Intensity Detection methods of Tropical Cyclone in Western North Pacific with Deviation Angle Variance Technique

In this paper, standardized infrared cloud images from Fengyun (FY) Series geostationary satellites and Best-Track Data from China Meteorological Administration (CMA-BST) within 2015-2017 are used to investigate the effects of two multi-factor models, generalized linear model (GLM) and Long Short-Term Memory (LSTM) model, for tropical cyclone (TC) intensity estimation. The typical single-factor Sigmoid function model (SFM) with map minimum value (MMV) of deviation angle variance (DAV) is also reproduced for comparison. Through applying the sensitivity experiments to DAV calculation radius and different training data groups, the estimation precision and their optimum calculation radius for DAV in Western North Pacific (WNP) are analyzed. The results show that the root mean square error (RMSE) of single-factor SFM is between 8.79 and 13.91 by using individual years as test sets and the remaining two years as training sets with the optimum calculation radius of 550 km. However, after selecting and using high-correlation factors by GLM, the RMSE of GLM and LSTM model decreases to 5.93~8.68  and 4.99~7.00 , respectively with their own optimum calculation radius of 350 km and 400 km. All sensitivity experiments indicate that the estimation results of SFM can be significantly influenced by DAV calculation radius and the characteristics of training set data, while the results of multi-factor models appear more stable. Furthermore, the multi-factor models reduce the optimum radius within the process of DAV calculation and improve the precision of TC intensity estimation in WNP, which can be an effective way for TC intensity estimation in marine area.