Global expansion of tropical cyclone precipitation footprint

The study addresses how the spatial footprint of heavy tropical cyclone (TC) precipitation has changed under recent climate warming. While prior work and theory (Clausius-Clapeyron) indicate increasing TC rain rates, observations have shown declining inner-core rain rates and increases in outer rain bands, suggesting complex spatial reorganizations of rainfall. Rainfall area is also influenced by relative sea surface temperature (SST). Many existing metrics average rainfall within arbitrary radii from the TC center, which may obscure spatial variability and complicate intercomparisons. To better capture the spatial structure and potential inland flood risk from heavy TC rainfall, the authors introduce a new metric, DIST30, defined as the rainfall-rate-weighted mean radial distance of heavy rainfall clusters (>30 mm/3 h) from the TC center. The work quantifies global trends in this footprint, explores regional patterns, and identifies environmental and storm characteristics controlling DIST30.

Prior studies project fewer but more intense TCs globally with warming, and theory suggests rain rates should increase with temperature following Clausius-Clapeyron scaling. Satellite observations have revealed decreasing inner-core TC rain rates with increases in outer bands, potentially linked to atmospheric stability changes and higher water vapor with warmer SSTs. Rainfall area has been shown to expand with higher relative SST (local SST minus tropical mean), affecting rainfall distribution. Research has often relied on metrics within fixed radii (e.g., 100–500 km), which may not represent heterogeneous rainfall structures and hinder model comparisons. Studies also find that inner-core and outer-region precipitation are controlled by both TC intensity and environmental conditions, including vertical wind shear and storm motion, which drive rainfall asymmetry. Collectively, these findings motivate a robust, spatially explicit metric to characterize heavy rainfall footprints and their climate drivers.

Data: The study uses IBTRACS v04 best-track data (1980–2020) for TC position, minimum sea-level pressure, and maximum sustained wind speed, and MSWEP v2 global precipitation (3-hourly) for 1980–2020. Land–sea boundary shapefiles were obtained from UCLA Geoportal.

TC precipitation extraction: For all TCs (over land and ocean) at 3-hourly synoptic times (00:00, 03:00, 06:00, 09:00, 12:00, 15:00, 18:00, 21:00 UTC), TC-related precipitation is defined as accumulated 3-hour MSWEP precipitation within 500 km of the TC center. To compute distances, the precipitation field is resampled to 25 km resolution in an Albers projection, with standard parallels set to the adjusted latitude of the TC center (rounded up in NH, down in SH) ±10°. The resulting grid is 41×41 with the TC center at (21,21).

Metrics: DIST30 is defined as the rainfall-rate-weighted mean radial distance from the TC center to clustered heavy rainfall grid cells exceeding 30 mm/3 h within 500 km. DIST50 is similarly defined with a 50 mm/3 h threshold. Spatial aggregation for visualization uses 4° grid cells; coastal cells are those containing both land and sea.

Trend analysis: Linear regression estimates trends and 95% confidence bounds (e.g., Fig. 1a). Spatial differences (ADIST30) are computed between 2001–2020 and 1980–1999. Significance of temporal trends at grid locations is assessed with a Mann-Whitney test on the regression model at the 99% significance level.

Machine learning: Monthly averaged DIST30 and coincident environmental variables are modeled using XGBoost with 5-fold cross-validation. Features include TC characteristics (maximum sustained wind speed, central pressure), location (latitude, longitude), SST (absolute and relative to tropical mean), relative 2 m air temperature, vertical wind shear, total column water vapor, and climate indices (e.g., AMO). Model performance is evaluated by R², RMSE, and MAE. SHAP values provide feature importance and partial relationships to interpret controls on DIST30 globally and by ocean basin. Spatial SHAP patterns are mapped (hexagons of 1° radius) for key features.

• Global trend: DIST30 increased significantly at 0.34 km/year from 1980 to 2020.
• Frequency shifts: For extreme TC rainfall (>30 mm/3 h), relative frequency within 200 km of TC center decreased by 5.02% (low latitudes ≤25°) and 22.60% (mid-latitudes >25°), while frequency beyond 200 km increased by 13.11% (low latitudes) and 43.33% (mid-latitudes) between 1980–1999 and 2000–2020. For DIST50, trend is 0.36 km/year; mid-latitude changes are +52.52% beyond 200 km and −21.54% within 200 km.
• Spatial distribution: DIST30 increased over 59.87% (8.79×10^7 km²) of global TC impact areas; decreases occurred over 40.13% (5.89×10^7 km²). Areas with positive ADIST30 exceeding thresholds represent 39.03% (>25 km), 22.06% (>50 km), 12.01% (>75 km), and 7.41% (>100 km) of TC-affected areas.
• Basin patterns: Western North Pacific (WNP) shows the largest area of increase (2.20×10^7 km²; 25.06% of global increased area). Northern Atlantic is second (1.92×10^7 km²; 21.80%). Northern Indian basin exhibits a general reduction in DIST30; other basins mostly increase or show weak trends.
• Coastal impacts: Of 4.73×10^7 km² land–sea boundary areas, 54.02% (2.56×10^7 km²) show increasing DIST30. Elevated growth in populated coasts: WNP boundary areas 61.84% (7.33×10^6 km²) and South Pacific 63.55% (4.93×10^6 km²) with increases.
• Machine learning performance: Global XGBoost R²=0.51, RMSE=70.48 km, MAE=47.34 km. Basin models R²=0.46–0.57, RMSE=66.66–77.90 km, MAE=43.74–50.77 km; no overfitting indicated.
• Feature importance (global): Maximum sustained wind speed (VMAX) and latitude (LAT) are most important, followed by central pressure (PRES), vertical wind shear (WS), relative 2 m temperature (RT2M), longitude (LON), relative SST (RSST), and total column water vapor (TCWV).
• Relationships:
  • VMAX: Generally negative relationship with DIST30; high VMAX (>70 kt) associates with small, stabilized DIST30 (inner-core convection near eyewall). Low VMAX associates with larger DIST30, consistent with unorganized systems and extratropical transition (ET).
  • Latitude: Higher absolute latitude linked to larger DIST30; mid-latitudes have more low-VMAX cases and faster storm motion, enhancing precipitation asymmetry.
  • Vertical wind shear: Positive, nonlinear relation with DIST30; strongest influence in mid-latitudes. Low shear (<10–15 kt) shows mixed effects, especially in warm SSTs.
  • Thermodynamics: RT2M shows nonmonotonic relation (negative below ~1 K relative, weak positive above). RSST positively relates to DIST30 for mature tropical systems; absolute SST has mixed effects. Higher TCWV generally associates with larger DIST30.
• Latitudinal trends: Tropics (25°S–25°N) show slight, low-confidence increase in DIST30 (p=0.11) with decreasing vertical wind shear; observation frequency shows weak increase. Northern mid-latitudes (>25°N) show strong increasing DIST30 (p=0.00), weakly increasing shear (p=0.33), and strong rise in DIST30 observation frequency (p=0.00). Southern mid-latitudes (>25°S) show no significant trends.
• Mechanism: Reduced tropical shear favors TC survival; more TCs translate poleward into regions of higher shear, undergo ET, and develop larger, more asymmetric rain fields, expanding DIST30—particularly in the Northern Hemisphere (notably WNP and North Atlantic).

The new DIST30 metric reveals a global expansion of the heavy rainfall footprint of TCs, especially away from the storm center and in mid-latitudes of the Northern Hemisphere. This finding reconciles observed increases in outer rainband rainfall with decreases near the core by quantifying where heavy precipitation occurs relative to the center. The interpretable machine learning analysis identifies vertical wind shear, together with storm intensity and latitude, as key controls shaping the precipitation footprint. Decreasing shear in the tropics likely enhances TC survival and intensity, enabling more storms to reach higher latitudes where shear tends to be stronger, storm translation speeds higher, and ET more common—conditions that increase rainfall asymmetry and extend heavy rainfall farther from the center. These mechanisms explain the strong regional increases in DIST30 in the WNP and North Atlantic and the elevated risk in coastal and inland mid-latitude regions. The results emphasize that flood risk from TCs is increasingly associated with heavy precipitation occurring far from the storm center, requiring adaptation of risk assessment and preparedness strategies in regions historically less exposed to TCs.

The study introduces DIST30, a spatially explicit metric capturing the radial footprint of heavy TC rainfall, and shows that it has increased globally at 0.34 km/year from 1980–2020. Heavy rainfall has shifted outward, with substantial increases beyond 200 km from TC centers, especially in mid-latitudes of the Northern Hemisphere. Spatial analyses indicate that nearly 60% of TC-affected areas experienced increases, with particularly strong growth in the Western North Pacific and North Atlantic, including populated coastal zones. Interpretable XGBoost modeling demonstrates skillful prediction of monthly DIST30 and identifies maximum sustained wind speed, latitude, vertical wind shear, relative 2 m temperature, relative SST, and water vapor as dominant drivers. Poleward migration of TCs, coupled with environmental changes (notably vertical wind shear), underpins the global increase in precipitation footprint. Future work should refine precipitation structure metrics by jointly considering rainfall area and intensity, and investigate regional mechanisms across basins to project how TC rainfall footprints will evolve in a warming climate.

DIST30 captures the rainfall-rate-weighted distance of heavy precipitation but cannot fully represent the complex TC precipitation structure (e.g., full rain area and intensity distributions across inner core and outer bands). Trends in some regions (e.g., tropics) have low statistical confidence. The approach relies on gridded precipitation (MSWEP v2) and best-track data, which may carry observational and reanalysis uncertainties. The machine learning interpretation, while informative, reflects associations that may vary regionally and seasonally; causality should be inferred cautiously.