Introduction

Tropical cyclones (TCs) are severe natural disasters. How TCs will change under anthropogenic climate change is of great concern1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29. As climate warms, it has been projected that TCs may move towards higher latitudes (poleward migration)30,31,32,33,34,35,36,37,38,39,40, how poleward migration can impact future TC’s intensification is therefore an important question to ask. Because poleward migration may involve different TC metrics, e.g., genesis location, lifetime maximum intensity (LMI) location, intensification track (genesis to LMI) or the entire TC track30,31,32,33,34,35,36,37,38,39,40, this research focus on the intensification track.

As TCs move towards higher latitudes, one would naturally anticipate that poleward migration could negatively impact future TC’s intensification, since TCs are also shifting towards a colder (less favourable) ocean environment as climate warms. Nevertheless, how to quantify this impact is a complex issue. Though high-resolution climate models simulate the projected intensity of TCs and include the effect of poleward migration, this is only one of the contributing factors to TC’s intensity change. In other words, poleward migration’s effect is embedded, but not readily separable. To separate its effect, a reference case is required, such that the only difference between the two cases is the effect of poleward migration. However, it is challenging to set up that properly in climate model simulations without modifying other factors in such complex models.

This research thus proposes to use an alternative approach, the Potential Intensity41,42,43,44,45 (PI) approach, to address the issue.

As will be shown, the advantage is that both the ‘with’ and ‘without’ poleward migration cases can be cleanly generated to separate and quantify poleward migration’s effect on future TCs. PI is TC’s intensity theoretical upper bound, given the thermal-dynamic environmental conditions. Though PI is not the actual intensity of TCs, it is the closest possible intensity estimate with strong theoretical underpinning13,41. It has also been suggested to be used as a proxy for the actual intensity as well13. Thus the information provided is highly relevant, especially understanding how PI may change in the future is by itself important.

Our study region is the Western North Pacific Ocean (WNP), as it is among the global oceans where poleward migration is the most evident, in both present-day observation as well as in future projections30,31,32,33,34,35,36,37,38,39,40. The WNP is also the largest TC basin on earth: 60% of the world’s most intense TCs (i.e., Category-5 TCs) are located in the WNP46,47.

Before proceeding, we first perform a latitude analysis to re-examine the poleward migration over the WNP by the end of the 21st Century, to ensure that a robust poleward migration projection exists. Three TC track groups are examined. The first two groups are from the CMIP5 (Coupled Model Intercomparison Project 5) explicitly-simulated TCs, under the Representative Concentration Pathway (RCP) 4.5 and 8.5 emission pathways, as in Camargo 20135. The third group of tracks is from synthetic TCs downscaled from the CMIP5 RCP 8.5 environmental fields, as in Emanuel 20158. The first and the 2nd groups are named C-Track-4.5, and C-Track-8.5, respectively, while the 3rd group is named E-Track-8.5. Both C-Track-4.5 and C-Track-8.5 include 6 different climate models, while E-Track-8.5 includes 7 models. Although more models are included in Camargo 20135, some generate too few TCs and their climatology is too different from observations, therefore those models are excluded from analysis (Methods). All 7 models from Emanuel 20158 are considered here.

As we are interested in centennial trends, a 10-year low-pass running filter is applied to remove high-frequency (e.g. interannual) variability. Additional tests using 5-year and 20-year filters are also conducted. We arrange the data according to each consecutive 10-year period till century’s end (2006–2015, 2007–2016… 2091–2100). We consider both a Multi-Model Ensemble (MME), as well as individual models in our analysis. Because our interest is on TC intensification, we define an intensification track (IT) for each TC. IT is defined as the track points from the genesis location to the Lifetime Maximum Intensity (LMI) location. The IT points for C-Track-4.5/ C-Track-8.5 have 6-hourly intervals, while the E-Track-8.5 track points have 2-hourly intervals. All analyses in this research are based on the IT points (i.e., point-based analyses). From here onwards, ‘track points’ refer only to the IT points, and ‘track locations’ refer only to their locations.

All track locations are analysed to assess whether poleward migration is a general behaviour rather than just existing at the locations of the LMI30. Figure 1 illustrates the MME-average results for the latitude analysis. Statistically significant northward trends for all 3 track groups are found, and the migration rate is ~ 0.07°N, 0.13°N, and 0.16°N per decade for C-Track-4.5, C-Track 8.5, and E-Track 8.5, respectively. It can also be seen that the migration rate for the weaker emission group, i.e., C-Track-4.5, is smaller than the stronger emission group (C-Track 8.5), though still significant statistically. Individual-model results also show consistency, and the majority of the models have statistically significant northward trends (≥95%, 5/6, 4/6, 7/7 in Supplementary Figs. 1–3). These results confirm the necessity to explore the effect of poleward migration on future TCs’ intensification.

Fig. 1: Latitude time series.
figure 1

MME-averaged latitude time series of the 3 TC track groups, with trends and p-values. 95% confidence trend bounds are shaded.

Full size image

Our idea is illustrated in Fig. 2. We construct and compare two types of future PI projection scenarios: one with and the other without the poleward migration impact. The projection with (without) poleward migration is called TC_Move (TC_Stay). Both projections start at the present-day climate (defined as 2006–2015, Time 1 in Fig. 2). At Time 1, PIs are calculated at each of the present-day track locations, according to the corresponding Time-1 environmental input fields. Four inputs are required, namely sea surface temperature (SST), atmospheric temperature profile, atmospheric humidity profile, and sea level pressure41. At Time 2 (i.e., 2007–2016), environmental fields are updated, and the 2 scenarios split. For TC_Move, PIs are calculated at the new track locations of Time 2. For TC_Stay, however, PIs are still calculated at the same locations as in Time 1, since no spatial movement is involved. In other words, the track locations for TC_Stay are fixed throughout (Fig. 2_lower part), while the track locations for TC_Move are updated dynamically (Fig. 2_upper part).

Fig. 2: Schematic diagram comparing the 2 PI projection scenarios.
figure 2

TC_Stay (i.e., no poleward migration) vs. TC_Move (poleward migration included) projections. For simplicity, only one TC track location (in blue) is shown to represent the TC locations at any given time.

Full size image

Both types of PI projections are constructed for individual-model members as well as MME. There are different ways to generate the MME, we consider four approaches, called MME1-MME4, to ensure the robustness of our results and to explore their uncertainty. Because there are 3 track groups (C-Track-4.5, C-Track-8.5, E-Track-8.5), and each has 4 MMEs, we have 12 MMEs for comprehensive assessment. As will be shown, the results are consistent across the four approaches. For simplicity, the discussion will be primarily based on MME1 with the results from the other MMEs shown in Supplementary Figs. 4, 5. Also as mentioned, the main results are based on a 10-year low-pass filter, 5-year and 20-year filters are also tested. The results are similar, as in Supplementary Figs. 6–9.

Results

Proportion of Category-5 PI trends

For each consecutive 10-year time period, we have a group of PI samples, i.e., one corresponding PI value for each TC track location. We define “Proportion of Category-5 PI” as the percentage of Category-5 PI samples relative to the total PI samples. This represents the chance of intensification to the highest intensity category 5, among all the samples. Figure 3a–c depict the results, based on MME1. Without poleward migration (i.e., the TC_Stay projection), robust increasing trends for all three RCP/track groups are found, as climate warms (red curves). When poleward migration is included (the TC_Move projection), the increasing trends are still evident and statistically significant, though weaker (black curves). For instance in Fig. 3c (E-Track-8.5), the trend slope is 2.92 ± 0.13% decade−1 for TC_Stay and 1.94 ± 0.11% decade−1 for TC_Move. The difference between these two slopes is 0.98, equivalent to ~ 34% (0.98/2.92) decrease compare to TC_Stay.

Fig. 3: Time series of the portion of Category-5 PI (i.e. Category-5 PI count over total PI count).
figure 3

Top (a, b, c): Proportion of Category-5 PI time series of the 3 track groups, with trends and p-values, based on MME1. 95% confidence trend bounds are shaded. The regression lines are aligned at the zero point for trend comparison. Bottom (d, e, f): The difference between TC_Move and TC_Stay, i.e., poleward migration’s negative effect, for the 3 track groups. All 4 MMEs are depicted.

Full size image

In Table 1, the overall decrease (weakening) of the trend slope (including MME1 to MME4) for C-Track-4.5, C-Track-8.5, and E-Track-8.5 is −4 to −37%, −30 to −87%, and −30 to −47%. These results suggest that poleward migration does have an appreciable and considerable negative impact on future PIs, by weakening the increasing trends though not reversing their sign. The mean weakening in trend slope for RCP 8.5 (i.e., including the 8 MMEs in C-Track-8.5 and E-Track-8.5) is 42% and for RCP 4.5 is 22% (Table 1). In other words, the increase in the proportion of Category-5 PI under RCP 8.5 (RCP 4.5) may only be about 58% (78%) of what is projected when the poleward migration of TCs is not taken into account.

Table 1 Trend results of the Proportion of Category-5 PI.
Full size table

Because the difference between TC_Move and TC_Stay is poleward migration’s impact, we depict this difference in the lower panels of Fig. 3. It can be seen that the stronger the emission, the stronger the negative effect of poleward migration (Fig. 3d, e, i.e. comparing RCP 4.5 and 8.5 for C-Track). Furthermore, the poleward migration’s negative impact can be seen in both RCP 4.5 and RCP 8.5, albeit with different magnitude, echoing the latitude analysis that the amount of poleward migration is larger in RCP 8.5 than in RCP 4.5 (Fig. 1a, b).

Average PI trends

We now consider the average PI, defined as the average from all PI samples in each consecutive 10-year time period. Similar to the results above, the average PI trends are smaller, when the poleward migration is included, though their signs are still positive (Fig. 4a–c). Consistently, the stronger the emission pathway, the stronger the negative effect of the poleward migration on average PI (Fig. 4d, e). The overall trend weakening for C-Track-4.5, C-Track-8.5, and E-Track-8.5 is −7 to −44%, −67 to −97%, and −56 to −65% (Table 2). The mean weakening for RCP 8.5 (i.e., including the 8 MMEs in C-Track-8.5 and E-Track-8.5) is 68% and for RCP 4.5 is 30% (Table 2). In other words, the increase in the average PI under RCP 8.5 (RCP 4.5) may only be about 32% (70%) of the PI projections which do not consider the poleward migration of TCs.

Fig. 4: Time series of Averaged PI.
figure 4

Top (a, b, c): Averaged PI time series of the 3 track groups, with trends and p-values, based on MME1. 95% confidence trend bounds are shaded. The regression lines are aligned at the zero point for trend comparison. Bottom (d, e, f): The difference between TC_Move and TC_Stay, i.e., poleward migration’s negative effect, for the 3 track groups. All 4 MMEs are depicted.

Full size image
Table 2 Trend results of averaged PI.
Full size table

It should be noted that though the average PI increases little in the TC_Move projection (e.g., black curve in Fig. 4c), it does not mean that the increase in the high PIs is as little. As in the black curve of Fig. 3c, the proportion of Category-5 PI still increases considerably. The reason for the less increase in the averaged PI w.r.t. current is because PI distribution change is involved, and the increase in the high and low PI samples compensate each other (see next section and Fig. 5d, f). Another point to note is that recent report suggested that the observed emission so far appears to match more closely with the RCP 8.5 pathway48. Kossin’s (2015)45 study in the present climate also suggested a considerable modulation by poleward migration of the observed average PI.

Fig. 5: PI distribution and poleward migration.
figure 5

Comparison between current (a) and the 2 future PI scenarios (b, c), based on E-Track-8.5 MME1. The background map in colour is the TC-Season averaged PI map. TC location contour, based on intensification-track points, is overlaid. These points are first interpolated to 2 by 2 grid before contouring, 95% of points are enclosed. a Current: present-day PI map with present-day TC location contour overlaid. b Future TC_Stay: future PI map with present-day TC location contour (in black) overlaid. This contour is the same as in (a), because there is no TC movement involved. c Future TC_Move: future PI map with future TC location contour (in white) overlaid. The contour in (b) is also shown in dashed black, for comparison. d–f PI probability distribution histogram for Current (d), future TC_Stay (e), and future TC_Move (f). Additional discussions see Supplementary Note 2.

Full size image

Lastly, it can be seen in Fig. 4 that multi-decadal variability is also visible. As in Fig. 1, multi-decadal variability plays a part in modulating the TC track as warming progresses. These suggest that for the western North Pacific TCs, besides the global warming signal, multi-decadal signal is likely to continue to play a considerable role in jointly modulating TC variability in the future, including track and PI.

The offsetting effect of poleward migration

Figure 5a presents the TC-Season averaged PI map in the current climate, with present-day TC locations overlaid as enclosed by the black contour (enclosed 95% of TC IT points). As the climate warms, PI increases (Fig. 5b). If the TC location shifts are not included, there is a net increase in PI, because the PI values enclosed by the black contour in Fig. 5b increase, as compared to the PI values enclosed by the same black contour in Fig. 5a. This is the well-known increase in PI due to anthropogenic climate change (e.g. Sobel et al. 2016)13. However, if poleward migration is considered, future TCs will concurrently migrate northward into a region of lower PI values (enclosed by the white contour in Fig. 5c). Therefore, the PI values enclosed by the white contour in Fig. 5c have a higher contribution from the lower PI values to the north, when comparing with the PI values enclosed by the black contour in Fig. 5b. In other words, the poleward migration has a negative influence on PI, offsetting global warming’s positive influence, and PI projection TC_Move is the result of these 2 competing factors (Fig. 5c). Comparing the two PI projections, the TC_Stay projection only includes global warming’s influence in increasing future PI values (Fig. 5b). In contrast, TC_Move PI projection (Fig. 5c), both the global warming’s positive influence and poleward migration’s negative influence are at work to determine future PIs. This can also be seen in the PI distribution where increases in both the high and low-ends of the spectrum w.r.t. current are found (Fig. 5d, f).

Discussion

This research quantifies the impact of poleward migration on future WNP TC’s intensification via two contrasting PI projections. Without poleward migration, at the end of the 21st century, the estimated increase in the proportion of Category-5 PI (average PI) w.r.t. current is ~ 18.3% (2.7 m s−1) under RCP 8.5 (Supplementary Table 1). With poleward migration, the estimated increase in the proportion of Category-5 PI (average PI) at century’s end w.r.t. current would only be ~11.6% (0.8 m s−1). For RCP 4.5, without and with poleward migration, the estimated increase in the proportion of Category-5 PI at century’s end w.r.t. current is 12.8% vs. 10.2%, and 2.3 m s−1 vs. 1.6 m s−1 for average PI (Supplementary Table 2).

Outside the PI framework analysed above, there are other environmental factors which may also change with poleward migration and could affect future TCs, such as ocean subsurface stratification9 and vertical wind shear49. Although these effects are not main scope of this study, we conducted some preliminary assessment. Environmental fields of two E-Track-8.5 models, SP-CCSM4 and HADGEMS2-ES, are examined. The results from the 2 models are consistent (Supplementary Note 1, Supplementary Fig. 10–12). A brief summary based on SP-CCSM4 is given below.

We found that there is an increase in ocean subsurface stratification as TCs move northwards (Supplementary Note 1). Using ocean profiles averaged along the IT points during the TC-Season (June-November) in with and without poleward migration situation, we employed the 3D Price-Weller-Pinkel (3DPWP) ocean model50 to estimate the TC-induced ocean cooling effect46,50,51. The cooling effect from both entrainment mixing and upwelling are included in the 3DPWP. 3DPWP is run at TC translation speed of 5 m s−1. After obtaining the cooling effect, we used the Ocean Coupling Potential Intensity (OCPI)52,53, i.e., a revised PI index, which incorporates the subsurface contribution to PI. The results show that ‘with’ and ‘without’ subsurface contribution, poleward migration’s negative effect (i.e., the difference between TC_Stay and TC_Move) on average PI is −0.95 m s−1 vs. −0.86 m s−1. This suggests that subsurface factor does contribute to the negative effect, though may not be too large (−0.09 m s−1, Supplementary Note 1). It also suggests that the effect of poleward migration estimated in the PI analysis should be smaller than in the actual intensity, because there are additional negative factors such as ocean subsurface contribution to add-on to the negative effect. Nevertheless, PI should still have captured the main effect, because the ‘add-on’ effect from ocean subsurface does not appear to be overly large. Further investigation using more models is needed.

In the case of vertical wind shear, it can be seen in Supplementary Fig. 11 that even when poleward migration (grey contour) is considered, most TC locations are still distant from the region of very high shear at 30–50˚N. With and without poleward migration, the average shear values are 9.8 and 9.9 m s−1, respectively. Again, further investigation using more models is needed.

Nature is an extremely complex system. The occurrence of poleward migration can partially offset global warming’s positive impact on future WNP TCs, the stronger the global warming, the larger the poleward shift, the stronger the negative impact from poleward migration to offset corresponds. These results suggest that at least over the largest TC basin on Earth, future TCs may not intensify without constraint as warming progresses, because a concurrent negative process (poleward migration) also projected to exist to jointly control future WNP TC’s intensification. When simulating WNP TC intensity in the future, it is important to consider this effect from poleward migration, to avoid over-estimation of TC intensity increase.

Methods

Study area

120–180°E, 0–90°N.

Study period

2006–2100.

TC season

June-November (6 months). All analyses in this research are TC-Season based.

Current and future conditions

Current: 2006–2015, Future: 2091–2100. The reason to use the first and the last 10 years to represent current and future is because CMIP5’s projection period is from 2006 to 2100 (95 years). If taking the first and the last 10 years, there are still 76 year’s gap between current and future. If we take the first and the last 20 years (i.e. 2006–2025 and 2081–2100), the gap between current and future will be too short (i.e., only 56 years). Nevertheless, as in Supplementary Fig. 6 and Supplementary Fig. 7, we also test the 20-year results, and the results are consistent.

Saffir–Simpson Tropical Cyclone Scale

Saffir–Simpson tropical cyclone scale based on the 1-min maximum sustained winds: Category 1: 64–82 kts (33–42 m s−1), Category 2: 83–95 kts (43- <49 m s−1), Category 3: 96–112 kts (49- <58 m s−1), Category 4: 113–136 kts (58- <71 m s−1), and Category 5: ≧ 137 kts (71 m s−1). Related classifications for Tropical Depression (TD): ≦ 33 kts (17 m s−1), Tropical Storm (TS): 34–63 kts (18–32 m s−1).

TC track groups and CMIP-5 model members

As in text, three groups of projected TC tracks are used: C-Tracks-4.5, C-Tracks-8.5 from Camargo 2013;5 and E-Tracks-8.5 from Emanuel 20158. As in Camargo 20135, some CMIP5 models generate too few TCs and their climatology is too different from observations. Therefore, for C-Tracks-4.5 and C-Tracks-8.5, models with fewer than 2 TCs per year in the current climate (2006–2015) or with more than 15 years of no TCs in the entire projection period (2006–2100) are excluded from our analysis. The final 6 models for C-Track-4.5 are: CanESM2, CSIRO Mk3.6.0, GFDL-CM3, GFDL-ESM2M, HadGEM2-ES, MRI-CGCM3; for C-Track-8.5 are: CSIRO Mk3.6.0, GFDL-CM3, GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MRI-CGCM3. For E-Track-8.5, by construction this low bias is not present and all 7 models in Emanuel 20158 are used. They are: GFDL-CM3, HadGEM2-ES, IPSL-CM5A-LR, MIROC5, MPI-ESM-MR, MRI-CGCM3, SP-CCSM4.

For C-Track-4.5 and C-Track-8.5, there are 2 model members from the same institution, i.e., GFDL (Geophysical Fluid Dynamic Laboratory). Therefore, concerning the issue of possible over-representation from a single institution40,54,55,56 (models not being fully independent), we first combine the 2 GFDL models (GFDL-CM3 and GFDL-ESM2M) as one entity, called GFDL-MME (Multi-Model Ensemble), to represent GFDL.

Latitude and longitude analyses

For each consecutive 10-year period (2006–2015, 2007–2016… 2091–2100), there is a group of TC locations, and a latitude average is calculated. After obtaining individual-model latitude time series, the MME average is obtained via averaging the individual time series. Similar analysis was conducted for the longitude57 time series (Supplementary Figs. 14, 15).

PI projections and multi-model ensembles (MMEs)

As in the text, 2 contrasting PI projections, i.e. TC_Stay (no poleward migration) and TC_Move (poleward migration included) are constructed. For convenience, we first generate a series of TC-season, 10-year average PI maps (2006–2015, 2007–2016,…2091–2100) for each model (called ‘model PI maps’). For each track group, we also generate a series of MME PI maps (called ‘MME PI maps’) via averaging the respective set of the model PI maps. The methodology is similar to Sobel et al. 201613. Monthly PI map is first built using monthly input fields from CMIP5 environmental atmospheric temperature/humidity profiles, sea level pressure, and SST. The grid size is 1 by 1. The original PI program is from K. Emanuel at MIT (ftp://texmex.mit.edu/pub/emanuel/TCMAX/pcmin_2013.f)41, and is run at each grid to build the monthly map.

After constructing the PI maps, we extract PI values at TC track locations from the PI maps. For TC_Stay, the same present-day (2006–2015) track locations are used throughout updated PI maps (Supplementary Fig. 16). For TC_Move, PI values are extracted from updated PI maps at updated TC locations (Supplementary Fig. 17).

Because there can be different approaches to generate the MME, for each track group, four approaches are applied. We consider two aspects, pre-rescaling of TC track points and sequence in PI map extraction. As each aspect has 2 options, there are 4 MMEs (Supplementary Table 3). The 2 options for pre-rescaling of track points are: ‘Yes’ and ‘No’. The option for ‘No’ (i.e., no pre-rescaling) is the direct composition using the original TC track points from individual-model members. However, because the original point counts among individual models can be vastly different (see rows 2/3 in Tables a/b of Supplementary Tables 4–6), this type of direct composition can have a bias towards model members with more track points, while overwhelming the contributions from members with fewer track points. Thus, we also consider the ‘Yes’ option, i.e. pre-rescaling the track points to equalize the contributions from individual-model members to MME (i.e., equal-weighting). Technical details see section below.

As for the sequence in PI map extraction, there are 2 options: Option 1 and Option 2. Option 1 extracts PI values from the MME PI maps, based on the MME composited track locations (e.g., see Supplementary Fig. 13). Option 2 first extracts PI values from individual-model PI maps using individual-model track locations, then combines together to form MME.

Pre-rescaling of TC Track Points

Pre-rescaling of track points is summarized in Supplementary Tables 4–6. Using C-Track-8.5 as an example, we explain the concept. As in Supplementary Table 5, the original track point counts among the individual-model members are vastly different. For example, the CSIRO model alone contributes ~37.1% to MME, 7 times the contribution from IPSL (5.7%) (row 3 in Supplementary Table 5b). Though direct composition using original track points (i.e. no pre-rescaling) is intuitive, it is a non-equal-weighting approach, and there is a bias towards big models such as CSIRO. To establish an equal-weighting approach, TC track points from individual models are pre-rescaled to equal base, before compositing.

We first choose a count reference close to the mean count among the model members. For example, the count reference for C-Track-8.5 is 2600, as the original counts are between 960 to 4523 (row 2 of Supplementary Table 5a). The rescaling is thus to scale the individual model counts towards 2600 (rows 5 in Supplementary Table 5a). Using the HadGEMS2-ES model as an example, our approach is illustrated. We first place the TC IT points over a map of grid size 1 by 1 (Supplementary Fig. 18). Because the original TC count is 1582, the rescaling factor is 1.643 (2600/1582). Therefore for each grid, we multiply the count by the rescaling factor. As in Supplementary Fig. 18b, the spatial features and patterns are still preserved after rescaling, because it is only a linear scaling. After rescaling, the count is 2681, i.e. ~2600. The reason why it can not be exactly 2600 is because at each grid, the count has to be an integer after rescaling, thus there is some minor difference.

After rescaling for current (2006–2015, Supplementary Fig. 18a, b), rescaling is also done for all subsequent 10-year tracks (2007–2016,…2091–2100). The result for 2091–2100 is shown in Supplementary Fig. 18c, d. This approach is applied to all individual models. After rescaling, all 5 members (CSIRO Mk3.6.0, GFDL-MME, HadGEM2-ES, IPSL-CM5A-LR, MRI-CGCM3) have ~2600 points, corresponding to ~20% equal contribution to the C-Track-8.5 MME (row 6 in Supplementary Table 5a, b). This approach thus enables feature preservation and yet able to bring models with large initial point count difference to equal basis. In addition, because current and future track points are both rescaled to ~2600, it has the additional benefit of ensuring that the change in track points between current and future is due to change in location alone, and not from TC frequency change between future and current33.

TC-induced ocean cooling effect

The TC-induced ocean cooling effect is estimated using the 3DPWP ocean model50. The wind forcing is based on the intensification-track averaged TC-Season climatological (1980–2019) wind speed from the US Joint Typhoon Warning Center. TC translation speed used is 5 m s−1.

Vertical wind shear

Vertical wind shear (VWS) is calculated based on the 850 and 200 hPa wind difference over the WNP domain, at each 1 by 1 grid.