Research map / Time series / Long-term forecasting
Computational Imaging for Long-Term Prediction of Solar Irradiance
Leron Julian, Haejoon Lee, Soummya Kar, Aswin C. Sankaranarayanan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use hyperboloidal instead of hemispherical mirror (ablation) | Transformer + GHI · GHI forecast, 10 min horizon | nRMSE: 0.250 -> 0.230 | Improved |
| use hyperboloidal instead of hemispherical mirror (ablation) | Transformer + GHI · GHI forecast, 15 min horizon | nRMSE: 0.258 -> 0.246 | Improved |
| use hyperboloidal instead of hemispherical mirror (ablation) | Seq2Seq + GHI · GHI forecast, 30 min horizon | nRMSE: 0.270 -> 0.277 | No clear effect |
| use hyperboloidal mirror for sun occlusion prediction (AUC 90% confident window) (ablation) | learning-based occlusion predictor · Simulated sun occlusion | forecast horizon: about 3 minutes -> about 18 minutes | Improved |
| add digital warping to hyperboloidal images (ablation) | Transformer + GHI · GHI forecast, 10 min horizon | nRMSE: 0.230 -> 0.266 | Hurt |
| add digital warping to hemispherical images (ablation) | Transformer + GHI · GHI forecast, 30 min horizon | nRMSE: 0.284 -> 0.341 | Hurt |
| add hyperboloidal sky imagery to GHI input (ablation) | Seq2Seq · GHI forecast, 10 min horizon | nRMSE: 0.265 -> 0.248 | Improved |
| add sky imagery to GHI input at 1 min horizon (ablation) | Seq2Seq · GHI forecast, 1 min horizon | nRMSE: 0.169 -> 0.179 | Hurt |
| combine both mirrors' space-time slices in channel dimension (ablation) | Transformer + GHI · GHI forecast, 30 min horizon | nRMSE: 0.289 -> 0.268 | Improved |
| use Transformer instead of Seq2Seq (ablation) | hyperboloidal + GHI · GHI forecast, 10 min horizon | nRMSE: 0.248 -> 0.230 | Improved |
| use Transformer instead of Seq2Seq (ablation) | hyperboloidal + GHI · GHI forecast, 30 min horizon | nRMSE: 0.277 -> 0.289 | Hurt |
| use learned Transformer with hyperboloidal imagery instead of persistence | GHI forecast · GHI forecast, 30 min horizon | nRMSE: 0.336 -> 0.289 | Improved |
Closest papers by meaning
- SolarCube: An Integrative Benchmark Dataset Harnessing Satellite and In-situ Observations for Large-scale Solar Energy Forecasting (NeurIPS 2024 · similarity 0.83)
- Cloud4D: Estimating Cloud Properties at a High Spatial and Temporal Resolution (NeurIPS 2025 · similarity 0.82)
- Solar Power Generation Forecasting via Multimodal Feature Fusion (Student Abstract) (AAAI 2024 · similarity 0.81)
- LiDAR-RT: Gaussian-based Ray Tracing for Dynamic LiDAR Re-simulation (CVPR 2025 · similarity 0.80)
- Scaling Storm-Resolving Atmospheric AI Simulation to the Entire Planet (NeurIPS 2026 · similarity 0.80)
- WeatherGen: A Unified Diverse Weather Generator for LiDAR Point Clouds via Spider Mamba Diffusion (CVPR 2025 · similarity 0.80)
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.
