Understanding uncertainties in projections of western North Pacific tropical cyclogenesis

Reliable projections of tropical cyclone (TC) activities in the western North Pacific (WNP) are crucial for climate policy-making in densely-populated coastal Asia. Existing projections, however, exhibit considerable uncertainties with unclear sources. Here, based on future projections by the latest Coupled Model Intercomparison Project Phase 6 climate models, we identify a new and prevailing source of uncertainty arising from different TC identification schemes. Notable differences in projections of detected TCs and empirical genesis potential indices are found to be caused by inconsistent changes in dynamic and thermodynamic environmental factors affecting TC formations. While model uncertainty holds the secondary importance, we show large potential in reducing it through improved model simulations of present-day TC characteristics. Internal variability noticeably impacts near-term projections of the WNP tropical cyclogenesis, while the relative contribution of scenario uncertainty remains small. Our findings provide valuable insights into model development and TC projections, thereby aiding in adaptation decisions.


Introduction
The western North Pacific (WNP) is the most active oceanic basin for tropical cyclone (TC) activity, witnessing around one-third of global TCs annually.These TCs, as one of the most devastating natural disasters on Earth, pose a threat to large populations in Asia and western Pacific countries, and the vulnerable ecosystems in the Pacific islands (Peduzzi et al 2012, Widlansky et al 2019).Consequently, reliable projections of the WNP TC activities are crucial for risk adaptation and mitigation decisions.
Previous studies have made much effort to project future changes in TCs using various climate models.Nevertheless, current projections still show large uncertainties in both the signs and the magnitudes of tropical cyclogenesis changes (Knutson et al 2020).For example, under a 2 • C anthropogenic warming, the WNP TC genesis frequency (TCGF) detected from climate models is projected to decrease by ∼10%, but with the 10th to 90th percentile range of about −26% to +11% among different modeling studies (Cha et al 2020).Moreover, to understand changes in TC frequency, TC genesis potential indices (GPIs) have been developed based on favorable large-scale environmental factors controlling tropical cyclogenesis (Gray 1979, Emanuel and Nolan 2004, Emanuel 2010, Tippett et al 2011, Menkes et al 2012).These GPIs are widely used in climate studies as proxies for TC formation to represent variabilities in TC activities (Camargo 2013, Choi et al 2015, Takahashi et al 2017, Yan et al 2019, Cao et al 2021).However, in future projections, while most of these empirical GPIs suggest more favorable conditions for tropical cyclogenesis (Camargo 2013, Camargo et al 2014, Wehner et al 2015), the majority of modeling studies based on direct TC-detection schemes project a decrease in WNP TCGF under global warming (Tory et al 2013c, Yoshida et al 2017, Roberts et al 2020, Tang et al 2021, Murakami and Wang 2022), albeit with a few exceptions (Emanuel 2013, 2021, Bhatia et al 2018).The large inter-model uncertainty in future changes, along with the contradiction between projections based on GPIs and the detected TCGF, has challenged the usefulness of projection information in climate adaptation strategies and resiliency efforts for densely-populated coastal areas (Seneviratne 2021).
Understanding sources of projection uncertainties improves scientific interpretation of future changes and provides useful information for model development (Deser et al 2012, Hall et al 2019, Huang et al 2020).Usually, uncertainties in climate projections are classified into three main sources: scenario uncertainty due to different future radiative forcing, model uncertainty arising from structure, parameter and resolution choices, and internal variability intrinsic to the climate system (Hawkins and Sutton 2009).For projections of tropical cyclogenesis, another uncertainty source arising from different TC identification schemes, i.e. different projection results derived from detected TCGF and GPIs, respectively, should also be taken into consideration.While some previous studies have examined TC projection uncertainties arising from different cumulus convection schemes, changes of sea surface temperature patterns, anthropogenic greenhouse warming, etc (Murakami et al 2012, Chu et al 2020, Zhao et al 2020), a systematic quantification of different sources of uncertainty in the WNP tropical cyclogenesis is still needed.Detailed analyses of contributions of different uncertainty sources will improve TC projections by focusing on the key and reducible uncertainty in different projection periods (Chen et al 2020, Huang et al 2020, Dong et al 2021).
Decision-makers need climate change information based on projections from the latest Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-model ensemble covering near-term, middleterm and long-term future under multiple new scenarios (O'Neill et al 2016).Here, we investigate future changes and uncertainties in projections of the WNP tropical cyclogenesis based on 20 CMIP6 models (supplementary table S1; Eyring et al 2016).To detect TCs in the CMIP6 model simulations, we apply the Okubo-Weiss-Zeta (OWZ) TC detection scheme (Tory et al 2013a) which has been so far validated in reanalysis datasets and used in the CMIP3/5 models (Tory et al 2013b, Bell et al 2018, 2019, Chand et al 2022).The OWZ scheme shows advantages in the low-to-medium resolution climate models as it identifies the immediate large-scale conditions in which TCs form instead of fully resolving mesoscale TC structures (McBride and Zehr 1981, Davidson et al 1990, Dunkerton et al 2009).Furthermore, we also use two GPIs to project tropical cyclogenesis changes, including one of the most widely-used GPI developed by Emanuel (2010;hereafter 'EGPI') and a new dynamic GPI (hereafter 'DGPI') developed by Wang and Murakami (2020).By using these three different schemes as a proxy of TC formations, respectively, we quantify the projected changes and uncertainties in the WNP tropical cyclogenesis.Through partitioning different uncertainty sources, we find the large contributions of scheme uncertainty and model uncertainty.We also reveal the large potential to reduce uncertainties in the projections of tropical cyclogenesis.To evaluate performance of the OWZ TC detection scheme in reproducing the observed TCGF, we use hourly data from the latest ERA5 reanalysis dataset produced by the European Centre for Medium-Range Weather Forecast, covering the period of 1979 to present with a horizontal resolution of 0.25 • latitude × 0.25 • longitude (Hersbach et al 2020).

Model simulations
We use both daily and monthly out puts from 20 CMIP6 models (supplementary table S1).The historical simulations are integrated from 1850 to 2014 driven by observed changes in natural and anthropogenic radiative forcing.For future projections performed from 2015 to 2100, the CMIP6 designed new scenarios (O'Neill et al 2016) combining the Shared Socioeconomic Pathways (SSPs) and the Representative Concentration Pathways.In this study, we analyze the outputs under the SSP2-4.5 and the SSP5-8.5 future scenarios.The first available realization for each model and each scenario is used here for an equal weight of the models.

Statistical analysis
All results from observations, reanalyses and model simulations are interpolated to a uniform grid of 2.5 • latitude × 2.5 • longitude for comparison.We analyze the WNP tropical cyclogenesis during the typhoon season from May to November (MJJASON) in each year.To focus on interdecadal variability, the anomalies of the detected TCGF and the GPIs are smoothed with a 20 year running mean.Considering the limited observations of TCs during the presatellite era, we focus our analyses starting from 1980.The historical period of 1980-2014 is defined as a baseline time period.Three specific future periods in model projections are provided, including the near-term (2021-2040), middle-term (2041-2060) and long-term (2080-2099).Additionally, we employ the Taylor skill score (TSS; Taylor 2001) to evaluate the performance of the CMIP6 models in characterizing the observed climatological TCGF, which expressed as: where R denotes the pattern correlation coefficient (PCC) between the model simulation and the observation.R 0 is the maximum potentially realizable correlation (taken as 0.99).σ denotes the standard deviation of the model simulation normalized by the observational standard deviation.The TSS combines the PCC and the normalized standard deviation, providing a more comprehensive evaluation of the model performance.

TC detection scheme
We apply an innovative TC detection scheme developed by Tory et al (2013a  Davidson et al 1990, Dunkerton et al 2009).On the other hand, the algorithm can be conveniently applied to multiple CMIP6 models without any further tuning or adjustment of detection thresholds to accommodate different models (Bell et al 2019).
Supplementary text S1 provides more details of its algorithm.
Based on the OWZ detection scheme, the TCs in the ERA5 well reproduces the present-day climatology and variability of the observed TCGF in the WNP (supplementary figures S1(a) and (b)).The detected TCs derived from the CMIP6 multimodel mean also perform well in capturing the climatological spatial distributions of the observed TCGF with a PCC of 0.87 (supplementary figure S1(c)).
To understand changes in the detected TCGF, we also investigate the projected changes in the environmental parameters used in the OWZ detection scheme (supplementary text S1).The dynamic environmental parameters include the OWZ variables at both the 850 hPa and the 500 hPa, and the vertical wind shear between the 850 and 200 hPa.The thermodynamic parameters include the relative humidity at both the 950 hPa and the 700 hPa, and the specific humidity at the 950 hPa.

TC genesis potential index 2.5.1. EGPI
The widely-used EGPI developed by Emanuel (2010) is expressed as ( where v s is the magnitude of vertical wind shear between 850 hPa and 200 hPa, η represents the absolute vorticity at 850 hPa, v pot is the potential intensity representing the theoretical upper bound of TC intensity under a given set of atmospheric and oceanic conditions (Bister and Emanuel 2002), which defined as where C k and C D are the exchange coefficient for enthalpy and the drag coefficient, respectively, T s is the sea surface temperature (SST), T 0 is the outflow temperature, h * o and h * are the saturation moist static energy at the sea surface and saturation moist static energy of air above the boundary layer, respectively.
The nondimensional parameter χ in equation ( 2) is a measure of the moist entropy deficit of the middle troposphere defined as where s b , s m and s * 0 represent the entropies of the boundary layer and middle troposphere, and the saturation moist entropy of the sea surface, respectively (Emanuel et al 2008).A larger value of χ means that the middle troposphere becomes drier (Emanuel 2010).In equation ( 2), we use the notations Vort and V s to denote the two dynamic environmental terms, and use the notations X and PI to denote the two thermodynamic terms of the EGPI for compactness.

DGPI
Recently, Wang and Murakami ( 2020) revisited the TC genesis potential index by objectively comparing the role of environmental factors that affect tropical cyclogenesis and developed a new dynamic GPI (hereafter 'DGPI') based on four leading dynamic factors that stably control TCGF from present to future.The DGPI is expressed as where v s is the magnitude of vertical wind shear between 850 hPa and 200 hPa; ∂ y u is the meridional gradient of zonal wind at 500 hPa; ω is the vertical pressure velocity at 500 hPa; and η represents the absolute vorticity at 850 hPa.In particular, the DGPI shows high skills in characterizing the observed climatology and interannual-to-interdecadal variability of TC frequency in the WNP (Wang et al 2022, Zhan et al 2022).The dynamic environmental conditions are documented in literatures as key to TC development in the WNP basin where the sea surface temperature keeps warm throughout the year and the thermodynamic criteria are easy to be satisfied (Chan 2009, Fu et al 2012, Murakami et al 2013).
Besides, the influences of the thermodynamic environmental factors are not unimportant here (Wang and Murakami 2020, Murakami and Wang 2022), because they are implicitly included in the calculation of DGPI by incorporating the vertical velocity term ω and the vertical wind shear term v s which are significantly correlated with midlevel relative humidity, sea surface temperature and maximum potential intensity v pot .Here, we also use the notations V s , U y , W and Vort to denote the four terms in equation ( 5) for compactness.

Diagnose contributions of dynamic and thermodynamic environmental factors
To quantitatively diagnose contributions of the dynamic and thermodynamic environmental factors to future changes in the GPIs, we separate GPI = GPI + ∆GPI, where the GPI denotes the EGPI or DGPI.The two terms on the right side indicate climatological mean and departure, respectively.Following Murakami (2022), the equation of DGPI is linearized as where res denotes residual term including the nonlinear interaction among the factors.We define the anomalous change relative to the climatology as Then, the equation is simplified as where V ′ s , U ′ y , W ′ and Vort ′ are anomalies of the four dynamic terms relative to their climatology, respectively.Sum D ′ denotes the sum of the anomalous dynamic terms.
Similarly, for the anomalous EGPI ′ = ∆EGPI EGPI , the equation ( 2) is linearized as where Sum E ′ denotes the sum of the four contributing terms on the right side of the equation.

Quantification of different sources of projection uncertainty
To quantify projection uncertainty arising from different sources, we first partition the total uncertainty T c (t) of each TC identification scheme c into its own internal variability uncertainty (I c ), model uncertainty (M c (t)) and scenario uncertainty (S c (t)) following Hawkins and Sutton (2009).Here, c denotes the TCGF detected using the OWZ scheme, the EGPI or the DGPI, respectively.Then, the overall total uncertainty in projections of the tropical cyclogenesis (T (t)) is further separated into the overall internal variability uncertainty (I), model uncertainty (M (t)), scenario uncertainty (S (t)) and scheme uncertainty (C (t)) in this study.See supplementary text S2 and figure S2 for more technical details.

Large uncertainties in projections of the WNP tropical cyclogenesis
We first examine the projected changes in the WNP tropical cyclogenesis during the typhoon season from May to November (MJJASON) over the period of 2015-2099.A decreasing trend in the detected TCGF and the DGPI is seen over the WNP's core genesis area (5 • -23 • N, 110 • E-155 • ) in the multi-modelmean (MMM) of the CMIP6 models, while a notable increasing trend is shown in the projected EGPI (figure 1(b)).To quantify these changes, we calculate the means and the 10th to 90th percentile range simulated by the 20 CMIP6 models in the near-term (2021-2040), middle-term (2041-2060), and longterm (2080-2099) projections under two different future scenarios (figure 1(c)).In the near term, the detected TCGF decreases by about −2.0% (−11.3% to +10.7%) and −3.9% (−12.6% to +6.3%) under the SSP2-4.5 and the SSP5-8.5 scenarios, respectively, with respect to the 1980-2014 baseline period.
To explore sources of the projection uncertainties, we partition the evolutions of total uncertainty into the four components, i.e. internal variability, model uncertainty, scenario uncertainty and scheme uncertainty (supplementary text S2 and figure S2).We calculate fractional contributions of each source to the total uncertainty for different projection lead times to clarify their relative importance (figure 2(a)).Overall, both the scheme uncertainty and the model uncertainty play an important role throughout the century.Quantitatively, the scheme uncertainty is the dominant uncertainty source for projections of the WNP tropical cyclogenesis at all lead times, explaining about 51%, 54% and 61% of the total uncertainty in near-, middle-and long-term projections, respectively.The model uncertainty shows the secondary large contribution of about 40%, 41% and 37%, respectively, of the total variance from the nearterm to the long-term future.At short projection lead times, the role of internal variability is also important and non-negligible, contributing to about 9% of the total uncertainty in the near-term.It suggests that key internal variability affecting tropical cyclogenesis need to be analyzed and considered in the next 10-30 years to ensure a more robust projection of TC activities (Huang et al 2020).Compared with other uncertainty sources, the contribution of scenario uncertainty is relatively small throughout the 21st century.
A credible projection for decision making depends, to a large extent, on the Noise-to-Signal ratio which measures how large the projection uncertainties are compared to the expected changes.Thus, we examine the Noise-to-Signal ratios of the projected WNP tropical cyclogenesis induced by the total uncertainty and each of the uncertainty sources (figure 2(b)).It is evident that the total Noise-to-Signal ratio is always much greater than 1.0, indicating the changing signals of TCs are eclipsed by noises.Actually, the large noises induced by the model uncertainties or scheme uncertainties alone is able to confuse the projected changes.The Noise-to-Signal ratio related to internal climate variability is remarkable in the near term, but it continually declines with increasing lead times and falls to ∼0.56 by the end of the century.For the scenario uncertainty, although the corresponding Noise-to-Signal ratio gradually increases with lead times, its contribution to projection uncertainty is relatively weak throughout the time period.

Understanding of the large scheme uncertainty
To understand the dominant role of scheme uncertainty, we compare the CMIP6 MMM projected trends in the detected TCGF and the GPIs during the period of 1980-2099 under the SSP5-8.5 scenario.Spatially, although all the three TC identification schemes indicate a poleward tendency of tropical cyclogenesis in the WNP basin under global warming, as suggested in previous studies (Sharmila and Walsh 2018, Sun et al 2019, Knutson et al 2020), they show different trends over the present-day core genesis area (figures 3(a)-(c)).The different signed responses of the detected TCGF, the DGPI and the EGPI contribute to the large scheme uncertainty.Overall, the detected TCGF exhibits a decreasing trend in the WNP's core genesis area (figure 3(a)).We investigate the projected changes in the dynamic and thermodynamic parameters used in the OWZ TC detection scheme to understand the decreases in TCGF (figure 3(d)).The decreasing trends in the OWZ parameter at both the 850 and 500 hPa and the increasing trend in vertical wind shear indicate an unfavorable dynamic environments to support TC formations.For the thermodynamic parameters, while the decreasing relative humidity in the lower-middle troposphere will also inhibit tropical cyclogenesis, the increasing specific humidity at the 950 hPa under global warming tends to enhance TC formation in the  c), orange and dark orange denote the TCGF detected in the CMIP6 models using the OWZ scheme under the SSP2-4.5 and SSP5-8.5 scenarios, respectively.Green and dark green denote the DGPI scheme under the SSP2-4.5 and SSP5-8.5 scenarios, respectively.Blue and dark blue denote the EGPI scheme under the SSP2-4.5 and SSP5-8.5 scenarios, respectively.
Figure 2. Sources of projection uncertainty in the WNP tropical cyclogenesis.(a) Evolution of fractional contributions of different uncertainty sources to total variance of the projected changes in the WNP tropical cyclogenesis.Units: %.The total uncertainty is separated into four different components, including internal variability (orange), scenario uncertainty (green), model uncertainty (blue) and scheme uncertainty (purple) for the period of 2015-2099. (b) The Noise-to-Signal ratios of the projected changes in the WNP tropical cyclogenesis during the period 2015-2100, represented as the standard deviation of projected changes in the CMIP6 models divided by the multi-scheme multi-model multi-scenario mean projection.Black, orange, blue, green and purple lines denote the 'noise' is considered as the contribution of the total uncertainty, internal variability, model uncertainty, scenario uncertainty and scheme uncertainty, respectively.(c) Potential Noise-to-Signal ratios in projections of the WNP tropical cyclogenesis by assuming that the scheme uncertainty (purple dashed line), model uncertainty (blue dashed line), and model uncertainty dashed line) could be improved to be 'perfect' .Units: MMM projected trends in the dynamic and thermodynamic parameters used in the OWZ TC detection scheme during 1980-2099 under the SSP5-8.5 scenario, including OWZ variables at both the 850 and 500 hPa (OWZ850 and OWZ500; units: 10 6 s −1 century −1 ), vertical wind shear between 850 and 200 hPa (vs; units: m s −1 century −1 ), relative humidity at both the 950 and 700 hPa (rh950 and rh700; units: % century −1 ), specific humidity at 950 hPa (sph950; units: 5 × 10 2 kg kg −1 century −1 ).Dots in (a) to (d) denote areas where less than 3/4 (15 out of the 20) models agree on the sign of change in the CMIP6 MMM.

WNP (figure 3(d))
. The results imply that environmental factors affecting tropical cyclogenesis in the WNP will change inconsistently under global warming, complicating future projections of the TCGF.The unfavorable dynamic environmental conditions for TC formation are also indicated by the projections of DGPI (figure 3(b)).Quantitatively, changes in the 500 hPa vertical velocity term (W ′ ), caused by the anomalous descending motion in the southwestern WNP (figure 4(a)), dominates the decreasing trend of DGPI from near-term to longterm (figure 4(b)).The decreasing trend of the V ′ s term and the Vort ′ term, induced by enhanced vertical wind shear and decreased 850 hPa absolute vorticity, respectively, also suppress tropical cyclogenesis in the WNP, while the strengthened meridional gradient of 500 hPa zonal wind (U ′ y ) tends to slightly offset the decreasing trend (figures 4(a) and (b)).
The projected EGPI, however, shows an increasing trend in the WNP's core genesis area (figure 3(c)), which implies a more favorable TC genesis potential there.The increasing EGPI is dominated by the two thermodynamic environmental factors PI and χ , especially the PI ′ term due to its sensitivity to the warming sea surface temperature (figures 4(a) and (c)).The above results indicate that the while the overall dynamic environmental fields tend to decrease the WNP TCGF, most of the thermodynamic factors affecting tropical cyclogenesis prefer to increase TCs in a warmer climate.As expected, the results are also confirmed in future projections under the SSP2-4.5 scenario (supplementary figure S3).The disparity in projected changes of the related dynamic and thermodynamic environmental factors challenges a reliable projection of future TC changes.under the SSP5-8.5 scenario, including vertical wind shear (vs; units: m s −1 century −1 ), meridional gradient of 500 hPa zonal wind (∂yu; units: 10 6 s −1 century −1 ), vertical velocity at 500 hPa (ω; units: 10 2 Pa s −1 century −1 ), absolute vorticity at 850 hPa (η; units: 10 6 s −1 century −1 ), potential intensity (PI; units: m s −1 century −1 ) and moist entropy deficit of the middle troposphere (χ ; units: 10 3 century −1 ).Dots in denote areas where less than 3/4 (15 out of the 20) models agree on the sign of change in the CMIP6 MMM.CMIP6 MMM projected changes in the (b) DGPI and (c).EGPI under the SSP5-8.5 scenario, along with the contributing dynamic and thermodynamic terms, in near-term (left; light blue and pink bars), middle-term (middle; blue and red bars) and long-term (right; dark blue and dark red bars) relative to the 1980-2014 baseline period.Units: %.In (b) and (c), the vertical lines denote the 10th to 90th percentile range among different CMIP6 models.
In figure 2(c), we calculate the 'potential Noise-to-Signal' ratio for a 'perfect' scheme to illustrate the potential gains from a better representation of modeling TC formations (supplementary text S2).We notice a considerable decrease in the Noise-to-Signal ratio if scheme uncertainty is reduced.It suggests that the spread in TC projections could be evidently narrowed with improved observations and process understanding for drivers of climate changes in tropical cyclogenesis.

The potential to reduce projection uncertainties
For a given TC identification scheme, model uncertainty is the most essential uncertainty source (supplementary figure S4).Thus, the reliability of TC projections could be further improved by identifying the reducible model uncertainties.Previous studies have shown that reliability of future projections could be affected by model's ability to reproduce the presentday behavior (Murakami et al 2014, Chen et al 2020, Zhou et al 2020, Chan et al 2021).To examine whether the model uncertainty in projections of the WNP tropical cyclogenesis could be reduced with model improvement, we separate the 20 CMIP6 models into 10 high-skill models and 10 low-skill models, respectively, based on their performance in reproducing the observed distributions of climatological TCGF.We first categorize the high-and low-skill models based on the detected TCGF as this scheme presents the largest contribution (about 61%-100%) to the total projection uncertainties throughout the 21st century (supplementary figure S5).The model skills in simulating the WNP TCGF are measured by an objective metric, i.e. the Taylor skill score (TSS in equation (1); figures S6 and S7).The TSS for the selected 10 high-skill models ranges from 0.83 to 0.65, while it reduces to 0.33-0.64 for the other 10 low skill models (supplementary table S1).Compared to the Then, we further confirm the above results using the other two GPI schemes (figure 6).As the GPIs demonstrate the environmental potentials for tropical cyclogenesis, we assess the performances of the simulated DGPI and EGPI by their PCCs with the observed climatological TCGF (supplementary figures S8 and S9).As indicated in the supplementary table S1, models chosen for the high-and low-skill groups differ across the various TC identification schemes.Despite some quantitative differences in the total uncertainties, the conclusion that projections derived from the high-skill models show smaller model uncertainties are supported by each TC identification scheme (figure 6).Thus, there is high potential to reduce uncertainties in WNP TC projections through improving TC modeling skills.The 'potential Noise-to-Signal' ratio for a 'perfect' scheme plus a 'perfect' model, assuming progresses in both theoretical understandings and modeling skills in tropical cyclogenesis, becomes less than 1.0 after ∼2040 (figure 2(c)).Hence, a reliable middle-to-long-term projection could be expected with future efforts.

Conclusions and discussion
In this study, we investigate the projected changes and uncertainties in the WNP tropical cyclogenesis by analyzing the multi-scenario projections of the CMIP6 multi-model ensemble.The large projection uncertainties are dominated by the scheme uncertainty, represented by the different projections based on the detected TCGF and the GPIs, at all projection lead times.Further investigation into the associated physical processes reveals that the more hostile dynamic environmental conditions and the more favorable conditions suggested by most thermodynamic factors with global warming result in the large spread among different TC identification schemes.In addition, the model uncertainty is identified as the secondary important source in projections of the WNP tropical cyclogenesis, which can be effectively reduced by improving model performance in simulating the present-day climatological TC frequency.Besides, the contribution of internal variability is also appreciable at short lead times but becomes negligible in the long-term, while the scenario uncertainty remains small compared to the other uncertainty sources.
Given the limited theoretical understanding of tropical cyclogenesis currently, the selection of environmental factors in existing TC identification schemes usually stems from their observed relationships with TC formations in the present climate.While GPIs capture the environmental potential for tropical cyclogenesis, changes in TCGF detected directly by the OWZ scheme, using higher temporalresolution data, may be more convincing for the CMIP models with relatively coarse spatial resolutions.Nonetheless, it remains challenging to determine the most appropriate scheme for representing future TC changes.Thus, the method employed to identify TC formations in climate models introduces notable projection uncertainties.Our research underscores the urgency to enhance TC identification techniques in climate models to mitigate these uncertainties.
These findings provide new insights to TC projections that understanding relative contributions of different uncertainty sources helps devote future efforts to addressing the key uncertainties.We acknowledge that the study is just a step towards reducing TC projection uncertainties.Considering the relatively low resolutions of the CMIP6 models and the relatively few future scenarios considered here, we may still underestimate the total uncertainties in future projections.With coordinated efforts in high-quality observations, improved process understanding and high-resolution multi-model ensemble covering a long future period with same forcings, we may expect a further increased confidence in projections of TC activities.

Figure
Figure Future projections of tropical cyclogenesis in the western North Pacific (WNP).(a) Observational climatology of May-June-July-August-September-October-November (MJJASON) mean tropical cyclone genesis frequency (TCGF) in the WNP during the period of 1980-2014.Units: 1.The dashed rectangle denotes the core genesis area in the WNP over the region (5 • -23 • N, 110 • E-155 • ).(b) Time series of the 20-year running mean anomalies relative to the period of 1980-2014 for the WNP tropical cyclogenesis represented by different schemes.Units: %.Thin lines denote individual CMIP6 models.Thick lines represent the CMIP6 multi-model-mean (MMM).(c) The means (lines) and the 10th to 90th percentile ranges (bars) of the WNP tropical cyclogenesis changes relative to the 1980-2014 mean, represented by different schemes, in near-term (2021-2040), middle-term (2041-2060) and long-term (2080-2099) projections under the SSP2-4.5 and SSP5-8.5 scenarios.Units: %.In (b) and (c), orange and dark orange denote the TCGF detected in the CMIP6 models using the OWZ scheme under the SSP2-4.5 and SSP5-8.5 scenarios, respectively.Green and dark green denote the DGPI scheme under the SSP2-4.5 and SSP5-8.5 scenarios, respectively.Blue and dark blue denote the EGPI scheme under the SSP2-4.5 and SSP5-8.5 scenarios, respectively.

Figure 4 .
Figure 4. Quantitative contributions of the related dynamic and thermodynamic environmental factors.(a) CMIP6 MMM projected trends in dynamic and thermodynamic environmental factors considered in the two GPI schemes during 1980-2099under the SSP5-8.5 scenario, including vertical wind shear (vs; units: m s −1 century −1 ), meridional gradient of 500 hPa zonal wind (∂yu; units: 10 6 s −1 century −1 ), vertical velocity at 500 hPa (ω; units: 10 2 Pa s −1 century −1 ), absolute vorticity at 850 hPa (η; units: 10 6 s −1 century −1 ), potential intensity (PI; units: m s −1 century −1 ) and moist entropy deficit of the middle troposphere (χ ; units: 10 3 century −1 ).Dots in denote areas where less than 3/4 (15 out of the 20) models agree on the sign of change in the CMIP6 MMM.CMIP6 MMM projected changes in the (b) DGPI and (c).EGPI under the SSP5-8.5 scenario, along with the contributing dynamic and thermodynamic terms, in near-term (left; light blue and pink bars), middle-term (middle; blue and red bars) and long-term (right; dark blue and dark red bars) relative to the 1980-2014 baseline period.Units: %.In (b) and (c), the vertical lines denote the 10th to 90th percentile range among different CMIP6 models.

Figure 5 .
Figure 5.Comparison of projected TCGF derived from different model groups.Climatology of MJJASON mean TCGF in the WNP during the period of 1980-2014 simulated in the (a) high-skill MMM and (b) low-skill MMM.Units: 1.The Taylor skill score (TSS) of the high-and low-skill MMM with the observed TCGF over the region (0 • -45 • N, 100 • E-180 • ) is 0.81 and 0.62, respectively.Time series of the 20 year running mean TCGF anomalies relative to the period of 1980-2014 simulated by the (c) high-skill models and (d) low-skill models under the SSP2-4.5 (blue) and SSP5-8.5 scenarios (red), respectively.Units: %.Thin lines denote individual models and thick lines represent the MMM of each group.Evolution of uncertainty in projections of the TCGF anomalies for the (e) high-skill and (f) low-skill models.The total uncertainty in TCGF projections is separated into three different components, including internal variability (orange), scenario uncertainty (green) and model uncertainty (blue) for the period of 2015-2099.Units: %.

Figure 6 .
Figure 6.Comparison of projection uncertainties in the GPIs derived from different model groups.Evolution of uncertainty in projections of the DGPI anomalies for the corresponding (a) high-skill and (b) low-skill model groups.Evolution of uncertainty in projections of the EGPI anomalies for the corresponding (c) high-skill and (d) low-skill model groups.The total uncertainty in TCGF projections is separated into three different components, including internal variability (orange), scenario uncertainty (green) and model uncertainty (blue) for the period of 2015-2099.Units: %.
Observed TCs are obtained from the International Best Track Archive for Climate Stewardship database version v04r00 (Knapp et al 2010).We use the TC best-track dataset in the WNP provided by the Shanghai Typhoon Institute, China Meteorological Administration (Ying et al 2014, Lu et al 2021).