Attribution of 2020 hurricane season extreme rainfall to human-induced climate change

Hurricanes are among the costliest and deadliest geophysical extremes, with damage driven by high winds, storm surge, and intense rainfall. The 2020 Atlantic hurricane season set records with 30 named storms and 13 U.S. landfalls, leading to tens of billions of dollars in losses. Anthropogenic greenhouse gas emissions have driven a detectable increase in global land-ocean surface temperatures, exceeding 1 °C relative to 1850 by 2020, which in turn has warmed North Atlantic sea surface temperatures (SSTs). Quantifying how this warming alters tropical cyclone rainfall remains challenging due to competing influences (e.g., wind shear, atmospheric stability) on storm genesis and development. Building on advances in event attribution, the study’s research question is to objectively quantify the impact of human-induced climate change on extreme rainfall across the entire 2020 North Atlantic hurricane season. The purpose is to estimate changes in extreme 3-hourly rainfall rates and 3-day accumulations for storms at tropical-storm strength or greater, thereby informing risk for coastal communities and improving operational attribution capabilities.

Recent assessments indicate increases in tropical cyclone intensity, the proportion of stronger hurricanes, and the occurrence of storms producing extreme precipitation. Prior attribution studies have analyzed individual high-impact hurricanes (e.g., Maria and Florence) using hindcast and other attribution frameworks, finding anthropogenic increases in storm rainfall ranging roughly from a few percent to about 20% depending on the metric and event. The Clausius–Clapeyron (C–C) relationship provides a thermodynamic benchmark for expected increases in moisture and precipitation with warming, often around 6–7% per °C. Observational work has suggested a global increase in tropical cyclone rainfall on the order of ~1–2% per year in recent decades. The study leverages these foundations, extending attribution from individual events to an entire hurricane season.

The study applies a hindcast attribution framework using the Community Atmospheric Model (CAM, version 5) in a variable-resolution configuration with ~28 km grid spacing over the North Atlantic. Two ensembles are generated: an “actual” ensemble meant to simulate the 2020 season conditions, and a “counterfactual” ensemble representing a world without human influence. Anthropogenic warming fingerprints for three-dimensional thermodynamic fields (temperature and specific humidity) and SSTs are derived from the CESM Large Ensemble (CESM-LE), which includes a 1500-year control and a 40-member transient ensemble from 1920 onward. For the counterfactual simulations, the anthropogenic SST fingerprint for June–November 2020 (estimated at ~0.4–0.9 °C across the North Atlantic) is removed from boundary conditions, three-dimensional atmospheric temperature and humidity fields are adjusted accordingly at initialization, and greenhouse gas concentrations are set to 1850 values. Ozone and solar forcing are ignored in the actual ensemble as used here. Initializations are performed every 3 days from June 1 through November 30, 2020. Each hindcast runs for 7 days, but only the first 3 days are used in accumulated rainfall analyses to avoid overlap across successive initializations. Twenty-member ensembles are generated at each initialization time using perturbed parameters, yielding a large sample; in total, 1200 individual simulations were completed for both the counterfactual and actual ensembles. Storm detection and rainfall attribution to tropical cyclones are performed with the TempestExtremes software, which detects and tracks candidate cyclones at 3-hourly timesteps. Analyses include storms at least tropical storm strength (generally >18 m/s; detection used >17 m/s). For each detected storm, precipitation within a defined radius and wind threshold is attributed to the cyclone. The study computes distributions of 3-hourly rainfall rates and 3-day accumulated rainfall amounts, with emphasis on extremes (e.g., 95th and 99th percentiles). Percentage differences between actual and counterfactual ensembles and 95% confidence intervals are estimated using large-sample resampling approaches. Model bias adjustments are applied consistently to minimize ensemble differences unrelated to anthropogenic forcing.

- Anthropogenic warming increased extreme rainfall across the entire 2020 North Atlantic hurricane season. For storms at least tropical storm strength (>18 m/s): extreme 3-hourly rainfall rates increased by about 10%, and extreme 3-day accumulated rainfall amounts increased by about 5%. - For hurricane-strength storms (>33 m/s), increases are larger: extreme 3-hourly rainfall rates up by about 11%, and extreme 3-day accumulated rainfall amounts up by about 8%. - Global mean surface temperature in 2020 exceeded preindustrial (1850) by over 1 °C; the anthropogenic contribution to North Atlantic SSTs during the 2020 season is estimated at approximately 0.4–0.9 °C (mean ~0.6 °C) based on CESM-LE. - Extreme 3-hourly rainfall increases, especially for hurricanes, approach nearly twice Clausius–Clapeyron scaling, whereas storm-total (3-day) accumulation increases are broadly consistent with C–C expectations (~6–7% per °C) given the estimated basin warming. - Actual and counterfactual ensembles produce similar storm track distributions, enabling robust comparison of rainfall changes attributable to thermodynamic forcing. - Large ensembles (20-member, 3-daily initializations) and 3-hourly diagnostics support probability distributions of rainfall; for some distribution analyses, sample size noted as 500.

The findings directly address the research question by quantifying how human-induced warming altered extreme rainfall across the entire 2020 hurricane season. The attribution signal is evident in both short-duration (3-hourly) extremes and multi-day (3-day) accumulations, with larger percentage increases for hurricane-strength storms. This implies that the strongest storms, which already pose the greatest flood risk, are experiencing disproportionately higher extreme rainfall rates in a warmer climate. The consistency of 3-day accumulation increases with Clausius–Clapeyron scaling supports a thermodynamic mechanism: warmer SSTs and a moister atmosphere supply more water vapor, enhancing storm-total rain. The greater-than-C–C increases in extreme 3-hourly rates suggest additional storm-structure or dynamical contributions, potentially related to intensification and eyewall processes that concentrate rainfall. These results reinforce prior event-specific attribution studies and extend them season-wide, thereby strengthening the evidence base that anthropogenic warming is amplifying tropical cyclone rainfall hazards. The implications for coastal and inland flooding risk are substantial, indicating the need for updated risk management and infrastructure planning that account for higher rainfall extremes during tropical cyclone landfalls.

This study extends the hindcast attribution framework from single events to an entire hurricane season and demonstrates that human-induced climate change increased extreme rainfall during the 2020 North Atlantic hurricane season. Increases of approximately 10% (3-hourly) and 5% (3-day accumulations) for all tropical-storm-strength systems, and 11% and 8% respectively for hurricanes, indicate heightened flood risk in a warming climate. Accumulation changes align with Clausius–Clapeyron expectations given ~0.4–0.9 °C anthropogenic SST warming, while sub-daily extreme increases likely reflect storm-structural and dynamical effects superimposed on thermodynamic forcing. Future research should explore higher-resolution modeling to resolve storm structure and precipitation extremes, examine large-scale circulation changes affecting storm genesis and frequency, and further develop operational, season-wide attribution systems for real-time risk communication and planning.

- The counterfactual anthropogenic fingerprint is derived from CESM-LE and may include a small component of natural variability; using a single model pattern limits exploration of forced response uncertainty. - CAM5 resolution (~28 km) may not fully resolve inner-core storm structure and the most intense precipitation features; higher resolution could alter extreme tail behavior. - Large-scale circulation and storm frequency/genesis changes due to climate change are not evaluated within this hindcast attribution framework. - Ozone and solar forcing are ignored in the actual ensemble configuration described, which could introduce small forcing biases. - Hindcasts are initialized every 3 days and analyzed for only the first 3 days of each 7-day forecast to avoid overlap, potentially omitting some longer-lived rainfall contributions. - Storm detection thresholds and tracking choices (e.g., >17–18 m/s criteria, attribution radii) can influence the sample and rainfall attribution. - Computational constraints precluded comprehensive exploration of uncertainty in response patterns and magnitudes across multiple climate models.