Solar flares produce radiation that can have an almost immediate effect on the near-Earth environment, making it crucial to forecast flares in order to mitigate their negative effects. The number of published approaches to flare forecasting using photospheric magnetic field observations has proliferated, with varying claims about how well each works. Because of the different analysis techniques and data sets used, it is essentially impossible to compare the results from the literature. This problem is exacerbated by the low event rates of large solar flares. The challenges of forecasting rare events have long been recognized in the meteorology community, but have yet to be fully acknowledged by the space weather community. During the interagency workshop on “all clear” forecasts held in Boulder, CO in 2009, the performance of a number of existing algorithms was compared on common data sets, specifically line-of-sight magnetic field and continuum intensity images from the Michelson Doppler Imager, with consistent definitions of what constitutes an event. We demonstrate the importance of making such systematic comparisons, and of using standard verification statistics to determine what constitutes a good prediction scheme. When a comparison was made in this fashion, no one method clearly outperformed all others, which may in part be due to the strong correlations among the parameters used by different methods to characterize an active region. For M-class flares and above, the set of methods tends toward a weakly positive skill score (as measured with several distinct metrics), with no participating method proving substantially better than climatological forecasts.

Solar flares are extremely energetic phenomena in our solar system. Their impulsive and often drastic radiative increases, particularly at short wavelengths, bring immediate impacts that motivate solar physics and space weather research to understand solar flares to the point of being able to forecast them. As data and algorithms improve dramatically, questions must be asked concerning how well the forecasting performs; crucially, we must ask how to rigorously measure performance in order to critically gauge any improvements. Building upon earlier-developed methodology of Paper I (Barnes et al. 2016), international representatives of regional warning centers and research facilities assembled in 2017 at the Institute for Space-Earth Environmental Research, Nagoya University, Japan to, for the first time, directly compare the performance of operational solar flare forecasting methods. Multiple quantitative evaluation metrics are employed, with the focus and discussion on evaluation methodologies given the restrictions of operational forecasting. Numerous methods performed consistently above the “no-skill” level, although which method scored top marks is decisively a function of flare event definition and the metric used; there was no single winner. Following in this paper series, we ask why the performances differ by examining implementation details (Leka et al. 2019), and then we present a novel analysis method to evaluate temporal patterns of forecasting errors in Paper IV (Park et al. 2019). With these works, this team presents a well-defined and robust methodology for evaluating solar flare forecasting methods in both research and operational frameworks and today’s performance benchmarks against which improvements and new methods may be compared.

A workshop was recently held at Nagoya University (2017 October 31–November 2), sponsored by the Center for International Collaborative Research, at the Institute for Space-Earth Environmental Research, Nagoya University, Japan, to quantitatively compare the performance of today’s operational solar flare forecasting facilities. Building upon Paper I of this series, in Paper II we described the participating methods for this latest comparison effort, the evaluation methodology, and presented quantitative comparisons. In this paper, we focus on the behavior and performance of the methods when evaluated in the context of broad implementation differences. Acknowledging the short testing interval available and the small number of methods available, we do find that forecast performance: (1) appears to improve by including persistence or prior flare activity, region evolution, and a human “forecaster in the loop”; (2) is hurt by restricting data to disk-center observations; (3) may benefit from long-term statistics but mostly when then combined with modern data sources and statistical approaches. These trends are arguably weak and must be viewed with numerous caveats, as discussed both here and in Paper II. Following this present work, in Paper IV (Park et al. 2019) we will present a novel analysis method to evaluate temporal patterns of forecasting errors of both types (i.e., misses and false alarms). Hence, most importantly, with this series of papers, we demonstrate the techniques for facilitating comparisons in the interest of establishing performance-positive methodologies.

A crucial challenge to successful flare prediction is forecasting periods that transition between “flare-quiet” and “flare-active.” Building on earlier studies in this series in which we describe the methodology, details, and results of flare forecasting comparison efforts, we focus here on patterns of forecast outcomes (success and failure) over multiday periods. A novel analysis is developed to evaluate forecasting success in the context of catching the first event of flare-active periods and, conversely, correctly predicting declining flare activity. We demonstrate these evaluation methods graphically and quantitatively as they provide both quick comparative evaluations and options for detailed analysis. For the testing interval 2016–2017, we determine the relative frequency distribution of two-day dichotomous forecast outcomes for three different event histories (i.e., event/event, no-event/event, and event/no-event) and use it to highlight performance differences between forecasting methods. A trend is identified across all forecasting methods that a high/low forecast probability on day 1 remains high/low on day 2, even though flaring activity is transitioning. For M-class and larger flares, we find that explicitly including persistence or prior flare history in computing forecasts helps to improve overall forecast performance. It is also found that using magnetic/modern data leads to improvement in catching the first-event/first-no-event transitions. Finally, 15% of major (i.e., M-class or above) flare days over the testing interval were effectively missed due to a lack of observations from instruments away from the Earth–Sun line.