Presentation #111.01(A) in the session Solar Flare Prediction. Not to be confused with presentation #111.01(B)
In recent head-to-head evaluations of operational flare forecasting facilities (Leka et al 2019 a, b; Park et al 2020) we saw that methods generally score below 0.5 (on a 0.0–1.0 scale) across numerous standard metrics, and all methods pretty much fail to correctly identify and predict upcoming variations in flaring activity (the first flare / last flare challenge). So, innovations are needed.
Forecasting is a stringent test of understanding; in the case of solar flares, this implies the need for physical models or physics-based empirical approaches. But forecasting solar flares has also been tackled by machine learning methods that avoid physics or deploy previously-identified physical parameters. To complicate matters, the timescales and performance benchmarks thus far are defined by humans, rather than by physics.
In this talk I will summarize where forecasting stands both operationally (see above) and in the current research. I will summarize recent (numerous) machine-learning attempts to improve forecasting outcomes and highlight other approaches such as avalanche models, being invoked for the same goal. I will finally outline our present work to address the first flare / last flare and limb-flare challenges, and to incorporate the chromosphere and corona into physics-based approaches to flare forecasting.