Real-time attribution of the influence of climate change on extreme weather events: a storyline case study of Hurricane Ian rainfall

As the Earth continues to warm due to human greenhouse gas emissions, there is a growing need to efficiently communicate the effect that global warming has on individual extreme weather events. Using a storyline approach, we present a rapid attribution of the increase in rainfall over Florida during Hurricane Ian in 2022 due to climate change as a case study. We find that extreme accumulated rainfall amounts associated with Hurricane Ian increased by 18%, far in excess of what would be expected by Clausius–Clapeyron scaling. The study offers a blueprint for rapid operational climate change attribution statements about extreme storms and other very rare weather events.


Increasing demand for rapid extreme weather event attribution
Tropical cyclones (TCs) are among the most devastating and deadliest weather events worldwide [1][2][3].As a result, when strong storms threaten to make landfall they typically command public interest and media coverage regionally, nationally, and, even, internationally.Human induced climate change has likely already increased the severity of TCs [4][5][6][7] and their impacts [8,9].Not surprisingly then, during such particularly threatening events, the media, general public, and political leaders turn to scientists with the question of 'How did climate change affect this event?'As extreme weather events become more severe, this desire for such information necessitates the need to operationalize extreme weather event attribution [10,11].Indeed, several nations are currently exploring such services that must span a wide variety of extreme weather types [12].
In this paper we present a 'storyline' methodology for attribution that we applied in real-time to the rainfall of the 2022 TC, Hurricane Ian.Storylines are well suited for events such as TCs, that require high resolution modeling, as they are performed with ensembles of short forecast (or hindcast) simulations rather than lengthy, computationally prohibitive calculations [13].However, storylines directly inform only about the attributable change in magnitude of an extreme weather event and not about any change in the probability of an event [14].Furthermore, storyline based attribution statements are highly conditional.That is, they describe a hypothetical storm that would have taken a similar track in a cooler world with similar large-scale meteorological conditions and interpret the difference from the actual storm as the attributable human influence.Despite this highly conditional formulation, storylines can provide useful information about the attributable change in magnitude of the event's physical and socioeconomic impacts [8,9].

Case study: Hurricane Ian (2022)
Hurricane Ian made landfall in Florida in the U.S. on 28 September 2022 as one of the strongest TCs to make landfall in the U.S. [15].It battered the coast of Florida with category 4 (approx.67 m s −1 ) winds on the Saffir-Simpson Hurricane Wind Scale and a storm surge exceeding 12 ft (approx.3.7 m) in some areas [16].Rainfall from Hurricane Ian stretched across the region with estimates of 10-20 inches (about 250-500 mm) throughout much of Florida [16] causing extensive inland flooding.Early estimates suggest that Hurricane Ian was responsible for over 112 billion U.S. dollars in damages and over 150 deaths in Florida [16].
Methods to quantify the influence of anthropogenic global warming on TCs have advanced significantly in recent years.The first storm with multiple attribution assessments was Hurricane Harvey in 2017, which dumped up to 60 inches (1.5 m) of rain over parts of Texas in the U.S [17].Three independent attribution studies using different methodologies estimated that climate change increased Harvey's rainfall accumulations by 13%-28% [18,19] and rainfall intensity by 8%-19% [20].Earlier TC attribution work also explored the influence of climate change on Hurricane Sandy's intensity and track [21] and on the storm surge associated with Typhoon Haiyan [22].Since these initial studies, event attribution analysis has been performed for a large number of tropical storms, including Hurricanes Katrina, Irma and Maria [23], Hurricane Florence [24], Typhoon Hagibis [25] and even the full 2020 North Atlantic Hurricane Season [26].However, often times it took months, even years, for these scientific studies to be completed, long after the impacts of the extreme event has been felt and public interest has waned.
This work shifts that paradigm by providing a TC rainfall attribution methodology for usage by operational attribution centers.By exploiting the hindcast attribution method [13] as used previously for the full 2020 North Atlantic Hurricane season [26], we present a rapid attribution analysis for Hurricane Ian that was performed as the storm made landfall.In the day leading up to Hurricane Ian's landfall we completed a 20-member ensemble of 'actual' forecasts initialized with the observed conditions on 27 September at 12Z over 30 h before the observed landfall on 28 September at about 19Z.In addition, we completed a 20-member ensemble of 'counterfactual' forecasts with the Community Atmosphere Model (CAM) [27] in which the large-scale fingerprint of human-induced climate change is removed from those same observed initial conditions and sea surface boundary conditions.The greenhouse gas and aerosol composition of the atmosphere were also set to preindustrial values.The methodology followed, exactly, our previous work for the full 2020 Hurricane Season analysis, except to update the initial and boundary conditions to appropriate for Hurricane Ian [26].The sea surface temperature varied between 0.6 • C-1.0 • C warmer in the Gulf of Mexico in the actual ensemble compared to the counterfactual, with the largest differences near shore.These forecasts used the same variable-resolution version CAM [28] configured with a high-resolution domain with a grid spacing of 28 km over the North Atlantic as in our previous study of the 2020 hurricane season [26].
Figure 1 displays the 3-day accumulated rainfall amounts for the actual and counterfactual initialized on 27 September 2022 the day before Ian's observed landfall.Notice that in the actual ensemble rainfall accumulations above 10 inches (about 250 mm) are simulated for a large part of Florida, including well inland, consistent with observations [16].However, the areas of heavy rain are shift northward compared to observations due to a more northward simulated landfall location, consistent with operational forecasts at similar initialization times [16].
As found in the previous TC rainfall attribution studies cited above, the impact of human-induced climate change on Hurricane Ian's simulated rainfall is clear, with larger accumulation magnitudes across Florida, including far inland from where Hurricane Ian made landfall.Large increases are also observed over the Gulf of Mexico in the ensemble with human-included climate change included.Focusing on a histogram of 3-day accumulated precipitation shown in figure 2, the actual ensemble shows a higher likelihood of 3-day accumulated precipitation amounts above 14 inches (356 mm) compared to the counterfactual ensemble.More directly, the 99th percentile 3-day accumulated precipitation is 17.9% (95% confidence interval: 15.7%-19.8%)larger in the actual ensemble than in the counterfactual without human-induced climate change.The percentage difference and 95% confidence interval was calculated using a bootstrap analysis of 10 000 samples, following the approach of the hindcast attribution method [26].Focusing on precipitation only over land (not shown), where the societal impact of extreme rainfall is most directly felt, the percentage change is the same but with a slightly larger uncertainty (95% confidence interval: 15.3%-20.7%).
While the storm was still over eastern Florida, we made the real-time attribution statement on social media of 'climate change increased Ian's rainfall by at least 10%' .This statement was deliberately conservative (hence at least) pending the more careful analysis of the model output presented here.
These percent increases in rainfall accumulations due to climate change for Hurricane Ian are consistent with previous estimates of different storms, such as Hurricane Harvey [18,19].The finding that TC rainfall increases are larger than might be expected from the Clausius-Clapeyron scaling of saturation specific humidity (∼7% • C −1 ) is robust across multiple previous studies of other TCs and is likely a result of increased wind speeds and changes in storm structure [23,29].
The impacts of Hurricane Ian were not confined to the freshwater flooding from high rainfall amounts.Saltwater flooding from the large storm surge caused extensive damage to Fort Myers, Florida region.It would be possible to ascertain the influence of climate change on storm surge by using the output of the actual and counterfactual simulations presented here as input to a storm surge model, following the approach of [8] who performed a storyline attribution of Hurricane Harvey's flood by perturbing a hydraulic flood [30] based on rainfall analyses.Wind damages can also be substantial.However, simple analyses based on Saffir-Simpson Hurricane Wind Scale metrics do not reveal that a climate change signal on instantaneous peak winds has emerged from the noise [23].However, other more integrated wind speed metrics may be more informative.

Rapid, real-time assessments of the effect of climate change on TCs
Using a previously developed and tested hindcast attribution methodology [26] in near real-time, we demonstrated that storyline rapid assessments of TC can be completed during an event.Such rapid assessment offers scientists a useful tool in answering questions about the climate change effect on individual extreme weather events as they unfold.
We note that Hurricane Ian was an ideal candidate for this type of imposed global warming framework (also known as pseudo-global warming [31]), as it path was reasonably predictable.Indeed, the stability of the storm track spread to perturbation is a requirement for this method to be useful.This is not always the case as simulated paths of some storms, such as Superstorm Sandy, are extremely sensitive to perturbations.Interpretation of the counterfactual ensemble only makes sense if the track locations are comparable to those of the actual ensemble, even if there are slight biases in the forecasts compared to the observed track.Another important factor to consider is the time of the simulation initialization.If it is too early to the period of interest, the path and magnitude of simulated storms may not resemble reality at all.If it is too close to the period of interest, the counterfactual hindcast model may not have enough time to adjust to the perturbed climate.We find there to be no hard and fast rule regarding the timing initialization, as the predictability of tropical storms varies.Rather, expert judgement, obtained by varying initialization times, must be used.
Storyline attribution methods can be useful when traditional event attribution methods [32] are not tractable.As we note in the introduction, such attribution statements are highly conditional on story definition.While validation of such attribution statements against traditional attribution statements [32] is limited, a storyline based attribution statement of Hurricane Harvey's rainfall [19] is consistent with traditional event attribution methods [18,20].Other classes of severe storms that also require high model horizontal resolution have been analyzed via storylines [33][34][35].Storylines can also inform about very rare events, such as the 2021 Pacific Northwest heatwave, as even lengthy climate model simulations are not long enough to provide meaningful statistics [36,37].Finally, combining storyline techniques with traditional event attribution methods increases confidence in attribution statements [38].
While the variable resolution CAM performed adequately in forecasting Hurricane Ian, national operational attribution centers would have access to operational forecast models and approaches, as well as superior human resource forecast expertise to interpret their output.Such expertise would increase confidence in extreme weather event attribution statements and lend an air of authority.Furthermore, the operationalization extreme weather event attribution would greatly increase the number of events and regions available for analysis, and lead to better insight into how climate change influences extreme weather.
Finally, we recognize that the distribution of funds through the planned 'loss and damage' fund of the United Nation's 27th Conference of the Parties Sharm el-Sheikh Implementation Plan [40] will be an inherently political process.However, widespread rapid extreme weather event attribution can provide important information to those decisions.

Figure 1 .
Figure 1.Ensemble average 3-day accumulated rainfall in inches (in) for the Hurricane Ian for CAM forecasts initialized on 27 September 2022 at 12Z for the (a) counterfactual and (b) actual ensembles.Each ensemble has 20 members.

Figure 2 .
Figure 2. Histogram of 3-day rainfall accumulated amounts in inches (in) associated with actual and counterfactual ensembles of the Hurricane Ian for CAM forecasts initialized on 27 September 2022 at 12Z.