Space-time clustering of flash floods in a changing climate (China, 1950-2015)

Nan Wang, Luigi Lombardo, Marj Tonini, Weiming Cheng, Liang Guo, Junnan XiongarXiv 2020 · DOI · arXiv

Abstract

The persistence over space and time of flash flood disasters -- flash floods that have caused either economical or life losses, or both -- is a diagnostic measure of areas subjected to hydrological risk. The concept of persistence can be assessed via clustering analyses, performed here to analyse the national inventory of flash floods disasters in China occurred in the period 1950-2015. Specifically, we investigated the spatio-temporal pattern distribution of the flash floods and their clustering behavior by using both global and local methods: the first, based on the Ripley's K-function, and the second on scan statistics. As a result, we could visualize patterns of aggregated events, estimate the cluster duration and make assumptions about their evolution over time, also with respect precipitations trend. Due to the large spatial (the whole Chinese territory) and temporal scale of the dataset (66 years), we were able to capture whether certain clusters gather in specific locations and times, but also whether their magnitude tends to increase or decrease. Overall, the eastern regions in China are much more subjected to flash floods compared to the rest of the country. Detected clusters revealed that these phenomena predominantly occur between July and October, a period coinciding with the wet season in China. The number of detected clusters increases with time, but the associated duration drastically decreases in the recent period. This may indicate a change towards triggering mechanisms which are typical of short-duration extreme rainfall events. Finally, being flash floods directly linked to precipitation and their extreme realization, we indirectly assessed whether the magnitude of the trigger itself has also varied through space and time, enabling considerations in the context of climatic changes.

Introduction

Flash floods are among the most lethal natural hazards because they occur rapidly, often leaving mere hours for evacuation. While intense rainfall is the primary trigger, the "persistence" of these disasters—their tendency to aggregate in specific regions during specific time windows—is a critical diagnostic of hydrological risk. Most previous studies have analyzed floods as either spatial or temporal events in isolation. This research addresses the challenge of spatio-temporal clustering to identify high-risk zones and determine if climate change is altering the nature of these disasters.

Methodology

Researchers employed a two-step statistical approach to move from a broad overview to specific local hotspots:

  1. Global Analysis (Ripley’s K-function): A spatio-temporal version of the Ripley’s K-function was used to quantify the interaction between space and time across the entire Chinese territory, determining if floods were distributed randomly or showed significant "clumpiness."
  2. Local Analysis (Spatio-temporal Permutation Scan Statistics): Using SaTScan software, a "moving cylindrical window" (where the base represents space and the height represents time) was applied to pinpoint statistically significant clusters where flood frequency exceeded chance expectations.

Data Sources

The study utilized two primary datasets:

  • Flash Flood Disaster Inventory (1950-2015): A subset of 32,473 events with precise timestamps and coordinates collated by the Chinese Institute of Water Resources and Hydropower Research.
  • Rainfall Data: Daily records from 699 rain gauges (1955-2015) provided by the China Meteorological Administration to correlate clusters with climatic shifts.

The analysis revealed a striking evolution in cluster behavior. In the mid-20th century, flood clusters were fewer but lasted longer (median ~46 days). In recent decades, the frequency of clusters has spiked, while their median duration has plummeted to approximately 17–20 days. This shift suggests a transition from prolonged wet periods to short-duration extreme rainfall events, such as convective "cloudbursts," which are characteristic of a warming climate.

Spatial and temporal distribution of flash flood disasters in China (1960-2015) and the increasing trend in event frequency.

Impact and Regional Vulnerability

Eastern China remains the primary hotspot due to the East Asian summer monsoon, with clusters predominantly occurring between July and October. However, new clusters are emerging in traditionally arid northwestern regions. The study also categorized disasters by "Impact," a composite metric of fatalities and economic loss, to identify the most significant clusters along major river systems and in high-altitude regions.

Classification matrix for flash flood disaster impact levels (F1-F5) based on economic loss and fatalities.

Conclusion and Future Work

While the increase in detected clusters after 1980 may be partially influenced by improved data reporting, the clear trend toward shorter, more frequent clusters points to changing hydrological risks. Future work aims to integrate terrain attributes and anthropogenic factors like urbanization into multivariate models to enhance predictive capabilities for flood risk management.

Summary updated

Abstract

The persistence over space and time of flash flood disasters -- flash floods that have caused either economical or life losses, or both -- is a diagnostic measure of areas subjected to hydrological risk. The concept of persistence can be assessed via clustering analyses, performed here to analyse the national inventory of flash floods disasters in China occurred in the period 1950-2015. Specifically, we investigated the spatio-temporal pattern distribution of the flash floods and their clustering behavior by using both global and local methods: the first, based on the Ripley's K-function, and the second on scan statistics. As a result, we could visualize patterns of aggregated events, estimate the cluster duration and make assumptions about their evolution over time, also with respect precipitations trend. Due to the large spatial (the whole Chinese territory) and temporal scale of the dataset (66 years), we were able to capture whether certain clusters gather in specific locations and times, but also whether their magnitude tends to increase or decrease. Overall, the eastern regions in China are much more subjected to flash floods compared to the rest of the country. Detected clusters revealed that these phenomena predominantly occur between July and October, a period coinciding with the wet season in China. The number of detected clusters increases with time, but the associated duration drastically decreases in the recent period. This may indicate a change towards triggering mechanisms which are typical of short-duration extreme rainfall events. Finally, being flash floods directly linked to precipitation and their extreme realization, we indirectly assessed whether the magnitude of the trigger itself has also varied through space and time, enabling considerations in the context of climatic changes.
BibTeX
@misc{wang2020space,
  author = {Nan Wang and Luigi Lombardo and Marj Tonini and Weiming Cheng and Liang Guo and Junnan Xiong},
  title = {{Space-time clustering of flash floods in a changing climate (China, 1950-2015)}},
  year = {2020},
  eprint = {2006.16865},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2006.16865},
}
Figure