spaceml-org / expanding-sdo-capabilities

ML pipeline for Solar Dynamics Observatory (SDO) data

This repo contains a configurable pipeline to train ML models on the SDO Dataset described in Galvez et al. (2019, ApJS) and retrievable from here.

The available models cover two main use-cases:

• learning spatial patterns of the Sun features to arrive at a self-calibration of EUV instruments
• synthesis of one EUV channel from other 3 channels for the design of a AI-enhanced solar telescope

Publications

The above use cases have been explored in the following publications:

• "Exploring the Limits of Synthetic Creation of Solar EUV Images via Image-to-Image Translation" ApJ 937 100 (2022)

• "Multi-Channel Auto-Calibration for the Atmospheric Imaging Assembly using Machine Learning" A&A 648, A53 (2021)

• "Auto-Calibration of Remote Sensing Solar Telescopes with Deep Learning" NeurIPS 2019 - ML4PS Workshop

• "Using U-Nets to Create High-Fidelity Virtual Observations of the Solar Corona" NeurIPS 2019 - ML4PS Workshop

All the results can be reproduced with the code contained in this repo.

The data uncorrected for degradation used in the autocalibration paper is available here.

How to use the repo

1. Reusable code lives inside src in the form of a package called sdo that can be installed.

   In order to install the package:

    1) cd expanding-sdo-capabilities
    2) pip install --user -e .
   

   Please note the core components of this package can be used to design a ML pipeline for use-cases beyond what described above.

2. The pipeline to train and test the autocalibration model can be started by running:

    1) export CONFIG_FILE=./config/autocal_paper_config.yaml 
    2) ./src/sdo/main.py -c $CONFIG_FILE 
   

   it requires access to a SDOML dataset in numpy memory mapped objects format.

3. The pipeline to train and test the virtual telescope model can be started by running:

    1) export CONFIG_FILE=./config/virtual_telescope_default.yaml 
    2) ./src/sdo/main.py -c $CONFIG_FILE 
   

   it requires access to a SDOML dataset in numpy memory mapped objects format.

4. Available models can be found in src/models

5. Some scripts for data pre-processing are contained in scripts/data_preprocess.

6. Notebooks with some analysis of the results live in the folder notebooks.

More on this project

This project started as part of the 2019 Frontier Development Lab (FDL) SDO team. A description of this program is available here.

Citations

If you decide to reuse this code, please cite