Establishment And Its Verification Of Regional Short-term Prediction Model Of Potato Late Blight Based On Meteorological Conditions In China

Potato late blight is not only the most serious disease in potato cultivation,but also the most harmful diseases in food crops.According to estimation of the international potato center(CIP),the lose of the global economic are $17 billion every year.China's potato yield and cultivation area are the first in the world.The economic loss of the potato late blight is about 8 billion yuan every year in China.According to existing research results,The effective prevention and control methods of potato late blight are the breeding of resistant varieties,the prevention and control of cultivation techniques,the prevention and control of chemical pesticides and the real-time forecast.Although the breeding disease resistant varieties is the best way,such thing is the most economical and effective,the best for environment.We could only improve cultivation technique,chemical pesticide control and real-time forecasting,as the existing potato varieties are not fully resistant to disease.The prevention and control of cultivation technique has a certain effect on the spread of epidemic disease,but the effect is not outstanding.Pesticide has the advantages of quick effect.However,that the forecast is not accurate,which may cause the spray to be sprayed too early or too late.On one hand,it increases the cost.On another hand,it causes the environmental pollution and have not a good control effect.Therefore,it is imperative to establish accurate and real-time forecasting methods.In this paper,by comparing the prediction model of potato late blight at home and abroad,and according to the pathogenesis of potato late blight,the pathogenic variables of potato late blight were selected.Then the data of potato late blight disease and national weather station data were combined and sorted and the potato growing area in China is divided into the northern planting area and the southern planting area.Then,according to the distance between the stations,the northern data sets and the southern data sets are further divided.The classification rule was established and the optimal threshold was filtered by ROC to establish the classification prediction model.In this process,due to the heavy workload,the software is programmed to organize and operate the data.Research results of northern data sets1)L < 35 km data set(76 samples):Autoregression,the sensitivity was 76%,the specificity was 76%,and the accuracy was 76%;Test,sensitivity 75%,specificity 80%,accuracy 75%;2)35?L < 60 km data set(64 samples):Autoregression,the sensitivity was 76%,the specificity was 76%,and the accuracy was 76%.Test,sensitivity 75%,specificity 70%,accuracy 76%.3)60?L km data set(40 samples): the day is sensitive when the rainfall and relative humidity meet certain conditions in this date set.Autoregression,the sensitivity was 75%,the specificity was 87% and the accuracy was 85%.Test,sensitivity 100%,specificity 80%,accuracy 85%.The average accuracy of the three sample sets was 78% and the sensitivity and the specificity of the three samples were above 70%.Research results of Southern data sets1)L < 35 km data set(27 samples):Autoregression,the sensitivity was 75%,the specificity was 79%,the accuracy was 78%;Test,the sensitivity was 50%,the specificity was 71%,the accuracy was 67%;2)35?L < 60 km data set(35 samples):Autoregression,the sensitivity was 61%,the specificity was 50%,and the accuracy was 58%.Test,Sensitivity was 80%,specificity was 100%,accuracy was 81%.3)60?L km data set(22 samples): the day is sensitive when the rainfall and relative humidity meet certain conditions in this date set.Autoregression,the sensitivity was 75%,the specificity was 100% and the accuracy was 86%.Test,the sensitivity was 100%,the specificity was 67%,the accuracy was 86%.The average accuracy rate of the three sample sets was 74%.Furthermore,we discussed about the relationship between regional and land parcel predictors,the relationship between day and time meteorological data,the southern model is inferior to the north,determination of sensitive days,the relationship between point model and regional model.