PROLA: Principal-Orthogonal Low-rank Adaptation for Predictive Spatiotemporal Weather Downscaling
Abstract
High-resolution weather prediction is important for resolving local atmospheric patterns missed by coarse global forecasts. Large pretrained weather backbones have improved global forecasting, but adapting them to produce future high-resolution fields from recent coarse states remains challenging and costly. We formulate this setting as Predictive Spatiotemporal Weather Downscaling (PSWD), a task that jointly refines spatial and temporal resolution while predicting future high-resolution trajectories. To this end, we propose a framework that decomposes prediction into spatial and temporal stages using a shared pretrained backbone, fixed numerical scaffolds, and lightweight residual heads. For backbone adaptation, we introduce PROLA (PRincipal-Orthogonal Low-rank Adaptation), which allocates a fixed low-rank budget between pretrained weights' principal directions and their orthogonal complement. PROLA further rescales rank-1 optimizer updates using gradient signals, improving adaptation without increasing trainable rank. Experiments in the PSWD setting show that PROLA outperforms representative low-rank adaptation baselines. Further experiments across diverse datasets and tasks support the broader applicability of PROLA, showing that its benefits extend beyond the primary weather downscaling setting.
Community
PROLA introduces Predictive Spatiotemporal Weather Downscaling (PSWD), a new task that jointly addresses future weather prediction and spatiotemporal resolution refinement. We propose a scaffold-and-correct framework and Principal-Orthogonal Low-rank Adaptation (PROLA), which allocates a fixed rank budget between principal and orthogonal weight subspaces and adaptively scales optimizer updates using gradient signals. PROLA consistently outperforms representative PEFT baselines across global and regional weather benchmarks, while requiring less than 0.4% of the trainable parameters used in full fine-tuning on the main benchmark. Additional experiments demonstrate its effectiveness across different weather tasks and pretrained backbones.
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper