Research On The Projection Of Tropical Cyclone Activities Under Climate Change Based On HadRM3P Model And Machine Learning Methods

There are significant changes in global tropical cyclone(TC)activity in the context of climate change.The current study of the influence of climate change on TC activity mainly relies on global climate model(GCM)or dynamical downscaling product.Considering the uncertainty of future climate projection by GCM,the dynamical downscaling product with coarse resolution,the bias in dynamic downscaling,therefore,there certainly exists bias in projected TC activity under future scenario when using downscaling product.Moreover,the significant problem for GCM is that the intensity of modeled TC is much less than observation and simulations on TC frequency,track and precipitation have some flaws,it is also a big challenge for TC projection.In recent years,machine learning(ML)technique is applied in TC research particularly as the auxiliary tool of TC forecast.However,the research that combining dynamic downscaling and ML for TC projection under climate change is still poor.In order to play their respective advantages,a high-resolution regional climate model(RCM): Had RM3 P,and a series of ML models are combined and used for TC projection over the Western North Pacific(WNP),also for bias-correction of projected TC activity.At last,the influence of warming climate on TC activity is quantified.The main research contents and relevant conclusions are listed as follows.(1)To assess the ability of Had RM3 P in producing historical TC activities,the ERA-Interim reanalysis data are used as initial and boundary conditions.Results show that the Had RM3 P model can reproduce TC genesis number well,but it has poor ability in simulating TC intensity,with the top intensity is just 48 m/s.TC tracks are more north than observations,and active TCs locate in South China Sea(SCS).These TCs are all concerned with large-scale environments,such as the subtropical high shifts further east and its intensity is weaker,the vorticity(positive value)area is wider over SCS basin,the Southeast Monsoon is stronger,the easterly flow is weaker.Further,a GCM,Had GEM2-ES selected from CMIP5 model is used to drive Had RM3 P model and project TC activity over WNP in RCP4.5 and RCP8.5 scenario.Results show two RCP scenarios produce less TCs than historical benchmark period,and Had RM3 P is not capable of simulating severe TCs with wind speed greater than 50 m/s.Therefore,the model is deficient in estimating the impact of climate change on TC genesis number.(2)Abnormal TC tracks are detected using initial storm tracking algorithm: TCs with short lifetime,TCs with multiple low-pressure centers,TCs with strong wind speed at initial time,and TCs with fractured track.In view of this,a new tracking model is built with implanting a ML method: GBDT.The new model considers TC's ambient synoptic condition,TC's climatology feature and TC's movement persistence feature.Results show that less abnormal TC tracks are detected,annual storm numbers decline by 4~7,TC lifetimes extend by 2 days,TC wind speeds at initial time reduce by 1 m/s.Further analysis shows that in RCP4.5 and RCP8.5 two scenario,there is no significant increase or decrease in storm genesis numbers between RCP scenario and historical benchmark period,this conclusion is different from the previous one that TC genesis number in RCP scenario is less than in historical benchmark period.(3)The TC intensity in Had RM3 P projection is excessively low,it is a challenge for TC projection in most RCMs.Consequently,a bias-correction scheme for TC intensity was built by integrating 4 ML methods: BP neural network,partial least square,support vector machine and random forest.It considers TC's ambient synoptic conditions which are closely related to TC movement as independent variables,including these thermal factors: sea surface temperature(SST),air temperature,specific humidity,moisture flux;and these dynamic factors: relative vorticity,divergence,zonal wind,10-m wind,geopotential height and vertical wind shear.Results show that different levels of intensity are corrected well in various degrees;the genesis number of tropical depression and tropical storm have greatly decrease.This scheme solves the problem of Had RM3 P to produce no severe TC(with wind speed > 50 m/s).Further comparison also suggests there is a slight decrease in the number of severe TC from historical benchmark period to warming climate.(4)There are certain biases in simulation of TC precipitation,where Had RM3 P underestimates TC precipitation of below rainstorm level,and overestimates TC precipitation of above rainstorm level comparing to observation.Thus,a bias-correction model for TC precipitation was built based on a ML method: Xgboost.Results show that TC precipitations at the 21 st century are greater than historical benchmark period under RCP4.5;however,under RCP8.5 scenario,TC precipitations at the mid-21st(the later 21st)century are greater(smaller)than historical benchmark period.(5)This study used the TC maximum potential intensity(MPI),the genesis potential index(GPI)and the accumulated cyclone energy(ACE)to quantify the impact of climate change on TC activity.Results show that under warming climate,the effects of SST in different gears on TC intensity are also different.In the environment as SST smaller than 27?,there is an increase in TC intensity as SST increase;on the contrary,in the environment as SST greater than 27?,there is a decrease in TC intensity as SST increase.Storm genesis numbers are different in different warming climates;under RCP4.5 scenario,storm number at the mid-21 st century are roughly equal to the later 21 st century;under RCP8.5 scenario,storm numbers at the mid-21 st century are much greater than the later 21 st century.Moreover,MPI are well correspondent to storm genesis numbers in different greenhouse-gas emission scenarios or different periods.In addition,higher SST may not necessarily stimulate higher MPI and GPI.With the SST increase,the storm ACE do not increase obviously.These results preliminarily show that the impact of climate change on TC activity could be manifested through not only the increasing SST,but also those physical mechanisms(e.g.large-scale environment,water vapor condition and mesoscale disturbance)which are closely related to TC movements.