Abstract: The occurrence degree of rice planthoppers is nonnormal and nonlinear related to climatic parameters, while classical statistic methods can hardly reach stable predictions for the occurrence degree. Artificial neural network (ANN) is an efficient way to estimate the occurrence degree, but fails to work when only a few training samples are available. Projection pursuit regression (PPR) model is an efficient way to solve nonnormal and non-linear problems, which projects high-dimensional data onto lowdimensional subspaces and analyzes the structure of the data on lowdimension. In this study, PPR method was applied to predict the occurrence degree of rice planthoppers on single cropping late rice in Xinchang of Zhejiang Province. The prediction derived from PPR was also compared with those from BP ANN and linear regression model. The results showed that both ANN and PPR model achieved historical coincidence rate of 100%, but the prediction bias of PPR was larger than that of ANN. The linear regression model showed less historical coincidence rate and larger bias than ANN and PPR model, while PPR model could access more satisfied prediction than ANN and linear regression model, being a potential method on the prediction of the outbreak of rice planthoppers.

Key words: Poplar clone, Osmotic stress, Photosynthesis photoinhibition, Reactive oxygen species, Protecting enzymes, 6-BA, AsA