A Multiple Linear Regression Model for Tropical Cyclone Intensity Estimation from Satellite Infrared Images

Yong Zhao,Ruyao Sun,Chaofang Zhao,Zhixiong Wang

doi:10.3390/atmos7030040

Abstract

An objectively trained model for tropical cyclone intensity estimation from routine satellite infrared images over the Northwestern Pacific Ocean is presented in this paper. The intensity is correlated to some critical signals extracted from the satellite infrared images, by training the 325 tropical cyclone cases from 1996 to 2007 typhoon seasons. To begin with, deviation angles and radial profiles of infrared images are calculated to extract as much potential predicators for intensity as possible. These predicators are examined strictly and included into (or excluded from) the initial predicator pool for regression manually. Then, the “thinned” potential predicators are regressed to the intensity by performing a stepwise regression procedure, according to their accumulated variance contribution rates to the model. Finally, the regressed model is verified using 52 cases from 2008 to 2009 typhoon seasons. The R2 and Root Mean Square Error are 0.77 and 12.01 knot in the independent validation tests, respectively. Analysis results demonstrate that this model performs well for strong typhoons, but produces relatively large errors for weak tropical cyclones.

Highlights

Tropical Cyclones (TCs) are some of the most damaging natural hazards [1]
The results showed that the WV-IRW-to-IRW ratio (WIRa)-based indicator gives a more accurate estimation of TC intensities
The intensities within this time period observably underestimated. By examing both the best-track and the model estimated intensities, we are observably underestimated. By examing both the best-track and the model estimated intensities, found the intensity of Melor did remain stable from 0:00 UTC 2 Oct to 6:00 UTC 3 October as recorded we found the intensity of Melor did remain stable from 0:00 UTC 2 Oct to 6:00 UTC 3 October as by the original Joint Typhoon Warning Center (JTWC) best-track, but it increased to 120 knot at 12:00 UTC Oct. 03 and later to 130 recorded by the original JTWC best-track, but it increased to 120 knot at 12:00 UTC Oct. 03 and later knot

Summary

Introduction

Tropical Cyclones (TCs) are some of the most damaging natural hazards [1]. The intensity (here refers to the maximum sustained wind) is generally used as a main indicator for quantifying the damage potential of TCs. Because it is difficult to obtain direct in-situ measurements of TC intensity [2], satellite remote sensing, especially geostationary meteorological satellite (GMS) remote sensing, has been heavily relied upon in related studies and operational forecasts. GMS could cover the vast tropical ocean continually, and thereby provide valuable datasets concerning the genesis, track, intensity and other well-concerned information about TCs throughout their life cycles. The Northwestern Pacific Ocean (NWP), which is characterized with frequent cyclone activities, is heavily dependent on GMS for TC intensity estimation and forecast [3,9,10]

Objectives

Results

Conclusion