Study On Classification And Forecast Of Precipitation Points

Precipitation is the cloud of water vapor accumulated to a certain extent a product,which closely related to the occurrence and development of clouds, cloud formations, the type of cloud, and the cloud moving speed and other aspects of cloud top brightness temperature, its features are directly related to the presence or absence of precipitation and even strength. Genting brightness temperature is acquired by meteosat infrared detection channel, which is to generate infrared cloud imagery and other display the most primitive quantitative data. Genting brightness temperature is relative to the blackbody temperature is concerned, its value is generally less than zero, if the lower blackbody temperature, indicating that the higher Genting, the more active convection, thereby potentially greater precipitation, so we can Genting bright temperature values and a series of related data, through learning network model, so as to provide the possibility of forecasting precipitation.In this paper, the main work of the research of the classification and forecast of precipitation points is mainly done:(1) analysis the influence factors related to precipitation, and then select 10 precipitation the simulation parameters, using the ground monitoring stations of 1 hour encryption precipitation data, and matched to the FY-2C infrared channel current IR1 data, before temporarily time IR1 data, current when the water vapor channel IR3 data by of two types of file read, combined with rainfall simulation parameters, to generate the training data.(2) Bayesian network and BP neural network of the two methods were studied and according to the same method, select the documents of other, generate the forecast data, finally, with two kinds of methods to forecast data classification and prediction, statistics of their classification results.(3) finally on the two methods were comparative analysis, selected six contrast factor: accuracy, kappa statistic, the root mean square error and ROC area, model were established time, draw the corresponding conclusions.Through the comparison of two methods of analysis can be drawn for the same forecast data, the accuracy of the prediction results of BP neural network was higher than the accuracy of a Bayesian network prediction results. However, the former in time than the latter to. And with the data in order to predict the time and training data time span is bigger, will also decline caused by the accuracy of the prediction results of both, of course, this paper also has some shortcomings, such as: data selection may not be perfect, as well as precipitation simulation parameters get enough to consider the surface aspects.