Abstract:Recent advances in deep learning (DL) have significantly improved the quality of precipitation nowcasting. Current approaches are either based on deterministic or generative models. Deterministic models perceive nowcasting as a spatiotemporal prediction task, relying on distance functions like -norm loss for training. While improving meteorological evaluation metrics, they inevitably produce blurry predictions with no reference value. In contrast, generative models aim to capture realistic precipitation distributions and generate nowcasting products by sampling within these distributions. However, designing a generative model that produces realistic samples satisfying meteorological evaluation indexes in real-time remains challenging, given the triple dilemma of generative learning: achieving high sample quality, mode coverage, and fast sampling simultaneously. Recently, diffusion models exhibit impressive sample quality but suffer from time-consuming sampling, severely hindering their application in nowcasting. Moreover, samples generated by the U-Net denoiser of the current denoising diffusion model are prone to yield poor meteorological evaluation metrics such as CSI. To this end, we propose a spatiotemporal transformer-based conditional diffusion model with a rapid diffusion strategy. Concretely, we incorporate an adversarial mapping-based rapid diffusion strategy to overcome the time-consuming sampling process for standard diffusion models, enabling timely nowcasting. In addition, a meticulously designed spatiotemporal transformer-based denoiser is incorporated into diffusion models, remedying the defects in U-Net denoisers by estimating diffusion scores and improving nowcasting skill scores. Case studies of typical weather events such as thunderstorms, as well as quantitative indicators, demonstrate the effectiveness of the proposed method in generating sharper and more precise precipitation forecasts while maintaining satisfied meteorological evaluation metrics.

Latent diffusion models for generative precipitation nowcasting with accurate uncertainty quantification

Precipitation nowcasting with generative diffusion models

Reliable precipitation nowcasting using probabilistic diffusion models

PreDiff: Precipitation Nowcasting with Latent Diffusion Models

LLMDiff: Diffusion Model Using Frozen LLM Transformers for Precipitation Nowcasting

Extreme Precipitation Nowcasting using Multi-Task Latent Diffusion Models

Advancing Realistic Precipitation Nowcasting With a Spatiotemporal Transformer-Based Denoising Diffusion Model

Generative Modeling of High-resolution Global Precipitation Forecasts

Skillful Precipitation Nowcasting Using Physical‐Driven Diffusion Networks

Generative emulation of weather forecast ensembles with diffusion models

A Generative Deep Learning Approach to Stochastic Downscaling of Precipitation Forecasts

Precipitation Nowcasting Using Physics Informed Discriminator Generative Models

Adverse Weather Conditions Augmentation of LiDAR Scenes with Latent Diffusion Models

Spacetime Separable Latent Diffusion Model With Intensity Structure Information for Precipitation Nowcasting

Experimental Study on Generative Adversarial Network for Precipitation Nowcasting

Integrating Spatio-Temporal and Generative Adversarial Networks for Enhanced Nowcasting Performance

DiffObs: Generative Diffusion for Global Forecasting of Satellite Observations

SRNDiff: Short-term Rainfall Nowcasting with Condition Diffusion Model

SCRD: A Spatiotemporal Cues-Guided Residual Diffusion Model for Precipitation Nowcasting

TimeLDM: Latent Diffusion Model for Unconditional Time Series Generation