Applicability Study of a Diffusion Transformer Forecasting Model in the Complex Terrain of Guizhou
Meteorological Applications · 2026-05 · 33 (3)
DOI: 10.1002/met.70186 ↗Kurz gefasst
Fachartikel in der Zeitschrift Meteorological Applications, online erschienen am 28.04.2026. Verfasst von Congying Li und 8 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Precipitation nowcasting in complex terrains like the Guizhou Plateau is challenged by the blurry forecasts and premature dissipation common in traditional and early deep learning models. To address this, this study evaluates HARE, a novel nowcasting model based on a state‐of‐the‐art Diffusion Transformer (DiT) architecture, against the operational Fenglei model benchmark. HARE generates 0–2 h forecasts by learning from historical radar sequences, and its performance was systematically assessed using a 142‐day macroscopic verification (including critical success index (CSI), probability of detection, false alarm rate, and bias) and microscopic diagnosis of two typical severe weather events. The results demonstrate HARE's clear superiority: its mean CSI (0.2842) was significantly higher than Fenglei's (0.1692), and it exhibited slower performance decay. Critically, case studies revealed that HARE successfully predicted key dynamic processes, such as new convective cell initiation and the full life‐cycle of a squall line, which the benchmark model failed to capture, instead showing an erroneous dissipation trend. These findings confirm the significant advantages of the DiT architecture, whose self‐attention mechanism and generative paradigm are key to capturing complex convective evolution with high fidelity. This work underscores the potential of next‐generation artificial intelligence models while also highlighting that future breakthroughs will likely depend on integrating such data‐driven approaches with atmospheric physics.
Zitieren
APA
Li, C., Kong, D., Yuan, C., Liao, T., Zhi, Y., Wang, B., Zhou, Y., Ge, G., & Pu, S. (2026). Applicability Study of a Diffusion Transformer Forecasting Model in the Complex Terrain of Guizhou. Meteorological Applications, 33(3). https://doi.org/10.1002/met.70186
Vancouver
Li C, Kong D, Yuan C, Liao T, Zhi Y, Wang B, et al. Applicability Study of a Diffusion Transformer Forecasting Model in the Complex Terrain of Guizhou. Meteorological Applications. 2026;33(3). doi:10.1002/met.70186
BibTeX
@article{Li2026,
author = {Li, Congying and Kong, DeXuan and Yuan, Chen and Liao, Tingting and Zhi, Yajing and Wang, Biao and Zhou, Yu and Ge, Gaojian and Pu, Shi},
title = {Applicability Study of a Diffusion Transformer Forecasting Model in the Complex Terrain of Guizhou},
journal = {Meteorological Applications},
year = {2026},
volume = {33},
number = {3},
doi = {10.1002/met.70186}
}
EUVIMED