Spacetime Separable Latent Diffusion Model with Intensity Structure Information for Precipitation Nowcasting
The growing volume of meteorological data and advancements in computing performance have made the ap- plication of deep learning technology in short-term rainfall prediction crucial. However, existing learning approaches strug- gle to accurately predict detailed spatial location information, particularly obvious in predicting extreme rainfall events, leading to inadequate prediction accuracy and subpar performance in meteorological assessment indicators, limiting the effectiveness and applicability of deep learning models in rainfall prediction. To address these challenges, we propose a Spacetime Separable Latent Diffusion Model with Intensity Structure Information (SSLDM-ISI) to capture spatial and temporal information more efficiently. SSLDM-ISI incorporates two key strategies to solve the spatiotemporal information issue. First, a Spatiotemporal Conversion Block within the backbone network effectively ex- tracts and integrates spatiotemporal information. Second, our proposed latent space coding technique based on rainfall intensity structural information, enhances the information representation ability of extreme rainfall. In addition, an examination of the impact of various conditions is conducted on the prediction results to enhance the model’s prediction accuracy and stabil- ity. Through comparative analysis of meteorological evaluation and image quality evaluation indicators on two datasets, our proposed approach outperforms existing advanced technologies in short-term rainfall prediction, achieving current state-of-the- art results.
pre-training you can find this address https://drive.google.com/file/d/1IJPlnGeL_JZbALS6Iidr0sylnlFsIhdQ/view?usp=drive_link
cd SSLDM-ISI
python train.py
python sample.py
There is no special environment, the only thing required is cuda>=11.7 torch>=1.13
- Provide clean training code
- Provide clean sampling code
- Provide operating environment