|
Analysis of influence factors on pest emergence size is the prerequisite for pest forecast and integrated control. In China, cotton bollworm (.Helicoverpa armigerd) has harm to cotton and non-cotton host crops such as corn, peanuts, etc. Previous systematic study had investigated population dynamics of cotton bollworm on8kinds of hosts, in100stations,6provinces of China from1992to2006. Based on linear model it thinks that Bt cotton planting area expansion is the key factors for decreasing of cotton bollworm larvae density on non-cotton hosts, while temperature and rainfall have less affect. Linear model is difficult to depict the complex nonlinear relationship between the response variable and the independent variables, besides, it has disadvantage of multicollinearity. On the basis of structure risk minimum, the support vector machine has advantages of nonlinear, better solved over-fitting, generalization ability, etc. Given this, after rearranging the previous data, a nonlinear support vector regression (SVR) model was built between cotton bollworms larvae density and possible influence factors, then, based on the early developed SVR nonlinear explanatory system, various influence factors were being reanalyzed and reexplained, the results are as follows:1. Based on analysis the source of second generation cotton bollworm adults and the distribution of their eggs, single province in each year deeming as a sample, we take third generation cotton bollworm larvae density on four kinds of non-cotton hosts such as corn in the six provinces of China from1997to2006as dependent variables, and chose23factors including second generation eggs density on cotton host, Bt cotton planting area, corn planting area etc. as independent variables, then the multivariate stepwise linear regression model and nonlinear SVR model were established, respectively. The results showed that the third generation cotton bollworm larvae density on four kinds of non-cotton hosts varies greatly between provinces and years, and is influenced by many factors such as host plant cultivation area and meteorological factors, especially the meteorological factors when take global warming tendency into consideration. The expansion of Bt cotton planting area indirectly reduce the third generation larvae density of non-cotton hosts by reducing the larvae density on cotton last year which further decrease the second generation egg density on cotton and the second generation larvae density on four kinds of non-cotton hosts.2. Based on analysis the source of first generation cotton bollworm adults and the distribution of their eggs, single province in each year deeming as a sample, we take second generation cotton bollworm larvae density on four kinds of non-cotton hosts such as corn in the six provinces of China from1997to2006as dependent variables, and chose29factors including10kinds of hosts planting area, third generation eggs density on cotton host of last year, etc. as independent variables, then the stepwise linear regression model and nonlinear SVR model were established, respectively. After the worst descriptor elimination multi-roundly method17variables were reserved, namely three generations eggs and three generations larvae density of last year; Bt cotton, peanuts, sorghum, melons plant area of last year; non-Bt cotton, Bt cotton, corn, sesame, beans plant area of this year; rainfall amounts in early June, mid-June and late June; average temperature in mid-June, late June and early July. The root of mean square error RMSE and determination coefficient R2of nonlinear model are0.2504and0.8476, respectively, then nonlinear interpretation system showed the significance of nonlinear model and17reserved independent variables as well as their single factor effect.3. In large spatial and long time scale field investigation, its observation or test conditions is often uncontrolled, so the results of single variable regression analysis have its disadvantages. In short, nonlinear and multivariate analysis of large and complicated data should arouse more attention.
|