Abstract: Simulating the relationship between temperature and developmental rate is an important content in entomology research. The traditional non-linear models, including Logan model, Lactin model and Wang model, however, have the disadvantage of utilizing information incompletely, over-fitting, etc. In the current paper, an improved support vector regression (SVR) model has been developed to analyze the relationship between temperature and pupal development of the cotton bollworm (Helicoverpa armigera). The results showed that the SVR had a higher performance on model-fitting and predict ability than other non-linear models based on the observed data (92 samples), with determination coefficients (R²) of 0.998 and 0.996, respectively. Estimation of the three fundemental points of temperature of the pupal stage with the improved SVR was more credible. On the basis of 20 samples, the Lactin model had the highest performance with R² of 0.958 among the mentioned traditional non-linear models, but it was still obviously lower than that of the improved SVR with R² of 0.981. When the number of samples was reduced to 12, the R² of SVR slightly declined to 0.964, while the traditional non-linear models were not applicable to the independent prediction any more. The results suggest that the improved SVR is superior in dealing with small sample set than traditional non-linear models, and the improved SVR may be useful in forecasting outbreaks of pests and artificial breeding of insects.

Key words: Helicoverpa armigera, support vector regression, pupal stage, temperature, developmental duration, non-linear model