms of maximiz- ing farmers’ profits. The common factor in the two strategies that will influence over- all farmer profit is the probability of an outbreak and the severity of attack. A moni- toring and forecasting system must improve on both the no control and the prophylaxis option if it is to be adopted by farmers, i.e. a monitoring and forecasting system must cor- rectly predict outbreak and non-outbreak years so that insecticide is applied only in outbreak years, and only when it is cost- effective to do so. Hence, the accuracy and consistency of the forecast will directly affect the yield and profits of the farmer. The value of a forecast scheme and the impor- tance of its accuracy can be assessed in eco- nomic terms (Watt, 1983).
The accuracy of a forecast will depend on the variability encountered in the fore- cast model. Forecast models are often based on regression models which are prone to a number of potentially serious defects. Firstly, the regression model should explain a large proportion of the variance if it is to be at all reliable. Secondly, regression models are really only reliable within the range of data from which they were originally derived. They may work for many years but then a combination of circumstances outside the original conditions arises, causing a model to fail (Conway, 1984b). These con- siderations must be taken into account
when developing forecast schemes, so that the reliability that can be placed on forecasts can be evaluated. A farmer
would soon stop using a monitoring and forecasting system if it were shown to be unreliable.
3.1 Introduction
The amount of useful product that is obtained from crop plants or livestock is commonly referred to as ‘yield’. Estimates of yield may be: (i) quantitative, e.g. weight of grain, fruit or tubers per unit of land; or (ii) qualitative, e.g. the percentage of a product meeting certain cosmetic stan- dards which will vary for a given crop or livestock system according to weather, the levels and types of input and pest inci- dence. If all factors are optimal then the highest attainable yield is obtainable. However, since conditions are rarely opti- mal, actual yields are normally well below those that are theoretically obtainable. The best a farmer can usually hope to achieve is a yield that provides the highest possible return on his investment of inputs.
Yield loss assessments attempt to account for the difference between actual and attain- able yield as a result of the inputs, weather and pests in a given cropping system. Hence, the losses due to pests constitute only one of a number of complex production con- straints. Even so yield losses due to insects alone provide a multitude of problems for the scientist to unravel and evaluate.