Update README.md

README.md

@@ -1,12 +1,97 @@
 ---
 license: apache-2.0
+tags:
+- Pytorch
+- Heliophysics
+- Space Weather
+- Time Series
+- Foundation Model
+- NASA
+- IBM
+- SDO
 ---
 
 # Surya 1.0
 
-This repo contains the weights, configuration file and data scaling information for Surya. Surya is a 366M parameter foundation model for heliophysics pretrained on 9 years (218 TB) of SDO data. Code and examples can be found [on GitHub](https://github.com/NASA-IMPACT/Surya).
+NASA, IBM, and partners present **Surya**, the first open-source AI **foundation model for heliophysics**.  
+Surya is a 366M-parameter transformer model pretrained on **9 years (≈218 TB)** of multi-instrument data from NASA’s [Solar Dynamics Observatory (SDO)](https://sdo.gsfc.nasa.gov/), including 8 Atmospheric Imaging Assembly (AIA) channels and 5 Helioseismic and Magnetic Imager (HMI) products.  
 
-Contents:
-- [surya.366m.v1.pt](surya.366m.v1.pt): Weights
-- [config.yaml](config.yaml): Model configuration
-- [scalers.yaml](scalers.yaml): Parameters for data scaling
+By leveraging advances in AI and open science, Surya provides a powerful tool for **understanding solar dynamics** and **predicting space weather**—critical for protecting satellites, power grids, communication systems, and astronauts. The model is accessible on Hugging Face, enabling scientists, startups, and agencies worldwide to experiment, fine-tune, and build new applications.
+
+---
+
+## Highlights
+
+- **General-purpose foundation model** for heliophysics, trained at SDO’s native resolution (4096×4096).
+- **Pretraining objectives**: one-hour-ahead forecasting + autoregressive rollout tuning up to 12 hours.
+- **Data scale**: 13-channel, harmonized, ML-ready dataset spanning nearly a full solar cycle (2010–2019).
+- **Open science**: full weights, config, and preprocessing pipelines shared for reproducibility.
+
+---
+
+## Applications
+
+Surya can be fine-tuned for a wide range of heliophysics and space-weather tasks:
+
+- 🌞 **Solar flare forecasting** — surpasses existing benchmarks by **15%** in preliminary tests, with 24h binary classification (M/X-class flares).  
+- 🌬 **Solar wind speed prediction** — downstream fine-tuning achieves strong performance compared to physics-based models.  
+- ☀️ **Active region segmentation** — outperforms baseline UNet with **IoU 0.768** and **Dice 0.853**.  
+- 🔭 **EUV spectral forecasting** — accurate prediction of solar spectra.  
+
+---
+
+## Model Variants
+
+- **`surya.366m.v1`** — pretrained on 9 years of SDO AIA/HMI data with forecasting objective + rollout tuning.  
+- **Fine-tuned versions** (coming soon) — for flare forecasting, active region segmentation, and solar wind prediction.
+
+---
+
+## Example Visualizations
+
+**Solar Flare Prediction (Zero-Shot Rollout)**  
+
+![image/png](https://cdn-uploads.huggingface.co/production/uploads/6488f1d3e22a0081a561ec8f/dJUMwJFE81sI5-3_AdFXR.png)
+
+
+Left two columns are the inputs. Top right two images the outputs, bottom right the ground truth.
+
+---
+
+## Architecture
+
+Surya’s architecture integrates **spectral gating** (frequency-domain filtering) with **long–short range attention** to efficiently model both local and global solar dynamics.  
+
+**Architecture Diagram:**  
+
+<p align="center">
+  <img src="https://cdn-uploads.huggingface.co/production/uploads/6488f1d3e22a0081a561ec8f/bjVv_iDXj9w7VK6S_xlL7.png" alt="Surya Architecture" width="550"/>
+</p>
+
+
+---
+
+## Contents
+
+- [surya.366m.v1.pt](surya.366m.v1.pt) – Model weights  
+- [config.yaml](config.yaml) – Configuration file  
+- [scalers.yaml](scalers.yaml) – Preprocessing & normalization parameters  
+
+Code and training examples available on [GitHub](https://github.com/NASA-IMPACT/Surya).
+
+---
+
+## Citation
+
+If you use Surya in your research, please cite:
+
+```bibtex
+@misc{roy2025surya,
+      title={Surya: Foundation Model for Heliophysics}, 
+      author={Sujit Roy and Johannes Schmude and Rohit Lal and Vishal Gaur and Marcus Freitag and Julian Kuehnert and Theodore van Kessel and Dinesha V. Hegde and Andrés Muñoz-Jaramillo and Johannes Jakubik and Etienne Vos and Kshitiz Mandal and Ata Akbari Asanjan and Joao Lucas de Sousa Almeida and Amy Lin and Talwinder Singh and Kang Yang and Chetraj Pandey and Jinsu Hong and Berkay Aydin and Thorsten Kurth and Ryan McGranaghan and Spiridon Kasapis and Vishal Upendran and Shah Bahauddin and Daniel da Silva and Nikolai V. Pogorelov and Campbell Watson and Manil Maskey and Madhulika Guhathakurta and Juan Bernabe-Moreno and Rahul Ramachandran},
+      year={2025},
+      eprint={XXXX.XXXXX},
+      archivePrefix={arXiv},
+      primaryClass={astro-ph.SR},
+      url={https://arxiv.org/abs/XXXX.XXXXX}, 
+}