Research For Application Of Wavelet Analysis In Pest Forecasting

How to improve pest forecast, especially the accuracy of long-term pest forecast, always was the core of the work and hot spots at pest forecast. The pest population dynamics showed heterogeneity, difference, diversity, and other characteristics of sudden and random because of the ecosystem complexity inherent characteristics, these characteristics led to that the accuracy of pest forecast was unsatisfactory. In recent years, thanks to development of forecasting theory and information technology, crop pest forecast technology has made great progress. Among them, it is the new direction of development of pest forecast that combining the traditional linear theory and nonlinear science including fractal, neural network, chaos theory and wavelet analysis etc, establishing the methodology of merging various disciplines, realizing the analysis and description of pests and establishing the fast and accurate prediction forecasting system. Currently, neural networks, chaos, fractal prediction have been applied in the pest forecasting, but wavelet analysis theory, as an important component of non-linear theory application has not been applied in pest forecasting.This paper, based on development of wavelets analysis theory,-introduced wavelets analysis method into the area of pest forecasting for the first time, analyzed pest occurrence in detail with the good advantage of wavelets in time-frequency localization. This paper, studied how to use wavelets method to analyze the rules and characteristics of pest occurrence based on historical investigation data of pest, explored deeply how to combine wavelets analysis and time series analysis, neural network. This paper built the new models of pest forecasting based on instances analysis, tested the models. In summary, the major findings are as follows:Firstly, based on wavelet analysis method, this paper analyzed inter-annual and inter-decadal rules of pests. According to the cases studying for annual oviposition quantity time series of the first, second, third generation Helicoverpa armigera in Yuncheng, annual moths quantity time series of the first, second generation Ostrinia furnacalis in Yantai, annual moths quantity time series of the second generation Mythimna separata in Jining and Wendeng, this paper indicated that wavelet analysis can dig out inter-annual and inter-decadal cycle of pests clearly, determine which of the main cycle, show the evolution and fluctuation of the cycle, cycle features such as mutation points and turnning points. On the other hand, this paper analyzed short-term rules of pests based on multi-resolution analysis. According the cases studying for daily moths quantity time series trapped by light of H. armigera at1995in Linqing and at1994in Wenshang, daily moths quantity time series trapped by light of Lithocolletis ringoniella at2008in Zhaoyuan, this paper indicated that wavelet analysis can locate features of pests such as generation change and occurrence peaks.Secondly, by using Lipschitz index, the singularity of some cycle mutation points of moth amount time series of the first generation O. furnacalis in Yantai and oviposition amount time series of the second generation H. armigera in yuncheng was analyzed, the different singular characteristics of cycle mutation points of pest time series were revealed, the sizes of singularity of the cycle mutation points were measured.Thirdly, this paper applied wavelet decomposition to analyze and forecast the non-stationary time series of pests occurrence. The non-stationary time series was decomposed into several components with wavelet decomposition. Then, every component was analyzed and model was established. Finally, the models of all components were combined to obtain the model of the original non-stationary time series. At case study, the non-stationary time series of the first generation O. furnacalis occurrence degree in Yantai and the first generation O. furnacalis moths amount in Wendeng was used to establish forecasting model, the results showed that this method was suitable to analysis and forecast for pest non-stationary time series.Fourthly, this paper introduced and applied radial basis wavelet network into the area of pest forecasting for the first time, modified the learning algorithms of radial basis wavelet network for application in pest forecast. At case study, the investigation data of H. armigera in Huimin, was used to establish forecasting model of oviposition peak day in the second generation of H.armigera based on radial basis wavelet network, the investigation data of H. armigera in Western shandong was used to establish forecasting model for occurrence degree of the second generation H. armigera based on radial basis wavelet network, the investigation data of O. furnacalis in Yantai was used to establish forecasting model for occurrence degree of the first generation O. furnacalis based on radial basis wavelet network. The test result showed that the forecasting result of three models was a great satisfaction.Fifthly, based on complex wavelet transform and correlation analysis in statistic al, multi-scale correlation between pests occurrence and meteorological factors was st udied and explored preliminary. At case study, multi-scale correlation between the o viposition quantity time series of the second generation H. armigera in Hunmin and meteorological factors, multi-scale correlation between the moths quantity time serie s of the first generation O. furnacalis in Yantai and meteorological factors was anal yzed. The result showed complicated correlation existed between pests occurrence an d meteorological, wavelet analysis can analyzed this kind of complicated correlation and provide reference for forecasting models.In summary, in this study, new theories and methods were applied to the area of crop pests forecasting and some active explorations were done to improve the ac curacy of crop pests forecasting.