Coastal flooding from storm surges, driven by extratropical cyclones (ETCs) and tropical cyclones (TCs), poses a significant threat to coastal communities worldwide. Approximately 100 million people live below current high tide lines, placing substantial socioeconomic activity and infrastructure at risk. High-impact events like Hurricanes Dorian and Michael, and storms Xaver and Ophelia, highlight the devastating consequences. While ETCs are prevalent in mid-latitudes, TCs affect tropical regions, and some areas experience both. Accurate assessment of storm tide return periods (RPs), combining surge and tide, is crucial for flood risk management and designing coastal defenses. Existing global-scale studies on storm tide RPs suffer from inaccuracies, particularly in TC-prone regions, due to two main limitations: (1) coarse resolution meteorological data leading to TC intensity underestimation, and (2) short record lengths (typically 40 years) insufficient for robustly estimating low-probability TC events. While improvements in climate modeling (e.g., ERA5) have occurred, the record length remains short for high RPs in TC-prone areas. Regional studies have used synthetic TC tracks to extend historical records, but a global-scale study integrating this approach has been lacking. This study fills this gap by creating a new global dataset, COAST-RP, which combines TC and ETC storm tide RPs, addressing the limitations of previous studies. The methodology involves separate modeling of TC and ETC-induced surges using different meteorological forcing: STORM dataset synthetic TCs for TCs and ERA5 for ETCs. The extreme value analysis employs an empirical approach based on large sample sizes rather than fitting extreme value distributions to estimate rare storm tide RPs. Finally, the study investigates how fully including TCs impacts global population exposure to low-probability coastal flooding.

Numerous studies have examined coastal flooding risks, assessing vulnerabilities and developing adaptation strategies. Global-scale analyses have modeled coastal flooding and societal risks under various scenarios, including sea level rise and socioeconomic development. However, these studies often lack accuracy in TC-prone regions due to limitations in meteorological data resolution and record length for TC events. Regional studies employing synthetic TC track generation have offered improved estimates, but their scope remained regional. The novelty of this study lies in applying the synthetic TC approach globally, creating a comprehensive dataset that incorporates both TC and ETC effects. Previous datasets like GTSR have underestimated the global exposure to coastal flooding, particularly in regions frequently impacted by tropical cyclones. The study utilizes these existing studies as a basis for comparison and aims to improve upon the accuracy of global flood risk assessments.

The study developed COAST-RP, a combined global dataset of TC and ETC storm tide RPs. The methodology consists of three steps: (1) Separate simulation of TC surge levels, ETC surge levels, and tidal levels; (2) Stochastic combination of TC and ETC surge levels with random tidal levels to obtain storm tide levels; (3) Calculation of exceedance probabilities for TC and ETC storm tides using Weibull's plotting position formula and probabilistic combination to obtain storm tide RPs. The Global Tide and Surge Model (GTSM) was used for hydrodynamic simulations. TC-induced storm surges were simulated using the STORM dataset, generating synthetic TC tracks representing 3000 years of current climate conditions. The Holland parametric wind model was used to derive wind and pressure fields from STORM data, with several adjustments implemented to optimize computation time and accuracy. A bias correction was applied to TC storm surge levels, comparing results against observed TC data from IBTRACS to account for potential overestimation or underestimation in certain regions. ETC-induced storm surges were simulated using 38 years of data from ERA5, with TC-induced surges removed to avoid double-counting. The extreme value analysis employed an empirical approach, generating stochastic storm tide event sets and estimating low exceedance probabilities from the large sample size. To validate COAST-RP, the study compared its storm tide RPs with regional studies in Australia and the United States, showing good agreement. Finally, flood exposure was assessed using a static flood model with MERIT elevation data and GPWv4 population data for 2020. The model incorporates water level attenuation and assumes no coastal flood protection.

The study's key findings include: (1) TC-induced storm surges show RP25 exceeding 2.0 m in several regions (e.g., Yellow Sea, Gulf of Carpentaria, Bay of Bengal, US Gulf Coast). Differences between STORM-based and IBTRACS-based RP25 estimates were mostly within 0.25 m, though larger discrepancies were observed in some regions, which was attributed to the inherent stochastic nature of TCs and the limited length of the IBTRACS record. (2) ETC-induced storm surges show RP25 exceeding 2.0 m in other regions (e.g., Argentina, Uruguay, Australia, Arctic Ocean, North Sea). Removing TC surges from ERA5 data significantly impacted RP25 estimates, particularly in regions also affected by TCs. (3) The global RP1000 storm tide level exceeded 5.0 m at 3% of locations, with different forcing mechanisms (TCs, ETCs, tides) driving high storm tides in different areas. The crossover RPs, where TC storm tide levels surpass ETC levels, varied geographically, highlighting the importance of considering both TCs and ETCs. (4) Validation against regional studies showed good agreement between COAST-RP and existing research. (5) The global population exposed to ETC RP1000 flood levels was 77.8 million, rising to 191.6 million when TCs were included. Exposure varied significantly among countries, with Bangladesh, India, China, and Vietnam highly exposed. The contribution of TCs to the total exposure was large in several countries. (6) Comparing COAST-RP-based exposure estimates to previous studies (e.g., Aqueduct, GTSR), the study found a 31% higher exposure estimate for RP1000 floods. This indicates a significant underestimation of global exposure in previous studies due to the inadequate accounting of low-probability TCs.

The study's findings address the research question by demonstrating the significant impact of incorporating low-probability TCs in global coastal flood risk assessments. The substantially higher exposure estimates (a 31% increase for RP1000 floods) compared to previous studies highlight a critical underestimation of global risk. This underestimation is attributed to the limitations of previous datasets and methodologies in fully capturing the low-probability, high-impact nature of TCs. The results underscore the necessity of using high-resolution data, advanced modeling techniques, and robust statistical methods in assessing coastal flood risk. The COAST-RP dataset provides a substantial improvement by utilizing synthetic TC data and a comprehensive modeling framework. This work offers valuable insights into global coastal flood vulnerability, informing risk management strategies and adaptation policies. The differences between COAST-RP and other datasets, particularly for higher RPs, highlight the limitations of previous studies and the improved accuracy of the current approach.

This study presents COAST-RP, a novel global dataset of storm tide RPs that for the first time comprehensively accounts for low-probability TCs. The use of synthetic TCs and a robust statistical method enabled accurate estimation of high return periods. The analysis revealed a significant increase in the global population exposed to coastal flooding when TCs are included, highlighting a previous underestimation of flood risk. Future research will apply this framework to future climate scenarios, considering the projected changes in TC and ETC intensity and frequency. Further improvements to the model may incorporate tide-surge interactions, wave setup, coastally trapped waves, and mean sea level variability.

Several limitations exist: the STORM dataset's statistical resampling doesn't account for all physical TC characteristics; the Holland wind model simplifies TC structure; coastal processes (tide-surge interactions, wave setup, coastally trapped waves, mean sea-level variability) are simplified. These simplifications might lead to overestimation or underestimation of storm tide RPs in certain regions. While a bias correction was applied, some inaccuracies remain. The use of a 2D barotropic model also limits the accuracy of the results. Despite these limitations, COAST-RP represents a significant advance over previous datasets and improves large-scale flood risk assessments.