The FLARECAST architecture is based on a four-step pipeline. In each step multiple algorithms can be plugged in to perform the required work. Between the steps data structures (or databases) with common interfaces are used to exchange data.
A management infrastructure takes care to execute the algorithms according to arrival of new input data.
The four steps are
Step 1: Data acquisition. Load data from different remote archives. This includes observational data (such as SDO/HMI magnetograms) as well as catalog data (such as HSWPC event lists).
Step 2: Feature Property Extraction. Image analysis algorithms will be used to extract solar features (such as active regions) from observational data.
Step 3: Machine learning. Machine learning algorithms will be trained to find patterns in large collections of solar properties in order to forecast solar eruptions.
Step 4: Verification. Continuously verify the outcome of the forecast algorithms.
The system is publicly available. To use it got to api.flarecast.eu.