1. Introduction
Tropical cyclones (TCs) are among the most destructive and costliest weather systems, threatening lives and inducing economic losses along coastal regions. The long-term variability, including the interdecadal and decadal variability and long-term trends, in TC activities has drawn increasing attention in recent decades under anthropogenic greenhouse gas concentration forcing (Webster 2005; Emanuel 2013; Knutson et al. 2008, 2019, 2020). Although they are intensely discussed, the future projections of TC genesis frequency (TCGF) and tracks still have large discrepancies, which partially originate from the lack of fundamental and theoretical understanding of historical variability of TC activity and partially stem from the uncertainties of observational TC records and model simulations.
TC variability in the Northern Hemisphere (NH) exhibits multiple time scale changes, which has been well documented in previous studies (Chan 2005; Walsh et al. 2016). On an interannual time scale, El Niño–Southern Oscillation (ENSO) can significantly affect TC genesis locations over the western North Pacific (WNP; Wang and Chan 2002; Camargo and Sobel 2005; Guo and Tan 2018a). During the warming phases of ENSO, the TCGF anomalies over the WNP show a dipole distribution, with an increase over the southeast part of the WNP and a decrease over the northwest part (Chen et al. 1998; Chan 2000; Wang and Chan 2002). Furthermore, El Niño events tend to suppress TCGF over the North Atlantic (NA; H. Wang et al. 2014; Patricola et al. 2016) but promote TC genesis over the eastern North Pacific (ENP; Jin et al. 2014). The sea surface temperature (SST) anomalies over the eastern Indian Ocean (EIO; Zhan et al. 2011a,b) and tropical North Atlantic (TNA; Yu et al. 2016; Huo et al. 2015) can modulate TCGF over the WNP, with the latter also influencing NA TC genesis. The prolonged SST warming over the EIO and the TNA has been shown to be favorable for a sharp decrease in TCGF over the WNP, while the warming TNA tends to enhance TC activity over the NA. Other climate factors such as the Pacific meridional mode (Gao et al. 2018; Hong et al. 2018) and some atmospheric modes (Cao et al. 2015; Huangfu et al. 2017; Zhang et al. 2020; Guo and Tan 2018b) also significantly affect interannual variability of TC activities in the NH.
Beyond interannual variability, TCGF and TC tracks undergo complicated long-term variabilities due to the tangled impacts of internal and external forcing (Murakami et al. 2020; Zhao et al. 2018, 2020a,b). The global TCGF shows an insignificant upward trend, with a significant increase in the NA and a significant decrease in the WNP since 1990s (Klotzbach et al. 2022). A substantial decrease in WNP TC track density (TCTD) and a significant increase in NA TCTD have also been observed since 1980s (Murakami et al. 2020). Recently, Zhao et al. (2020a,b) found that the nonuniform SST warming induced by greenhouse gas (GHG) concentrations could also cause changes in basin-dependent TCTD, as well as TCGF, based on historical TC records. The interdecadal Pacific oscillation (IPO) has been revealed to strongly affect TCGF/TCTD on the decadal time scale over the NA (Li et al. 2015), ENP, and WNP (Li et al. 2015; Zhao et al. 2018, 2020b), with more TCs observed over the North Pacific and fewer over the NA during positive IPO phases. Meanwhile, TC activity over the NA and WNP is also largely modulated by the Atlantic multidecadal oscillation (AMO), with an increase in TCGF over the NA but a decrease over the WNP during its positive phases, which has been well addressed in previous studies (Zhang et al. 2018; Colbert and Soden 2012; Vimont and Kossin 2007; Zhang et al. 2017).
A majority of previous studies have focused on projecting future TC changes based on global climate numerical models (Yamada et al. 2021; Yoshida et al. 2017; Knutson et al. 2010; Lin and Chan 2015; Emanuel 2005; Zhang and Wang 2017; Sugi et al. 2012, 2015), dynamical downscaling, or statistical–dynamical hybrid methods (Lee et al. 2020; Knutson et al. 2015; Emanuel 2013, 2021). These models have projected either long-term decreasing or increasing trends in TCGF with medium-to-low confidence levels (Knutson et al. 2020). The partial reason for such large discrepancy is that the long-term change in TC activity is not only influenced by GHG concentrations but also largely modulated by internal climate modes (Zhao et al. 2018; Murakami et al. 2020; Zhao et al. 2020b). The internal climate modes can further increase the uncertainties of long-term changes in TC activity under GHG forcing. Untangling the external and internal modes and evaluating their relative contributions to TCGF and TCTD (Zhao et al. 2020b) is of great importance (Li et al. 2022) but also a huge challenge. In addition to the internal climate variability, the uncertainty of long-term variability in TCs can also originate from the model itself. For instance, although the Madden–Julian oscillation (MJO) skill has been much improved in CMIP6 models (Ahn et al. 2020; Chen et al. 2022), there is still a big gap between synoptic and climate simulations that causes intraseasonal uncertainties and feedbacks to other time scale variability. The uncertainty is tightly related to convective schemes (Jiang et al. 2015, 2016; Klingaman et al. 2015) and exerts a large modulation on the intraseasonal activity and interactions with longer time scale variability in TCs (Arakane and Hsu 2021; Hsu and Li 2011). Further, although the Pacific SST zonal gradient and SST warming trend in atmospheric general circulation models are consistent with the observations, the responses of circulations and TC activities to the SST anomalies could be diverse across models. Moreover, the uncertainty of future warming patterns or SST biases (Seager et al. 2019; Zhao et al. 2020a; Heede and Fedorov 2021; Watanabe et al. 2021; Zhang et al. 2021) could also affect projections of future TC activities. Here, an important issue for the model projections is to understand the model’s confidence and uncertainty in each specific region, which will help better understand model projections for future TC changes and improve the performance of the climate models. However, it has seldom been discussed in previous studies and remains unclear.
Numerous previous studies have concluded that an improvement in model resolution could improve the model’s skill in simulating the MJO, monsoon, and TC genesis (Miura et al. 2007; Miyakawa et al. 2014; Oouchi et al. 2009; Satoh et al. 2012; Rajendran et al. 2012; Yamada and Satoh 2013; Miyamoto et al. 2014; Kodama et al. 2015), which to some extent increases the confidence of future TC change projections (Roberts et al. 2020a). The High Resolution Model Intercomparison Projection (HighResMIP; Haarsma et al. 2016) from phase 6 of the Coupled Model Intercomparison Project (CMIP6) provides a good opportunity to project future changes in TC activity (Roberts et al. 2020a,b). A natural question is to what extent we can trust these model projections for future TC activity. In fact, where results are consistent across ensembles and observations, a high degree of confidence can be assumed in the projected change. When model results are divergent or inconsistent with the observations, these discrepancies need to be noted with caution by both the scientific community and the user community (Dosio et al. 2021). For the latter, we also need to further understand why the differences occur. Thus, the main aim of this work is to compare the confidences and uncertainties of long-term changes in TCGF and TCTD from new simulations produced as part of CMIP6-HighResMIP and to understand the possible reasons. In this study, we will use the CMIP6-HighResMIP highresSST-present (Tier 1) simulations, which are the atmospheric model runs forced by the observed SST and sea ice from 1950 to 2014, to quantify the model biases in presenting TCGF and TCTD, and further evaluate the relative contributions of internal and external forcing to TC activities.
The rest of the paper is organized as follows. The data and methodology are described in section 2. TC activities in the CMIP6 HighResMIP models are presented in section 3a. The model’s confidence and uncertainty in simulating long-term TC variabilities over each subregion and the possible reasons are discussed in sections 3b and 3c. Section 3d gives the quantitative analysis of model’s biases. The conclusions and discussion are provided in the last section.
2. Data and methodology
a. Data
The historical TC data from 1950 to 2014 were obtained from the International Best Track Archive for Climate Stewardship (IBTrACS; Knapp et al. 2010, 2018), which contains the locations (longitude and latitude) and intensity (surface wind speed) of TC records at 3- and 6-h intervals. Here, we defined the TC genesis location as the TC surface wind speed equal to or larger than 35 kt (17 m s−1) for the first time and calculated the TC tracks at 6-h intervals. TCs in the boreal summer season [June–November (JJASON)] in the NH were investigated in this work. Since the annual TC frequency is much smaller over the Indian Ocean (IO) where TCs usually form during the pre- and post-monsoon seasons, we only focused on TC activity over three NH basins, the NA, ENP, and WNP.
We also derived the TC data from the fifth-generation European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA5) for the same period, which contains a horizontal resolution of 0.25° at 6-h temporal interval (Hersbach et al. 2020). The data of sea level pressure (SLP), wind speed at 850 hPa, and air temperature between 500 and 300 hPa were used to derive the TC information.
Meanwhile, the monthly atmospheric data from 1950 to 2014, mainly including winds, vertical velocity, and relative humidity, were obtained from the National Oceanic and Atmospheric Administration (NOAA) twentieth-century reanalysis datasets (NOAA 20C; Compo et al. 2006, 2011), the National Centers for Environmental Prediction–National Centers for Atmospheric Research Reanalysis I (NCEP/NCAR 1; Kalnay et al. 1996) and the ERA5. These three reanalysis datasets are averaged when analyzing the circulation patterns. The monthly SST data were downloaded from the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST; Rayner et al. 2003) with a horizontal resolution of 1° × 1°. The IPO (Henley et al. 2015; https://psl.noaa.gov/data/timeseries/IPOTPI/) and AMO (Enfield et al. 2001; https://psl.noaa.gov/data/timeseries/AMO/) indices are downloaded from the Physical Sciences Laboratory (PSL; https://psl.noaa.gov) at National Oceanic and Atmospheric Administration (NOAA). Based on the global SST averaged between the latitude 45°S and 60°N, we decomposed the SST time series into nine components using ensemble empirical mode decomposition (EEMD; Wu and Huang 2009; Wu et al. 2016) and took the last component of SST trend as the global warming (GW) index (Zhao et al. 2020b). The JJASON EIO index is defined as the average of the SST over the region 10°S–22.5°N, 75°–100°E following Zhan et al. (2011a); the ENSO index is defined as the average of the JJASON SST over the region 5°S–5°N, 120°–170°W (Zhao et al. 2016); and the TNA index is defined as the average of the SST over the region 5°–25°N, 30°–70°W (Ham et al. 2013; Yu et al. 2016).
In this study, the time series of interannual variation are derived by employing a high-pass fast Fourier transform (FFT) filter with an 11-yr cutoff period. The interdecadal/decadal variation is defined as follows: the linear trend of TCGF and TCTD is first removed, and then the remaining TCGF and TCTD are filtered by a low-pass FFT filter with an 11-yr cutoff period. The linear trend is obtained by linear regression. The long-term variability in this study includes both interdecadal variation and linear trends.
b. HighResMIP datasets
The modeled TC records in the same period were calculated based on 6-hourly model output data from six different models that participated in the CMIP6 HighResMIP (Table 1). Here, we used the CMIP6-HighResMIP Tier 1 experiments, that is, historical forced atmospheric experiments for 1950–2014 (Haarsma et al. 2016), which are also called the “highresSST-present” runs. The highresSST-present runs are forced by daily SST and sea ice datasets from HadISST with the initial state derived from the twentieth-century dataset of ERA (ERA-20C), while the anthropogenic aerosol and GHG concentrations forcing and other settings are the same as historical simulations in CMIP6 models. Among the six HighResMIP models, the ECMWF model contains five members, the EC_Earth model has three members, and the MRI has one, all having a relatively high resolution (HR; approximate 0.5°). The other three models, HadGEM3-HM, NICAM, and FGOALS, have a very high resolution (VHR) of approximately 0.25°. For multimember HR models including the ECMWF and EC_Earth, we first conducted the average of all members in the individual models, and then calculated the multimodel ensemble mean for the six models. So, the six models have equal weights. For the VHR models, we only used the dataset of the first member even if they have multiple members. Note that the model from the Max Planck Institute only has approximately 10 TCs annually over the NH based on the TC detection schemes (Roberts et al. 2020a), so we did not include it in this study. In this study, the climatological annual TC number based on our tracker is comparable with the observations. We also compared the NH TCs derived from different detection schemes (Roberts et al. 2020a,b) and extracted using different thresholds; the methods show little impact on the main results of this study.
Details of variables used to extract TC-like structures in the ERA5 and CMIP6-HighResMIP with SST and sea ice forcing and atmospheric initialization introduced.
c. TC detection algorithm
The TC detection algorithm developed by the Geophysical Fluid Dynamics Laboratory (GFDL; https://www.gfdl.noaa.gov/tstorms/; Zhao et al. 2009; Zhao et al. 2020a,b) was used to explicitly extract TC information from the high-resolution model datasets and ERA5. The extraction criteria are as follows: First, there was a low pressure center on the ground. Considering that the TC intensity is initially weak and the structure is asymmetrical, the ground isobars were not to be closed. Second, for the vortex to reach a TC strength, the surface wind speed was not less than 17 m s−1 for the 0.25° model and the 850-hPa wind speed was not less than 17 m s−1 for the 0.5° model in its lifespan, and the absolute value of the vorticity was greater than 5 × 10−5 s−1 or 10−5 s−1 for the respective VHR and HR models at the same time. Third, the temperature in the range of 1200 km in the TC center was higher than the temperature in the range of 1200–2400 km at 300–500 hPa to ensure a warm core structure for the ECMWF data and ERA5. Finally, the TC lifespan should have been at least 1.5 days (HR models) or 2 days (VHR models). The details are listed in Table 1.
d. Genesis potential index
e. Relative importance analysis
The LMG(xk) represents the contribution of regressor xk in the models in this study. To highlight the simulation bias of the model relative to the observation, if the regression coefficient of the simulated TCGF/TCTD to the regressor xk is opposite to the regression coefficient of the observed TCGF/TCTD to the regressor xk, it means that this regressor has a negative contribution to the simulated TCGF/TCTD, and its relative importance is also expressed as a negative value.
f. Definitions of TCGF and TCTD
TC activity includes many aspects, such as number, intensity, and track. This paper mainly focuses on TCGF and TCTD. Inspired by the kernel density method of calculating TCGF (Lu and Xiong 2019), here the TCGF (TCTD) on each grid was counted as the number of TCs generated in (passed by) an area of 20° in the zonal direction and 10° in the meridional direction away from the center of this grid box, and the grid spacing was 1°. This method could not only offset the uncertainty and deviation of TC generation locations, but also establish a stable relationship between the TCGF/TCTD and related environmental factors. We choose a 20° × 10° box because the synoptic waves (such as equatorial Rossby waves, mixed Rossby–gravity waves, and easterly waves, or other types of synoptic disturbances) that trigger TC genesis have a zonal scale of about 2000 km and a meridional e-folded scale (Chen and Huang 2009). We also tested 30° × 15° and 10° × 5° domains. Generally, the result from a 30° × 15° domain shows a similar correlation with that using a 20° × 10° domain, while the correlations are weak and contain much noise using 10° × 5° domains.
3. Results
a. General features of the simulated TCGF and TCTD in CMIP6-HighResMIP models
Figure 1 shows the climatological distributions of JJASON TCGF in the observations, reanalysis (ERA5), and highresSST-present simulations averaged during 1950–2014 over the NH. In general, both the ERA5 and highresSST-present runs of HighResMIP models reproduced the observational climatological distribution of TCGF (Fig. 1) and TC tracks well (Fig. S1 in the online supplemental material). A prominent feature is that the TCGFs over the three basins are lower in the HR models (∼0.5°; left panel) than in the VHR models (∼0.25°; right panel), with TCGFs in the VHR models closer to observational data. Another important feature is that the high-resolution models except the MRI and FGOALS models capture the observed highest TCGF density over the ENP among the three basins in the NH. Note that the results in the ECMWF model are from the five-member ensemble mean and those in the EC_Earth model are from the three-member ensemble mean.
Climatological distributions of the annually June–November-averaged TCGF (a) in observations, and derived from (b) ERA5 6-hourly datasets and (c1–c6) six CMIP6-HighResMIP models in the Northern Hemisphere for 1950–2014. The black dotted boxes indicate TC MDRs: SCS (herein “S”: 7°–25°N, 110°–122°E), WNP_W (“W”: 7°–25°N, 122°–140°E), WNP_E (“E”: 7°–25°N, 140°–175°E), WNP (7°–25°N, 110°–175°E), ENP (7°–20°N, 125°–90°W), Caribbean ( “Ca”: 7°–20°N, 85°–65°W), NA_T (7°–20°N, 85°–20°W), and NA_N (20°–37°N, 90°–55°W).
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
Since the six models show reasonable distributions of climatological TCGFs over the NH, we further examine the correlation of TCGF in each basin between observation and simulations. Figure 2 shows the time series of annual TCGF in observation, ERA5, and six HighResMIP models for 1950–2014 period and correlation coefficients among them. As expected, the TCGF derived from ERA5 shows a high consistency with the observation. The correlation coefficients of TCGF between observation and ERA5 are 0.68 over the WNP, 0.49 over the ENP, and 0.77 over the NA, all reaching 99% confidence level based on Student’s t test. In sharp contrast, there are the low correlations of TCGF between observation and the multimodel ensemble (MME) from six HighResMIP models over the three basins, with correlation coefficients of 0.27 over the WNP, 0.1 over the ENP, and 0.4 over the NA. The correlation is significant above 99% confidence level only over the NA, but much weaker compared with that from the ERA5 over the NH. Moreover, the correlations between MME and ERA5 over the WNP and ENP are also insignificant, but the correlation is significant over the NA.
Time series of annual TCGF summed over June–November over the (a) WNP, (b) ENP, and (c) NA during 1950–2014. The gray bars represent the observational TCGF provided by the IBTrACS; the black lines indicate the TCGF derived from the ERA5 6-hourly dataset, the red lines stand for the MME of the six models, and the green shading indicates the model spread of TCGF. The correlation coefficients of TCGF between observation and MME (ERA5), as well as ERA5 and MME, are marked in each panel over three basins. The asterisks represent the correlation coefficients are significant above 99% confidence level based on the Student’s t test.
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
To further extract the interdecadal and decadal variabilities and linear trends simulated by the models, we calculated the correlation of TCGF between observation and modeling (reanalysis) with the interannual signal removed, as shown in Fig. 3. The interannual correlations between the observed and simulated TCGF (TCTD) are shown in Fig. S2. It is clear that the ERA5 reproduced the TCGF and TCTD variabilities above interannual scale very well over the whole NH (Figs. 3a and 3h). For the HighResMIP models, almost all of them can capture the TCGF and TCTD variabilities over the eastern part of the WNP (WNP_E) since the correlations are significant and high over the WNP_E, and the three HR models and HadGEM model also reproduced the TC variability over the tropical NA (NA_T). However, over the western part of the WNP (WNP_W) and the South China Sea (SCS), defined as Fig. 1a, the models had a poor performance in simulating the observed TCGF and TCTD changes (Figs. 3b–g and 3i–n; see also Fig. S2), which could be linked to the complicated interactions over the SCS region between the MJO, monsoon, IO SST anomaly, and the local topography impacts (Hsu and Li 2011; Zhan et al. 2011b). In particular, the correlations of both the TCGF and TCTD over the ENP with the observations on a long-term time scale were negative for five models except for a weak positive relationship with HadGEM, indicating large differences in the TC variability over the ENP from 1950 to 2014 between the models and observations (Fig. 3). Such large differences in the ENP TC simulations might be associated with the poor simulations of the long-term variability beyond the interannual time scale (Fig. 3; see also Fig. S3). We also tested the correlations of the TCGFs between the observations and models since the satellite era (1975–2014), and the results were similar to the above those during 1950–2014 (figure not shown). In general, the HighResMIP models can better reproduce the interannual change in TC activities over most of the NH, except that over the WNP_W (SCS) with low interannual correlation coefficients. Another concern is that over the WNP_W and ENP, the simulations also show the opposite long-term linear trends to the observations (Fig. S3).
The spatial distribution of the correlation coefficients (shading) between observed and simulated (left) TCGF and (right) TCTD on a long-term time scale (here, the TCGF and TCTD are filtered by a low-pass FFT filter with an 11-yr cutoff period), derived from (a),(h) ERA5, (b),(i) ECMWF, (c),(j) EC-Earth, (d),(k) MRI, (e),(l) FGOALS, (f),(m) NICAM, and (g),(n) HadGEM. Contours are for the observed TCGF as in Fig. 1a. The dots indicate the area where correlation coefficients are statistically significant at the 90% confidence level by the Student’s t test after adjusting the degrees of freedom.
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
b. Confidence and uncertainty in simulating long-term TC variability
Since the models generally show large biases in reproducing long-term changes in TC activity, we will focus on examining confidence and uncertainty of the CMIP6 HighResMIP models in simulating long-term TC variability including interdecadal variabilities and linear trends. To better clarify and quantify the model’s abilities in different ocean basins and regions, the NH was divided into eight main TC development regions (MDRs), as shown in Fig. 1a.
Figure 4 shows the standardized time series of observed and MME TCGF/TCTD on interdecadal time scale in the eight MDRs. The observed TCGF and TCTD show an apparent interdecadal to decadal variability over the ENP with a decrease before the mid-1960s and after the late 1990s and an increase between them. However, the ensemble-mean TCGF and TCTD show a weaker and opposite change to the observations, both with negative correlation coefficients (r = −0.44 and −0.16, respectively) (Fig. 4a). In particular, the ENP TCGF/TCTD in the observations was increased in both the positive IPO and negative AMO phases, as shown by the green and orange shading in Fig. 4a, respectively, while that in the MME was decreased. For the northern North Atlantic (NA_N), unlike the observation, the MME TCGF shows considerably weak interdecadal variability, and exhibits an insignificant correlation (r = −0.13) with the observations (Fig. 4b). The MME also cannot well reproduce the decadal variability of the TCGF/TCTD over the SCS and WNP_W, and the simulated TCGF and TCTD were weakly correlated with the observations there (Figs. 4c and 4d).
Standardized time series of observed (black) and MME-mean (red) TCGF/TCTD on an interdecadal time scale over the (a) ENP, (b) NA_N, (c) CSC, (d) WNP_W, (e) Caribbean, (f) NA_T, (g) WNP_E, and (h) WNP. The red shading indicates the range between the maximum and minimum values of the six models. The correlation coefficients between the observed TCGF (TCTD) and MME-mean TCGF (TCTD) are shown at the top (middle) of the figure. The green background indicates the positive phase of the IPO, and the orange background indicates the negative phase of the AMO. Asterisks indicate that correlation coefficients are statistically significant at the 90% confidence level by the Student’s t test after adjusting the degrees of freedom.
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
In sharp contrast, most of the models could capture the decadal variabilities of TC activities over the Caribbean Sea, the NA_T, and the WNP_E, as shown in Figs. 4e–h. More interestingly, the TCGF and TCTD over the Caribbean Sea and NA_T decreased during the negative AMO phase but increased during the positive AMO phase, consistent with the observation. The correlation coefficients between the MME and the observation reached 0.73 and 0.74 over the Caribbean Sea and 0.61 and 0.56 over the NA_T, respectively, both significantly over the 90% confidence level based on the Student’s t test after adjusting the degrees of freedom. Additionally, the TCGF and TCTD over the Caribbean Sea and NA_T increased in the recent negative IPO phase, suggesting that the IPO might also modulate TC activities over these two regions. Nevertheless, the relative importance of the AMO and IPO to the TCGF/TCTD over these two regions was unknown. Over the WNP_E or WNP, there exist also high correlations between the observed and simulated TCGF/TCTD.
Figure 5 shows the trends of observed and ensemble-mean TCGF/TCTD in the eight MDRs. Over the ENP, there are dramatically opposite trends between the observed and MME-mean TCGF/TCTD during 1950–2014. This indicates that we should be especially cautious about the future changes in ENP TCs projected by the CMIP6 HighResMIP, since the historical simulations had large biases in reproducing long-term trends in TC activities (Fig. 5a). Similarly, the variability of the TCGF over the SCS was hardly captured by the models (Fig. 5c). There was no significant trend in TC activities over the NA_N and WNP_W (Figs. 5b and 5d). However, it was encouraging that over the Caribbean Sea, NA_T, and WNP_E (Figs. 5e–g), the simulated long-term trends of MME-mean TCGF/TCTD were consistent with the observations from 1950–2014.
Standardized time series of observed (thin black line) and MME (thin red line) TCGF/TCTD after linearly removing the interdecadal signals and the corresponding trends (thick line) over the (a) ENP, (b) NA_N, (c) CSC, (d) WNP_W, (e) Caribbean, (f) NA_T, (g) WNP_E, and (h) WNP. The solid thick line represents the trend significant at the 90% confidence level by the Mann–Kendall trend analysis.
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
Figure 6 summarizes the correlation coefficients between the simulated TCGF/TCTD in each model and the observations on interannual and interdecadal time scales and their linear trend amplitudes. In general, most models can simulate the interannual variability of TCGF/TCTD over the NH, except that the SCS remains a challenging region to simulate (Figs. 6a and 6d). Among these models, the ECMWF had the best performance in simulating the interannual variability of TC activities over the WNP_E and NA_T.
Box plots for the correlation coefficients between observed and simulated (left) TCGF and (right) TCTD on (a),(d) interannual and (b),(e) interdecadal time scales. (c),(f) Box plots for the linear trend increments of simulated (left) TCGF and (right) TCTD. The box indicates the range from the 25th to 75th percentiles of the distributions, with a line at the median, and the whiskers indicate the range of the distributions. The colored dots indicate the correlation coefficients or the linear trend increments for each of the models. The black dots in (c) and (f) indicate the linear trend increments in observed TCGF/TCTD. The solid dots (hollow dots) indicate that the correlation coefficients or the linear trend increments are statistically significant (insignificant) at the 90% confidence level by the Student’s t test in (a), (b), (d), and (e) or the Mann-Kendall trend analysis in (c) and (f). For the interdecadal components, the degrees of freedom were adjusted.
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
On the decadal time scale, four of the six models simulated an opposite decadal change in the ENP TCGF/TCTD compared to the observation, which was represented by the negative correlation coefficients; only the HadGEM and NICAM models simulated relatively consistent changes in the ENP TCGF/TCTD with the observations, with higher correlation for the HadGEM (Figs. 6b and 6e). Over the NA_N, SCS, and WNP_W, these models also performed poorly in simulating the TCGF/TCTD, and showed great differences. Specifically, over the NA_N and the SCS, the ECMWF model performed best in simulating the TCGF among all models, while the MRI model had a higher correlation with the observed TCTD than other models. The FGOALS model showed best performance in reproducing TCTD over the NA_T and Caribbean Sea, while the NICAM model simulated the interdecadal variability of TCGF/TCTD very well over the WNP_W. This indicates that for a given region we need to first evaluate the model performances and then choose the best model to describe regional variability in TC activity. Additionally, most of the models can capture the decadal variability of TCGF/TCTD over the WNP_E and NA_T (Caribbean Sea).
For the trend increments of TCGF/TCTD during 1950–2014 (Figs. 6c and 6f), the MME-mean linear trends in the TCGF over the ENP and the SCS were inconsistent with the observations, as well as the TCTD trends over the ENP. Over the WNP_W and NA_N, the observed trend in the TCGF/TCTD was weak, and the trends in the models were diverse. Over the WNP_E, five of the six models simulated a visibly decreasing trend in the TCGF/TCTD. Although the TCGF/TCTD showed an increasing trend over the NA_T (or the Caribbean Sea) from 1950 to 2014, the simulated trends were diverse in models, and the MME-mean trend was very weak (Figs. 6c and 6f).
c. Possible reasons for poor and good simulations
Internal variabilities in climate, such as ENSO, EIO, and TNA, can significantly modulate the interannual variation in TC activity in the NH, while the AMO, IPO, and anthropogenic external forcing (GW) can modulate the long-term variation in TCGF. Thus, one can seek possible factors affecting the model skills in simulating TC activity by evaluating the simulated relationship between the climate variability and TC activity. Here, we first evaluated the interannual variability of TCGF and TCTD in models influenced by ENSO, EIO, and TNA SST anomalies. The ENSO, EIO, and TNA events were defined by a threshold of 0.8 times the standard deviation. There were 14 ENSO (12 EIO and 14 TNA) warm years and 13 ENSO (14 EIO and 17 TNA) cold years. The 1978–97 period was defined as the positive phase of the IPO, and the 1961–96 period was considered to be the negative phase of the AMO, as shown in Fig. S4. From the above analysis, it can be seen that the model can reproduce the interannual change in TC activity very well, with high interannual correlation coefficients between the observed and modeled TCGFs (Fig. S2). Specifically, the responses of MME-mean TCGF to interannual variabilities have the same patterns as the observed responses, except that the response of the MME-mean TCGF to EIO has east–west deviation on WNP_W (Figs. S5a–f), which means that the MME can reproduce the response of TCGF to ENSO, EIO, and TNA well. Correspondingly, the simulated responses of large-scale environmental factors to interannual variabilities in the MME were consistent with the observed responses (Fig. S6 for ENSO; the EIO and TNA related circulations are not shown).
As mentioned above, the HighResMIP showed a poor performance in simulating TC activities over the ENP, which was mainly due to interdecadal variability and linear trends (Fig. 3). Therefore, we mainly compare the interdecadal variability and long-term linear trend of TC activities over different basins and regions between the models and observations in the following. Figure 7 shows the composite differences in the reanalysis-averaged and MME-mean large-scale environments between the positive and negative phases of the IPO and the AMO over the ENP and their corresponding trends. Here, large-scale environments include 850-hPa wind fields, 500-hPa vertical velocity and relative humidity, and vertical wind shear. As expected, the MME-mean large-scale environmental factors over the ENP induced by the IPO, AMO, and GW showed different and even opposite changes from the reanalysis datasets. In the reanalysis datasets, during the IPO positive phases, there was an anticyclone in the MDR of the ENP and a cyclone anomaly south of 10°N, and the negative center of the vertical wind shear was concentrated south of 10°N, which was more favorable for the development of storm activity in the south of the ENP (Fig. S5g). In the MME, however, the large-scale anticyclonic circulation anomaly in response to the IPO positive phases covers the whole ENP with the center shifting northward. Moreover, the anomalous sinking largely strengthens over the eastern MDR (Fig. 7b). All were not conducive to TC genesis and development over the ENP (Fig. S5h). Similar to the IPO, the MME-mean and observed TCGF and environmental factors had significant differences in response to the AMO. For the long-term linear trend, the observed TCGF showed an increase, whereas the MME-mean TCGF tends to decrease over the ENP (Figs. S5k,l), along with dramatically opposite trends between the observed and MME-mean large-scale environmental factors. The observed atmospheric conditions showed cyclonic circulation, ascending motion, and a moist middle atmosphere, while the MME-mean atmospheric conditions showed the opposite characteristics, which is conducive (not conducive) to the development of observed (MME-mean) TC activities (Fig. 7).
The composite differences in large-scale environment over the ENP averaged in (a),(c),(e),(g),(i),(k) the three reanalysis datasets (NOAA 20C, NCEP1, and ERA5) and (b),(d),(f),(h),(j),(l) the MME between the different phases of the (a)–(d) IPO and (e)–(h) AMO, and (i)–(l) their linear trend increments. (a),(b),(e),(f),(i),(j) 850-hPa wind field (arrow) and 500-hPa vertical velocity (10−2 Pa s−1; shading) and (c),(d),(g),(h),(k),(l) relative humidity (%; shading) and vertical wind shear [m s−1; contours; solid (dashed) lines indicate positive (negative) values]. In composite fields, a plus sign (+) indicates the area where the differences are significant at the 90% confidence level based on the Student’s t test. In trend fields, a plus sign (+) indicates the area where the trends are significant at the 90% confidence level based on the Mann–Kendall test.
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
We further analyze the simulated large-scale conditions associated with TC activities over the WNP and the NA (Figs. 8 and 9). For the interdecadal time scale, the models can reproduce TCGF (Figs. S5g–j) and large-scale environmental characteristics over the WNP_E and NA_T, with increased TCGF and westerly anomaly over the WNP_E and decreased TCGF and easterly anomaly over the NA_T during the positive IPO phase. However, there are great differences between the simulated and observed TCGF over the WNP_W, SCS, and NA_N, as well as the large-scale environmental conditions. For the long-term linear trend, the observed TCGF showed an east–west distribution over the WNP, with a reduced trend east of the Philippines and an increasing trend in the SCS, while the MME-mean TCGF showed a consistent decrease over the WNP (Figs. S5k,l). The models can capture large-scale environmental characteristics that were consistent with the reanalysis datasets over the WNP_E, such as anticyclonic circulation, increased VWS, and a dry middle atmosphere, while there were significant differences between models and reanalysis datasets over the WNP_W (Figs. 8i–l). There was no significant trend in the observed TCGF and the MME-mean TCGF over the NA (Figs. S5k,l).
Since the GPI can reflect the combined effects of the large-scale conditions related to TC genesis, we quantitatively compared the GPI and the relative importance of five terms in the formula between the reanalysis and the MME. Figure 10 shows the composite differences in GPI anomalies and contributions of the five terms from the reanalysis and the MME between the different phases of the IPO and the AMO. During positive IPO phases, in the reanalysis, positive (negative) GPI anomalies occur over the ENP (NA), while a quasi-tripole pattern is found over the WNP, with a negative center over the region 5°–15°N, 120°–150°E and positive ones over the northeast and northwest sides of that negative center. The GPI response to the IPO over the WNP is not quite consistent with the observed TCGF response, in which a dipole pattern exists over the WNP with positive in the east and negative in the west of about 140°E (Li et al. 2015). It could be due to poor skill of the GPI in representing WNP TC genesis (Wang and Murakami 2020). The MME shows large biases over the WNP and ENP, but good skill over the NA. The large biases over the WNP are mainly induced by the poor responses of simulated vertical wind shear, relative humidity, and vertical velocity to the IPO, while those over the ENP are due to poor model performance in reproducing the responses of vertical velocity, low-level vorticity, and relative humidity to the IPO. During negative AMO, positive GPI anomalies in the reanalysis occur over the ENP, with negative ones over the NA. Similarly, there are large biases over the ENP and the WNP in the MME, but relatively good skill over the NA. The large discrepancy over these two regions might be attributed to the inconsistency of all large-scale factors except the PI between the reanalysis and the MME. Note that the MME can well reproduce the responses of PI to the IPO and the AMO since the highresSST-present runs are forced by the observed SSTs.
The composite differences in GPI and contributions of the five terms (a)–(f) between positive and negative IPO phases and (g)–(l) between negative and positive AMO phases. (a),(b),(g),(h) Spatial distributions of GPI anomalies from (a),(g) the reanalysis and (b),(h) the MME. (c)–(f),(i)–(l) Total GPI anomalies and contributions of the five terms to GPI anomalies over different regions. The white crosses in (a), (b), (g), and (h) represent areas where the differences are significant above 90% confidence level based on the Student’s t test. The bars with crosses represent the GPI contributions are opposite between reanalysis and MME.
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
The reanalysis-based GPI responses to GW are generally characterized by a dipole over the three basins, with positive (negative) values over the eastern (western) ENP, positive (negative) over the western (eastern) WNP, and positive (negative) over the tropical NA (NA_N) (Fig. 11). In contrast, large biases remain over the ENP, WNP_W, and SCS. There is also a large bias over the Caribbean. Comparing the contributions of five terms, the biases of GPI over the ENP between the reanalysis and the MME are more likely to be induced by the distinct contributions from vertical velocity, relative humidity, and vorticity, while those over the WNP_W and SCS might be related to relative humidity.
As in Fig. 10, but for the trend magnitude of GPI during 1950–2014.
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
The above results strongly suggest that the model skills in simulating TC activities are closely associated with their ability to reproduce the observed relationship between TC activities and climate variabilities. Those that can reproduce well such relationship show good skill in simulating TC activities. There are generally large biases in the HighResMIP models in reproducing the long-term changes in TCGF and TCTD over the ENP, the WNP_W, and the SCS. Such biases might be attributed to poor model performance in simulating the responses of large-scale conditions related to TC activities to the interdecadal variability and the GW, with large contributions from the simulations of relative humidity and vertical velocity.
d. Quantitative analysis of model biases
In the above analysis, we evaluated model performance in simulating TC activity and the associated large-scale conditions on different time scales over the eight MDRs. Here, we further quantified the simulation skills induced by external forcing and internal variabilities, including the GW, IPO, AMO, ENSO, EIO, and TNA, and explored the sources of the model biases in simulating TC activity.
Figures 12 and 13 show the relative importance of external forcing (GW) and internal variabilities (IPO, AMO, ENSO, EIO, and TNA) for the observed and MME-mean TCGF and TCTD over the eight MDRs with the relative importance for each model shown in Figs. S7 and S8. Over the ENP, among these climate factors, the GW played a dominant role in the observed versus modeled TCGF variation, contributing 17.2% to 16.4% of the total variance. However, the observed and simulated TCGF showed an opposite change under the GW, similar to the TCGF trends (Fig. 5a). The AMO and IPO contributed relatively equal to the observed ENP TCGF, and the positive AMO or negative IPO suppressed the observed TCGF over the ENP, but the contributions are reversed in the models. The relative contributions of these climate indices to TCTD (Fig. 13a) are similar to those in Fig. 12a except that the AMO contribution becomes stronger.
The relative importance of external variability (GW) and internal variabilities (IPO, AMO, ENSO, EIO, and TNA) for the observed (red) and MME-mean (black) TCGF over the (a) ENP, (b) WNP_W, (c) NA_N, (d) SCS, (e) NA_T, (f) WNP, (g) Caribbean, and (h) WNP_E. The error bars in gray show 90% confidence levels based on the bootstrap resampling test.
Citation: Journal of Climate 35, 19; 10.1175/JCLI-D-21-0875.1
It should be mentioned that over the WNP_W, TCGF was mainly affected by ENSO and EIO SST anomalies in the observations (Fig. 12b), while the TCTD was mostly modulated by the IPO and ENSO (Fig. 13b). This strongly suggests that the model bias over the WNP_W may originate from the interannual time scale signals. For the SCS (Figs. 12d and 13d), the AMO has the largest contribution to the observed TCGF (TCTD), reaching 12.8% (17.9%), with fewer SCS TCs during its positive phases, while ENSO is the most important in models. In this sense, the AMO and ENSO were the main sources of simulation biases of the models over the SCS.
The GW and internal variabilities explained a relatively small amount of the variances of the observed TCGF over the NA_N, which was probably due to the complicated impacts from synoptic-scale events or other climate variabilities there. In sharp contrast, the AMO had the largest contribution to the TCTD over the NA_N, reaching 16.5%, and it caused the largest simulation bias. For NA_T TC activities (Figs. 12e and 13e), the AMO and TNA SSTs contributed the most to the observed TCGF/TCTD, and more TCs formed during the positive phase of the AMO/TNA than the negative phase, indicating that the local Atlantic SSTs have the greatest impact on NA_T TC activities. Comparing these contributions between the MME and observation, the sources of the model bias over the NA_T were more likely from the longer-time variability, including the AMO, the IPO, and GW. Finally, for the WNP_E (Figs. 12h and 13h), the IPO and ENSO contributed the most to TC activities. The models could simulate the IPO, ENSO, and GW influences very well, resulting in better simulation performance over the WNP_E.
We also compared the relative importance model by model, as shown in Figs. S7 and S8. Over the ENP, the contributions of these longer-term SST climate indices to TCGF/TCTD in the observations and models are different or even opposite to each other. For example, GW has a positive contribution to the TCGF/TCTD in the observation, while a negative contribution exists in all of the models. There are similar results for the IPO and the AMO, although very few models can represent the sign of the observed contributions but with much smaller values relative to the observation. Over the NA_T, Caribbean, and WNP_E, the relative importance of these climate indices in the individual models is relatively consistent with the observations. In general, the results from most of the individual models are consistent with those from the MME, and none of them shows an absolute advantage in reproducing the relative importance analysis. It should be mentioned that in the observation, the IPO contribution to TCGF over the ENP is close to the AMO contribution, both showing an important impact on TCGF. However, it is not well represented by the MME based on the HighResMIP. It is also very interesting that the IPO/AMO impacts on TCGF are constrained from 160°E to 120°W in simulation, while the IPO/AMO impacts are almost uniform over the Pacific basin in the observation. It could be an important clue to explore the model biases of TCGF over the ENP in the future.
4. Conclusions and discussion
At present, most studies on future changes in TC activity under global warming scenarios mainly rely on high-resolution global climate numerical models, dynamical downscaling approaches, or statistical–dynamical hybrid models (Lee et al. 2020; Knutson et al. 2015; Emanuel 2013, 2021), but few have studied the confidence and uncertainty levels between historical simulations and observations. Based on the model data in the CMIP6 HighResMIP with high temporal and spatial resolutions, both the uncertainty and confidence levels of the TC activity simulation were evaluated in this study. Generally, the model had a good ability to simulate TC activities on the interannual time scale, but there were systematic biases on the interdecadal time scale and long-term trends. On the interdecadal time scale, the modeled and observed TCGF had opposite changes over the ENP and WNP_W, and significant differences over the WNP (including the SCS) and NA_N. For the linear trends, there were opposite changes between the modeled and observed TCGF over the ENP and SCS. These simulation biases will reduce the credibility of the model simulations due to internal variabilities (IPO and AMO) and trends (GW). Moreover, we found that the long-term variabilities of TCGF and TCTD over the WNP_E and NA_T in the models were highly consistent with observations, indicating that the interannual, interdecadal, and long-term linear trends induced by ENSO, IPO, AMO, and GW impacts are well captured by the models there.
We further quantitatively evaluated the relative contributions of anthropogenic external forcing (GW) and internal climate variabilities (IPO, AMO, ENSO, EIO, and TNA) to TC activities in the NH. Over the ENP, the TC frequency from observations and MME averages showed diametrically opposite changes on the interdecadal time scale and linear trends, which may have been related to the westward shift of climate patterns simulated by most climate models. This implies that even if the future ENP TC frequency predicted by multiple models is significantly reduced or increased, it may not be credible because the model’s historical simulations and observations show opposite changes over the ENP. For the WNP, the TCGF was mainly controlled by the IPO and ENSO over the WNP_E, and this response was well simulated by the models; however, the TC generation mechanism was more complicated over the WNP_W and the SCS, and the simulation bias was also relatively large. For the NA, the models have a good ability to simulate the interdecadal variabilities of TCGF in the southern part of the NA (especially the Caribbean), which is dominated by the AMO, but they poorly simulated the TC activity over the NA_N.
It should be noted that the simulated TC activity could be sensitive to the specific choice of tracking criteria (Murakami et al. 2015). In this study we used the TC detection algorithm developed by the GFDL. To confirm that our results in this study are robust and not sensitive to TC detection algorithm, we redid the analyses using two different tracking algorithms from Roberts et al. (2020a) including TRACK and TempestExtremes (Figs. S9 and S10). In general, the results based on the three detection algorithms are consistent with each other, although some small differences exist over the SCS and the WNP. Therefore, the results in this study are robust across the detection algorithms. It is interesting that when investigating hurricane-strength storms, when the TC should reach at least 33 m s−1 for one time (Fig. S11), we found that the correlation between observed and HadGEM-simulated ENP TCTD becomes significant (Fig. S11c), implying that the HadGEM model could be commendable for studying the intense TCs over the ENP.
It should be mentioned that TC genesis is characterized by strong regionalism. Only improving the basinwide spatial pattern of positive and negative phases of the IPO and AMO cannot simulate TC activity well. Improving representation of regional processes, especially over the ENP, is of great importance. For example, a cold tongue bias is common among most air–sea coupled models (C. Wang et al. 2014; Seager et al. 2019), which could lead to the distinct variability of ENP TCs in simulations from observation. However, the uncertainty of ENP TCs might not only originate from the oceanic biases, but also be influenced by atmospheric internal variability. This deserves our further studies in the future. It is also important to realize that much is to be done for a better understanding and representation of the relationship between TC activity and climate variability in the remote regions in climate models for better climate projections. Finally, historical simulations and observations of TCs have significant differences in different basins and regions. This strongly suggests that we need to be cautious when using numerical models to project future changes in TCs. Although it may not be fully true that a model that has issues simulating current climate is definitely worse in simulating the future climate, considering the confidence and uncertainty in present-day climate will definitely help reduce or constrain the uncertainty and improve the reliability when projecting future TC changes.
Acknowledgments.
We are grateful to three anonymous reviewers for their constructive comments and suggestions and to Prof. Yongqiang Yu for providing the FGOALS dataset. This work has been supported by National Key Research and Development Program on Monitoring Early Warning and Prevention of Major Natural Disaster (Grant 2019YFC1510004), National Natural Science Foundation of China (Grants 42088101, 42105022, and 42075015), and the Science and Technology Commission of Shanghai Municipality, China (Grant 20dz1200700).
Data availability statement.
The TC best-track data during 1950-2014 can be downloaded from https://www.ncdc.noaa.gov/ibtracs/. The reanalysis data for atmosphere are available from NOAA at https://psl.noaa.gov/data/20thC_Rean/ and from ECMWF at https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5. The SST dataset can be obtained from https://www.metoffice.gov.uk/hadobs/hadisst/. The HighResMIP data of the CMIP6 are openly available at https://data.ceda.ac.uk/badc/cmip6/data/CMIP6/HighResMIP. The FGOALS modeled data, including the TCGF and TCTD, can be downloaded from https://doi.org/10.6084/m9.figshare.13535249.v1.
REFERENCES
Ahn, M. S., and Coauthors, 2020: MJO propagation across the Maritime Continent: Are CMIP6 models better than CMIP5 models? Geophys. Res. Lett., 47, e2020GL087250, https://doi.org/10.1029/2020GL087250.
Arakane, S., and H.-H. Hsu, 2021: Tropical cyclone footprints in long-term mean state and multiscale climate variability in the western North Pacific as seen in the JRA-55 reanalysis. J. Climate, 34, 7443–7460, https://doi.org/10.1175/JCLI-D-20-0887.1.
Camargo, S. J., and A. H. Sobel, 2005: Western North Pacific tropical cyclone intensity and ENSO. J. Climate, 18, 2996–3006, https://doi.org/10.1175/JCLI3457.1.
Camargo, S. J., K. A. Emanuel, and A. H. Sobel, 2007: Use of a genesis potential index to diagnose ENSO effects on tropical cyclone genesis. J. Climate, 20, 4819–4834, https://doi.org/10.1175/JCLI4282.1.
Cao, X., S. Chen, G. Chen, W. Chen, and R. Wu, 2015: On the weakened relationship between spring Arctic Oscillation and following summer tropical cyclone frequency over the western North Pacific: A comparison between 1968–1986 and 1989–2007. Adv. Atmos. Sci., 32, 1319–1328, https://doi.org/10.1007/s00376-015-4256-y.
Chen, G., and R. Huang, 2009: Interannual variations in mixed Rossby–gravity waves and their impacts on tropical cyclogenesis over the western North Pacific. J. Climate, 22, 535–549, https://doi.org/10.1175/2008JCLI2221.1.
Chen, G., J. Ling, R. Zhang, Z. Xiao, and C. Li, 2022: The MJO from CMIP5 to CMIP6: Perspectives from tracking MJO precipitation. Geophys. Res. Lett., 49, e2021GL095241, https://doi.org/10.1029/2021GL095241.
Chan, J. C. L., 2000: Tropical cyclone activity over the western North Pacific associated with El Niño and La Niña events. J. Climate, 13, 2960–2972, https://doi.org/10.1175/1520-0442(2000)013<2960:TCAOTW>2.0.CO;2.
Chan, J. C. L., 2005: Interannual and interdecadal variations of tropical cyclone activity over the western North Pacific. Meteor. Atmos. Phys., 89, 143–152, https://doi.org/10.1007/s00703-005-0126-y.
Chen, T.-C., S.-P. Weng, N. Yamazaki, and S. Kiehne, 1998: Interannual variation in the tropical cyclone formation over the western North Pacific. Mon. Wea. Rev., 126, 1080–1090, https://doi.org/10.1175/1520-0493(1998)126<1080:IVITTC>2.0.CO;2.
Colbert, A. J., and B. J. Soden, 2012: Climatological variations in North Atlantic tropical cyclone tracks. J. Climate, 25, 657–673, https://doi.org/10.1175/JCLI-D-11-00034.1.
Compo, G. P., J. S. Whitaker, and P. Sardeshmukh, 2006: Feasibility of a 100-year reanalysis using only surface pressure data. Bull. Amer. Meteor. Soc., 87, 175–190, https://doi.org/10.1175/BAMS-87-2-175.
Compo, G. P., and Coauthors, 2011: The Twentieth Century Reanalysis Project. Quart. J. Roy. Meteor. Soc., 137, 1–28, https://doi.org/10.1002/qj.776.
Dosio, A., and Coauthors, 2021: Projected future daily characteristics of African precipitation based on global (CMIP5, CMIP6) and regional (CORDEX, CORDEX-CORE) climate models. Climate Dyn., 57, 3135–3158, https://doi.org/10.1007/s00382-021-05859-w.
Emanuel, K. A., 2005: Increasing destructiveness of tropical cyclones over the past 30 years. Nature, 436, 686–688, https://doi.org/10.1038/nature03906.
Emanuel, K. A., 2013: Downscaling CMIP5 climate models shows increased tropical cyclone activity over the 21st century. Proc. Natl. Acad. Sci. USA, 110, 12 219–12 224, https://doi.org/10.1073/pnas.1301293110.
Emanuel, K. A., 2021: Response of global tropical cyclone activity to increasing CO2: Results from downscaling CMIP6 models. J. Climate, 34, 57–70, https://doi.org/10.1175/JCLI-D-20-0367.1.
Emanuel, K. A., and D. S. Nolan, 2004: Tropical cyclone activity and global climate. 26th Conf. on Hurricanes and Tropical Meteorology, Miami, FL, Amer. Meteor. Soc., 240–241
Enfield, D. B., A. M. Mestas-Nunez, and P. J. Trimble, 2001: The Atlantic multidecadal oscillation and its relationship to rainfall and river flows in the continental U.S. Geophys. Res. Lett., 28, 2077–2080, https://doi.org/10.1029/2000GL012745.
Gao, S., L. Zhu, W. Zhang, and Z. Chen, 2018: Strong modulation of the Pacific meridional mode on the occurrence of intense tropical cyclones over the western North Pacific. J. Climate, 31, 7739–7749, https://doi.org/10.1175/JCLI-D-17-0833.1.
Groemping, U., 2006: Relative importance for linear regression in R: The package relaimpo. J. Stat. Softw., 17, 1–27, http://hdl.handle.net/10.18637/jss.v017.i01.
Guo, Y.-P., and Z.-M. Tan, 2018a: Westward migration of tropical cyclone rapid-intensification over the northwestern Pacific during short duration El Niño. Nat. Commun., 9, 1507, https://doi.org/10.1038/s41467-018-03945-y.
Guo, Y.-P., and Z.-M. Tan, 2018b: Impacts of the boreal spring Indo-Pacific warm pool Hadley circulation on tropical cyclone activity over the western North Pacific. J. Climate, 31, 1361–1375, https://doi.org/10.1175/JCLI-D-17-0422.1.
Haarsma, R. J., and Coauthors, 2016: High Resolution Model Intercomparison Project (HighResMIP v1.0) for CMIP6. Geosci. Model Dev., 9, 4185–4208, https://doi.org/10.5194/gmd-9-4185-2016.
Ham, Y.-G., J.-S. Kug, J.-Y. Park, and F.-F. Jin, 2013: Sea surface temperature in the north tropical Atlantic as a trigger for El Niño/Southern Oscillation events. Nat. Geosci., 6, 112–116, https://doi.org/10.1038/ngeo1686.
Heede, U. K., and A. V. Fedorov, 2021: Eastern equatorial Pacific warming delayed by aerosols and thermostat response to CO2 increase. Nat. Climate Change, 11, 696–703, https://doi.org/10.1038/s41558-021-01101-x.
Henley, B. J., J. Gergis, D. J. Karoly, S. Power, J. Kennedy, and C. K. Folland, 2015: A tripole index for the interdecadal Pacific oscillation. Climate Dyn., 45, 3077–3090, https://doi.org/10.1007/s00382-015-2525-1.
Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10.1002/qj.3803.
Hong, C.-C., M.-Y. Lee, H.-H. Hsu, and W.-L. Tseng, 2018: Distinct influences of the ENSO-like and PMM-like SST anomalies on the mean TC genesis location in the western North Pacific: The 2015 summer as an extreme example. J. Climate, 31, 3049–3059, https://doi.org/10.1175/JCLI-D-17-0504.1.
Hsu, P.-C., and T. Li, 2011: Interactions between boreal summer intraseasonal oscillations and synoptic-scale disturbances over the western North Pacific. Part II: Apparent heat and moisture sources and eddy momentum transport. J. Climate, 24, 942–961, https://doi.org/10.1175/2010JCLI3834.1.
Huangfu, J., R. Huang, and W. Chen, 2017: Relationship between the South China Sea summer monsoon onset and tropical cyclone genesis over the western North Pacific. Int. J. Climatol., 37, 5206–5210, https://doi.org/10.1002/joc.5141.
Huo, L., P. Guo, S. N. Hameed, and D. Jin, 2015: The role of tropical Atlantic SST anomalies in modulating western North Pacific tropical cyclone genesis. Geophys. Res. Lett., 42, 2378–2384, https://doi.org/10.1002/2015GL063184.
Jiang, X., and Coauthors, 2015: Vertical structure and physical processes of the Madden–Julian oscillation: Exploring key model physics in climate simulations. J. Geophys. Res. Atmos., 120, 4718–4748, https://doi.org/10.1002/2014JD022375.
Jiang, X., M. Zhao, E. D. Maloney, and D. E. Waliser, 2016: Convective moisture adjustment time scale as a key factor in regulating model amplitude of the Madden–Julian oscillation. Geophys. Res. Lett., 43, 10 412–10 419, https://doi.org/10.1002/2016GL070898.
Jin, F.-F., J. Boucharel, and I.-I. Lin, 2014: Eastern Pacific tropical cyclones intensified by El Niño delivery of subsurface ocean heat. Nature, 516, 82–85, https://doi.org/10.1038/nature13958.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, https://doi.org/10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Klingaman, N. P., X. Jiang, P. K. Xavier, J. Petch, D. Waliser, and S. J. Woolnough, 2015: Vertical structure and physical processes of the Madden–Julian oscillation: Synthesis and summary. J. Geophys. Res. Atmos., 120, 4671–4689, https://doi.org/10.1002/2015JD023196.
Klotzbach, P. J., K. M. Wood, C. J. Schreck III, S. G. Bowen, C. M. Patricola, and M. M. Bell, 2022: Trends in global tropical cyclone activity: 1990–2021. Geophys. Res. Lett., 49, e2021GL095774, https://doi.org/10.1029/2021GL095774.
Knapp, K. R., M. C. Kruk, D. H. Levinson, H. J. Diamond, and C. J. Neumann, 2010: The International Best Track Archive for Climate Stewardship (IBTrACS). Bull. Amer. Meteor. Soc., 91, 363–376, https://doi.org/10.1175/2009BAMS2755.1.
Knapp, K. R., H. J. Diamond, J. P. Kossin, M. C. Kruk, and C. J. Schreck, 2018: International Best Track Archive for Climate Stewardship (IBTrACS) Project, version 4. NOAA/National Centers for Environmental Information, accessed 1 September 2019, https://doi.org/10.25921/82ty-9e16.
Knutson, T., J. J. Sirutis, S. T. Garner, G. A. Vecchi, and I. M. Held, 2008: Simulated reduction in Atlantic hurricane frequency under twenty-first-century warming conditions. Nat. Geosci., 1, 359–364, https://doi.org/10.1038/ngeo202.
Knutson, T., and Coauthors, 2010: Tropical cyclones and climate change. Nat. Geosci., 3, 157–163, https://doi.org/10.1038/ngeo779.
Knutson, T., J. J. Sirutis, M. Zhao, R. E. Tuleya, M. Bender, G. A. Vecchi, G. Villarini, and D. Chavas, 2015: Global projections of intense tropical cyclone activity for the late twenty-first century from dynamical downscaling of CMIP5/RCP4.5 scenarios. J. Climate, 28, 7203–7224, https://doi.org/10.1175/JCLI-D-15-0129.1.
Knutson, T., and Coauthors, 2019: Tropical cyclones and climate change assessment: Part I: Detection and attribution. Bull. Amer. Meteor. Soc., 100, 1987–2007, https://doi.org/10.1175/BAMS-D-18-0189.1.
Knutson, T., and Coauthors, 2020: Tropical cyclones and climate change assessment: Part II: Projected response to anthropogenic warming. Bull. Amer. Meteor. Soc., 101, E303–E322, https://doi.org/10.1175/BAMS-D-18-0194.1.
Kodama, C., and Coauthors, 2015: A 20-year climatology of a NICAM AMIP-type simulation. J. Meteor. Soc. Japan, 93, 393–424, https://doi.org/10.2151/jmsj.2015-024.
Lee, C.-Y., S. J. Camargo, A. H. Sobel, and M. K. Tippett, 2020: Statistical–dynamical downscaling projections of tropical cyclone activity in a warming climate: Two diverging genesis scenarios. J. Climate, 33, 4815–4834, https://doi.org/10.1175/JCLI-D-19-0452.1.
Li, S., W. Mei, and S.-P. Xie, 2022: Effects of tropical sea surface temperature variability on Northern Hemisphere tropical cyclone genesis. J. Climate, 35, 4719–4739, https://doi.org/10.1175/JCLI-D-21-0084.1.
Li, W., L. Li, and Y. Deng, 2015: Impact of the interdecadal Pacific oscillation on tropical cyclone activity in the North Atlantic and eastern North Pacific. Sci. Rep., 5, 12 358, https://doi.org/10.1038/srep12358.
Lin, I.-I., and J. C. L. Chan, 2015: Recent decrease in typhoon destructive potential and global warming implications. Nat. Commun., 6, 7182, https://doi.org/10.1038/ncomms8182.
Lindeman, R. H., P. F. Merenda, and R. Z. Gold, 1980: Introduction to Bivariate and Multivariate Analysis. Scott Foresman, 444 pp.
Lu, M., and R. Xiong, 2019: Spatiotemporal profiling of tropical cyclones genesis and favorable environmental conditions in the western Pacific basin. Geophys. Res. Lett., 46, 11 548–11 558, https://doi.org/10.1029/2019GL084995.
Miura, H., M. Satoh, T. Nasuno, A. T. Noda, and K. Oouchi, 2007: A Madden–Julian oscillation event realistically simulated by a global cloud-resolving model. Science, 318, 1763–1765, https://doi.org/10.1126/science.1148443.
Miyakawa, T., and Coauthors, 2014: Madden–Julian oscillation prediction skill of a new-generation global model demonstrated using a supercomputer. Nat. Commun., 5, 3769, https://doi.org/10.1038/ncomms4769.
Miyamoto, Y., M. Satoh, H. Tomita, K. Oouchi, Y. Yamada, C. Kodama, and J. Kinter, 2014: Gradient wind balance in tropical cyclones in high-resolution global experiments. Mon. Wea. Rev., 142, 1908–1926, https://doi.org/10.1175/MWR-D-13-00115.1.
Murakami, H., and B. Wang, 2010: Future change of North Atlantic tropical cyclone tracks: Projection by a 20-km-mesh global atmospheric model. J. Climate, 23, 2699–2721, https://doi.org/10.1175/2010JCLI3338.1.
Murakami, H., and Coauthors, 2015: Simulation and prediction of category 4 and 5 hurricanes in the high-resolution GFDL HiFLOR coupled climate model. J. Climate, 28, 9058–9079, https://doi.org/10.1175/JCLI-D-15-0216.1.
Murakami, H., T. L. Delworth, W. F. Cooke, M. Zhao, B. Xiang, and P.-C. Hsu, 2020: Detected climatic change in global distribution of tropical cyclones. Proc. Natl. Acad. Sci. USA, 117, 10 706–10 714, https://doi.org/10.1073/pnas.1922500117.
Oouchi, K., A. T. Noda, M. Satoh, B. Wang, S.-P. Xie, H. G. Takahashi, and T. Yasunari, 2009: Asian summer monsoon simulated by a global cloud-system-resolving model: Diurnal to intra-seasonal variability. Geophys. Res. Lett., 36, L11815, https://doi.org/10.1029/2009GL038271.
Patricola, C. M., P. Chang, and R. Saravanan, 2016: Degree of simulated suppression of Atlantic tropical cyclones modulated by flavour of El Niño. Nat. Geosci., 9, 155–160, https://doi.org/10.1038/ngeo2624.
Rajendran, K., A. Kitoh, J. Srinivasan, R. Mizuta, and R. Krishnan, 2012: Monsoon circulation interaction with Western Ghats orography under changing climate: Projection by a 20-km mesh AGCM. Theor. Appl. Climatol., 110, 555–571, https://doi.org/10.1007/s00704-012-0690-2.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res., 108, 4407, https://doi.org/10.1029/2002JD002670.
Roberts, M. J., and Coauthors, 2020a: Impact of model resolution on tropical cyclone simulation using the HighResMIP–PRIMAVERA multimodel ensemble. J. Climate, 33, 2557–2583, https://doi.org/10.1175/JCLI-D-19-0639.1.
Roberts, M. J., and Coauthors, 2020b: Projected future changes in tropical cyclones using the CMIP6 HighResMIP multimodel ensemble. Geophys. Res. Lett., 47, e2020GL088662, https://doi.org/10.1029/2020GL088662.
Satoh, M., and Coauthors, 2012: The intra-seasonal oscillation and its control of tropical cyclones simulated by high-resolution global atmospheric models. Climate Dyn., 39, 2185–2206, https://doi.org/10.1007/s00382-011-1235-6.
Seager, R., M. Cane, N. Henderson, D.-E. Lee, R. Abernathey, and H. Zhang, 2019: Strengthening tropical Pacific zonal sea surface temperature gradient consistent with rising greenhouse gases. Nat. Climate Change, 9, 517–522, https://doi.org/10.1038/s41558-019-0505-x.
Sugi, M., H. Murakami, and J. Yoshimura, 2012: On the mechanism of tropical cyclone frequency changes due to global warming. J. Meteor. Soc. Japan, 90A, 397–408, https://doi.org/10.2151/jmsj.2012-A24.
Sugi, M., K. Yoshida, and H. Murakami, 2015: More tropical cyclones in a cooler climate? Geophys. Res. Lett., 42, 6780–6784, https://doi.org/10.1002/2015GL064929.
Vimont, D. J., and J. P. Kossin, 2007: The Atlantic meridional mode and hurricane activity. Geophys. Res. Lett., 34, L07709, https://doi.org/10.1029/2007GL029683.
Walsh, K., and Coauthors, 2016: Tropical cyclones and climate change. Wiley Interdiscip. Rev.: Climate Change, 7, 65–89, https://doi.org/10.1002/wcc.371.
Wang, B., and J. C. L. Chan, 2002: How strong ENSO events affect tropical storm activity over the western North Pacific. J. Climate, 15, 1643–1658, https://doi.org/10.1175/1520-0442(2002)015<1643:HSEEAT>2.0.CO;2.
Wang, B., and H. Murakami, 2020: Dynamic genesis potential index for diagnosing present-day and future global tropical cyclone genesis. Environ. Res. Lett., 15, 114008, https://doi.org/10.1088/1748-9326/abbb01.
Wang, C., L. Zhang, S.-K. Lee, L. Wu, and C. R. Mechoso, 2014: A global perspective on CMIP5 climate model biases. Nat. Climate Change, 4, 201–205, https://doi.org/10.1038/nclimate2118.
Wang, H., and Coauthors, 2014: How well do global climate models simulate the variability of Atlantic tropical cyclones associated with ENSO? J. Climate, 27, 5673–5692, https://doi.org/10.1175/JCLI-D-13-00625.1.
Watanabe, M., J.-L. Dufresne, Y. Kosaka, T. Mauritsen, and H. Tatebe, 2021: Enhanced warming constrained by past trends in equatorial Pacific sea surface temperature gradient. Nat. Climate Change, 11, 33–37, https://doi.org/10.1038/s41558-020-00933-3.
Webster, P. J., 2005: Changes in tropical cyclone number, duration, and intensity in a warming environment. Science, 309, 1844–1846, https://doi.org/10.1126/science.1116448.
Wu, Z., and N. E. Huang, 2009: Ensemble empirical mode decomposition: A noise-assisted data analysis method. Adv. Adapt. Data Anal., 1, 1–41, https://doi.org/10.1142/S1793536909000047.