Journal of Geophysical Research: Atmospheres
Western North Pacific Tropical Cyclone Model Tracks
in Present and Future Climates
Jennifer Nakamura1, Suzana J. Camargo1 , AdamH. Sobel1,2 , Naomi Henderson1,
KerryA. Emanuel3, Arun Kumar4 , TimothyE. LaRow5,Hiroyuki Murakami6,Malcolm J. Roberts7 ,
Enrico Scoccimarro8,9, Pier Luigi Vidale10 , Hui Wang4, Michael F. Wehner11 , andMing Zhao6
1Lamont-Doherty Earth Observatory, Columbia University, Palisades, NY, USA, 2Department of Applied Physics and
Applied Mathematics, Columbia University, New York, NY, USA, 3Program in Atmospheres, Oceans and Climate,
Massachusetts Institute of Technology, Cambridge, MA, USA, 4NOAA/NWS/NCEP Climate Prediction Center, College Park,
MD, USA, 5Verato Inc., McLean, VA, USA, 6NOAA Geophysical Fluid Dynamics Laboratory, Princeton, NJ, USA, 7Met Office,
Hadley Centre, Exeter, UK, 8Istituto Nazionale di Geofisica e Vulcanologia, Bologna, Italy, 9Centro Euro-Mediterraneo sui
Cambiamenti Climatici, Bologna, Italy, 10NCAS-Climate, University of Reading, Reading, UK, 11Lawrence Berkeley National
Laboratory, Berkeley, CA, USA
Abstract Western North Pacific tropical cyclone (TC) model tracks are analyzed in two large multimodel
ensembles, spanning a large variety of models and multiple future climate scenarios. Two methodologies
are used to synthesize the properties of TC tracks in this large data set: cluster analysis and mass moment
ellipses. First, the models’ TC tracks are compared to observed TC tracks’ characteristics, and a subset of
the models is chosen for analysis, based on the tracks’ similarity to observations and sample size. Potential
changes in track types in a warming climate are identified by comparing the kernel smoothed probability
distributions of various track variables in historical and future scenarios using a Kolmogorov-Smirnov
significance test. Two track changes are identified. The first is a statistically significant increase in the
north-south expansion, which can also be viewed as a poleward shift, as TC tracks are prevented from
expanding equatorward due to the weak Coriolis force near the equator. The second change is an eastward
shift in the storm tracks that occur near the central Pacific in one of the multimodel ensembles, indicating
a possible increase in the occurrence of storms near Hawaii in a warming climate. The dependence of the
results on which model and future scenario are considered emphasizes the necessity of including multiple
models and scenarios when considering future changes in TC characteristics.
1. Introduction
There is a large body of research aiming to understand how tropical cyclones’ (TCs) characteristics are influ-
enced by climate change (Knutson et al., 2010; Walsh et al., 2016). Most studies have focused on changes
in global TC frequency and intensity in a warming climate (Camargo, 2013, Knutson et al., 2015, Murakami
et al., 2014). As computational resources have increased and global climatemodels’ ability to simulate TCs has
improved (Camargo &Wing, 2016), analyses of other aspects of TC characteristics, including regional studies,
have gained momentum in the modeling community (Dwyer et al., 2015; Scoccimarro et al., 2014; Villarini &
Vecchi, 2012).
A TC’s landfall location depends on its track. There is large element of inherent randomness (from a climate
perspective) in each TC’s track, as it is a function of the steering winds, which can be highly variable on a
range of time scales. Some tracks can diverge from the historical record, as in the case of Hurricane Sandy (Hall
& Sobel, 2013). Nevertheless, climatologically there are typical track types that occur in each TC basin. The
possibility that there may be changes in the properties of these typical TC tracks due to climate change is of
great interest is due to thepossibility of changes in landfall occurrence. However, in order for theseprojections
of track changes to be credible, they need to be statistically significant and robust across a large number of
models and climate change scenarios.
We focus here on TC tracks over thewesternNorth Pacific (WNP) basin. TheWNP, climatologically, is the region
with the largest number of TCs per year. Typhoons in the WNP can have large impacts in many Asian coun-
tries including the Philippines, China, Taiwan, Japan, Vietnam, and South Korea. A tragic example of the large
RESEARCH ARTICLE
10.1002/2017JD027007
Key Points:
• Western North Pacific tropical
cyclone tracks’ characteristics in two
multimodel data sets are compared
with observed tracks
• Changes in TC tracks under a warming
climate are analyzed
• Track changes: a northward shift in
the most common track type and an
eastward shift in the tracks that can
potentially affect Hawaii
Correspondence to:
S. J. Camargo,
suzana@ldeo.columbia.edu
Citation:
Nakamura, J., Camargo, S. J.,
Sobel, A. H., Henderson, N.,
Emanuel, K. A., Kumar, A., …
Zhao, M. (2017). Western North
Pacific tropical cyclone model
tracks in present and future
climates. Journal of Geophysical
Research: Atmospheres, 122,
9721–9744, https://doi.org/10.1002/
2017JD027007
Received 24 APR 2017
Accepted 17 AUG 2017
Accepted article online 24 AUG 2017
Published online 29 SEP 2017
©2017. American Geophysical Union.
All Rights Reserved.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9721
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
impacts of a landfalling TC in this region was supertyphoon Haiyan, which devastated the Philippines in 2013
(Lander et al., 2014; Lin et al., 2014).
Over the last several decades, there has been a poleward shift in the average latitude of TC lifetime maxi-
mum intensities globally (Kossin et al., 2014). This shift is very robust in the WNP and is projected to continue
through the end of the century (Kossin et al., 2016). This poleward shift is expected to cause systematic shifts
in the areas at greatest TC risk in the region. On the other hand, Lin and Chan (2015) noticed a decrease in the
typhoondestructive potential in the Asia Pacific region and linked it to changes in the Pacific subtropical high,
which is strongly related to TC tracks. Mei and Xie (2016) noticed an increase in the observed intensities of the
TCsmaking landfall in Asia since late 1970s.More recently, Daloz andCamargo (2017) found a significant pole-
ward shift in themean genesis position over the Pacific basins, associatedwith a poleward shift in the genesis
indices in the region.While Liang et al. (2017) showed a connected poleward shift in typhoon-induced rainfall
over Taiwan.
Currently, there is no clear consensus on the projections of track changes in this region.While in somemodels
there is a poleward (northward in the WNP) shift (Wu et al., 2014), in others there is an eastward shift toward
the central North Pacific (Li et al., 2010; Mori et al., 2013; Murakami et al., 2011; Yokoi et al., 2013), a combi-
nation of both (Colbert et al., 2015; Murakami et al., 2012; Roberts et al., 2015; Zhao & Held, 2012), or even a
southeastward shift (Manganello et al., 2014). Given these results, it is important to consider a uniform sta-
tistical approach across multimodel data sets to this problem, so that we can investigate the robustness and
statistical significance of track changes in the WNP under global warming.
Our analysis here considers the WNP tracks in current and future climates in two multimodel data sets. The
first data set is the U.S. CLIVAR Hurricane Working Group (HWG), with contributions from multiple modeling
groups. Each modeling group performed high-resolution (0.25∘ to 1.25∘) global climate model simulations
using the same forcings for the current climate, as well as for highly idealized future climate change scenarios.
Various aspects of the HWG simulations have been analyzed in the literature, and a summary of these results
appeared in Walsh et al. (2015). Of particular interest are the results of Daloz et al. (2015), who analyzed the
TC tracks over the North Atlantic basin, using a similar methodology as that applied here to the WNP.
The secondmultimodel data set consideredhere is that from theFifthCoupledModel IntercomparisonProject
(CMIP5) (Taylor et al., 2012). Fourteen models were analyzed in the historical and one warming scenario,
namely the Representative Concentration Pathway 8.5 (RCP8.5). Most CMIP5 global climate models have low
horizontal resolution (1.2∘ to 3.0∘), and the TC activity climatology in these models have well-known biases,
such as TC intensities lower than observations (Camargo, 2013). Despite these biases, it is possible to obtain
useful information from the TC projections from the CMIP5 models as shown in Camargo (2013), Tory et al.
(2013), Tang and Camargo (2014), and Kossin et al. (2016).
In addition to the TC tracks obtained by detecting tropical cyclone-like features directly in the model output,
we also include in our analysis TC synthetic tracks obtained by a statistical-dynamical downscaling method-
ology (Emanuel et al., 2008) using the large-scale environmental fields simulated by the models as inputs.
Synthetic tracks have beengenerated using thismethod for a subset of themodels from theHWG (Daloz et al.,
2015) and CMIP5 (Emanuel, 2013; Dwyer et al., 2015) data sets.
Although there aremanypapers analyzingpossible track changes in theWNPdue to climate change, this is the
first time that a comprehensive analysis is performed using the same methodology in two large multimodel
data sets, as well as synthetic tracks generated from these data sets by statistical-dynamical downscaling.
Our analysis of the TC tracks will be based on two statistical methods. The first is a cluster analysis, which
has been extensively applied to observed (Camargo et al., 2007a, 2007b, 2008; Kossin et al., 2010) and model
tracks (Camargo, 2013; Daloz et al., 2015). The second is a method previously applied to North Atlantic TC
tracks (Nakamura et al., 2009) which synthesizes multiple track characteristics into a few parameters.
By using these two methodologies across two large multimodel data sets, we determine which type of track
changes occurmost robustly under climate change. Beforewe examine the climate change question, though,
we will use the same methods to determine the capabilities of these models to reproduce the climatological
characteristics of observed tracks in the WNP.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9722
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Table 1
Model Simulations Analyzed
Type Name Abbreviation SST CO2
HWG Control ctl climatology present
HWG plus 2K p2K clim. + 2K present
HWG 2×CO2 CO2 climatology 2× present
HWG plus 2K & 2×CO2 p2KCO2 clim. + 2K 2× present
MRI HWG Present pres present present
MRI HWG A1B SST FSST future present
MRI HWG A1B CO2 FCO2 present future
MRI HWG plus 1.83K & A1B CO2 p2KFCO2 pres. + 1.83K future
CMIP5 Historical hist coupled observed
CMIP5 RCP8.5 RCP8.5 coupled 8.5W by 2100
Note. The HWG simulations are forced with fixed SST (climatology or climatology plus 2K) and
CO2 values (present climate or twice present climate), for the present (Control) and idealized
future simulations (plus 2K, 2× CO2, plus 2K and 2× CO2. The CMIP5 historical and future pro-
jection RCP8.5 are coupled simulations. These simulations are described in detail in Walsh et al.
(2015) and Taylor et al. (2012), respectively.
The observed and model data are described in section 2. Section 3 covers our methods, sections 4 and 5
present the results for the historical and future scenarios, respectively, and we summarize those results in
section 6.
2. Data and Model Simulations
We analyzed WNP TC tracks from two multimodel data sets. The first is that from the U.S. CLIVAR HWG inter-
comparison. The HWG multimodel data set consists of a set of highly idealized experiments using a suite of
high-resolution global and regional climatemodels with the same forcings, most importantly prescribed CO2
and sea surface temperatures (SSTs) (Walsh et al., 2015), inspired by Yoshimura and Sugi (2005) and Held and
Zhao (2011). These idealized experimentswere chosen inorder togain abetter understandingof the response
of TC activity to different forcings. Here we consider four different experiments: (i) a control simulation forced
with climatological seasonally varying SSTs and sea ice concentrations (1985–2001) and atmospheric gas con-
centrations from 1992 (called “ctl”); three idealized future simulations, consisting of (ii) a uniform addition of
2K to the control experiment SSTs (plus 2K or “p2K”); (iii) a doubling of the CO2 concentration (CO2) with the
same SSTs; and (iv) 2K added to the SSTs and a doubling of CO2 (p2KCO2). A summary of these simulations is
given in Table 1. Many aspects of these simulations have already been examined (Camargo et al., 2016; Horn
et al., 2014; Patricola et al., 2014; Scoccimarro et al., 2014; Shaevitz et al., 2014; Villarini et al., 2014; Wehner
et al., 2014), but their focus was not in the WNP TC tracks, as considered here. The HWG models included in
our analysis are listed in Table 2. The TC tracks were generated by each modeling group, using their standard
tracking routines, and also given in Table 2.
In the case of the MRI model (H8), the simulation designs are not exactly the same as those used in the HWG
simulationswith theothermodels, but they are close enough thatwedecided to incorporate thismodel in our
analysis nonetheless. The MRI simulations are similar to those described in Sugi et al. (2012). For the present
climate, the MRI model is forced with monthly observed SST for the period 1979–2003, instead of monthly
climatological SST, i.e., the SST varies from year to year, instead of having the same value in a given calen-
dar month and location in all years. The MRI team defined future SST (FSST) and future CO2 (FCO2) scenarios
based on the average SST and greenhouse gas changes projected by phase 3 of the Coupled Model Inter-
comparison Project (CMIP3) data set in the period 2075–2100 for the A1B scenario (Meehl et al., 2007). The
methodology for the construction of FSST and FCO2 is explained in Sugi et al. (2012), Murakami and Wang
(2010), and Murakami et al. (2011). Three future simulations were performed with the MRI using different SST
and CO2 forcings as follows: (i) future SST (FSST) and current climate CO2, (ii) present climate SST and future
climate CO2 (FCO2), and (iii) 1.83K added uniformly to the present observed SST and future CO2 (p2KFCO2).
These simulations were constructed to examine the effect of greenhouse gases and CO2 separately, as done
in the other simulations of the HWGmultimodel ensemble.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9723
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Table 2
HWGModels’ Characteristics, References for Models and Tracking Schemes, and Number of Simulation Years in Each Scenario
Model Name Resolution Reference Tracking Scheme # Years
CAM5.1 LR H1L 1∘ Wehner Vitart/Prabhat 24
CAM5.1 HR H1 0.25∘ Wehner Vitart/Prabhat 16
CMCC/ECHAM5 H2T 0.75∘ Rockner/Scoccimarro Vitart/Walsh 9
CMCC/ECHAM5 H2 0.75∘ Rockner/Scoccimarro Vitart/Zhao 9
FSU H3 1∘ LaRow Vitart/Zhao 5
GFS H4 1∘ Saha Vitart/Zhao 20
GISS H5 1∘ Schmidt Camargo and Zebiak 20
HadGEM3 LR H6L 1.87∘ Walters Hodges/Bengtsson 20
HadGEM3 MR H6M 0.83∘ Walters Hodges/Bengtsson 20
HadGEM3 HR H6 0.35∘ Walters Hodges/Bengtsson 10
HiRAM H7 0.5∘ Zhao Vitart/Zhao 20
MRI H8 1.25∘ Mizuta/Murakami Murakami 25
Note. LR, LowResolution;MR,MediumResolution; HR, HighResolution. References:Wehner,Wehner et al. (2015); Prabhat,
Prabhat (2012); Rockner/Scoccimarro, Roeckner et al. (2003) and Scoccimarro et al. (2011); Walsh, Walsh (1997); LaRow,
LaRow et al. (2008); Vitart, Vitart et al. (2003); Saha, Saha et al. (2014); Zhao, Zhao et al. (2009); Schmidt, Schmidt et al.
(2014); Camargo and Zebiak, Camargo and Zebiak (2002); Walters, Walters et al. (2011); HB, Hodges (1995) and Bengtsson
et al. (2007a, 2007b); Mizuta and Murakami, Mizuta et al. (2012) and Murakami et al. (2012); and Murakami, Murakami
et al. (2012).
We also considered CMIP5models and simulations. These include the historical runs and one future scenario,
RCP8.5, in which greenhouse gas concentrations reach relatively high values in the later years of the 21st
century. Only one ensemble member was analyzed for each CMIP5 model and scenario. The models and TCs
considered here are the same as those included in Camargo (2013) and Tang and Camargo (2014). The TCs
were tracked using the Camargo-Zebiak tracking algorithm (Camargo & Zebiak, 2002). TheWNP TCs in a sub-
set of these models have already been discussed in Kossin et al. (2016). The list of the CMIP5models included
in our analysis is given in Table 3. The horizontal resolutions in the CMIP5models are overall much lower than
those in the HWG models. It is well known that low-resolution global climate models are able to generate
TC-like structureswithmany similarities to those of observed TCs (Bengtsson et al., 1982; Camargo et al., 2005;
Table 3
List of the CMIP5Models Analyzed, Including References and Their Horizontal Resolution
Model Name Resolution Reference
CanESM2 M1 2.9∘ von Salzen et al. (2013)
CCSM4 M2 1.2∘ Gent et al. (2011)
CSIRO-Mk3.6.0 M3 1.9∘ Rotstayn et al. (2012)
FGOALS-g2 M4 3.0∘ Bao et al. (2013)
GFDL-CM3 M5 2.5∘ Donner et al. (2011)
GFDL-ESM2M M6 2.5∘ Donner et al. (2011)
HadGEM2-ES M7 1.9∘ Jones et al. (2011)
INM-CM4.0 M8 2.0∘ Volodin et al. (2010)
IPSL-CM5A-LR M9 3.7∘ Voldoire et al. (2013)
MIROC-ESM M10 2.8∘ Watanabe et al. (2011)
MIROC5 M11 1.4∘ Watanabe et al. (2010)
MPI-ESM-LR M12 1.9∘ Zanchettin et al. (2013)
MRI-CGCM3 M13 1.2∘ Yukimoto et al. (2012)
NorESM1-M M14 2.5∘ Zhang et al. (2012)
Note.TCsare trackedusing theCamargo-Zebiak tracking routine (Camargo&Zebiak,
2002), as described in Camargo (2013).
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9724
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Table 4
DownscaledModels From the HWG and CMIP5Multimodel Ensembles
Using Emanuel’s Technique (Emanuel et al., 2006; Emanuel, 2006)
Name Type Original Model
dH1 HWG CAM5.1 LR
dH2 HWG CMCC/ECHAM5
dH5 HWG GISS
dH7 HWG HiRAM
dM2 CMIP5 CCSM4
dM5 CMIP5 GFDL-CM3
dM7 CMIP5 HadGEM2-ES
dM11 CMIP5 MIROC5
dM12 CMIP5 MPI-ESM-LR
dM13 CMIP5 MRI-CGCM3
Note. The downscaledmodels are the same as in Daloz et al. (2015)
and Emanuel (2013). The names of the downscaled models corre-
spond to the original model names (Tables 2 and 3).
Camargo &Wing, 2016; Manabe et al., 1970). However, these TC-like structures are
weaker and larger than observed storms or from high-resolution climate models
such as the HWG multimodel data set. By including the CMIP5 models, however,
we are able to span a broader range of future scenarios andmodels in our analysis,
and we judged this sufficient motivation to do so.
The tracking routinesused in theHWGandCMIP5are very similar. They look for fea-
tures in themodel output with aminimum sea level pressure, maximum low-level
vorticity andwind speed, and awarm core. All CMIP5models used the same track-
ing algorithm, but with thresholds dependent on model resolution (Camargo,
2013). In contrast, eachmodeling group applied their own tracking scheme to the
HWG models (Shaevitz et al., 2014). In the case of the HWG models Horn et al.
(2014) showed that the differences in TC frequency due to tracking algorithm
decrease as model resolution increases and TC intensity increases. We examined
some specific cases for HWG model tracks, similar to what was done in Daloz
et al. (2015), and we could not find any dependence of our results on the tracking
routine considered.
In addition to the TC tracks from the explicit simulations from the HWG and the
CMIP5 models, we also analyzed tracks produced by statistical-dynamical down-
scaling from a subset of these models. The downscaling uses the method devel-
oped by Emanuel et al. (2006) and Emanuel (2006). The main benefit of this downscaling technique is that
it can generate a very large number of synthetic TC tracks with realistic intensities based on environmental
fields from reanalyses and climatemodels. This technique has been successfully applied to generate TC tracks
from both reanalysis (Emanuel, 2010) and climate models (Emanuel et al., 2008) and has been coupled with
storm surge models (Lin et al., 2012).
Table 5
Number of WNP Storms (or Tracks) in EachModel and Scenario for the HWGMultimodel Ensemble
Name Type ctl p2K CO2 p2kCO2 Total <ctl>
H1L HWG 24 29 33 29 115 1
H1 HWG 153 105 169 157 584 9
H2T HWG 482 404 436 418 1740 48.2
H2 HWG 354 272 343 313 1282 35.4
H3 HWG 145 133 135 105 518 29.0
H4 HWG 92 80 83 80 335 3.7
H5 HWG 579 528 681 637 2425 28.9
H6L HWG 190 175 218 – 583 8.6
H6M HWG 138 109 119 – 366 13.8
H6 HWG 128 100 126 – 354 16
H7 HWG 677 648 591 482 2398 33.8
pres FSST FCO2 p2KFCO2 Total <ctl>
H8 HWG 747 627 528 728 2630 6.2
ctl p2K CO2 p2kCO2 Total
dH1 downs. HWG 2987 2267 2434 2184 9872
dH2 downs. HWG 2858 2744 2764 2738 11104
dH5 downs. HWG 2799 2979 2888 2997 11663
dH7 downs. HWG 2575 2576 2711 2705 10567
Observed climatology per year: 28.5
Note.Models in boldface were selected to be used when comparing present and future climates using two
criteria: number of storms available and similarity of the storm tracks by clusters with observations. In cases
that more than one version of tracks were available per model type (using different model resolution or
tracking scheme) only one version of the model tracks was considered. See text for more details of the
selection criteria. For the explicit models, the median number of storms per year is shown in <ctl>, as well
the median number of named storms per year in observations for the period 1981–2010.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9725
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Table 6
Number of WNP Storms (or Tracks) in EachModel and Scenario for the CMIP5
Multimodel Ensemble
hist RCP85 Total <hist>
M1 CMIP5 346 443 789 17.3
M2 CMIP5 62 144 206 3.1
M3 CMIP5 1009 1568 2577 50.4
M4 CMIP5 252 140 392 12.6
M5 CMIP5 697 903 1600 34.8
M6 CMIP5 1520 1258 2778 76.0
M7 CMIP5 255 344 599 12.7
M8 CMIP5 13 111 124 0.6
M9 CMIP5 85 297 382 4.2
M10 CMIP5 27 16 43 1.3
M11 CMIP5 354 402 756 17.7
M12 CMIP5 974 991 1965 48.7
M13 CMIP5 2330 832 3162 116.5
M14 CMIP5 20 31 51 1.0
dM2 downs. CMIP5 10413 18241 28654
dM5 downs. CMIP5 8748 14966 23714
dM7 downs. CMIP5 7332 13641 20973
dM10 downs. CMIP5 7006 11950 18956
dM11 downs. CMIP5 9004 14808 23812
dM12 downs. CMIP5 7253 13064 20317
Observed climatology per year: 28.5
Note. Models in boldface were selected to be used when comparing present
and future climates using two criteria: number of storms available and similar-
ity of the storm tracks by clusters with observations. See text for more details
of the selection criteria. For the explicit models, themedian number of storms
per year is shown in <hist>, as well the median number of named storms per
year in observations for the period 1981–2010.
The Emanuel’s downscaling technique is described in detail in Emanuel (2006) and Emanuel et al. (2006);
here we only give a brief summary. First, synthetic track origin points are generated by seeding randomly the
smoothed space-time observed probability distribution function of tropical cyclone genesis. The survival of
these seeds depends on its environment. Once the storm is generated, it moves according to the environ-
mental winds vertically averaged over a deep layer of the troposphere, with a correction for the “beta drift”
(Holland, 1983), similar to the well-known “beta and advectionmodel” (Marks, 1992). Once the track is gener-
ated, the Coupled Hurricane Intensity Prediction System (CHIPS) (Emanuel et al., 2004) is run along each track
and determines the storm intensity, as well as when the storm dissipates. The environmental fields necessary
to generate the synthetic tracks used here are from the CMIP5 and HWGmodel simulations.
The CMIP5 synthetic tracks analyzed here have been previously discussed in Emanuel (2013), Dwyer et al.
(2015), and Kossin et al. (2016) and were generated from a subset of the CMIP5 models above. Similarly, syn-
thetic TC tracks were generated from a subset of the HWG models, as discussed in Daloz et al. (2015) for the
case of the North Atlantic. The list of downscaledmodels is given in Table 4. Tables 5 and 6 show the numbers
of TC tracks in each model and scenario analyzed here.
There are two important caveats in our analysis that should be clearly stated. The first is that when comparing
the CMIP5 and HWG explicit tracks, the differences between the HWG and CMIP5 simulations are convolved
with the differences in model resolution, which affects TC simulation. The second is that, while the CMIP5
simulations are coupled, the HWG are forced with fixed SSTs; therefore, the HWG experiments cannot enforce
surface energy balance, which could have potential consequences when simulating TCs, similar to the issues
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9726
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
due to SST bias in the coupled simulations. Therefore, there is no reason to expect that the track changes in
the HWG experiments should be consistent with those in the CMIP5 simulations.
We compared the model TC data with WNP observed TC tracks from the Joint Typhoon Warning Center
best-track data set for the period 1950–2013 (Chu et al., 2002; JTWC, 2017).
3. Methods
3.1. Cluster Analysis
We use a cluster analysis method that has been extensively used to analyze TC tracks, both in observations
(Camargo et al., 2007a, 2007b, 2008; Kossin et al., 2010; Ramsay et al., 2012) andmodels (Camargo, 2013; Daloz
et al., 2015). This method is described in detail in Gaffney (2004) and was first applied to extratropical cyclone
tracks (Gaffney et al., 2007). The cluster technique is based on a mixture of polynomial regression models
(quadratic here), which are used to fit the shape of the TC tracks. The log likelihood is a goodness of fit metric
for probabilistic models. Here the best fit is obtained by maximizing the likelihood that these polynomials fit
the data, in this case the longitude and latitude of the tracks. Each model is described by a set of parameters,
including regression coefficients and a noise matrix.
The strength of the cluster analysis technique is that it easily fits tracks of different lengths. As is typical in clus-
ter analysis, however, the number of clusters is not uniquely determined but must be specified a priori. Here
we use the same number of clusters that was chosen for observed WNP typhoon tracks, i.e., seven (Camargo
et al., 2007a, 2007b). By choosing the same number of clusters in models and observations, we can make a
direct comparison.
Each model track is assigned to a specific cluster. In the case of the explicit model tracks, there are cases in
which there are not many storms per model and scenario (a typical bias of low-resolutionmodels). Therefore,
in order to increase the data sample size used in the cluster analysis in each case, we considered the tracks of
all scenarios simultaneously for each model as an input of the cluster algorithm. Once each track is assigned
to a specific cluster, we can identify to which scenario it belongs.
3.2. Track Moments
A method to distill track shape and length down to a few physically relevant parameters was developed by
Nakamura et al. (2009). The entire track shape and length are taken into account to define mass moments of
the open curve that defines a storm track. These moments can be used to summarize the statistical charac-
teristics of the storm tracks. The centroid is the first mass moment defining the longitude (X) and latitude (Y)
of the center of mass of an individual track or collection of tracks. In the case of an individual track, this cen-
troid lies in the interior of the curve, but not on the curve itself. This first moment determines the location of
the effective center of gravity of the individual track or group of tracks. The secondmassmoments are amea-
sure of the shape of the track or tracks considered. They are defined by the variance or the average squared
differences of the weighted distances from the centroid and can be expressed geometrically as a covariance
ellipse. The variance is then represented by the orientation and length of the principal axes of the ellipse and
is a measure of the extent of the tracks in three directions X , Y , and XY . By analyzing the location of the cen-
troids and the shape of the ellipses, one is able to synthesize a large amount of information about the tracks
in a very simplified manner. For instance, a rounded variance ellipse implies that the variance in directions X
and Y is very similar, while the tilt of the ellipse points to the dominant direction of the track. Thismethodwas
applied to the North Atlantic hurricane tracks in Nakamura et al. (2009), where it is described in detail.
Here we use the ellipses for two purposes: first, to compare the model tracks to the observed ones and sec-
ond, to determine the existence of shifts in model tracks under climate change scenarios. The strength of this
method is that it uses a simple feature to represent the characteristics of the tracks, either for the whole basin
or in each cluster, which makes the comparison with observations and analysis of tracks’ shifts simpler than
using many tracks or track density.
3.3. Statistical Significance of Track Changes
We tested various characteristics of the tracks to determine if their differences are statistically significant in
present and future climates. In order to do that, first a kernel smoothing function estimator (KE) was applied
to the distributions of variables in the analysis. KE can increase the signal-to-noise ratio by making visible the
signal thatmatches the size and shape of the KE.We used theMatlab2012 default KEwhich employs a normal
kernel with an optimized bandwidth. Use of the KE before testing ensures that the continuous distribution of
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9727
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
a variable is tested rather than a difference in sampling. Future distributions are estimated at the same points
along the axis of the 20C distributions. The future distributions are then renormalized by multiplying by the
ratio of the future KE by the 20C KE.
The Kolmogorov-Smirnov (KS) test is then applied to the control and future scenarios to determine if they are
from the same underlying probability distributions at the 0.1 level. The KS test is nonparametric and com-
pares the location and shape of the empirical cumulative distribution functions of the two samples. Once
statistical significant changes in the full PDF are identified, the type of the change (e.g., westward/eastward
or larger/smaller) across the multimodel ensembles is examined based on the distribution mean. In order for
a track change to be considered statistically significant and robust at least half of the models in each of the
multimodel data sets are required to have the same type of statistically significant shift.
4. Present Climate Tracks
4.1. Observations
In Camargo et al. (2007a, 2007b) cluster analysis was applied to the observed WNP TC tracks for the period
1950–2005. Here we summarize an updated version of their analysis for the period 1950–2013. The tracks (in
grey), genesis positions (red circles), and track ellipses (in black) for all clusters (a–g) and all TCs (h) are shown
in Figure 1.
The clusters were originally labeled in order of occurrence (Camargo et al., 2007a, 2007b), from themost pop-
ulated cluster A (361 TCs) to the least populated cluster G (117 TCs). Clusters D and E had a very similar number
of storms in the original analysis, 178 and 175 TCs, respectively. In the updated version, cluster D (207 TCs) has
slightly fewer TCs than cluster E (216 TCs). Clusters A, C, and E are dominated by recurving TCs, while clusters B,
D, and F TCs are mostly straight moving, and G has a combination of both. These clusters strongly depend on
the storms’ genesis positions. Some track types are modulated by the El Niño–Southern Oscillation (ENSO):
Cluster E TC tracks occur more often in eastern Pacific El Niño events, Cluster G in central Pacific or Modoki El
Niño seasons, and Cluster A in La Niña events (Camargo et al., 2007b). Furthermore, TC tracks in clusters A, B,
and E occur more often when the Madden-Julian Oscillation is active over the western North Pacific basin.
The slopes and sizes of the variance ellipses, as well as their centroid locations, emphasize the characteris-
tics of the different clusters in Figure 1. Straight-moving clusters D and F have very elongated ellipses, while
recurving clusters ellipses are more rounded. The slopes of the ellipses differ among the recurving clusters
as well. The ellipse of cluster C has a centroid north of 20∘N and a northeastward slope, while the ellipses of
cluster E and G have centroids south of 20∘N and tilt in the northwestward direction.
4.2. Present Climate
The first question we want to examine is whether the models are able to reproduce the observed tracks in
the current climate. Given the high number of models, it is impossible to show the tracks of all models and
scenarios here, so only the tracks of a few chosen models are shown in Figure 2. On the left are tracks from
the explicit models, while on the right are the tracks of the corresponding downscaled models. Figures 2a
and 2c show HWG model tracks, with CMIP5 model tracks in Figures 2e and 2g for the control and historical
simulations, respectively. The centroid of the observed tracks is located near 138∘E and 20∘N, and the mod-
els of each type reproduce this well. The explicit models match the slight southeast to northwest tilt of the
observed tracks, while the downscaled tracks have a distinct southwest to northeast tilt. In the tracks this tilt
is reflected as a predominately eastward vs. westward movement. For instance, model dH2 in Figure 2d has
more downscaled tracks above 30∘N than the corresponding explicit tracks (H2) in Figure 2c, enhancing the
southwest to northeast tilt of the recurving tracks. The explicit model ellipses are smaller both because of the
shorter lifetime of tracks as in the case of model H2, as well as themuch smaller sample size of the data. There
aremanymore tracks of the downscaledmodels (see Tables 5 and 6), allowing awider variance, as the ellipses’
variance increases with frequency. Ideally, wewould have similar sample sizes; however, given the huge com-
putational resources necessary to generate more explicit tracks, this is not possible, and we consider in our
analysis all tracks available from all cases. Furthermore, the differences between the explicit and downscaled
ellipses could include a contribution from their different termination criteria, as the downscaled tracks allow
for extratropical transition taking the storms to higher latitudes than the explicit tracks.
We compare the mass moment ellipses of all the models’ tracks (shown in Figure 3) with the observations
(Figure 1h). The explicit HWGmodels have higher horizontal resolution and aremore closely grouped than are
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9728
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Figure 1. Western North Pacific observed tracks for the period 1950–2013. (a–g) The tracks (in grey) in individual
clusters based on the classification of Camargo et al. (2007a, 2007b). The initial positions are marked in red circles. The
mean mass moment ellipses are shown in black, with the centroids marked with a black cross.
the CMIP5 explicit models. Some of the CMIP5 explicit models have mass moments that are significantly dif-
ferently shaped than those in observations, indicating tracks that are not realistic, as was seen for the Atlantic
and eastern North Pacific in Camargo (2013). This indicates, in general, that the higher horizontal resolutions
of theHWGmodels lead tomore realistic tracks or couldbe a result of the inexistenceof SSTbiases, as theHWG
simulations are forced with fixed climatological SSTs. It is interesting to notice, though, that the downscaled
HWG and CMIP5 models have ellipses with very consistent sizes and shapes. The southwest to northeast tilt
in the downscaled tracks ellipses occurs in all but one of the models (dH1).
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9729
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Figure 2. Western North Pacific model tracks (grey), genesis (red circles), and mass moment ellipses and centroids
(black) for the current climate in selected models. (a, c, e, g) Tracks from the explicit models. (b, d, f, h) The
corresponding downscaled models. HWG (CMIP5) models are shown in Figures 2a and 2c (Figures 2e and 2g) in the
control (historical) simulations. Two hundred randomly selected tracks are shown in each panel.
Some of the models have an unrealistically low number of tracks in the present and/or future climates. We
need a reasonable sample size in order for the cluster analysis to yield statistically significant results. Similar to
what was done in Camargo (2013) and Kossin et al. (2016), we exclude the models with very few tracks from
the analysis. Themodels that fall in this category areMIROC-ESM (M10, total of 43 tracks), NorESM1 (M14, total
of 51 tracks), and CAM5.1 LR (H1L, total of 115 tracks) (see Table 6).
We performed a few sensitivity tests on subsets of the model tracks as well. The first test was to examine the
role of horizontal resolution in the WNP tracks of the HadGEM3model, which was available in three different
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9730
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Figure 3. Western North Pacific mass moment ellipses for the current climate in all models. (a and c) The ellipses from
the explicit models. (b and d) The downscaled models.
resolutions: H6L, H6M, andH6 (see Table 2). Therewere no significant differences in themassmoment ellipses
among these different versions (not shown). Therefore, for the rest of our analysis we considered only the
version with the highest horizontal resolution (H6). Even the lowest resolution version of this model has a
higher resolution, though, than all the CMIP5 models. This seems to indicate that models with resolutions
as low as the CMIP5 models tend to have unrealistic tracks, as indicated by the comparison of Figure 3c and
Figures 3a, 3b, and 3d. Once the model resolution is above a certain threshold (in this case 1∘), using even
higher resolutions will not lead to further improvements in the track characteristics. This issue should be
further investigated using more models with multiple horizontal resolutions.
We also compared the tracks obtained by different tracking routines for the model CMCC/ECHAM5 (H2T and
H2; not shown). Although the number of tracks generated in each case is different, the overall characteristics
of the tracks do not depend on the tracking routine, similar to the result obtained in Daloz et al. (2015) for the
North Atlantic tracks. Therefore, for the rest of our analysis we will only consider model H2.
We applied the cluster analysis to the remaining models, i.e., excluding M10, M14, H1L, H2T, H6L, and H6M.
A test to judge model fitness is the similarity of the model tracks to the seven observed clusters. For a model
to be considered well suited for this analysis, identification of at least four of the seven observed clusters was
required. In order to do that, we compared the ellipses of the models’ and observed clusters. Primarily, the
maximum overlapping area of the observed and model ellipses was used to determine to which observed
cluster the model cluster corresponded. Second, geographic location and ellipse tilt were taken into consid-
eration. Of the 12 explicit CMIP5models considered here, only 6models passed these criteria, namelymodels
M1, M2, M3, M7, M10, and M11. In contrast, all HWG models examined and all of the downscaled CMIP5 and
HWGmodels passed this test. These results corroborate our previous conclusion that high-resolutionmodels
generate more realistic model tracks.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9731
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Figure 4. Western North Pacific model tracks in individual clusters for CMIP5 model M12 in the historical simulation.
Model clusters that do not correspond to any of the observed clusters are marked with an asterisk (∗).
The resulting clusters can be seen in Figures 4–6 for tracks from one model of each type, i.e., explicit CMIP5,
explicit HWG, and downscaled (from CMIP5). As could be expected from our discussion above, the CMIP5
clusters have some track types that do not occur in reality (Figure 4). Both the HWG (Figure 5) and the
CMIP5 downscaled (Figure 6) cluster tracks are more realistic and more similar to observations, even though
some clear biases and differences with observed clusters can still be noted. For instance, both models have
problems reproducing the South China Sea storms (straight-moving tracks in observed clusters B and D)
(Figure 1). In any case, the large improvement that can be achieved in model tracks by using either higher
horizontal resolution or downscaling techniques is very clear in these figures.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9732
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Figure 5. Western North Pacific model tracks in individual clusters for HWG model H7 in the present climate simulation.
Two hundred randomly selected tracks are shown in each panel.
5. Future Climate
5.1. Cluster Occurrence
The next question we examine is whether there are statistically significant changes in the tracks in the future
climate scenarios compared to the historical climate. In addition to assessing statistical significance, we want
to determine which changes are robust across many models. We first consider changes in the occurrence of
a cluster in the future. Do specific track types becomemore or less common in the future, and if so, are these
changes robust across models?
No statistically significant changes in frequency in the future scenarioswere found for theHWGexplicitmodel
tracks using the rank sum test at the 0.1 level for all clusters. The rank sum test was chosen as it can be used for
testing significance of small populations of unknown distributions. We repeated the same statistical test with
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9733
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Figure 6. Western North Pacific model tracks in individual clusters for the downscaled CMIP5 model dM12 in the
present climate simulation. Two hundred tracks are shown in each panel.
the CMIP5 explicit model tracks and the HWG and CMIP5 downscaled tracks. None of these models showed
a statistically significant change in the cluster assignment occurrence in future climates, as shown in Figure 7.
We also examined whether the total number of storms in the WNP in each model was statistically different
in the future and present climates, and again, no model passed the rank sum significance test, even though
there is an increase in the number of tracks in the downscaled CMIP5models aswas shown in Emanuel (2013).
5.2. Track Changes
Next we examine possible changes in the characteristics of the tracks in the future. These changes could
be related to shifts in the tracks, or tracks’ shape or length. In order to test those possibilities we applied a
Kolmogorov-Smirnov (KS) test in present and future climate distributions for each characteristic of the tracks
(e.g., longitude of the ellipse centroid), to determine if they belong to different probabilistic distributions.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9734
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Figure 7. Percentage of storms assigned to each cluster per model and scenario for the HWG explicit tracks. Clusters not
corresponding to observed clusters are marked with a star. None of the models showed a statistically significant change
in the cluster assignment occurrence in future climates.
For a change on a specific direction, e.g., northward or eastward, to be considered statistically significant for
a specific cluster or the whole basin, it needs to pass the KS test at the 0.1 level for at least half of the models
available for that type of model (HWG or CMIP5) for that cluster, or six or more models for the whole basin.
The number six was chosen as it corresponds to half of the number of CMIP5 models (explicit and down-
scaled) and HWGmodels (explicit and downscaled) considered in our analysis. However, as discussed above,
we could not identify all clusters in all models; therefore, the number ofmodels necessary for significance test
in specific clusters needs to take that into account.
As an example, we show in Figure 8 the ellipse centroid X kernel distributions for cluster E in the CO2 and
control simulations in selected HWG models, as well as cluster F in the RCP85 and historical simulations in
selected CMIP5 and downscaled CMIP5 models. Eastward and westward shifts in the means of the distribu-
tions can be clearly seen. Some distributions show shifts of the peak westward (model H1), while in others
shifts occur in the tails of the distribution (model H7), and in still others shifts are found in both peak and tails
(model H6). A similar analysis was performed for all models, clusters, and scenarios for various characteristics
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9735
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Figure 8. Kernel smoothed centroid X probability distributions estimates (PDEs) as a function of longitude for (i) cluster
E for ctl and CO2 scenarios for selected (a–f ) HWG models; (ii) cluster F for hist and RCP85 scenarios for selected CMIP5
and downscaled (g–l) CMIP5 models. The vertical lines mark the median in each probability distribution.
of the distributions, namely the locations of the centroid ellipses (centroids X and Y), the variances of the
ellipses (variances X , Y , and XY), their seasonalities, and track lengths.
The variance in the direction X is ameasure of the west to east extent of the tracks. The variance in direction Y
is ameasure of the south to north extent, and the variance in the direction XY is ameasure of the tilt, described
as a southwest to northeast or positive tilt and as southeast to northwest or negative tilt extent. These three
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9736
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Figure 9. Track ellipses in cluster A for selected models that have a statistically significant increase in the variance of Y ,
for (a–d) HWG and (e–j) CMIP5 models.
directional variances have by far the most number of significant changes in the future distributions when
compared with the control or historical simulations. As an example of our analysis, Figure 9 shows that there
are changes in cluster A, with significant changes in the variances of Y .
When all tracks in the basin are considered together, there is a net northward movement, in particular in the
RCP85 scenario, and a net eastward movement of the straight-moving tracks. However, only in one scenario
the changes in ellipse characteristics are statistically significant, namely all tracks in p2K scenario (variance
X and variance Y), with no statistically significant change for the other scenarios. This could potentially be
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9737
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Table 7
Statistical Significant Changes (0.1 Level) in Variance in Y in Future Scenarios
With a Northward (N) Shift, ComparedWith the Present Climate in the
Recurving Clusters A, C, E, and G
Model A C E G
H1 1N 2N 3N 1N 2N
H2 1N 2N 3N 2N
H3 1N 2N 1N
H4 1N 2N 3N 1N 2N 3N 1N 2N 1N 3N
H5 2N 1N
H6 1N 2N 1N 1N
H7 3N 3N
dH1 2N 3N
dH2 1N 3N
M1 4N
M2 4N
M3 4N 4N
M7 4N
M11 4N 4N
M12 4N 4N
dM2 4N
dM5 4N 4N
dM7 4N
dM12 4N 4N
Note. Future scenarios p2K, CO2, p2KCO2, and RCP85 are indicated as 1, 2,
3, and 4 in the table.
because changes in one track type cancels changes in other track types. Therefore, in order to examine this
possibility, we need to consider track changes in specific clusters.
Given the very large number of models, clusters, and scenarios analyzed, only the statistically significant and
robust results from our analysis will be discussed here. The most dominant recurving track type (cluster A)
has an increase in the variance of Y , which is statistically significant in two HWG scenarios (p2K and CO2) and
in the RCP85 CMIP5 scenario. This is consistent with the northward movement noticed for all the tracks in
the basin noted above, given that TCs do not form very close to the equator. Table 7 shows the models and
scenarios that have a significant increase in variance of Y for the recurving clusters A, C, E, and G. In contrast,
the straight-moving cluster D has a smaller variance in Y in the HWG scenarios, as well as the straight-moving
cluster F in the RCP85 scenario. Overall, significant changes in the N-S direction of the tracks were the most
frequent in our analysis, though not always consistent across the HWG and CMIP5 data sets.
Another interesting result is that the tracks in cluster F, which are westward straightmoving and can originate
in the central Pacific, have an eastward shift in centroid X for the CMIP5 RCP85 scenario (5 out of 11 models),
as well as shorter life-span (5 out of 11 models), as shown in Table 8. Furthermore, some of the HWGmodels
have a decrease of variance in X (6 out of 11 models) and a decrease in the life-span (6 out of 11 models). As
cluster F tracks are straight moving fromwest to east, both these changes would also result in a net eastward
displacement of these tracks. Taking all three changes (centroid X , variance in X , and life-time) into account,
the eastward shift in cluster F is clear, though not statistically significant when only considering centroid X
changes. As cluster F has genesis locations very close to thedate line, this eastward shiftwould lead to ahigher
occurrence of central Pacific storms in the future, as previously discussed in the literature (Colbert et al., 2015;
Li et al., 2010; Mori et al., 2013; Murakami et al., 2011, 2012; Roberts et al., 2015; Yokoi et al., 2013; Zhang et al.,
2017). This track type is alsomodulated by the central Pacific orModoki ENSO. In recent years, there have been
very active central Pacific seasons (e.g., Sobel et al., 2016), perhaps with a contribution from anthropogenic
climate change (Murakami et al., 2015, 2017). Cluster G, which can also affect Hawaii, is also the only cluster
which has consistent and statistically significant changes for X and Y variances for HWG and CMIP5 scenarios.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9738
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Table 8
Statistical Significant Changes (0.1 Level) in Centroid X, Variance of X, and Life-Span in Future
Scenarios ComparedWith the Present Climate in Cluster F
Model Variance X Life-Span Model Centroid X Life-Span
H1 1S 2S 3S 1S 2S 3S M1 4S
H2 1S 2S 3S 1S 3S M2 4E 4B
H3 3S 1B 2S 3S M11 4E 4S
H5 1S 3S M12 4E
H6 1B 2B 1B 2B dM2 4E
H7 1B 2B 1S 2S 3S dM5 4W 4S
H8 1S 2S 3B 2S 3B dM7 4E 4B
dH1 2S dM11 4W
dH5 1S 3S dM12 4S
dM13 4W 4S
H4 ⋆ ⋆ M7 ⋆ ⋆
Note. Variance of X and life-span are labeled B for bigger and S for smaller. Centroid of X
is labeled E for East and W for West. Also shown in the table with a star are the models for
which cluster F could not be identified.
In theWNP the variance of XY plays a large role in landfall potential. Themain landmass in the basin is located
to the west and northwest. The recurving track shapes of A, C, E, and G tilt toward land when moving from
southeast to northwest (negative tilt) and away from landwhenmoving southwest to northeast (positive tilt).
In two of the HWG scenarios (p2K and p2KCO2), there is an eastward shift in cluster A, the most dominant
track type. In contrast, in the CMIP5models, there is a larger tilt (variance XY) in the RCP85 scenario in three of
the clusters (A, B, and F), with the corresponding tracks, therefore, having a tendency for moving away from
land. While the types of shifts are different in both multimodel groups, they lead to a similar consequence.
The location of lifetime maximum intensity (LMI) is another metric of interest. Kossin et al. (2014) showed
that in observations this metric is less sensitive to nonmeteorological data issues. In observations there is a
poleward shift in the LMI in some regions, in particular the WNP (Kossin et al., 2014, 2016), and this pole-
ward shift in the WNP is projected to continue in the future under anthropogenic climate change (Kossin
et al., 2016).
In the case of the dominant cluster A in CMIP5 there were five models with a statistically significant LMI east-
ward shift. This eastward shift in the CMIP5 models’ cluster A is coherent with the ellipses’ eastward shift
discussed above. Furthermore, in our analysis overall (including significant and nonsignificant cases) there
were 24 cases (cluster and scenario) of a LMI northward shift out of 47 possible cases, including all of the
CMIP5 cases. However, in spite of being a clear dominant shift in the northward direction, very few were
statistically significant, including when all tracks in the basin are considered. This northward LMI shift is in
qualitative agreement with Kossin et al. (2016). It should be noted though that the chosen subset of CMIP5
models in Kossin et al. (2016) is different from the one here, as different criteria were applied. Second, here we
used a kernel smoother prior to constructing a probability distribution function and KS statistical test, while in
Kossin et al. (2016) the probability distribution functions of the latitude of LMI were constructed with the raw
model output.
5.3. Environmental Field Changes
In theprevious sectionwe found twoprimary robust track changes: a poleward shift and an increase in Central
Pacific tracks. Both of these changes are coherent with large-scale environmental changes in the models.
There is large body of literature discussing projections of a poleward shift in multiple aspects of the climate
system under global warming, mainly in extratropical clouds and storm tracks (e.g., Barnes & Polvani, 2013;
Chen & Held, 2007; Tselioudis et al., 2016; Yin, 2005), associated with the weakening and poleward expansion
of the Hadley cell under global warming (Lu et al., 2007; Vecchi & Soden, 2007). Kossin et al. (2014) showed
that the observed LMI poleward shift could be related to changes in the large-scale environment over the
past 30 years. Kossin et al. (2014) found that changes in vertical wind shear and potential intensity—the
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9739
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
latter being the theoretical maximum intensity that a TC can achieve under specified environmental condi-
tions (Emanuel, 1988)—have resulted in an expansion in the regions most favorable for TC development.
Similarly, in the CMIP5 multimodel mean there is an increase in potential intensity in the whole Northern
Hemisphere, a decrease in the vertical wind shear in the northern part of the basin, and an increase in the
genesis potential index (Camargo, Emanuel, & Sobel, 2007; Emanuel & Nolan, 2004) in the northern part of
the basin (see Camargo, 2013, Figures 12–14), which leads to a poleward expansion of the region favorable
for TC genesis and intensification. This favorable region also expands into the central North Pacific, making
that region more prone to the occurrence of TCs.
Similar analysis of the HWG multimodel ensemble environmental fields is currently in progress and will be
the topic of a future publication. Results from the Goddard Institute for Space Studies (GISS) model show that
there is an increase in the potential intensity in the western and central North Pacific for the p2K and p2KCO2
scenarios, accompaniedbyadecreaseof theverticalwind shear andan increase in the tropical cyclonegenesis
index (Camargo et al., 2014; Tippett et al., 2011) in the eastern part of the basin, leading to an expansion of the
area that is favorable for TC occurrence poleward and eastward (Camargo et al., 2016, Figures 10, 11, and 13).
Anothermetric of the environment’s favorability for TC occurrence and intensification is the ventilation index,
which combines vertical wind shear (between 850 and 250 hPa), potential intensity, and entropy deficit
(defined using the ratio of the differences of the saturated and moist entropy value at 700 hPa, and the sea
surface and boundary layer) (Tang & Emanuel, 2012).
In the CMIP5 models there is a general tendency toward an increase in the seasonal ventilation index with
warming in most basins, including the deep tropical region of the western North Pacific, which would inhibit
both tropical cyclogenesis and intensification (Tang & Emanuel, 2014). In the CMIP5 multimodel mean this
increase has a maximum around 10∘N and 160∘W, decreasing poleward and eastward of there. This change
pattern would lead to a reduction of TC activity in the southern part of the basin and an increase poleward.
There is also a decrease in the ventilation index in the central North Pacific, helping to explain the increase
of TC activity near Hawaii. In summary, the large-scale environment in the CMIP5 projections and in the HWG
GISS model simulations are coherent with the poleward and eastward track shifts discussed above.
6. Conclusions
We analyzed TC tracks in the Western North Pacific (WNP) basin in two large multimodel ensembles. These
ensembles span a variety of model types (low and high horizontal resolution models, models forced with
fixed SST, and coupledmodels) and tracks (explicit and downscaled). We used two primarymethodologies to
examine the tracks’ characteristics: a cluster analysis and mass moment ellipses. We applied these methods
first to compare the model tracks with observed tracks, and second to examine if there are changes in the
tracks in a warming climate that are statistically significant and robust across the ensembles. The impact of
trackingmethodologies on our analysis was explored, and our results do not depend on the trackingmethod
for the cases analyzed. Furthermore, it should be noted that changes in genesis locations cannot be separated
from the track changesby thismethodology, as thegenesis locations are inherently part of thedetected tracks
and the thresholds used in the different tracking algorithms.
The HWG models’ explicit tracks are much more similar to observed tracks than are the CMIP5 explicit
tracks. This indicates that, all else equal, higher horizontal resolution yields more realistic tracks. However, an
improvement with resolution was not apparent when comparing the tracks from three versions of an HWG
model in three resolutions (even the lowest resolution of this model has a finer resolution than the CMIP5
models), with no additional modifications in the model. The downscaled tracks have a northeastward bias
which is present in both the HWG and CMIP5 downscaled model tracks, indicating that these biases were
not dependent on themodels’ large-scale environments but rather appear to be features of the downscaling
methodology.
We examinedmany characteristics ofWNP tracks to determine if therewere statistically significant and robust
changes in future scenarios. There is an increase in variance of Y or south to north extent of the range over
which the tracks occur for several models and clusters. As WNP tropical cyclones are bound on the southern
end by the vanishing of the Coriolis parameter at the equator, this can be interpreted as a northern shift of
the WNP TC tracks. This northern shift is not statistically significant at a particular point, such as the mean
(centroid) or the LMI, but is very robust in the variance or extent of the model tracks.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9740
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
There were also many models and scenarios that show eastern and northeastern shifts. As the WNP basin is
bound on the west side by the Asian landmass, an extension in the variance of X can be interpreted as an
eastern movement and an extension in the variance of XY as a northeastward movement. However, in most
cases, the shifts in the centroid location are too small to be statistically significant, even when the variance
shifts are statistically significant.
Some of the track changes described here have been previously noticed in the literature, to the extent that
they are apparent in the statistics of the set of all WNP tracks. Here we pinpoint which track types, as defined
by cluster analysis, are involved in specific track shifts. In some clusters, there is an increase in the variance in
the latitudinal direction, while in others there is an eastward shift.
For the most frequent track type, recurving cluster A, while the centroid shifts are small, there is an increase
in the south-north extent of the tracks with warming in both the HWG and CMIP5 simulations, effectively cor-
responding to a northward shift in the tracks. This is an important result, as cluster A has impacts throughout
the region and occurs more commonly in La Niña events. Shifts in cluster A tracks could lead to significant
changes in the landfall occurrences, as discussed in Kossin et al. (2016).
Another interesting case is the straight-moving Cluster F, which has an eastward mean shift in the centroid
for CMIP5 models, which could lead to more storms in the Central Pacific and Hawaii. The other cluster with
potential influence in Hawaii is the recurving cluster G. While there was no significant mean centroid location
change for cluster G, the variance in both longitudinal and latitudinal directions increased in two of the HWG
scenarios, which could be interpreted as an eastward (toward Hawaii) shift in the storms’ preferred formation
region accompanied by a poleward shift in recurvature when compared to the twentieth century control
simulation.
Changes in the large-scale environment in the CMIP5 multimodel mean and in the GISS model in the HWG
data set are coherentwith the statistically significant and robust changes in track properties in theWNP. These
were, for instance, a poleward expansion of the areas with high potential intensity and increased values of
the ventilation index in the CMIP5 models. Our results highlight the complexity of potential track changes
in future climates, with different shifts occurring simultaneously for different track types. Furthermore, these
track shifts are model and scenario dependent, highlighting the value of considering multiple models and
scenarios when inferring robust changes in TC tracks in future climates. The upcomingmulti-resolutionmulti-
model simulations planned for CMIP6will be a good opportunity to explore robustly the future track changes
using high-resolution coupled models (Haarsma et al., 2016).
References
Bao, Q., Lin, P., Zhou, T., Liu, Y., Yu, Y., Wu, G., … Zhou, L. (2013). The Flexible Global Ocean-Atmosphere-Land systemmodel, spectral version
2: FGOALS-s2. Advances in Atmospheric Sciences, 30, 561–576. https://doi.org/10.1007/s00376-012-2113-9
Barnes, E. A., & Polvani, L. M. (2013). Response of the midlatitude jets, and of their variability, to increased greenhouse gases in the CMIP5
models. Journal of Climate, 26, 7117–7135. https://doi.org/10.1175/JCLI-D-12-00536.1
Bengtsson, L., Böttger, H., & Kanamitsu, M. (1982). Simulation of hurricane-type vortices in a general circulation model. Tellus A, 34, 440–457.
Bengtsson, L., Hodges, K. I., & Esch, M. (2007a). Tropical cyclones in a T159 resolution global climate model: Comparison with observations
and re-analysis. Tellus A, 59, 396–416.
Bengtsson, L., Hodges, K. I., Esch, M., Keenlyside, N., Kornblueh, L., Luo, J.-J., & Yamagata, T. (2007b). Howmany tropical cyclones change in a
warmer climate? Tellus A, 59, 539–561.
Camargo, S. J. (2013). Global and regional aspects of tropical cyclone activity in the CMIP5 models. Journal of Climate, 26, 9880–9902.
https://doi.org/10.1175/JCLI-D-12-00549.1
Camargo, S. J., & Wing, A. A. (2016). Tropical cyclones in climate models. Wiley Interdisciplinary Reviews Climate Change, 7, 211–237.
https://doi.org/10.1002/wcc373
Camargo, S. J., & Zebiak, S. E. (2002). Improving the detection and tracking of tropical storms in atmospheric general circulation
models.Weather and Forecasting, 17, 1152–1162.
Camargo, S. J., Barnston, A. G., & Zebiak, S. E. (2005). A statistical assessment of tropical cyclones in atmospheric general circulation models.
Tellus A, 57A, 589–604. https://doi.org/10.1111/j.1600-0870.2005.00117.x
Camargo, S. J., Robertson, A. W., Gaffney, S. J., Smyth, P., & Ghil, M. (2007a). Cluster analysis of typhoon tracks. Part I: General properties.
Journal of Climate, 20, 3635–3653.
Camargo, S. J., Robertson, A. W., Gaffney, S. J., Smyth, P., & Ghil, M. (2007b). Cluster analysis of typhoon tracks. Part II: Large-scale circulation
and ENSO. Journal of Climate, 20, 3654–3676.
Camargo, S. J., Emanuel, K. A., & Sobel, A. H. (2007). Use of a genesis potential index to diagnose ENSO effects on tropical cyclone genesis.
Journal of Climate, 20, 4819–4834.
Camargo, S. J., Robertson, A. W., Barnston, A. G., & Ghil, M. (2008). Clustering of eastern North Pacific tropical cyclone tracks: ENSO and MJO
effects. Geochemistry, Geophysics, Geosystems, 9, Q06V05. https://doi.org/10.1029/2007GC001861
Camargo, S. J., Tippett, M. K., Sobel, A. H., Vecchi, G. A., & Zhao, M. (2014). Testing the performance of tropical cyclone genesis indices in
future climates using the HIRAMmodel. Journal of Climate, 27, 9171–9196.
Acknowledgments
The authors acknowledge the
support from the following
grants: NASA NNX09AK34G and
NNX13AM18G, NSF AGS1143959, and
NOAA NA11OAR4310093. M.R. was
supported by the Joint UK BEIS/Defra
Met Office Hadley Centre Climate
Programme (GA01101). M.F.W.
contribution to this work is supported
by the Regional and Global Climate
Modeling Program of the Office
of Biological and Environmental
Research in the Department of Energy
Office of Science under contract
number DE-AC02-05CH11231.
P.L.V. acknowledges support from the
PRACE-UPSCALE project. We would
like to thank the members of the U.S.
CLIVAR Hurricane Working Group and
Naomi Henderson for making the
model data available for the Hurricane
Working Group and managing the
Hurricane Working Group data set. The
track model data can be made avail-
able by individual request for research
purposes by contacting the corre-
sponding author Suzana Camargo
(suzana@ldeo.columbia.edu). The
JTWC best-track data set is available at
https://metoc.ndbc.noaa.gov/web/guest/
jtwc/best_tracks/western-pacific.
NAKAMURA ET AL. TC TRACKS IN PRESENTAND FUTURE CLIMATES 9741
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Camargo, S. J., Sobel, A. H., Genio, A. D. D., Jonas, J. A., Kelley, M., Lu, Y., … Henderson, N. (2016). Tropical cyclones in the GISS ModelE2.
Tellus A, 68, 31494. https://doi.org/10.3402/tellusa.v68.31494
Chen, G., & Held, I. M. (2007). Phase speed spectra and the recent poleward shift of Southern Hemisphere surface westerlies. Geophysical
Research Letters, 34, L21805. https://doi.org/10.1029/2007GL031200
Chu, J.-H., Sampson, C. R., Levine, A. S., & Fukada, E. (2002). The Joint Typhoon Warning Center tropical cyclone best-tracks, 1945–2000
(Tech. Rep. NRL/MR/7540-02-16), Naval Research Laboratory. Retrieved from http://www.usno.navy.mil/NOOC/nmfc-ph/RSS/jtwc/
best_tracks/TC_bt_report.html.
Colbert, A. J., Soden, B. J., & Kirtman, B. P. (2015). The impact of natural and anthropogenic climate change on western North Pacific tropical
cyclone tracks. Journal of Climate, 28, 1806–1823.
Daloz, A. S., & Camargo, S. J. (2017). Is the poleward migration of tropical cyclone maximum intensity associated with a poleward migration
of tropical cyclone genesis? Climate Dynamics. https://doi.org/10.1007/s00382-017-3636-7
Daloz, A. S., Camargo, S. J., Kossin, J. P., Emanuel, K., Horn, M., Jonas, J. A., … Zhao, M. (2015). Cluster analysis of explicitly and downscaled
simulated North Atlantic tropical cyclone tracks. Journal of Climate, 28, 1333–1361. https://doi.org/10.1175/JCLI-D-13-00646.1
Donner, L. J., Wyman, B. L., Hemler, R. S., Horowitz, L. W., Ming, Y., Zhao, M., … Zeng, F. (2011). The dynamical core, physical
parameterizations, and basic simulation characteristics of the atmospheric component of the GFDL global coupled model CM3.
Journal of Climate, 24, 3484–3519. http: //dx.doi.org/10.1175/2011JCLI3955.1
Dwyer, J. G., Camargo, S. J., Sobel, A. H., Biasutti, M., Emanuel, K. A., Vecchi, G. A., … Tippett, M. K. (2015). Projected 21st century changes in
the length of the tropical cyclone season. Journal of Climate, 28, 1333–1361. https://doi.org/10.1175/JCLI-D-13-00646.1
Emanuel, K. (2006). Climate and tropical cyclone activity: A new model downscaling approach. Journal of Climate, 19, 4797–4802.
Emanuel, K. (2010). Tropical cyclone activity downscaled from NOAA-CIRES reanalysis, 1908–1958. Journal of Advances in Modeling Earth
Systems, 2, 1. https://doi.org/10.3894/JAMES.2010.2.1
Emanuel, K., DesAutels, C., Holloway, C., & Korty, R. (2004). Environmental control of tropical cyclone intensity. Journal of the Atmospheric
Sciences, 61, 843–858.
Emanuel, K., Ravela, S., Vivant, E., & Risi, C. (2006). A statistical deterministic approach to hurricane risk assessment. Bulletin of the American
Meteorological Society, 87, 299–314.
Emanuel, K., Sundararajan, R., & Williams, J. (2008). Hurricanes and global warming: Results from downscaling IPCC AR4 simulations.
Bulletin of the American Meteorological Society, 99, 347–367.
Emanuel, K. A. (1988). The maximum intensity of hurricanes. Journal of the Atmospheric Sciences, 45, 1143–1155.
Emanuel, K. A. (2013). Downscaling CMIP5 climate models shows increased tropical cyclone activity over the 21st century. Proceedings of
the National Academy of Sciences of the United States of America, 110, 12,219–12,224. https://doi.org/10.1073/pnas.1301293110
Emanuel, K. A., & Nolan, D. S. (2004). Tropical cyclone activity and global climate. Bulletin of the American Meteorological Society, 85,
666–667.
Gaffney, S. J. (2004). Probabilistic curve-aligned clustering and prediction with regression mixture models (PhD. Thesis), Department of
Information and Computer Science, Univ. of California, Irvine, CA.
Gaffney, S. J., Robertson, A. W., Smyth, P., Camargo, S. J., & Ghil, M. (2007). Probabilistic clustering of extratropical cyclones using regression
mixture models. Climate Dynamics, 29, 423–440. https://doi.org/10.1007/s00382-007-0235-z
Gent, P. R., Danabasoglu, G., Donner, L. J., Holland, M. M., Hunke, E. C., Jayne, S. R., … Zhang, M. (2011). The Community Climate System
Model version 4. Journal of Climate, 24, 4973–4991. https://doi.org/10.1175/2011JCLI4083.1
Haarsma, R. J., Roberts, M., Vidale, P. L., Senior, C. A., Bellucci, A., Bao, Q., … von Storch, J.-S. (2016). High resolution model intercomparison
project (HighResMIP). Geoscientific Model Development, 9, 4185–4208. https://doi.org/10.5194/gmd-2016-66
Hall, T. M., & Sobel, A. H. (2013). On the impact angle of Hurricane Sandy’s New Jersey landfall. Geophysical Research Letters, 40, 2312–2315.
https://doi.org/10.1002/grl.50395
Held, I. M., & Zhao, M. (2011). The response of tropical cyclone statistics to an increase in CO2 with fixed sea surface temperatures.
Journal of Climate, 20, 5353–5364.
Hodges, K. I. (1995). Feature tracking on the unite sphere.Monthly Weather Review, 19, 5686–5699.
Holland, G. J. (1983). Tropical cyclone motion: Environmental interaction plus a beta effect. Journal of the Atmospheric Sciences, 40, 328–342.
Horn, M., Walsh, K., Zhao, M., Camargo, S. J., Scoccimarro, E., Murakami, H., … Oouchi, K. (2014). Tracking scheme dependence of simulate
tropical cyclone response to idealized climate simulations. Journal of Climate, 27, 9197–9213.
Jones, C. D., Hughes, J. K., Bellouin, N., Hardiman, S. C., Jones, G. S., Knight, J., … Zerroukat, M. (2011). The HadGEM2-ES implementation of
CMIP5 centennial simulations. Geoscientific Model Development, 4, 453–470. https://doi.org/10.5194/gmd-4-543-2011
Joint Typhoon Warning Center (2017). Joint Typhoon Warning Center tropical cyclone best track data site. Retrieved from
https://metoc.ndbc.noaa.gov/web/guest/jtwc/best_tracks
Knutson, T. R., McBride, J., Chan, J., Emanuel, K. A., Holland, G., Landsea, C., … Sugi, M. (2010). Tropical cyclones and climate change.
Nature Geoscience, 3, 157–163. https://doi.org/10.1038/ngeo779
Knutson, T. R., Sirutis, J. J., Zhao, M., Tuleya, R. E., Bender, M., Vecchi, G. A., … Chavas, D. (2015). Global projections of intense tropical
cyclone activity for the late twenty-first century from dynamical downscaling of CMIP5/RCP4.5 scenarios. Journal of Climate, 28,
7203–7224. https://doi.org/10.1175/JCLI-D-15-0129.1
Kossin, J. P., Camargo, S. J., & Sitkowski, M. (2010). Climate modulation of North Atlantic hurricane tracks. Journal of Climate, 23, 3057–3076.
https://doi.org/10.1175/2010JCLI3497.1
Kossin, J. P., Emanuel, K. A., & Vecchi, G. A. (2014). The poleward migration of the location of tropical cyclone maximum intensity.
Nature, 509, 349–352.
Kossin, J. P., Emanuel, K. A., & Camargo, S. J. (2016). Past and projected changes in western North Pacific tropical cyclone exposure.
Journal of Climate, 29, 5725–5739. https://doi.org/10.1175/JCLI-D-16-0076.1
Lander, M., Guard, C., & Camargo, S. J., (2014). Tropical cyclones, super-typhoon Haiyan. In “State of the Climate in 2013”. Bulletin of the
American Meteorological Society, 95, S135–S137. https://doi.org/10.1175/2014BAMSStateoftheClimate.1
LaRow, T. E., Lim, Y.-K., Shin, D. W., Chassignet, E. P., & Cocke, S. (2008). Atlantic basin seasonal hurricane simulations. Journal of Climate, 21,
3191–3206.
Li, T., Kwon, M., Zhao, M., Kug, J.-S., Luo, J.-J., & Yu, W. (2010). Global warming shifts Pacific tropical cyclone location. Geophysical Research
Letters, 37, L21804. https://doi.org/10.1029/2010GL045124
Liang, A., Oey, L., Shiming, S. M., & Chou, S. (2017). Long-term trends of typhoon-induced rainfall over Taiwan: In situ evidence of
poleward shift of typhoons in western North Pacific in recent decades. Journal of Geophysical Research: Atmospheres, 122, 2750–2765.
https://doi.org/10.1002/2017JD026446
NAKAMURA ET AL. TC TRACKS IN PRESENTAND FUTURE CLIMATES 9742
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Lin, I.-I., & Chan, J. C. L. (2015). Recent decrease in typhoon destructive potential and global warming implications. Nature Communications,
6, 7182. https://doi.org/10.1038/ncomms8182
Lin, I.-I., Pun, I.-F., & Lien, C.-C. (2014). “Category-6” supertyphoon Haiyan in global warming hiatus: Contribution from sub-surface ocean
warming. Geophysical Research Letters, 41, 8547–8553. https://doi.org/10.1002/2014GL061281
Lin, N., Emanuel, K., Oppenheimer, M., & Vanmarcke, E. (2012). Physically based assessment of hurricane surge threat under climate change.
Nature Climate Change, 2, 462–467.
Lu, J., Vecchi, G. A., & Reichler, T. (2007). Expansion of the Hadley cell under global warming. Geophysical Research Letters, 34, L06805.
https://doi.org/10.1029/2006GL028443
Manabe, S., Holloway, J. L., & Stone, H. M. (1970). Tropical circulation in a time-integration of a global model of the atmosphere. Journal of
the Atmospheric Sciences, 27, 580–613.
Manganello, J. V., Hodges, K. I., Dirtmeyer, B., Kinter, J. L., Cash, B. A., Marx, L., … Wedi, N. (2014). Future changes int he western North
Pacific tropical cyclone activity projected by a multidecadal simulation with a 16-km global atmospheric GCM. Journal of Climate, 27,
7622–7646.
Marks, D. G. (1992). The beta and advection model for hurricane track forecasting (NOAA Tech. Memo. NWS NMC 70). p. 89, Camp Springs,
MD: National Meteorological Center.
Meehl, G., Covey, C., Delworth, T., Latif, M., McAvaney, B., Mitchell, J., … Taylor, K. (2007). The WCRP CMIP3 multimodel dataset: A new era
in climate change research. Bulletin of the American Meteorological Society, 88, 1383–1394.
Mei, W., & Xie, S.-P. (2016). Intensification of landfalling typhoons over the northwest Pacific since the late 1970s. Nature Geoscience, 9,
753–757. https://doi.org/10.1038/ngeo2792
Mizuta, R., Yoshimura, H., Murakami, H., Matsueda, M., Endo, H., Ose, T., … Kitoh, A. (2012). Climate simulations using the improved
MRI-AGCM with 20-km grid. Journal of the Meteorological Society of Japan, 90A, 235–260.
Mori, M., Kimoto, M., Ishi, M., Yokoi, S., Mochikuzi, T., Chikamoto, Y., … Koyama, H. (2013). Hindcast prediction and near-future projection
of tropical cyclone activity over the western North Pacific using CMIP5 near-term experiments with MIROC. Journal of the Meteorological
Society of Japan, 91, 431–452. https://doi.org/10.2151/jmsj.2013-402
Murakami, H., & Wang, B. (2010). Future change of North Atlantic tropical cyclone tracks: Projection by a 20-km-mesh global atmospheric
model. Journal of Climate, 23, 2569–2584. https://doi.org/10.1175/2010JCLI3338.1
Murakami, H., Wang, B., & Kitoh, A. (2011). Future change of western North Pacific typhoons: Projections by a 20-km-mesh global
atmospheric model. Journal of Climate, 24, 1154–1169. https://doi.org/10.1175/2010JCLI3723.1
Murakami, H., Wang, Y., Yoshimura, H., Mizuta, R., Sugi, M., Shindo, E., … Kitoh, A. (2012). Future changes in tropical cyclone activity
projected by the new high-resolution MRI-AGCM. Journal of Climate, 25, 3237–3260.
Murakami, H., Hsu, P.-C., Arakawa, O., & Li, T. (2014). Influence of model biases on projected future changes in tropical cyclone frequency of
occurrence. Journal of Climate, 27, 2159–2181.
Murakami, H., Vecchi, G. A., Delworth, T., Paffendorf, K., Gudgel, R., Jia, L., & Zheng, F. (2015). Investigating the anthropogenic forcing and
natural variability on the 2014 Hawaiian hurricane season. In “Explaining Extremes of 2014 from a Climate Perspective”. Bulletin of the
American Meteorological Society, 96, S115–S119.
Murakami, H., Vecchi, G. A., Delworth, T. L., Wittenberg, A. T., Underwood, S., Gudgel, S., … Zhang, W. (2017). Dominant role of
subtropical Pacific warming in extreme eastern Pacific hurricane seasons: 2015 and the future. Journal of Climate, 30, 243–264.
https://doi.org/10.1175/JCLI-D-16-0424.1
Nakamura, J., Lall, U., Kushnir, Y., & Camargo, S. J. (2009). Classifying North Atlantic tropical cyclones tracks by their mass moments.
Journal of Climate, 22, 5481–5494. https://doi.org/10.1175/2009JCLI2828.1
Patricola, C. M., Saravanan, R., & Chang, P. (2014). The impact of the El-Niño-Southern Oscillation and Atlantic Meridional Mode on seasonal
tropical cyclone activity. Journal of Climate, 27, 5311–5328.
Prabhat, O. R., Byna, S., Wu, K. S., Wehner, M., & Bethel, W. (2012). TECA: A parallel toolkit for extreme climate analysis. In H. Ali et al. (Eds.),
Proceedings International Conference on Computational Science (Vol. 9, pp. 866–876) ICCS 2012, Omaha, Nebraska: Procedia Computer
Science.
Ramsay, H. A., Camargo, S. J., & Kim, D. (2012). Cluster analysis of tropical cyclone tracks in the Southern Hemisphere. Climate Dynamics, 39,
897–917.
Roberts, M. J., Vidale, P. L., Mizienlinski, M. S., Demory, M.-E., Schiemann, R., Strachan, J., … Camp, J. (2015). Tropical
cyclones in the UPSCALE ensemble of high-resolution global climate models. Journal of Climate, 28, 574–596.
https://doi.org/10.1175/JCLI-D-14-00131.1
Roeckner, E., Bäuml, G., Bonaventura, L., Brokopf, R., Esch, M., Giorgetta, M., … Tompkins, A. (2003). The atmospheric general circulation
model ECHAM5. Part I: Model description (Tech. Rep. 349, p. 127). Hamburg, Germany: Max-Planck Institute for Meteorology.
Rotstayn, L. D., Jeffrey, S. J., Collier, M. A., Dravitzki, S. M., Hirst, A. C., Syktus, J. I., & Wong, K. K. (2012). Aerosol- and greenhouse gas-induced
changes in summer rainfall and circulation in the Australasian region: A study using single-forcing climate simulations. Atmospheric
Chemistry and Physics, 12, 6377–6404. https://doi.org/10.5194/acp-12-6377-2012
Saha, S., Moorthi, S., Wu, X., Wang, J., Nadiga, S., Tripp, P., … Becker, E. (2014). The NCEP climate forecast system version 2. Journal of
Climate, 27, 2185–2208.
Schmidt, G. A., Kelley, M., Nazarenko, L., Ruedy, R., Russell, G. L., Aleinov, I., … Zhang, J. (2014). Configuration and assessment of GISS
ModelE2 contributions to the CMIP5 archive. Journal of Advances in Modeling Earth Systems, 6, 141–184.
Scoccimarro, E., Gualdi, S., Bellucci, A., Sanna, A., Fogli, P. G., Manzini, E., … Navarra, A. (2011). Effects of tropical cyclones on ocean heat
transport in a high resolution coupled general circulation model. Journal of Climate, 24, 4368–4384.
Scoccimarro, E., Gualdi, S., Villarini, G., Vecchi, G. A., Zhao, M., Walsh, K., & Navarra, A. (2014). Increased precipitation events associated
with landfalling tropical cyclones in response to a warmer climate and increased CO2. Journal of Climate, 27, 4642–4654.
https://doi.org/10.1175/JCLI-D-14-00065-1
Shaevitz, D. A., Camargo, S. J., Sobel, A. H., Jonas, J. A., Kim, D., Kumar, A., … Henderson, N. (2014). Characteristics of tropical
cyclones in high-resolution models of the present climate. Journal of Advances in Modeling Earth Systems, 6, 1154–1172.
https://doi.org/10.1002/2014MS000372
Sobel, A. H., Camargo, S. J., Barnston, A. G., & Tippett, M. K. (2016). Northern hemisphere tropical cyclones during the quasi-El Niño of late
2014. Natural Hazards, 83, 1717–1729. https://doi.org/10.1007/s11069-016-2389-7
Sugi, M., Murakami, H., & Yushimura, J. (2012). On the mechanism of tropical cyclone frequency changes due to global warming.
Journal of the Meteorological Society of Japan, 90A, 397–408. https://doi.org/10.2151/jmsj.2012-A24
Tang, B., & Camargo, S. J. (2014). Environmental control of tropical cyclones in CMIP5: A ventilation perspective. Journal of Advances in
Modeling Earth Systems, 6, 115–128. https://doi.org/10.1002/2013MS000294
NAKAMURA ET AL. TC TRACKS IN PRESENTAND FUTURE CLIMATES 9743
Journal of Geophysical Research: Atmospheres 10.1002/2017JD027007
Tang, B., & Emanuel, K. (2012). A ventilation index for tropical cyclones. Bulletin of the American Meteorological Society, 93, 1901–1912.
https://doi.org/10.1175/BAMS-D-11-00165.2
Taylor, K. E., Stouffer, R. J., & Meehl, G. A. (2012). An overview of CMIP5 and the experiment design. Bulletin of the American Meteorological
Society, 93, 485–498.
Tippett, M. K., Camargo, S. J., & Sobel, A. H. (2011). A Poisson regression index for tropical cyclone genesis and the role of large-scale vorticity
in genesis. Journal of Climate, 24, 2335–2357. https://doi.org/10.1175/2010JCLI3811.1
Tory, K. J., Chand, S. S., McBride, J. L., Ye, H., & Dare, R. A. (2013). Projected changes in late-twenty-first-century tropical cyclone frequency
in 13 coupled climate models from phase 5 of the Coupled Model Intercomparison Project. Journal of Climate, 26, 9946–9959.
https://doi.org/10.1175/JCLI-D-32-00010.1
Tselioudis, G., Lipat, B. R., Konsta, D., Grise, K. M., & Polvani, L. M. (2016). Midlatitude cloud shifts, their primary link to the Hadley cell, and
their diverse radiative effects. Geophysical Research Letters, 43, 4594–4601. https://doi.org/10.1002/2016GL068242
Vecchi, G. A., & Soden, B. J. (2007). Global warming and the weakening of the tropical circulation. Journal of Climate, 20, 4316–4340.
https://doi.org/10.1175/JCLI4258.1
Villarini, G., & Vecchi, G. A. (2012). Twenty-first-century projections of North Atlantic tropical storms from CMIP5 models. Nature Climate
Change, 2, 604–607.
Villarini, G., Lavers, D. A., Scoccimarro, E., Zhao, M., Wehner, M. F., Vecchi, G. A., … Reed, K. A. (2014). Sensitivity of tropical cyclone rainfall
to idealized global scale forcings. Journal of Climate, 27, 4622–4641. https://doi.org/10.1175/JCLI-D-13-00780.1
Vitart, F., Anderson, D., & Stockdale, T. (2003). Seasonal forecasting of tropical cyclone landfall over Mozambique. Journal of Climate, 16,
3932–3945.
Voldoire, A., Sanchez-Gomez, E., y Mélia, D. S., Decharme, B., Cassou, C., Sénési, S., … Chauvin, F. (2013). The CNRM-CM5.1 global climate
model: Description and basic evaluation. Climate Dynamics, 40, 2091–2121. https://doi.org/10.1007/s00382-011-1259-y
Volodin, E. M., Dianskii, N. A., & Gusev, A. V. (2010). Simulating present-day climate with INMCM4.0 coupled model of the atmospheric and
oceanic general circulations. Atmospheric and Oceanic Physics, 46, 414–431. http: //dx.doi.org/10.1134/S000143381004002X
von Salzen, K., Scinocca, J. F., McFarlane, N. A., Li, J., Cole, J. N. S., Plummer, D., … Solheim, L. (2013). The Canadian fourth generation
atmospheric global climate model (CanAM4). Part I: Representation of physical processes. Atmosphere-Ocean, 51, 104–125.
https://doi.org/10.1080/07055900.2012.755610
Walsh, K. (1997). Objective detection of tropical cyclones in high-resolution analyses. Monthly Weather Review, 125, 1767–1779.
Walsh, K. J. E., Camargo, S. J., Vecchi, G. A., Daloz, A. S., Elsner, J., Emanuel, K., … Henderson, N. (2015). Hurricanes and
climate: The U.S. CLIVAR Working Group on hurricanes. Bulletin of the American Meteorological Society, 96, 997–1017.
https://doi.org/10.1175/BAMS-D-13-00242.1
Walsh, K. J. E., McBride, J. L., Klotzbach, P. J., Balachandran, S., Camargo, S. J., Holland, G., … Sugi, M. (2016). Tropical cyclones and climate
change.Wiley Interdisciplinary Reviews Climate Change, 7, 65–89. https://doi.org/10.1002/wcc.371
Walters, D. N., Best, M. J., Bushell, A. C., Copsey, D., Edwards, J. M., Falloon, P. D., … Williams, K. D. (2011). The Met Office Unified
Model Global Atmosphere 3.0/3.1 and JULES Global Land 3.0/3.1 configurations. Geoscientific Model Development, 4, 919–941.
https://doi.org/10.5194/gmd-4-919-2011
Watanabe, M., Suzuki, T., O’ishi, R., Komuro, Y., Watanabe, S., Emori, S., … Kimoto, M. (2010). Improved climate simulation by MIROC5: Mean
states, variability, and climate sensitivity. Journal of Climate, 23, 6312–6335. https://doi.org/10.1175/2010JCLI3679.1
Watanabe, S., Hajima, T., Sudo, K., Nagashima, T., Takemura, T., Okajima, H., … Kawamiya, M. (2011). MIROC-ESM 2010: Model description
and basic results of CMIP5-20c3m experiments. Geoscientific Model Development, 4, 845–872. https://doi.org/10.5194/gmd-4-845-2011
Wehner, M., Reed, K. A., Prabhat, Stone, D., Collins, W. D., & Bacmeister, J. (2015). Resolution dependence of future tropical cyclone
projections of CAM5.1 in the U.S. CLIVAR Hurricane Working Group idealized configurations. Journal of Climate, 28, 3905–3925.
https://doi.org/10.1175/JCLI-D-14-00311.1
Wehner, M. F., Reed, K. A., Li, F., Prabhat, Bacmeister, J., Chen, C.-T., … Jablonowski, C. (2014). The effect of horizontal resolution on
simulation quality in the Community Atmospheric Model, CAM5.1. Journal of Advances in Modeling Earth Systems, 6, 980–997.
https://doi.org/10.1002/2013MS000276
Wu, L., Chou, C., Chen, C.-T., Huang, R., Knutson, T. R., Sirutis, J. J., … Feng, Y.-C. (2014). Simulations of the present and
late-twenty-first-century western North Pacific tropical cyclone activity using a regional model. Journal of Climate, 27, 3405–3424.
Yin, J. H. (2005). A consistent poleward shift of the storm tracks in simulations of 21st century climate. Geophysical Research Letters, 32,
L18701. https://doi.org/10.1029/2005GL023684
Yokoi, S., Takayabu, Y. N., & Murakami, H. (2013). Attribution of projected future changes in tropical cyclone passage frequency over the
western North Pacific. Journal of Climate, 26, 4096–4111. https://doi.org/10.1175/JCLI-D-12-00218.1
Yoshimura, J., & Sugi, M. (2005). Tropical cyclone climatology in a high resolution AGCM—Impacts of SST warming and CO2 increase.
Scientific Online Letters of the Atmosphere, 1, 133–136. https://doi.org/10.2151/sola2005-035
Yukimoto, S., Adachi, Y., Hosaka, M., Sakami, T., Yoshimura, H., Hirabara, M., … Kitoh, A. (2012). A new global climate model of the
Meteorological Research Institute: MRI-CGCM3 - Model description and basic performance. Journal of the Meteorological Society of Japan,
90A, 23–64. https://doi.org/10.2151/jmsj.2012-A02
Zanchettin, D., Rubino, A., Matei, D., Bothe, O., & Jungclaus, J. H. (2013). Multidecadal-to-centennial SST variability in the MPI-ESM simulation
ensemble for the last millennium. Climate Dynamics, 40, 1301–1318. https://doi.org/10.1007/s00382-012-1361-9
Zhang, L., Karnauskas, K. B., Donnelly, J. P., & Emanuel, K. (2017). Response of the North Pacific tropical cyclone climatology
to global warming: Application of dynamical downscaling to CMIP5 models. Journal of Climate, 30, 1233–1243.
https://doi.org/10.1175/JCLI-D-16-0496.1
Zhang, Z. S., Nisancioglu, K., Bentsen, M., Tjiputra, J., Bethke, I., Yan, Q., … Jansen, E. (2012). Pre-industrial and mid-Pliocene simulations
with NorESM-L. Geoscientific Model Development, 5, 523–533. https://doi.org/10.5194/gmd-5-523-2012
Zhao, M., & Held, I. M. (2012). TC-Permitting GCM simulations of hurricane frequency response to sea surface temperature anomalies
projected for the late-twenty-first century. Journal of Climate, 2995–3009. http: //dx.doi.org/10.1175/JCLI-D-11-00313.1
Zhao, M., Held, I. M., Lin, S.-J., & Vecchi, G. A. (2009). Simulations of global hurricane climatology, interannual variability and response to
global warming using a 50 km resolution GCM. Journal of Climate, 22, 6653–6678.
NAKAMURA ET AL. TC TRACKS IN PRESENT AND FUTURE CLIMATES 9744