Changes in Tropical Cyclones Undergoing Extratropical Transition in a Warming Climate: Quasi‐Idealized Numerical Experiments of North Atlantic Landfalling Events

The current study extends earlier work that demonstrated future extratropical transition (ET) events will feature greater intensity and heavier precipitation to specifically consider potential changes in the impacts of landfalling ET events in a warming climate. A quasi‐idealized modeling framework allows comparison of highly similar present‐day and future event simulations; the model initial conditions are based on observational composites, increasing representativeness of the results. The future composite ET event features substantially more impactful weather conditions in coastal areas, with heavier precipitation and greater storm intensity. Specifically, a Category 2 present‐day storm attained Category 4 Saffir‐Simpson intensity in the future simulation and maintained greater intensity throughout the entire life cycle, although the storm undergoes less reintensification during the post‐ET process, a result of reduced baroclinic conversion. These findings suggest increased potential for coastal hazards due to stronger tropical cyclone winds and heavier rainfall, leading to more severe coastal flooding and storm surge.

Studies that have examined storm-scale ET changes in response to climate change find lower minimum sea level pressure (SLP) and substantially enhanced precipitation during the ET and post-ET phases in future transitioning storms (e.g., Jung & Lackmann, 2019;JL21;Michaelis & Lackmann, 2019. Specifically, JL21used a high-resolution quasi-idealized experiment based upon composites from a decades-long set of recurving oceanic (RCO) North Atlantic ET cases. It is noted that a similar strategy was employed by Grams and Archambault (2016) using a composite of Western North Pacific ET events, though they did not consider climate-change aspects. JL21 examined the responses of storm-scale ET characteristics as well as downstream impacts to climate change using small ensembles of convection-permitting, quasi-idealized simulations (Figure 1). TCs exhibiting these tracks did not directly threaten the United States and recurved into the open ocean, though some affected eastern Canada and Western Europe. They found that future ET events featured greater intensity, heavier precipitation, and stronger downstream Rossby wave train development relative to present-day counterpart simulations. While the future composite ET was substantially stronger over almost the entire ET and post-ET phases, it underwent less reintensification during the post-ET process due to a reduction in lower-tropospheric baroclinic conversion associated with Arctic amplification of warming.
While JL21 provides insight and better understanding into the evolution and underlying mechanism of future North Atlantic ET events, they did not take into account the storm-scale changes brought about by ET events affecting coastal areas or making landfall, which can be accompanied by heavy rain, damaging wind, and storm surge to the East Coast of the U.S. Here, we extend the analyses of JL21 by examining recurving landfalling (RCL) TCs (Figure 1), again using a composite-initialized, convection-permitting, quasi-idealized experimental design. The purpose of this study is to assess changes in risk and storm impacts for future landfalling ET events along the East Coast of the United States and eastern Canada. Recurving landfalling (RCL) TCs, shown as light pink tracks, cross the thick red boundary and threaten the U.S. East Coast. Recurving oceanic (RCO) TCs, shown as light blue tracks, did not directly threaten the United States. (a) The selected 21 RCL and 21 RCO TC tracks for use in compositing over the 40-year reanalysis period are shown in red contours and light blue contours, respectively, with dots indicating the location of the TC at 6-hourly intervals. Comparison of present-day simulation (thick blue lines) to future simulation (thick red lines) for (a) track, (b) minimum sea level pressure (hPa), and (c) maximum near-surface wind speed (m s −1 ). The onset and completion of ET are labeled ET B for the onset and ET E for the completion of the transition.
In Section 2, we describe the model simulations, data sets, and analysis methods. Section 3 presents projected future changes in intensity and precipitation of the composite North Atlantic RCL events from both synoptic scale and storm-scale points of view; Section 4 concludes with a summary of key results.

Data and ET Classification
Following JL21, we apply the track-based classification method of Colbert and Soden (2012) to the historical database  of North Atlantic RCL ET events; recurving landfalling TCs are defined as those which cross 70°W north of 25°N or cross 65°W north of 40°N and threaten the U.S. East Coast (Figure 1). We use the Atlantic Hurricane Database (HURDAT2) best-track data (Landsea & Franklin, 2013); the track-based classification procedure yields 37 RCL ET cases over the 40-year period. To minimize the spread of the selected ET cases in terms of time to complete transition, we apply a threshold of 36 hr as in JL21. As HURDAT2 only specifies when transition is complete but does not provide the timeline of the transition, we supplement it with the cyclone phase space (CPS) method of Hart (2003) to diagnose onset and completion times of ET; ET events are first identified by HURDAT2 and then we compute the CPS parameters to confirm the ET timeline for each event using the ERA5 reanalysis data set (Hersbach et al., 2020). This filtering process yields 21 RCL ET cases (Table S1 in Supporting Information S1).

Experimental Design and Numerical Model Simulations
The method of using quasi-idealized, convection-permitting model experiments with initial and lateral boundary conditions based on composites of past ET events is again used here. To generate an initial-condition ensemble, the simulations are initialized by randomized 15-case composites selected from the 21 RCL ET cases as in JL21 (Text S1 in Supporting Information S1).
Using the Weather Research and Forecasting (WRF) version 4.0 (Skamarock et al., 2019), we design the simulations with realistic land and orography to assess plausible impacts of landfalling ET events along the U.S. East Coast. We compute SST in an Earth-relative sense in order to reflect realistic ocean currents, such as the Gulf Stream, unlike the JL21 study where SST was composited in storm-relative coordinates. To simulate realistic TC structure, such as TC secondary circulations, we use a two-way, convection-permitting vortex-following nest at 4-km grid spacing within an outer domain featuring 12-km grid spacing. The dimension of the vortex-following nest is 1,636 km × 1,636 km, large enough to capture the transitioning system within the nest. The model top is set at 50 hPa with 50 vertical levels. Details on the model physics schemes are provided in Supporting Information S1 (Text S2).
For the projected future simulation, we use the pseudo global warming (PGW) approach that is nearly identical to that of Jung and Lackmann (2019) and JL21. Briefly, this method uses a 20-GCM ensemble to compute thermodynamic differences in July-September climatological averages derived from phase 5 of the Coupled Model Intercomparison Project (CMIP5). Thermodynamic differences are calculated by subtracting the 1980-1999 average from the 2080-2099 average following the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5) Representative Concentration Pathway (RCP) 8.5 emission scenario (Meinshausen et al., 2011). The thermodynamic differences (i.e., "delta" values), such as 2-m air temperature, SST, soil temperatures, and atmospheric temperature at each isobaric level are then added to the present-day composite initial and lateral boundary conditions. Relative humidity is held constant, resulting in increased specific humidity when warming is applied.
Use of past event composite initial conditions and the PGW method introduces some degree of imbalance in the initial fields. Therefore, we apply the WRF digital filter initialization (DFI) to all our simulations, which helps to balance the wind and mass fields at the initial time using adiabatic backward and diabatic forward time integration to enable application of the filter (e.g., Lynch & Huang, 1992, 1994Peckham et al., 2016). The DFI also populates hydrometeor arrays and reduces spin-up time.

Defining ET Onset and Completion
The CPS technique is a popular objective method to diagnose the onset and completion of ET in research and operational fields (J. L. Evans & Hart, 2003;Hart, 2003). We apply this technique to determine the onset time of ET; the onset time of ET is defined as the time when the thermal asymmetry of the storm starts to develop (see JL2021 for more details). However, for our high-resolution simulations, we find, consistent with Sarro and Evans (2022), that the CPS method of diagnosing ET completion has difficulty owing to the presence of a lingering warm core structure. Thus, we use an experimental energy-based method similar to that described in JL21, who diagnosed ET completion using metrics representing tropical and extratropical cyclone energy sources: surface latent heat flux and baroclinic conversion. The baroclinic conversion of eddy available potential energy (EAPE) to eddy kinetic energy (EKE) is defined as: is the dry-air gas constant, ′ is the temperature deviation, and is pressure (see Text S3 in Supporting Information S1 for details). Using a spatial average of the two metrics within 1,000-km radius relative to the storm center, this method defined ET completion as the time when the value of baroclinic conversion exceeds the value of surface latent heat flux; baroclinic conversion is a major energy source for the storm from this time onward. Although this method is reasonable for the transitioning TCs that move over the ocean, its application to landfalling TCs undergoing ET is complicated by large diurnal variations of surface latent heat flux over the land. To account for this, we apply linear regression to identify an increasing trend for baroclinic conversion and a decreasing trend for surface latent heat flux ( Figure S1 in Supporting Information S1). Based on examination of 12 North Atlantic RCL ET events, the completion of ET is determined by the time when the value of linear regression of baroclinic conversion exceeds that of surface latent heat flux by 100 W m −2 . The diagnosed ET completion time through this method is evaluated by subjective identification technique; see Text S4 in Supporting Information S1 for details. We emphasize that this is an experimental method of ET detection, and additional testing is required to evaluate its general applicability.

Initial Condition Sensitivity
Before the model simulations can be used to analyze future projections of ET features, it is necessary to examine how the initial condition variability impacts the results. To this end, two sets of simulations with different initial conditions, initialized by a randomized 15-case composite out of the 21 RCL ET cases, using the identical physics schemes, are run and compared to examine initial condition variability. We find that differences in track, intensity, and storm evolution are small between the two sets of experiments, implying that the evolution of the simulated TC and its interaction with environmental fields is robust to small-scale differences arising from the compositing process ( Figure S4 in Supporting Information S1); see Text S5 in Supporting Information S1 for details on the comparison.

Changes in Coastal ET Impacts
We analyzed several of the same processes presented in JL21 for these simulations, including duration of ET and large-scale energy transfer processes. The results of these calculations in the RCL simulations are consistent with those of the RCO simulations and will not be presented in detail here; see Jung (2020) for these results. The RCL runs show no change in ET duration, and enhanced future downstream energy dispersion as in JL21. In the remainder of this section our focus is on changes in near-shore storm impacts.

Near-Surface Winds
Both future and present simulated TCs undergo a period of reintensification after ET completion; the future storm indicates deeper minimum central sea level pressure over the entire simulation period (Figures 1b and 1c). Rapid deepening is evident during the TC stage in both simulations prior to landfall in coastal North Carolina; the storms become weaker due to TC-land interaction after this point. The future and present-day TCs both travel over the warm waters of the Gulf Stream ( Figure S6a in Supporting Information S1) and this may help them to maintain intensity as they move northward (Figures 2a and 2b). Frictional effects associated with TC-land interaction could delay or reduce reintensification compared with the RCO simulations presented in JL21. While the future storm reaches ∼16 hPa deeper in minimum SLP than that in the present-day simulation, the substantial difference in minimum SLP between the future and present-day storms persists during the TC phase but declines to ∼2 hPa by the end of the simulations (Figure 1b). This implies that the future storm again undergoes weaker reintensification after ET completion, as was the case for the RCO simulations. To support this speculation, Hovmöller diagrams for column-integrated conversion of APE to EAPE are displayed and they demonstrate stronger conversion in the vicinity of the present-day storm compared with the conversion in the future storm during ET and post-ET processes ( Figure S7 in Supporting Information S1). Given that the reintensification process is closely linked to the conversion of EAPE to EKE (e.g., JL21), this suggests that Arctic amplification ( Figure S6c in Supporting Information S1) results in a reduced APE reservoir, weakening conversion of APE to EAPE and thus, weakening the reintensification process, consistent with JL21 (see Text S6 in Supporting Information S1 for details). The difference in EAPE generation can occur because of a delayed interaction with the midlatitude flow (e.g., a poleward-shifted jet stream in a warmer climate). Our PGW method, however, imposes highly similar synoptic patterns in the present and future, which largely eliminates the role of such circulation changes from consideration.
To provide information regarding changes in hazardous near-surface conditions, 10-m wind speed is examined. An expansion of the radius of gale-force winds is obvious in both future and present-day storms as the storms move to higher latitudes (Figures 2a and 2b). For the future simulation, considerably stronger wind fields are evident, consistent with decreased SLP (Figures 1b and 2b). Specifically, the future storm is accompanied by 10-m wind speeds exceeding ∼30 m s −1 (yellow shading) to nearly 50°N latitude, along the coast of Newfoundland (Figures 2b and 2c).
The examination of maximum simulated 10-m wind speed on the U.S. East Coast with close-up figures reveals a significant increase in near-surface winds in the coastal zone, as depicted in Figures 2d-2f, highlighting the intricate changes in potential impacts on coastal areas. The difference in the near-surface wind speed associated with the transitioning storm between the future and present-day simulations exceeds ∼15 m s −1 over the ocean and ∼10 m s −1 along the East Coast (Figure 2f). Specifically, the maximum sustained near-surface wind speed of the present-day storm is ∼44 m s −1 , which corresponds to Category 2 in the Saffir-Simpson Hurricane wind scale while the future storm attains a maximum wind speed of ∼59 m s −1 , a Category 4 hurricane.
In order to examine changes in storm-scale characteristics, we use the 4-km convection-permitting vortex-following nest domain. The 10-m wind field derived from the nest domain displays a clear distinction in storm structure between the present-day and future storms; the future storm becomes noticeably stronger in terms of 10-m wind (and vertical velocity, not shown) with a smaller eye and a reduced radius of maximum wind (Figures 3a-3f and Figure S8 in Supporting Information S1). A smaller eye size in the future simulation could be consistent with mechanisms proposed by Shen (2006): A reduced area of strong winds leads to a decrease in surface frictional dissipation, reducing the offset of kinetic energy generation associated with the surface entropy flux, and thus a stronger storm. Of course, increased entropy flux from a warmer SST is likely the main reason for the stronger future storm.

Precipitation
Changes in 3-hourly precipitation with varying radius and life-cycle stage show that the future storm produces more precipitation than the present-day storm regardless of averaging radius and stage of transition, consistent with prior studies (Table 1). In addition, an outward shift of maximum precipitation rate is evident for both storms as they undergo ET, experiencing an expansion of asymmetric rainbands as previously shown (Table 1).
Comparison of precipitation change to the rate of water vapor increase predicted by the Clausius-Clapeyron (C-C) relation, and to the analyzed vapor increase, provides meaningful thermodynamic context. Future changes in precipitation relative to C-C scaling vary with averaging radius. An increase in precipitation is evident in the future simulation at all averaging radii, but precipitation increases at a rate which substantially exceeds C-C scaling in the inner-and mid-core regions (Table 1). Increases are close to that predicted by C-C scaling in the outercore regions regardless of TC phase (Table 1). For the 100-km inner-core region, the precipitation change greatly exceeds both the vapor increase and the C-C scaling for vapor increase, with rainfall doubling that predicted by C-C scaling. The computed average vapor increase also exceeds C-C scaling at all radii, but by much smaller margins relative to the increase in precipitation in the inner-core regions.
Temporally and spatially averaged differences in column-integrated moisture convergence, mass convergence, and specific humidity between the future and present-day simulations show that while the future storm consistently 8 of 10 exhibits increases in moisture convergence and specific humidity over the present-day simulation at all radii, a decrease in mass convergence is evident in the outer-core region (Table S2 in Supporting Information S1). This result is consistent with the intensified minimum SLP and strengthened radial inflow through the enhanced secondary circulation in the future storm; this circulation transports more moisture to inner-core regions, contributing to an increase in precipitation at a super Clausius-Clapeyron rate there.
Accumulated precipitation analysis also reveals substantially heavier precipitation associated with the transitioning storm along the East Coast in the future simulation. Specifically, regions where the accumulated precipitation increases by more than 120 mm compared to the present-day extend from eastern North Carolina all the way up to eastern Long Island, New York (Figures 2e-2j). Simulated radar reflectivity derived from the vortex-following domain is consistent with the accumulated precipitation plots, demonstrating that the future storm delivers heavier precipitation over a larger area compared with its present-day counterpart (Figures 3g-3l).

Summary
In this study, we extend the previous analysis of JL21 of North Atlantic ET and climate change by examining a set of recurving landfalling ET cases, and by shifting our focus to impact-based quantities in nearshore regions. Using 4-km convection-permitting simulations derived from a composite of North Atlantic RCL ET cases, we explore future changes in ET and landfall or postlandfall impacts of ET. Key findings include: 1. Substantially stronger near-surface winds attributed to the intensified transitioning storm are evident, directly impacting the U.S. East Coast in the future. Specifically, a Category 2 present-day event becomes a Category 4 storm in Saffir-Simpson scale in a warmed environment. 2. The intensity of the future storm remains higher than its present-day counterpart throughout its entire life cycle, however, it experiences reduced reintensification during the post-ET phase due to Arctic amplification. 3. Accumulated precipitation analysis features substantial future increases in total rainfall associated with the transitioning storm all along the East Coast. Specifically, regions where the accumulated precipitation increased by more than 120 mm extend along the East Coast from eastern North Carolina up to eastern Long Island, New York. 4. For the innermost 100-km TC-centered averaging radius, rainfall increased at a rate more than double that predicted by Clausius-Clapeyron scaling.
Furthermore, the combination of increased inner-core wind speed and lower surface pressure implies an increase in storm surge, which for this hypothetical quasi-idealized simulation would be dire for the barrier islands of North Carolina. Substantially heavier rainfall could contribute to increased nearshore flooding in compound events near rivers and estuaries.
The pseudo-global warming modeling strategy has some limitations, such as the inability to analyze changes in storm frequency (see, e.g., Lackmann, 2015). However, an advantage is the ability to isolate and quantify storm-scale changes due to differences in the thermodynamic environment. The use of composite-based initial conditions allows us to generalize results for this "generic" storm, which has elements of historical cases but without details present in specific case studies. It could be that some of these details are important to the storm impacts, and in this sense, our approach can be considered a conservative estimate of changes in ET events due to climate warming. Finally, we suggest that the composite approach could be useful for other extreme-weather attribution studies of the type discussed by Faranda et al. (2022).

Data Availability Statement
All data sets used in this study are freely available. The ERA-5 reanalysis data were obtained from Climate Data Store (CDS; available at https://cds.climate.copernicus.eu/). The HURDAT2 data are retrieved from https://www. nhc.noaa.gov/data/. CMIP6 model outputs are freely available from the World Climate Research Programme (WCRP), 2011: https://esgf-node.llnl.gov/projects/esgf-llnl/. The source code for WRF is freely available from https://github.com/wrf-model/WRF. The WRF model output from the simulations presented in this paper is located on the North Carolina State University Hazel cluster. Output that can create figures and tables used in the manuscript is available https://doi.org/10.5061/dryad.7sqv9s4x5 (Jung & Lackmann, 2023).