Scale-Adaptive Framework for Joint Spatiotemporal Super-Resolution with Diffusion Models
A new deep-learning framework from arXiv (2604.21903) proposes scale-adaptive joint spatiotemporal super-resolution for climate applications. Traditional video super-resolution methods typically handle either spatial or temporal upscaling separately, and existing joint models are limited to fixed pairs of super-resolution factors. The framework decomposes spatiotemporal SR into a deterministic conditional mean prediction with attention and a residual conditional diffusion model. It includes an optional mass-conservation transform to preserve aggregated precipitation totals. The approach reuses the same architecture across different spatial and temporal scaling factors, addressing the challenge that larger SR factors increase underdetermination and required context.
Key facts
- arXiv paper 2604.21903 introduces a scale-adaptive framework for joint spatiotemporal super-resolution.
- The framework decomposes SR into deterministic prediction and residual diffusion model.
- An optional mass-conservation transform preserves total precipitation amounts.
- The same architecture works across multiple spatial and temporal scaling factors.
- Larger SR factors increase underdetermination and require more context.
- Climate applications typically super-resolve either space or time separately.
- Existing joint models are limited to single pairs of SR factors.
- The framework uses attention for conditional mean prediction.
Entities
Institutions
- arXiv