Compound Climate Risk: Diagnosing Clustered Regional Flooding at Inter-Annual and Longer Time Scales

Yash Amonkar, James Doss‐Gollin, Upmanu LallHydrology 2023 · DOI

TL;DR

This study investigates how large-scale climate variability drives clustered regional flooding in the Ohio River Basin, revealing dominant inter-annual, decadal, and secular flood risk patterns. The findings highlight a significant east-to-west shift in flood incidence linked to global warming trends, suggesting a need for more sophisticated flood risk assessments that account for these compound climate risks.

What Problem Does This Paper Address?

Extreme climate events, like floods, don't happen in isolation; they often cluster in both space and time. This clustering significantly amplifies overall risk, a phenomenon known as "compound climate risk." While much research focuses on site-specific flood analysis, understanding how regional flood patterns emerge and persist due to broader climate influences has been less explored. This paper tackles this gap by investigating whether low-frequency climate variability (like ENSO or long-term warming trends) can explain the observed spatial and temporal clustering of flood risk in the Ohio River Basin. The goal is to move beyond simple site-level risk assessments to a more holistic understanding of regional flood dynamics.

Methods

The researchers employed two complementary analytical strategies to dissect the complex space-time patterns of flood risk:

  • PC-Wavelet: This method first uses Principal Component Analysis (PCA) to identify dominant spatial patterns of streamflow variability across the 24 gauging stations. Then, it applies wavelet analysis to the time series of these principal components to uncover common low-frequency variations over time. Essentially, it looks for spatial patterns first, then analyzes their temporal behavior.
  • Wave-Clust: This approach reverses the order. It begins by applying wavelet analysis to the time series of each individual stream gauge to understand its time-frequency structure. Then, it uses hierarchical clustering to group gauges with similar time-frequency behaviors, effectively identifying regions with similar flood dynamics. PCA is then applied within these clusters to find dominant modes.

Both methods rely heavily on wavelet analysis, a powerful tool for analyzing localized variations in time series across different frequencies and time scales. This allows researchers to identify not just the presence of a signal but also when and at what scale it's most prominent. To link these flood patterns to climate drivers, they used correlation analysis, linear regression (LASSO) for variable selection, and random forest models for non-linear relationships.

Datasets

The study utilized:

  • Daily Streamflow Data: From 24 USGS stream gauges located in the Ohio River Basin. These stations were selected based on criteria ensuring minimal data loss and significant regulation of flow, aiming to capture natural flood variability. The dataset covers the period from 1935 to 2021.
  • Climate Indices: Time series data for key climate drivers, including:
  • El Niño Southern Oscillation (ENSO) - specifically the Niño 3.4 index.
  • Pacific Decadal Oscillation (PDO).
  • North Atlantic Oscillation (NAO).
  • Atlantic Multidecadal Oscillation (AMO).
  • Interactions between these indices (e.g., ENSO-NAO).
  • Global annual temperature (as a proxy for anthropogenic warming trends).

Key Insights and Ideas

  • Floods Aren't Random: The paper strongly suggests that the observed clustering of extreme flood events in the Ohio River Basin is not due to chance. This implies underlying drivers that create these "epochs" of increased or decreased flood risk.
  • Low-Frequency Variability is Key: The study identifies that inter-annual (6-7 years), decadal (11-13 years), and secular (long-term) climate modes are the primary drivers of these flood clusters, rather than just short-term weather events.
  • Regional Differences Matter: Different sub-regions within the Ohio River Basin respond differently to these large-scale climate forcings, meaning a "one-size-fits-all" approach to flood risk is insufficient.
  • ENSO's Dominance (Inter-annual): The leading mode of flood variability is strongly linked to ENSO, particularly at the 6-7 year inter-annual scale, and appears to have a basin-wide impact.
  • Warming Trend's Spatial Shift (Secular): A significant finding is the association of a secular trend with an east-to-west shift in flood incidence. This suggests that as the climate warms, the storm tracks and moisture transport patterns influencing floods are changing, potentially leading to increased flood risk in western parts of the basin and decreased risk in the east. This is a crucial insight for future flood management.
  • Methodological Synergy: The PC-Wavelet and Wave-Clust methods, while having different starting points (spatial vs. temporal focus), provided consistent insights into the dominant flood patterns, reinforcing the robustness of the findings.

Key Results

  • Dominant Time Scales: Flood risk in the Ohio River Basin exhibits significant variability at inter-annual (6-7 years), decadal (11-13 years), and secular time scales.
  • PC-1: Inter-annual Variability: The first principal component (PC-1), explaining ~39% of the total variance, is strongly associated with ENSO and shows a dominant 6-7 year periodicity. This mode impacts the entire basin.
  • PC-2: Secular Trend & Spatial Shift: The second principal component (PC-2), explaining ~16% of the total variance, is characterized by a long secular trend and a west-east dipole pattern. This trend is most strongly correlated with global annual temperature, indicating a potential shift in flood incidence due to anthropogenic warming.
  • ENSO Leads PC-1: Wavelet coherence analysis revealed that ENSO leads PC-1 by approximately one year at the 12-year cycle frequency.
  • Temperature Predicts PC-2: Regression analysis showed that temperature is a key predictor for PC-2, highlighting its role in the long-term trend.

Limitations and Future Work

  • Data Constraints: The study focused on sites with relatively small drainage basins to minimize regulatory impacts. This limits the assessment of large-scale, basin-wide flood events where the entire Ohio River network might be inundated simultaneously.
  • Methodological Differences: While PC-Wavelet and Wave-Clust yielded similar results for the leading mode, they diverged for the second dominant mode, suggesting that the order of analysis (space-time vs. time-space) can influence insights.
  • Future Work: The authors emphasize the need for statistical methods that can handle multi-scale conditioning of compound extremes, specifically accounting for slower climate variations that influence synoptic event structures and lead to clustering. They also suggest that understanding the spatial structure of flood incidence over entire flood seasons, within the regional context of larger river basins, is crucial for future research and flood management. The identified east-west shift in flood risk due to warming also warrants further investigation and incorporation into climate change projections.

Summary updated

Abstract

The potential for extreme climate events to cluster in space and time has driven increased interest in understanding and predicting compound climate risks. Through a case study on floods in the Ohio River Basin, we demonstrated that low-frequency climate variability could drive spatial and temporal clustering of the risk of regional climate extremes. Long records of annual maximum streamflow from 24 USGS gauges were used to explore the regional spatiotemporal patterns of flooding and their associated large-scale climate modes. We found that the dominant time scales of flood risk in this basin were in the interannual (6–7 years), decadal (11–13 years), and secular bands and that different sub-regions within the Ohio River Basin responded differently to large-scale forcing. We showed that the leading modes of streamflow variability were associated with ENSO and secular trends. The low-frequency climate modes translated into epochs of increased and decreased flood risk with multiple extreme floods or the absence of extreme floods, thus informing the nature of compound climate-induced flood risk. A notable finding is that the secular trend was associated with an east-to-west shift in the flood incidence and the associated storm track. This is consistent with some expectations of climate change projections.
BibTeX
@inproceedings{amonkar2023compound,
  author = {Yash Amonkar and James Doss‐Gollin and Upmanu Lall},
  title = {{Compound Climate Risk: Diagnosing Clustered Regional Flooding at Inter-Annual and Longer Time Scales}},
  booktitle = {Hydrology},
  year = {2023},
  doi = {10.3390/hydrology10030067},
}
Figure