Research map / Architectures and efficiency / Pruning and efficient tuning
PROLA: Principal-Orthogonal Low-rank Adaptation for Predictive Spatiotemporal Weather Downscaling
Minseo Yoon, Minseong Bae, Sojin Lee, Hyunwoo J. Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes PROLA, a principal-orthogonal low-rank adaptation method for predictive spatiotemporal weather downscaling (PSWD), producing future high-resolution weather fields from recent coarse states.
Problem
Adapting large pretrained weather backbones to predict future high-resolution fields from recent coarse atmospheric states is challenging and costly.
Approach
Formulates predictive spatiotemporal weather downscaling as a task and adapts pretrained weather backbones via principal-orthogonal low-rank adaptation (details beyond the abstract not available).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use PROLA dual-branch adaptation instead of LoRA at rank 8 | Aurora PSWD · WeatherBench 2 3h Z500 | RMSE: 54.72 -> 50.58 | Improved |
| use PROLA instead of PiSSA | Aurora PSWD · ERA5-to-CERRA 3h Z500 | RMSE: 71.80 -> 68.44 | Improved |
| use PROLA instead of OPLoRA | Aurora · Regional precipitation 12h | SEEPS: 0.605 -> 0.590 | Improved |
| PROLA (3.92M params) vs full fine-tuning (1.26B) | Aurora PSWD · WeatherBench 2 Z500 | RMSE: 60.13 -> 50.58 (3h) | Improved |
| add spatial and temporal correction heads (ablation) | PROLA pipeline · WeatherBench 2 3h Z500 | RMSE: 87.73 -> 50.30 | Improved |
| add temporal correction head Ct (ablation) | PROLA pipeline · WeatherBench 2 3h Z500 | RMSE: 84.53 -> 50.30 (vs w/o Ct) | Improved |
| add spatial correction head Cs (ablation) | PROLA pipeline · WeatherBench 2 T2M/U10 6h | RMSE: 6h Z500 78.25 -> 73.07; T2M 1.209 -> 1.211 | Mixed |
| use scaffold-and-correct pipeline instead of direct prediction (ablation) | PROLA · WeatherBench 2 3h Z500 | RMSE: 53.13 -> 50.30 | Improved |
| add orthogonal branch to principal branch (4+4) (ablation) | PROLA · WeatherBench 2 3h Z500 | RMSE: 53.66 -> 51.36 | Improved |
| add gradient-guided update scaling (ablation) | PROLA 4+4 · WeatherBench 2 3h Z500 | RMSE: 51.36 -> 50.30 | Improved |
| use balanced 4+4 rank split instead of 6+2 or 2+6 (ablation) | PROLA · WeatherBench 2 6h Z500 | RMSE: 73.07 vs 74.19 and 76.35 | Improved |
| use both branch and component scores instead of one (ablation) | PROLA · WeatherBench 2 6h Z500 | RMSE: 76.06 / 75.45 -> 73.07 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PiSSA — PiSSA: Principal Singular Values and Singular Vectors Adaptation of Large Language Models (NeurIPS 2024) · better on 3 of 3 shared results
- DoRA — DoRA: Weight-Decomposed Low-Rank Adaptation (ICML 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Experimental setup
Builds on: LoRA
Closest papers by meaning
- Less Tokens, Better Forecasts: Sparse Residual Routing for Efficient Weather Prediction (ECCV 2026 · similarity 0.86)
- Pushing the Limits of High-Resolution Weather Forecasting through Data Scaling (ECCV 2026 · similarity 0.85)
- On the Integration of Spatial-Temporal Knowledge: A Lightweight Approach to Atmospheric Time Series Forecasting (NeurIPS 2025 · similarity 0.85)
- DAWP: A framework for global observation forecasting via Data Assimilation and Weather Prediction in satellite observation space (NeurIPS 2025 · similarity 0.85)
- Learning with Calibration: Exploring Test-Time Computing of Spatio-Temporal Forecasting (NeurIPS 2025 · similarity 0.85)
- Precipitation Downscaling with Spatiotemporal Video Diffusion (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
