Wang et al. (2026) P2I-GAN Benchmark: Deep Generative Framework for Spatio-Temporal Rainfall Reconstruction from Sparse Gauges

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Open MIND
• Year: 2026
• Date: 2026-04-11
• Authors: Bing-Zhang Wang, Li-Pen Wang
• DOI: 10.5281/zenodo.18623975

Research Groups

Not specified in the provided text.

Short Summary

P2I-GAN is a deep generative benchmark that reconstructs spatio-temporal rainfall by formulating interpolation from highly sparse and irregular rain-gauge observations as a video inpainting task.

Objective

• To develop and provide a deep generative benchmark (P2I-GAN) for spatio-temporal rainfall reconstruction, specifically addressing interpolation from sparse and irregular rain-gauge observations.

Study Configuration

• Spatial Scale: Spatio-temporal rainfall fields (implied by 'rainfall reconstruction' and 'rain-gauge observations').
• Temporal Scale: Spatio-temporal rainfall fields (implied by 'rainfall reconstruction' and 'video inpainting task').

Methodology and Data

• Models used: P2I-GAN (deep generative model).
• Data sources: Highly sparse and irregular rain-gauge observations.

Main Results

• P2I-GAN successfully formulates rainfall interpolation from sparse observations as a video inpainting task, providing a deep generative benchmark for spatio-temporal rainfall reconstruction. The framework includes training pipelines, HDF5 and Zarr-based dataloaders, evaluation utilities, visualization tools, and pretrained model weights.

Contributions

• P2I-GAN introduces a novel deep generative benchmark for spatio-temporal rainfall reconstruction, offering a reproducible framework for research in rainfall interpolation, spatio-temporal modeling, and nowcasting. It provides comprehensive tools including training pipelines, data loaders, evaluation utilities, visualization tools, and pretrained model weights.

Funding

Not specified in the provided text.

Citation

@article{Wang2026P2IGAN,
  author = {Wang, Bing-Zhang and Wang, Li-Pen and Auguste, Gires},
  title = {P2I-GAN Benchmark: Deep Generative Framework for Spatio-Temporal Rainfall Reconstruction from Sparse Gauges},
  journal = {Open MIND},
  year = {2026},
  doi = {10.5281/zenodo.18623975},
  url = {https://doi.org/10.5281/zenodo.18623975}
}