Open AccessEstimation of Hurricane Intensity from ATMS-Derived Temperature Anomaly using Machine Learning
Author(s) -
Lin Lin
Publication year - 2020
Publication title -
global journal of science frontier research
Language(s) - English
Resource type - Journals
eISSN - 2249-4626
pISSN - 0975-5896
DOI - 10.34257/gjsfrhvol20is4pg5
Subject(s) - mean squared error , support vector machine , tropical cyclone , sea surface temperature , intensity (physics) , anomaly (physics) , mathematics , wind speed , machine learning , meteorology , statistics , computer science , geography , physics , condensed matter physics , quantum mechanics
The warm-core structure is one of the basic characteristics that vary during the different stages of tropical cyclones (TCs). The warm core structure of the TCs during2016-2019 over the Atlantic Ocean was derived based on the observations of the ATMS onboard S-NPP. From linear regression, the mean prediction error (MPE) is 39.04 mph for Vmax and 14.47 hPa for Pmin. The root-mean-square error(RMSE) is 42.70 mph for the maximum sustained wind (Vmax) and 77.69 hPa for the minimum sea-level pressure (Pmin). Several machine learning (ML) techniques are used to develop the Atlantic TC intensity (Vmax and Pmin) estimation models. The support vector machine (SVM) model has the best performance with the MPE of 14.62 mph for Vmaxan 7.66 hPa for Pmin, and the RMSE of 19.91 mph for Vmax and 10.58 hPa for Pmin. Adding latitude and day of year (DOY) can further improve the estimation of Vmax by decreasing MPE to 13.01mph and RME to 17.33 mph using SVM. Best estimation of Pminoccurs when adding the day of year to the training process, as the MPE is 7.23 hPa and RMS is 9.88 hPa. Other TC information, such as longitude and local time, does not help to improve the performance of the hurricane intensity estimation models significantly.