Skip to main content

Flare Likelihood and Region Eruption Forecasting

Periodic Reporting for period 2 - FLARECAST (Flare Likelihood and Region Eruption Forecasting)

Reporting period: 2016-01-01 to 2017-12-31

TITLE: FLARECAST – Flare Likelihood and Region Eruption Forecasting
FUNDING AGENCY: EC Horizon 2020, RIA, Protec-1-2014: Space Weather
GRANT AMOUNT: EUR 2,400 000.00
DURATION: 3 years

START DATE: 1 January 2015
CONSORTIUM: 9 institutions from 6 countries
PROJECT COORDINATOR: Dr. Manolis K. Georgoulis, Academy of Athens, Greece
PROJECT SCIENTIST: Dr. D. Shaun Bloomfield, Trinity College Dublin, Ireland

KEY WORDS: Flare Forecasting, Space Weather, Heliophysics, Prediction Algorithms, Machine Learning

Space weather can have detrimental effects on astronaut safety and a multitude of technologies on which we rely on a daily basis. Accurate and reliable space-weather monitoring and forecasting helps those affected, such as satellite operators, to take timely impact-mitigation measures. FLARECAST will significantly advance our ability to predict flares prior to their occurrence in the Sun, aiming to assist existing and future mitigation efforts.

FLARECAST will first aim to understand the drivers of solar-flare activity to improve flare prediction. It will then aim to provide a globally and openly accessible flare prediction service that facilitates evolution and expansion. Finally, FLARECAST will aim to engage in a dialogue with space-weather stakeholders, policy makers, and the public on the societal benefits of a reliable solar-flare prediction.

Solar active-region properties will be extracted using advanced image-processing techniques applied to remote-sensing solar observations. Flare prediction algorithms, mostly relying on machine learning, will highlight a statistically rigorous set of the most promising of these predictors. Flare-forecast probabilities will be then verified and utilized to launch a fully automated, near real-time flare forecasting service.

FLARECAST will push the envelope of current understanding of solar active-region properties and their relation to flaring activity. In parallel, a functionally-expandable infrastructure will accommodate flare predictions allowing a simple, but suitably verified, transition of scientific research into an operational space-weather application. The resulting user-friendly, interactive facility will be freely accessible to researchers and operators in Europe and around the globe.

AA: Academy of Athens Greece
TCD: Trinity College Dublin Ireland
UNIGE: Universita Degli Studi Di Genova Italy
CNR: Consiglio Nazionale Delle Ricerche Italy
CNRS: Centre National de la Recherche Scientifique France
UPSud: Université Paris-Sud France
FHNW: Fachhochschule Nordwestschweiz Switzerland
MO: Met Office United Kingdom
UNN: University of Northumbria at Newcastle United Kingdom


CME : coronal mass ejection
DB : Database
LOS : Line of Sight
ML: : Machine Learning
SHARP : Space Weather HMI Active Region Patch
WP : Work Package


1. To understand the drivers of flare activity and improve flare prediction.

2. To provide a globally accessible flare prediction service that facilitates expansion.

3. To engage with space-weather end users and inform policy makers and the public.

WORK OVER FIRST OBJECTIVE IN YEARS 2 – 3: Wps 2, 3, 5 and 6:

WP2 (solar active region properties as predictors of flare activity) was the first to be finalized, because the prediction DB was a prerequisite for the prediction and verification DBs to be filled. WP2 gave rise to two of its three deliverables, with the latest one (D2.4) referring to the finalized predictor algorithms, metadata and documentation. The project produced a total of 171 predictors for each SHARP analyzed.

WP3 (flare prediction algorithms) exploited the prediction DB by implementing a total of 21 ML algorithms on top of 8 non-ML ones. Four out of five deliverables of WP3 was produced in Years 2 – 3 with the last one (D3.5) conveying the final state of the ML prediction algorithms, where the bulk of prediction work is made.

WP5 (data and forecast validation) was entirely implemented in Year 2 – 3. Work was structured and revolved around 3 deliverables that pertained to the production of the forecast verification software, associated uncertainties and the data monitoring software. A total of 20 metrics, skill scores and discriminants were used to verify data and forecasts.

WP6 (explorative research) was also entirely implemented in Years 2 – 3. Revolving around objectives to understand flares, to investigate the role of forecast window and latency in flare forecasting and to improve the flare – CME connection, it produced 3 deliverables and a number of refereed papers, with more manuscripts in various preparatory stages.


WP4 (data storage and processing cloud) aimed to prepare the FLARECAST technology element and to create the infrastructure to be used for processing, forecasting, and visualization. It combined the efficient exploitation of big data and archive technology and documented this work in 6 deliverables, five of which were completed in Years 2 – 3. The finalized infrastructure framework and prediction database schema are described in deliverables D4.5 and D4.6 respectively.


WP7 (dissemination) aimed toward communication with the scientific community, users and stakeholders, and the public. While the first and the third aims originated in the first period, the second aim was exclusively completed during the second period: in Year 3 the project organized two User / Stakeholder Workshops in the MO partner and during the 2017 European Space Weather Week, respectively. The Workshops are presented and discussed in deliverables D7.6 and D7.7 respectively. In addition, a survey capturing the understanding and level of knowledge of the user community toward flare prediction was implemented. Of a total of 9 deliverables in WP7, 5 were fullfiled in Years 2 – 3.
FLARECAST has managed to push the envelope of current understanding of solar active-region properties and their relation to flaring activity. This goal was decisively advanced in Years 2 – 3 with the accumulation of (1) the largest number of predictor parameters ever used for flare forecasting, (2) the largest number of assembled prediction algorithms , (3) a set of new flare, and even CME, predictors and (4) an interactive, customizable, fully verified flare prediction facility, open to scientists, operators and industry professional in Europe and around the globe.