Teleconnection between the early immigration of brown planthopper (BPH) Nilaparvata lugens Stl and ENSO indices from January 2 years previously until June of the current year was investigated. ENSO indices included the Southern Oscillation Index (SOI) and Sea Surface Temperature (SST) anomalies of four Nino regions: Nino1+2 region (N1+2, 0°-10°S, 90°-80°W), Nino3 region (N3, 5°N-5°S, 150°-90°W), Nino4 region (N4, 5°N-5°S, 160°E-150°W) and Nino3.4 region (N3.4, 0°-10° S, 170°-120° W). Brown planthopper data used in this study were light trap catches during 1977-2003 from 16 BPH monitoring stations in 6 different BPH occurrence areas.
Pearson correlation coefficients between monthly mean ENSO indices and the early immigration of BPH at different monitoring stations were calculated. The main correlation results were as follows: ENSO indices which were significantly correlated with the early immigration of BPH were primarily SST anomalies in N3, N4 and N3.4 regions, accounting for 718% of the total. Significant ENSO indices from two years and one year before the immigration events had a proportion of about 84%, while those in the current year only accounted for 16.7%. There was significant negative correlation between the early immigration of BPH and SST anomalies for each Nino region from two years before until the previous spring, whereas there was significant positive correlation between these two factors during the period from the previous winter to the current spring. The significant correlation between the early immigration of BPH and SST anomalies for each Nino region in the last summer and autumn did not show any obvious tendencies. The relationship between the SOI and the early immigration of BPH was opposite to that between the immigration and SST anomalies for each Nino region.
The above significant ENSO indices were used as key factors to build forecasting models for the early immigration of BPH by step-wise multiple linear regression analysis. BPH light trap data from the last three years were set aside for predictive validation. To evaluate the forecasting models, the early immigration of BPH was divided into 3 levels including small immigration (1st level,y<-σ_y), medium immigration (2nd level,-σ_y≤y≤+σ_y) and large immigration (3rd level,y>+σ_y). If the predictive immigration level was the same as the actual occurrence, it was classed as a correct prediction. On the contrary, if the predicted value was different from the actual data, it was classed as an error. The historical accordance (the percentage of correct predictions in the modeled years) and predictive accuracy (correct predictions in three predictive years) of each model was calculated. The models with more than 50% of historical accordance and two correct predictions in the last three years were screened out for integrated forecasts. Finally, 12 forecasting models were obtained, which can make prediction 3-27 months ahead and had a predictive accuracy of 88.9% (32 of 36 forecasts correct).