.png){kind=icon}
.png){kind=icon}
Return
Basin-scale typhoon cluster-matching using a large-ensemble dataset and deep learning under different climate scenarios: a case study of the Chikugo River Basin, southwestern Japan
.png){kind=icon}
OA
AI
P
V
S
B
M
T
DOI：10.1186/s40645-026-00825-8.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Understanding typhoon characteristics and identifying affected areas are crucial for effective preparedness and proactive response during extreme events. While many previous studies have focused on typhoon clustering at oceanic scales using historical best-track datasets, a research gap remains in applying such analyses to the database for Policy Decision-making for Future Climate Change (d4PDF) to assess typhoon impacts at the basin scale. This study addresses that gap by analyzing 32,517 typhoons from the d4PDF under both historical and +4K future climate scenarios that affect the Chikugo River Basin (CRB) in southwestern Japan. A novel deep learning-based cluster-matching model was developed to cluster typhoons affecting the CRB and to match forecast events with similar typhoons in the d4PDF. At the core of the model is an embedding vector generated by an encoder–decoder neural network, which effectively overcomes the limitations of conventional clustering and matching algorithms by handling high-dimensional features, large datasets, and the complex nature of typhoons. The analysis identified four distinct clusters, allowed for a clearer understanding of typhoon behavior in key typhoon characteristics (trajectory, maximum wind speed (Vmax), minimum central pressure (Pmin), and lifetime) and their changes under different climate scenarios. Notably, under the +4K future climate scenario (2051–2110), typhoons affecting the CRB were projected to increase significantly in intensity, with a 33% rise in average Vmax and a marked increase in the proportion of violent typhoons (Category 4 and Category 5). The real-time matching process demonstrated high reliability when validated using 280 historical typhoons (1951–2024) as forecast inputs. It achieved a high level of similarity, as indicated by overall median differences of + 9%, − 3.4%, and + 1.8% in Vmax, Pmin, and lifetime, respectively. Integrating this approach with precipitation data from similar typhoons improves flood forecasting and enhances the development of water resources management strategies under climate variability.
Keywords:
Cluster-matching typhoon
Deep learning
d4PDF
Climate change
Basin scale
Chikugo River Basin
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
2.8
Papers:
802
Citations:
2.0K
Organization
Cited Papers