The Artificial Neural Network Model Is Used To Predict The Occurrence Frequency And Probability Of Tropical Cyclones In The Northwest Pacific

This study uses the Artificial Neural Network(ANN)method and the Multiple Linear Regression(MLR)method to predict the numbers and probability of the tropical cyclone(TC)formed over the western North Pacific from June to October.The correlations between the frequency of TCs and the large scale environmental variables during the boreal spring(March-May)have been analyzed in 1950-2009,and then the eight highly correlated predictors are selected to predict TC frequency in2010-2017.The comparison between the ANN and MLR models shows that the ANN model exhibits a better performance.Specifically,the correlation coefficient(R)reaches 0.99 and mean absolute error(MAE)is 0.77 during the historical data simulation.During the prediction period,the R values of ANN and MLR models are0.8 and 0.46,respectively.The MAE of ANN and MLR models are 1.97 and 3.3,respectively.For the prediction of the 24 h and 48 h probability of TC formation in individual 5×5° subregions of the WNP(0°-30°N,100°E-180°)from June to October.First,a screening step removes all data points with environmental conditions highly unfavorable to TC formation.Then,a ANN model and a MLR model are applied to the screened dataset during 2000-2009 and predict TC probability in2010-2017.The comparison between the ANN and MLR models shows that the ANN model exhibits a better performance.Specifically,with 40 hit cases and no false alarm during the historical data simulation.The predicted quality in 2010-2017 is not as well as the simulation period,but still better than MLR model in statistical and cases analysis,which further confirms that the performance in the ANN model takes significant advantage over that in the MLR model in both frequency and probability prediction,and has a good potential for application in the operational forecast.