The Solar Dynamics Observatory (SDO) is equipped with the Helioseismic and Magnetic Imager (HMI) instrument, which measures the intensity of the Sun in the region of the Fe I spectral line at 6173Å on the solar surface. More information here.
The resultant "intensitygrams" give a clear picture of the solar surface, from which the origin and evolution of sunspots and their links with solar magnetic activity can be investigated.
A timeseries of white-light continuum data is useful for the investigation of phenomena such as umbral decay rate, the Wilson effect, sunspot velocity during flare activity, and other areas of inquiry.
While algorithms such as the Sunspot Tracking And Recognition Algorithm (STARA) already exist to perform image segmentation analyses of sunspots, this algorithm is computationally expensive to run on native resolution HMI continuum images (4096px
To solve this problem, one can train a machine learning classifer to detect regions of the solar surface with active sunspot/pore activity; in this case: a support vector machine. Subsequently, a segmentation analysis can be performed using STARA only on these smaller active regions. Using this strategy drastically reduces execution time for full resolution images.
Such is the purpose of this project - to provide an efficient way of building a sunspot catalog from intesitygram timeseries for use in scientific studies.
Kedro Custom dataset which reads/writes SunPy Map objects to/from a specified cloud service such as Amazon S3, or locally.
A class to assign labels to regions of training images which contain sunspot/pore activity.
Designed to be used with the matplotlib widget backend inside a Jupyter notebook.
A class designed to overplot target predictions on their corresponding Map as a visual verifiction that the model is functioning correctly.
Designed to be used with the matplotlib widget backend inside a Jupyter notebook.
The training pipeline trains a support vector machine capable of assigning targets to SunPy Map patches based on their sunspot/pore activity.
Before training can be performed, one needs a training dataset (which can be obtained via the SunPy API). This dataset can then be used with the SunspotSelector class while using the matplotlib widget backend inside a Jupyter notebook to assign labels to regions of solar activity.
Run training pipeline with:
kedro run --pipeline training_pipeline
The execution pipeline uses the trained SVM to classify SunPy Map patches based on their sunspot/pore contents.
Contigous regions of activity are then extracted and saved as sub-Maps, before being analysed by the STARA algorithm.
Run an execution pipeline using the date determined by the SunPy API when your dataset downloads. For example, for 20140910Timeseries:
kedro run --pipeline 20140910_execution
pip install -r src/requirements.txt
Optionally install the project in editable mode:
pip install -e src
Chris Osborne @goobley
- A revision of the STARA algorithm supporting JIT compilation: https://github.com/Goobley/stara
- Basis code for the utilities:
SunspotSelector,SunspotInspector - Region extraction functions:
select_region,extract_region - Plotting functions:
overplot_rect_from_coords,overplot_spots_from_mask
Fraser Watson @fraserwatson
- The original STARA algorithm: https://github.com/fraserwatson/stara
The template for this project was generated using Kedro 0.18.3:
Kedro documentation