A potential explanation for the global increase in tropical cyclone rapid intensification

The study investigates whether anthropogenic climate change has contributed to observed increases in tropical cyclone rapid intensification (RI), defined as extreme 24-hour intensity changes, and whether storm environments have become more conducive to RI. Recent years have seen improved RI forecasting skill in some basins, yet RI events still produce substantially larger forecast errors than non-RI cases. Observations suggest an upward trend in the proportion of storms undergoing RI, raising concerns that forecasting challenges may increase as the climate warms. However, attribution is complicated by natural multidecadal variability (e.g., AMV, PDO), the relatively short reliable satellite-era record, and dataset inhomogeneities. The paper’s purpose is to compare observed trends in intensification and environmental favorability with those expected from internal climate variability alone, using high-resolution climate model control simulations, and to assess anthropogenic influence via model ensembles and reanalyses.

Prior work documented increases in TC intensification rates across several basins during the satellite era and projected further increases with warming (e.g., Bhatia et al. 2019; Emanuel 2017). Studies link stronger thermodynamic environments (higher SST and potential intensity, PI) to more intense TCs and greater RI likelihood, while acknowledging the role of internal storm dynamics that are difficult to resolve and predict. Basin-scale multidecadal variability (AMV, PDO) can modulate TC activity and coincides with the period of improved intensity estimates, complicating attribution. Homogeneous satellite-derived datasets (ADT-HURSAT) have been used to reduce observational biases, while IBTRACS provides best-track intensities but with potential temporal inhomogeneities. Modeling studies (e.g., HiFLOR) have begun to explore responses of intensity and RI to anthropogenic forcing, but robust global attribution of observed RI trends has remained elusive.

- Observational datasets: IBTRACS v04r00 best tracks for intensity and positions; ADT-HURSAT for a temporally and spatially homogeneous satellite-based intensity record. Analysis focuses on 1982–2017 (first three ADT-HURSAT years omitted due to limited geostationary coverage). RI ratio is defined as the number of 24-h intensity changes exceeding 30 kt divided by the total number of 24-h intensity changes. Storm samples are restricted to TCs with at least 72 h duration, at least 36 h ≥34 kt, over ocean, <40° latitude for start/end, and maintaining ≥34 kt; extratropical transition regions (>40°) are excluded. Intensities are rounded to nearest 5 kt for consistency across datasets. - Model-based internal variability: Use the GFDL high-resolution coupled model HiFLOR control simulation with fixed preindustrial forcings (1860CTL; 1500 simulation years, first 50 discarded). TCs are tracked following established HiFLOR tracking and warm-core criteria. Because HiFLOR’s intensification distribution has biases, Quantile Delta Mapping (QDM) bias correction is applied basin-by-basin to ensure realistic RI ratio slopes. Overlapping 36-year periods (n=1,414 windows) in the corrected 1860CTL yield a distribution of RI ratio trends representing internal variability. - Comparison of observed trends to internal variability: Least-squares slopes of annual RI ratio for 1982–2017 from IBTRACS and ADT-HURSAT are compared to the 1860CTL trend distribution (raincloud/box plots). Statistical significance is inferred when observed slopes lie in extreme percentiles of the internal-variability distribution. - Environmental analysis (storm-local): Four environmental variables known to influence RI are derived from ERA5 and MERRA-2 reanalyses: vertically averaged RH (850/700/600 hPa), vertical wind shear (200–850 hPa vector difference), SST, and PI (computed from reanalysis profiles following Bister & Emanuel using published code; reversible ascent and dissipative heating allowed; flux coefficient ratio 0.9; output scaled by 0.8). To isolate background environments, RH/SHR/SST fields are spectrally filtered to T11 (removing ~95% of cyclonic circulation in ERA5), and along-track means within 5° radius are sampled at 6-hourly TC fixes. Potential intensity is computed without spectral filtering on data regridded to 1°. - Environmental thresholds and RI probability: For each basin and variable, a logistic regression is solved to find a critical threshold that yields the basin’s mean RI probability (e.g., Atlantic shear threshold ~9.2 m/s). RI probabilities are computed for cases exceeding (or, for shear, below) each threshold. RI probability is also examined as a function of the number of thresholds met (0–4). Temporal trends in the annual fraction of fixes satisfying 3–4 vs 0–1 thresholds are assessed (1982–2017), with Wald tests for significance. - Tropical-mean environments: As a proxy for what global climate models can resolve, tropical-mean seasonal (ASO for NH; FMA for SH) averages of RH, SHR, SST, and PI are computed over oceanic regions (10–30° in each hemisphere with specified longitude bounds) from ERA5 and MERRA-2 for 1982–2017. Linear trends and Wald-test significance are reported. - CMIP6 attribution: CMIP6 ensembles are analyzed for 1982–2014 for AllForc (historical), NatForc (hist-nat), and GHGforc (hist-GHG) experiments. For each ensemble member, linear trends in tropical-mean SST, PI, RH, and SHR are computed to form probability density functions (PDFs). Pairwise t-tests and Kolmogorov–Smirnov tests compare slope distributions (AllForc vs NatForc, AllForc vs GHGforc). Equally weighted ensemble means (to mitigate centers with many members) are used to compare mean trends; t-tests assess differences from zero and between forcing categories. - Uncertainty: Monte Carlo perturbations are applied to discretized intensity changes (±2×√5 kt uniform noise per value; 1000 subsamples) to estimate slope uncertainties; 5th–95th percentiles of derived slopes provide confidence bounds.

- Observed RI trends exceed internal variability: In multiple basins and globally, observed RI ratio trends (1982–2017) lie at or above the extreme tail of the HiFLOR 1860CTL internal-variability distribution. IBTRACS slopes exceed the 99th percentile in all analyzed basins; ADT-HURSAT slopes exceed the 95th percentile globally, in the Atlantic and West Pacific (Australian ~94th percentile), indicating a detectable anthropogenic contribution to increased RI rates. - Environmental favorability for RI is increasing: Storm-local environments show significant increases in the annual fraction of cases meeting 3–4 critical thresholds (RH high, SST high, PI minus current intensity high, shear low) and significant decreases in cases meeting 0–1 thresholds for the global, Atlantic, and West Pacific datasets (Australian basin shows weaker/insignificant changes). In the Atlantic, cases meeting 3–4 thresholds more than doubled, while 0–1-threshold cases declined by over 50%. - Tropical-mean environments corroborate: Reanalysis tropical-mean trends (1982–2017) show significant increases in SST and PI in both hemispheres: SST ~0.16–0.24 K per decade (ERA5 NH 0.17; ERA5 SH 0.16; MERRA-2 NH 0.23; MERRA-2 SH 0.24). PI increases ~1.0–1.56 m/s per decade (ERA5 NH 1.06; ERA5 SH 1.2; MERRA-2 NH 1.02; MERRA-2 SH 1.56). Shear shows small, generally insignificant declines (e.g., ERA5 NH −0.09; SH −0.12 m/s per decade). RH trends are dataset-dependent (ERA5 positive: NH 0.34%, SH 0.44% per decade; MERRA-2 negative: NH −0.66%, SH −0.45% per decade). - Anthropogenic attribution from CMIP6: CMIP6 AllForc trend PDFs for SST, PI, RH, and SHR are significantly different from NatForc, indicating detectable anthropogenic influence on tropical-mean environments. Observed reanalysis SST trends fall within AllForc PDFs but outside NatForc PDFs. Observed PI trends are outside NatForc and also generally larger than AllForc ensemble means, implying a detectable but under-explained anthropogenic component and potential model underestimation of PI increases. - Model–reanalysis discrepancy and implications: A discrepancy in upper-tropospheric warming (CMIP6 tends to over-warm aloft relative to reanalyses) helps explain differences in PI trends; this bias may cause models to underestimate future TC intensities and RI increases if not corrected. - Overall: The global and basin-specific increases in RI ratios, alongside more favorable thermodynamic environments and CMIP6 attribution signals, indicate that anthropogenic forcing has contributed to the observed rise in rapid intensification.

The study directly addresses whether recent increases in TC rapid intensification are attributable to anthropogenic climate change by comparing observed trends with those expected from internal variability. The alignment of increasing RI ratios with increasingly favorable storm-local and tropical-mean environments, coupled with CMIP6 AllForc vs NatForc contrasts, supports a positive anthropogenic contribution. The environmental changes—especially higher SST and PI—are consistent with theory that greater thermodynamic disequilibrium enables larger 24-h intensity changes. The robustness across spatial scales (local along-track and basin/tropical-mean) suggests that changes in storm tracks or transient weather variability are unlikely to negate continued increases in RI under warming. The results highlight the importance of improved high-resolution modeling of mesoscale processes and environmental interactions and underscore the urgency of enhancing coastal resilience to address increasing risks from rapidly intensifying storms.

Main contributions: (1) Detection of significant global and basin-scale increases in RI ratios that exceed internal variability, implying an anthropogenic signal; (2) Demonstration that storm-local and tropical-mean thermodynamic environments (SST, PI) have become more favorable for RI since 1982; (3) Attribution analysis showing CMIP6 historical forcing simulations produce trends distinguishable from natural-only forcing, indicating anthropogenic influence on key environmental drivers; (4) Identification of model–reanalysis discrepancies in vertical temperature trends that likely lead to underestimation of PI and potentially RI in models. Future directions: Improve high-resolution coupled models to better resolve TC mesoscale processes and RI; reconcile reanalysis–model discrepancies in upper-tropospheric warming and PI trends; extend attribution across basins with additional high-resolution models beyond HiFLOR; refine homogeneous intensity datasets and data assimilation to reduce observational uncertainties; explore how internal TC dynamics may evolve with warming.

- Internal variability estimated from a single high-resolution model (HiFLOR) after bias correction; conclusions may have model dependence and should be tested with other coupled models capable of simulating RI statistics. - Model resolution (0.25° atmosphere) limits explicit representation of small-scale processes crucial for RI; results emphasize environmental drivers rather than internal storm dynamics. - Observational dataset limitations: While ADT-HURSAT improves homogeneity, it has inherent satellite-based uncertainties; IBTRACS blends multiple agencies and may include time-varying biases. The definition of RI ratio (>30 kt) and sampling choices can influence trend estimates. - Reanalysis inconsistencies: RH trends differ between ERA5 and MERRA-2; shear trends are small and often insignificant; uncertainties in data assimilation affect environmental diagnostics. - Attribution discrepancies: Observed PI trends exceed CMIP6 AllForc ensemble means and differ from NatForc, indicating detectable anthropogenic influence but incomplete explanation of magnitude; vertical temperature profile biases in models complicate projections. - Scope: The study focuses on environments and does not quantify changes in internal dynamical processes of TCs or fully resolve how these may change under warming.