Abstract
Linear multiple regression models to predict the population of summer and autumn generations of the arrowhead scale, one of the important citrus tree insect pests in Japan, were constructed. The population density at each specified developmental stage of the insect, average temperature, precipitation in each month of observation, etc. were used as variables in these analyses. These data were accumulated from 1960 to 1972 at experiment stations in Kanagawa, Shizuoka, Hiroshima and Kumamoto. The following two types of model were built up by statistically and/or biologically significant variables. ‘Type 1’: 4 separate models constructed for respective locations by using data obtained at each of the experiment stations noted above. ‘Type 2’: A common model constructed by using all data from the 4 experiment stations. In these five cases, the multiple correlation coefficients of the regression attained at least 0.8. These five models were used to predict the arrowhead scale population density in the two years following the observations used in the analysis, i.e., 1973 and 1974. The results show that the Type 2 model had high predictability but the Type 1 models did not have high predictability. This is because the Type 2 model was constructed by using data covering various different environments. Although the Type 1 models fitted thier own specific situations well, they were not sufficient to forecast future conditions because these models did not incorporate diverse experiences which might occur in the future. Hence, it is concluded that a multiple regression type model can have very high predictability, if it is built by using data obtained in various locations and many seasons, as was the Type 2 model in this study.
