Recent increase in the potential threat of western North Pacific tropical cyclones

The study addresses how to better quantify and track the potential threat (PT) posed by western North Pacific (WNP) tropical cyclones beyond traditional intensity metrics. While intensity (maximum sustained wind) is widely used, damages are also tied to forecast skill, which is particularly low for storms with high intensification rates (IR) and rapid intensification (RI). The authors hypothesize that combining intensity with intensification characteristics provides a more meaningful measure of threat and that PT has increased in recent decades. They also seek to identify environmental drivers of PT, positing that ocean subsurface thermal conditions may play a dominant role. The work aims to objectively categorize TCs by PT using clustering on intensity and IR, assess long-term trends in high-PT storms, evaluate forecast errors across PT groups, and link PT changes to environmental variability and global warming.

- Forecasting skill for TC tracks has improved steadily, but intensity forecasts have progressed slowly; RI remains a key forecasting obstacle. - Prior TC threat assessments often rely on intensity, but impacts also depend on precipitation, storm size, storm surge, and minimum pressure; these metrics can vary independently of intensity. - Clustering methods (e.g., K-means) have been used in TC research to categorize tracks and behaviors; recent work has clustered intensity-related variables. - Observations and projections indicate increases in major TCs in some regions and changes in RI characteristics; coastal regions of East Asia have seen increases in activity and IR. - Environmental controls on TC intensity and RI include mid-level humidity, vertical wind shear (VWS), and ocean thermal structure; subsurface ocean heat content (e.g., TCHP) is known to modulate intensification and surface cooling feedbacks. - Long-term variations in WNP TCs have been linked to climate modes (ENSO, PDO, AMO) and global warming, but their influence on a combined PT metric had remained unclear.

Data and study domain: - TC best-track data: IBTrACS v4r00 (primary agency JTWC) for 1980–2020. Additional agency intensity datasets for comparison: CMA, HKO, JMA. - Inclusion criteria: tropical storms with lifetime maximum intensity (LMI) ≥ 34 kt; ocean-only tracks within 0°–40°N; records at 00/06/12/18 UTC. - Sample sizes: 1073 TCs (JTWC). CMA, JMA, HKO recorded 956, 736, and 903 TCs, respectively. - Intensity definitions: JTWC uses 1-min sustained wind; CMA (2-min), JMA and HKO (10-min). Converted 2- and 10-min winds to 1-min using factors 0.93 and 0.96. - Intensification metrics: Average IR (ΔV per 24 h over lifecycle) and ΔV24 within 24 h prior to LMI. RI defined as ΔV ≥ 30 kt in 24 h (alternate thresholds 25, 35, 40, 45 kt tested). Clustering and PT definition: - K-means clustering applied to three inputs: LMI, average intensity change (ΔV; kt/24 h), and ΔV24 before LMI (kt/24 h). - Optimal number of clusters determined by silhouette and Davies–Bouldin indices (details in Supplementary Note 1). - The resulting clusters define PT levels; groups with both high intensity and high intensification are labeled as high PT. Forecast error analysis: - ECMWF forecasts from TIGGE used to compute 24-h and 48-h forecast errors in maximum wind (Vmax) and minimum pressure (Pmin) for 2007–2020 across PT groups and by Saffir–Simpson categories. Environmental composites: - Atmospheric reanalysis: ERA5 (0.25°). Variables: 600-hPa relative humidity (RH), vertical wind shear (VWS; 200–850 hPa vector difference), 850-hPa relative vorticity, 200-hPa divergence. - Ocean reanalysis: GLORYS daily (0.083°) for pre-storm ocean conditions 5–2 days prior to TC approach. Variables: sea surface temperature (SST) and tropical cyclone heat potential (TCHP; heat content integrated from surface to 26 °C isotherm depth). - Period for composites: 1994–2020 (availability of daily ocean reanalysis). For each TC, atmospheric variables were averaged along tracks prior to LMI within a 20°×20° box centered on the TC; oceanic variables sampled 2 days before TC approach. Statistical significance assessed by two-tailed t-test at p=0.01. Trend analyses and attribution: - Basin-mean TCHP trends evaluated using NCEP GODAS monthly ocean temperatures (1°) for 1980–2020 over 5–20°N, 120–180°E and two comparison regions from prior literature. - Climate indices: Global mean SST (60°S–75°N), Niño3 (ENSO), PDO, and AMO from standard sources. Linear regression used to remove the influence of each index (or combinations) from time series of Group A TC counts to isolate residual trends. Statistical methods: - Linear regression for trends; significance via two-tailed t-test (p=0.01 threshold for composites/trends). Bootstrap with 1000 resamples of 35-, 30-, or 25-year subsets from 1980–2020 to assess robustness of Group A trend. Quality control and comparisons: - Cross-agency clustering performed (CMA, JMA, HKO) with optimal cluster counts per dataset to test robustness; intensity measurement differences noted. - Multiple RI thresholds examined to test sensitivity of RI counts by PT group.

- Objective clustering of 1073 WNP TCs (1980–2020) into four groups produced two severe groups (A: 114 TCs; B: 250 TCs) with high LMI (≥85 kt; ≥Category 2), and two less intense groups (C: 284; D: 425). - Despite similar mean LMI (Group A: 131.1 kt; Group B: 125.0 kt), Group A cyclones had significantly higher intensification metrics and poorer forecast performance: • Average 24-h intensity change (ΔV): 7.5 vs 4.7 kt/24 h (A vs B; p<0.01). • RI within 24 h before LMI: 100% of Group A vs 24% (59/250) of Group B. • Lifetime maximum intensification rate (LMIR): 60.0 vs 41.3 kt/24 h (p<0.01). - Forecast errors (ECMWF/TIGGE): • 24-h Vmax error: 23.3 kt (Group A) vs 20.1 kt (Group B), higher by 3.2 kt (p<0.01). • 48-h Vmax error: 32.1 vs 29.2 kt; Pmin errors at 24/48 h also larger for Group A. By contrast, forecast errors were similar when grouped solely by Saffir–Simpson categories (e.g., ~30 kt at 48 h for Categories 3 and 4), indicating that intensity alone does not capture forecast difficulty. Group A (≥Cat 2) had higher errors than Cat 3/4 groups for several lead times. - Spatial characteristics: Group A cyclones formed and intensified at lower latitudes and farther west than Group B (mean genesis 10.8°N,145.2°E vs 12.1°N,156.0°E; mean LMI 17.8°N,132.6°E vs 18.9°N,135.9°E) and occurred closer to the coast (12/114 formed within 400 km of coast vs 8/250 in Group B). - Lifecycle differences: • Intensification duration from 20 kt to LMI: 105 h (A) vs 152 h (B) (p<0.01). • Path length: 1685 km (A) vs 2460 km (B) (−32%, p<0.01). Translation speeds did not differ significantly between A and B (~4.2 m/s). • RI events per life cycle: 6.2 (A) vs 4.2 (B). Results robust across alternative RI thresholds and agencies. - Environmental drivers: • Atmospheric differences (mid-level RH, VWS, vorticity, divergence) between A and B were small and mostly not statistically significant. • Ocean subsurface conditions differed markedly: pre-storm SST was modestly higher for A (29.3 °C vs 29.0 °C), but TCHP was substantially higher for A by ~15×10^7 J m−2 (81.6 vs 66.6×10^7 J m−2), significant over most of the domain, implicating subsurface heat as the dominant driver of stronger intensification and higher PT. - Long-term trends (1980–2020): • Total WNP TC count decreased by ~0.9 per decade (p=0.1); counts of major TCs (≥Category 3) and RI TCs remained roughly constant. • High-PT Group A increased by 0.8 TCs per decade (p<0.01), and its fraction rose by 3.2% per decade (p<0.01). Across agencies, annual counts of high-PT TCs increased by 0.4–0.6 per decade (p<0.05). • Basin-mean TCHP (5–20°N, 120–180°E) increased by 6.3×10^7 J m−2 per decade (7.3%/decade; p<0.01). This increase aligned with the rise in Group A activity better than trends based solely on LMI or IR. • Abstract-level synthesis: annual number of high-PT TCs increased by 22% per decade over 41 years; approximately 10% of TCs pose great PT and have high forecast errors. - Attribution: The annual count of Group A TCs and seasonal-mean TCHP correlate more with global SST warming than with PDO or AMO. After regressing out global SST, the residual trend in Group A counts was not significant (0.14/decade, p=0.58), whereas trends remained significant after removing PDO (0.77/decade, p<0.01), Niño3 (0.76/decade, p<0.01), or AMO (0.46/decade, p=0.08), indicating a primary role for global warming in the observed increases.

By jointly clustering lifetime maximum intensity with intensification metrics, the study defines a potential threat (PT) framework that more effectively captures storms that are both intense and rapidly intensifying—events that are historically more difficult to forecast and thus pose heightened risk. The findings demonstrate that these high-PT TCs (Group A) suffer significantly higher short-range forecast errors than comparably intense storms categorized only by Saffir–Simpson scales, validating that intensity alone misses critical aspects of operational threat. Environmental composite analyses indicate that ocean subsurface heat content (TCHP) is the dominant differentiator between high-PT and other severe storms, consistent with the mechanism whereby higher subsurface heat buffers against storm-induced cooling and supports RI. The observed multi-decade increase in high-PT storms, alongside significant basin-wide TCHP increases, and the disappearance of the trend after removing global SST warming signals, connect the growing PT to ocean warming under climate change. These results emphasize the need to incorporate subsurface ocean metrics into monitoring and forecasting frameworks and suggest that continued ocean warming could further elevate operational challenges and coastal risks in East Asia.

This work introduces and operationalizes a new PT metric for WNP tropical cyclones by clustering intensity and intensification characteristics, revealing that roughly 10% of storms constitute high PT with systematically larger forecast errors, shorter intensification times, and greater proximity to coastlines. The study documents a robust rise in high-PT TCs over 1980–2020—about 22% per decade—concurrent with a significant increase in subsurface ocean heat (TCHP), and shows that the upward trend is primarily associated with global SST warming rather than internal climate modes. The identification of subsurface heat as a key environmental control highlights the value of monitoring ocean temperatures down to the 26 °C isotherm depth (~80–120 m) for predicting the most hazardous TCs. Future directions include: incorporating additional impact-relevant metrics (size, precipitation, storm surge) into PT, validating PT against observed damages and operational errors, testing alternative clustering frameworks, extending analyses with high-resolution modeling and downscaling, and improving data coverage and homogeneity to assess centennial-scale trends and the roles of climate regime shifts.

- The analysis period (1980–2020) is constrained by data quality; 41 years may be insufficient to fully resolve multi-decadal variability and long-term climate signals. - Intensity measurements differ across agencies (1-, 2-, 10-min winds) despite conversion factors; residual inconsistencies may affect clustering and RI diagnostics. - PT focuses on intensity and intensification; other important hazard metrics (size, precipitation, storm surge) are not included and may change independently of intensity. - Statistical attribution is limited by potential confounding from spatial/seasonal distributions and internal variability (e.g., regime shifts around 2000, warming hiatus ~2005–2015). - Forecast error evaluation uses one modeling system (ECMWF/TIGGE) over 2007–2020; generalization to other forecast systems and longer periods warrants caution. - Composite analyses, while robust, may not capture case-specific processes; more targeted case studies and numerical experiments are needed to elucidate mechanisms.