The eastern United States is highly susceptible to floods due to various factors including heavy precipitation, snowmelt, and cyclones. With climate change and population growth expected to increase flood exposure, understanding how flood characteristics (magnitude, frequency, duration) will change is crucial for effective flood preparedness and mitigation of economic losses and casualties. While research has focused on flood magnitude and frequency, the impact of climate change on flood duration remains understudied. This study addresses this gap by examining historical flood duration data and projecting future changes using global climate models (GCMs) and Shared Socioeconomic Pathways (SSPs). The study focuses on minor and moderate flood events due to data limitations associated with major flood events. The goal is to develop a robust understanding of how climate change will influence the duration of flooding, a crucial factor in assessing overall flood risk and implementing effective mitigation strategies.

Existing literature primarily focuses on the magnitude and frequency of annual maximum peak discharges to understand flood changes. However, this approach has limitations, as it may include values too small to represent actual flooding. Analyzing peaks-over-threshold (POT) data, exceeding a specific flood magnitude, offers a valuable alternative to capture changes not evident in annual maximum series. While some studies have examined trends in POT-driven flood characteristics, extrapolating these findings for future projections is challenging because observed trends may not persist. To address this, statistical relationships between flood characteristics and physical drivers are established, allowing the use of projected drivers to assess future flood characteristics. This study builds on this approach, focusing specifically on flood duration, a previously under-researched aspect of flood risk.

The study uses data from 378 USGS streamgages across the eastern US, focusing on minor and moderate flood events as defined by NWS severity levels. Binomial regression models were developed for each season (fall, winter, spring, summer) to describe the observed changes in flood duration. Basin- and season-averaged precipitation and temperature were used as predictors, accounting for both concurrent and lagged (previous season) values to capture the impact of antecedent conditions. Model performance was assessed using Spearman correlation coefficients between observed and modeled flood durations, and cross-validation was conducted using leave-one-out techniques. To project future changes, precipitation and temperature data from 36 CMIP6 GCMs were used as inputs to the developed statistical attribution models, under four SSPs (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5). Empirical quantile mapping was applied to correct biases in GCM outputs. GCM suitability was evaluated by comparing historical trends in the occurrence probability of flooding from GCMs and PRISM data. Finally, a Student's t-test was used to assess significant shifts in flood duration between historical (1985-2014) and future (2071-2100) periods.

The statistical attribution models showed good performance in capturing interannual variability in flood duration, with median correlation coefficients between observed and modeled durations around 0.6. Concurrent precipitation was consistently selected as the most important predictor across all seasons and locations, indicating its dominant role in driving flood duration. Other variables (lagged precipitation and temperature) showed more regional and seasonal variability in their influence. The analysis of projected changes in flood duration revealed a distinct seasonal difference. For both minor and moderate flooding, winter showed the highest percentage of sites with projected increases in flood duration, ranging from 18.7% to 53.3% depending on the emission scenario. This contrasts with spring and summer, which show a higher percentage of sites with projected decreases in flood duration. Higher emission scenarios generally lead to more sites experiencing longer annual flood durations. Both reference and non-reference sites exhibited similar patterns in projected changes, suggesting that the impact of human modifications on basins is less significant compared to climate drivers. The analysis of 36 CMIP6 GCMs revealed that 15 GCMs successfully reproduced the observed historical flood trends, forming the basis for ensemble projections of future flood duration.

The findings highlight the significant role of climate change in influencing flood duration across the eastern US. The consistent selection of concurrent precipitation as the dominant predictor underscores the importance of rainfall in flood generation. The projected increase in flood duration, especially in winter under higher emission scenarios, emphasizes the need for enhanced flood preparedness and mitigation strategies. The seasonal differences in projected changes warrant region-specific adaptation measures. The overall results support the need for both climate change mitigation to reduce greenhouse gas emissions and adaptation measures such as raising dikes or creating detention areas to manage longer flood durations.

This study provides valuable insights into the projected changes in flood duration in the eastern US, emphasizing the significant influence of climate change. The findings show that floods are projected to last longer, particularly in winter and under higher emission scenarios. This information is crucial for effective flood risk management and adaptation strategies. Future research could explore the socioeconomic impacts of prolonged flood durations and investigate the effectiveness of different adaptation measures in the context of climate change.

One limitation is the minimum requirement of five years of flood occurrence for model fitting, potentially leading to an underestimation of projected increases in flood duration. Another limitation is the reliance on basin-averaged climate variables which may not fully capture the spatial heterogeneity of flood generation processes within individual basins. The study primarily focuses on minor and moderate floods, limiting the analysis of major floods due to data scarcity.