Final Report Summary - PREDICTING FLARES (Statistics of Solar Flare Activity for Space Weather Predictions)
Solar flares are spectacular demonstrations of solar activity, releasing large amounts of energy from the complex magnetic fields of sunspot groups. Flares are among the most energetic events in the solar system, influencing physical systems all the way from the solar surface, through the solar atmosphere, to planetary atmospheres, and further out into the heliosphere. Significant progress has been made in understanding the Sun's structure and the physical processes that support this complex system. However, those processes that have a direct impact on the near-Earth environment and upon life on Earth, such as solar flares, are still poorly understood. Along with coronal mass ejections, solar flares are a major contributor to space weather the interaction of magnetic fields and particles (accelerated on or near the Sun) with the Earth's magnetic field and upper atmosphere. Understanding the precise physical conditions that result in flare energy release is necessary to develop reliable space-weather forecasts. Accurate warnings of impending solar activity are important to reduce the risk of human exposure to damaging solar radiation (e.g. astronauts and polar-route airline crew and passengers) as well as the continued smooth running of many key modern technologies for civilian society (e.g. telecommunications, global positioning system (GPS) navigation, and electric power grids). The main objective of this project was to provide improved methods for predicting solar flares.
The work carried out in this project has lead to significant advances in our understanding of the origins of solar activity and our ability to forecast flares. The basis of the project lies in the likelihood of flare activity being directly related to the magnetic field complexity on the solar surface, such that large complex sunspot groups produce more frequent and more intense flares than small simple sunspot groups. However, despite qualitative knowledge that complex highly twisted magnetic field structures result in solar activity, we have lacked in quantitative understanding of the physical processes at work. The main results from this research are summarised below.
Appropriate forecast measures
One of the key achievements of this project was the identification of a fundamental issue with the forecast comparison measure commonly used by the flare forecast community (i.e. the Heidke skill score), which critically depends on the relative number of flaring and non-flaring days. In this work we propose the adoption of a forecast measure that is independent of the relative flaring statistics (the true skill statistic, alternatively known as Peirce's skill score or the Hanssen and Kuipers discriminant). This approach results in the first quantification of forecast performance that permits the correct and accurate comparison of different flare forecasting techniques.
Benchmark forecast performance
Having defined an appropriate forecast measure, the benchmark performance for the most basic method of flare forecasting was determined for the first time. The basic method assumes that flares are independent events and uses average flare rates from broad categories describing sunspot group structure and Poisson statistics to provide probabilities of flaring. This method achieves generally low levels of forecast performance, but is surprisingly similar to the levels achieved by some existing, much more sophisticated methods.
Improved forecasting capability
Research facilitated by this project has increased the maximum performance of flare forecasting schemes through a novel combination of solar feature detection and computer vision methods. The application of machine-learning techniques to the physical magnetic field properties of sunspot groups over a whole solar cycle has achieved a higher performance than any previous method, raising the bar for the state-of-the-art in flare forecasting.
The work carried out in this project has lead to significant advances in our understanding of the origins of solar activity and our ability to forecast flares. The basis of the project lies in the likelihood of flare activity being directly related to the magnetic field complexity on the solar surface, such that large complex sunspot groups produce more frequent and more intense flares than small simple sunspot groups. However, despite qualitative knowledge that complex highly twisted magnetic field structures result in solar activity, we have lacked in quantitative understanding of the physical processes at work. The main results from this research are summarised below.
Appropriate forecast measures
One of the key achievements of this project was the identification of a fundamental issue with the forecast comparison measure commonly used by the flare forecast community (i.e. the Heidke skill score), which critically depends on the relative number of flaring and non-flaring days. In this work we propose the adoption of a forecast measure that is independent of the relative flaring statistics (the true skill statistic, alternatively known as Peirce's skill score or the Hanssen and Kuipers discriminant). This approach results in the first quantification of forecast performance that permits the correct and accurate comparison of different flare forecasting techniques.
Benchmark forecast performance
Having defined an appropriate forecast measure, the benchmark performance for the most basic method of flare forecasting was determined for the first time. The basic method assumes that flares are independent events and uses average flare rates from broad categories describing sunspot group structure and Poisson statistics to provide probabilities of flaring. This method achieves generally low levels of forecast performance, but is surprisingly similar to the levels achieved by some existing, much more sophisticated methods.
Improved forecasting capability
Research facilitated by this project has increased the maximum performance of flare forecasting schemes through a novel combination of solar feature detection and computer vision methods. The application of machine-learning techniques to the physical magnetic field properties of sunspot groups over a whole solar cycle has achieved a higher performance than any previous method, raising the bar for the state-of-the-art in flare forecasting.