DPSformer: A long-tail-aware model for improving heavy rainfall prediction

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Huang, Zenghui, Shu, Ting, Wang, Zhonglei, Lu, Yang, Yan, Yan, Zhong, Wei, Wang, Hanzi
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912615328907264
author Huang, Zenghui
Shu, Ting
Wang, Zhonglei
Lu, Yang
Yan, Yan
Zhong, Wei
Wang, Hanzi
author_facet Huang, Zenghui
Shu, Ting
Wang, Zhonglei
Lu, Yang
Yan, Yan
Zhong, Wei
Wang, Hanzi
contents Accurate and timely forecasting of heavy rainfall remains a critical challenge for modern society. Precipitation exhibits a highly imbalanced distribution: most observations record no or light rain, while heavy rainfall events are rare. Such an imbalanced distribution obstructs deep learning models from effectively predicting heavy rainfall events. To address this challenge, we treat rainfall forecasting explicitly as a long-tailed learning problem, identifying the insufficient representation of heavy rainfall events as the primary barrier to forecasting accuracy. Therefore, we introduce DPSformer, a long-tail-aware model that enriches representation of heavy rainfall events through a high-resolution branch. For heavy rainfall events $ \geq $ 50 mm/6 h, DPSformer lifts the Critical Success Index (CSI) of a baseline Numerical Weather Prediction (NWP) model from 0.012 to 0.067. For the top 1% coverage of heavy rainfall events, its Fraction Skill Score (FSS) exceeds 0.45, surpassing existing methods. Our work establishes an effective long-tailed paradigm for heavy rainfall prediction, offering a practical tool to enhance early warning systems and mitigate the societal impacts of extreme weather events.
format Preprint
id arxiv_https___arxiv_org_abs_2509_25208
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DPSformer: A long-tail-aware model for improving heavy rainfall prediction
Huang, Zenghui
Shu, Ting
Wang, Zhonglei
Lu, Yang
Yan, Yan
Zhong, Wei
Wang, Hanzi
Machine Learning
Atmospheric and Oceanic Physics
Accurate and timely forecasting of heavy rainfall remains a critical challenge for modern society. Precipitation exhibits a highly imbalanced distribution: most observations record no or light rain, while heavy rainfall events are rare. Such an imbalanced distribution obstructs deep learning models from effectively predicting heavy rainfall events. To address this challenge, we treat rainfall forecasting explicitly as a long-tailed learning problem, identifying the insufficient representation of heavy rainfall events as the primary barrier to forecasting accuracy. Therefore, we introduce DPSformer, a long-tail-aware model that enriches representation of heavy rainfall events through a high-resolution branch. For heavy rainfall events $ \geq $ 50 mm/6 h, DPSformer lifts the Critical Success Index (CSI) of a baseline Numerical Weather Prediction (NWP) model from 0.012 to 0.067. For the top 1% coverage of heavy rainfall events, its Fraction Skill Score (FSS) exceeds 0.45, surpassing existing methods. Our work establishes an effective long-tailed paradigm for heavy rainfall prediction, offering a practical tool to enhance early warning systems and mitigate the societal impacts of extreme weather events.
title DPSformer: A long-tail-aware model for improving heavy rainfall prediction
topic Machine Learning
Atmospheric and Oceanic Physics
url https://arxiv.org/abs/2509.25208