The devastating series of nine atmospheric rivers (ARs) that hit California from late December 2022 to mid-January 2023 caused significant damage, including flooding, landslides, power outages, and fatalities. This event underscores the critical need to better understand and predict such compounding weather events, known as AR clusters. AR clusters, defined as multiple AR landfalls within a short period, pose heightened flood risk due to the lack of sufficient time for soil moisture and other land properties to recover between events. While previous research has touched upon AR families and subseasonal clustering, a comprehensive understanding of AR cluster activity and future projections remains limited. This research aims to address this gap by investigating the variations in AR cluster characteristics, circulation patterns, and impact levels. The central hypothesis is that AR cluster impacts vary significantly, influenced by factors such as landfall duration, spacing between events, and event intensity. This study builds on past work by integrating unique analytical techniques like cluster density and incorporating future climate projections.

Past studies have examined the impact of antecedent soil moisture and snowpack on runoff during ARs, showing higher runoff-to-precipitation ratios under preconditioned land surfaces. The concept of "AR families," representing periods of multiple AR landfalls, has been defined, along with analysis of associated synoptic patterns modulated by El Niño Southern Oscillation (ENSO). Research has also explored the subseasonal clustering of ARs across the western US, highlighting regional differences in cluster counts. While existing literature indicates a projected increase in AR intensity under warmer climates, research specifically focused on future changes in temporal clustering of ARs remains limited. Studies have shown increased intensity and duration of consecutive ARs with rising temperatures, leading to unprecedented monthly precipitation totals by the end of the century. Focusing on specific events like the 2017 Oroville Dam crisis, studies have projected increased precipitation from consecutive ARs under future climate scenarios. However, research suggests a varied response of AR precipitation to global warming, depending on the association with extratropical cyclones.

This study employs a data-driven, unsupervised machine learning algorithm called "Mean Shift" to identify AR clusters from landfalling ARs along the US West Coast. The algorithm automatically identifies temporal clusters from a time series of AR landfall flags. AR clusters are categorized as "solitary clusters" (single AR landfalls) and clusters with multiple AR landfalls. Cluster density is defined as the ratio of AR days to total cluster days. Clusters are further divided into "dense clusters" (≥50th percentile density) and "sparse clusters" (<50th percentile density), excluding solitary clusters. The analysis utilizes 6-hourly averaged integrated vapor transport (IVT) data from ERA5 reanalysis (1979-2023) to detect ARs using the Toolkit for Extreme Climate Analysis Bayesian AR Detector (TECA-BARD). Additional AR detection algorithms (Guan and Waliser (GW15) and Rutz, Steenburgh, and Ralph (RSR14)) were used to assess detection uncertainty. The study also uses ERA5 500 hPa geopotential height and 850 hPa winds to analyze circulation patterns associated with AR clusters. Extreme precipitation data from the Climate Prediction Center (CPC), soil moisture and runoff data from the National Centers for Environmental Prediction (NCEP) North American Regional Reanalysis (NARR), and PRISM gridded observations are used to assess impacts. To validate results, the study examined ARs in CMIP5/6 multi-model ensembles (ARTMIP) and CESM2-LENS simulations. Empirical Orthogonal Function (EOF) analysis was conducted on daily 500 hPa geopotential height anomalies to understand the interaction between AR clusters and large-scale circulation variability. Finally, pseudo-PCs were generated by projecting historical CESM2-LENS EOFs onto its future simulations (SSP370) to assess future changes in AR cluster characteristics under various warming levels (+1°C, +1.5°C, +2°C, +3°C, +4°C).

The analysis reveals significant differences in AR cluster characteristics based on density. Dense clusters show higher AR frequency, with AR frequency being 100-150% higher than in sparse clusters along the US West Coast (35° to 50°N). Dense clusters are more active from October to January, while sparse clusters are active throughout the season. The average AR category increases with increasing density, indicating a higher likelihood of high-category ARs occurring in close temporal proximity. Dense clusters contribute significantly to extreme precipitation (20-50% of top 2% precipitation days), whereas sparse clusters contribute less (10-20%). Precipitation intensity is significantly higher in dense clusters. Analysis of soil moisture and runoff shows that both dense and sparse clusters increase soil moisture saturation and runoff, but the increase is substantially larger for dense clusters (nearly 100% more in northern California and 150-300% more over the Cascade Range). The study identifies distinct large-scale circulation patterns associated with dense and sparse clusters. Dense clusters are linked to a tri-pole 500 hPa geopotential height anomaly pattern over the North Pacific, inducing cyclonic flow favorable for moisture transport from the subtropics. Sparse clusters are associated with a south-north dipole pattern and a zonally oriented eastward flow. Subseasonal variability is the dominant temporal frequency modulating AR cluster activity. EOF analysis identified three dominant modes of 500 hPa geopotential height variability. EOF3, strongly correlated with dense clusters, shows a tri-pole pattern similar to the dense cluster circulation pattern. The relationship between cluster density and PC3 is robust across historical and future simulations. Future climate projections (CESM2-LENS SSP370) suggest a decrease in the averaged pseudo-PC1, an increase in pseudo-PC2, but no significant trend in pseudo-PC3. While the linkage between AR clusters and the first two EOF modes weakens with warming, the relationship with EOF3 (linked to high-density clusters) is maintained. However, the number of high-density clusters increases significantly at higher warming levels (+4°C).

The findings highlight the importance of AR cluster density as a critical factor influencing the severity of AR impacts. The distinct circulation patterns associated with dense and sparse clusters offer valuable insights into the atmospheric dynamics driving these events. The continued strong relationship between EOF3 and high-density clusters in future climate projections suggests that the risk of severe, back-to-back AR events may increase with warming, despite potential changes in the overall frequency of ARs. This study emphasizes the need for improved forecasting capabilities that account for the clustering behavior of ARs and their association with large-scale circulation patterns. The results have significant implications for the development of climate adaptation and resilience strategies, particularly in regions vulnerable to AR-related flooding and extreme precipitation.

This study demonstrates that the impacts of atmospheric river clusters vary significantly depending on their density. High-density clusters are associated with more frequent high-category events, extreme precipitation, and greater land surface impacts. Subseasonal variability plays a major role in modulating AR cluster activity. While the relationship between AR clusters and some large-scale circulation patterns may weaken under warming, the connection to high-density clusters remains robust, suggesting increased risk of severe, back-to-back events in the future. These findings are critical for developing effective climate adaptation and resilience strategies.

The study acknowledges uncertainties associated with AR detection algorithms. While using multiple algorithms helps mitigate this, differences in the fraction of time steps identified as ARs exist. The study focuses on the US West Coast and its findings may not be directly generalizable to other regions. Future research could explore the spatial variability of AR cluster characteristics and investigate the impacts of other factors, such as topography and land use, on the severity of AR cluster events.