1. Sunspot Group Evolution and Flare Forecasting
Aoife McCloskey
Supervisors: D. Shaun Bloomfield & Peter T. Gallagher

2. Using Sunspot Group Evolution to Forecast Solar Flares
Most current forecasting techniques use point-in-time measurements of sunspot properties
Simplest forecasting method used is McIntosh-Poisson method (Gallagher et al., 2002)
Few studies have investigated sunspot group evolution and its application in flare forecasting
This work focuses on the evolution of McIntosh classifications and their associated flaring rates

3. Sunspot Classification Data
NOAA SWPC daily sunspot properties (C. Balch, private communication)
Data range (Solar Cycle 22)
Associated GOES 1-8 Å X-Ray flares

4. McIntosh Classification Scheme
White-light structure
Zurich
Longitudinal extent
penumbral
Size and symmetry of largest spot penumbra
compactness
Interior distribution or “filling factor” of spots A B C H D E F X R S A H K X O I C

5. Do sunspot groups produce more flares as they evolve upward or downward in class?

6. Evolution of Zurich Class
Percentage occurrence of region evolutions from a given starting class
No evolution dominates on 24-hr timescale
Preferential evolution of ±1 step
Most regions do not evolve
Preferential evolution is one step in zurich class
East limb - low numbers
West limb - strong evidence that evolution does not represent the underlying distribution

7. McIntosh Evolution Flaring Rates
Flaring rates produced using flares observed in following 24 hr after evolution step
Highest flaring rates associated with smallest occurrences
Majority of flaring rates are statistically significant
Most flaring occurs when regions evolve upward (flux emergence)
Highest flaring rates seen (C, M & X?)

8. McIntosh Evolution Flaring Rates
Most flaring occurs when regions evolve upward (flux emergence)
Highest flaring rates seen (C, M & X?)
McCloskey et al., Solar Physics, 2016

9. Forecasting Application: Poisson Probabilities
Now using full McIntosh classification evolution
𝑍𝑝𝑐 1 → 𝑍𝑝𝑐 2
These flaring rates (Cycle 22) can be applied to more recent data (Cycle 23) to produce flare probabilities:
𝑃 𝑁 = 𝜆 𝑁 𝑁! 𝑒 −𝜆
𝑃 𝑁>0 =1 −𝑃 𝑁=0 =1− 𝑒 −𝜆
λ = 24 hr flaring rate
N = No. of expected events
increase capabilities for forecasting flares. Different characteristics extracted from ARs used to forecast flares.

10. Reliability Plot (>C1.0)
Cycle 23
Cycle 22

11. Reliability Plot (>C1.0)
Cycle 23
Over-forecasting
Cycle 22

12. Reliability Plot (>C1.0)
Lower cycle 23 flaring rates?
Cycle 23
Over-forecasting
Cycle 22

13. Simplest Method: Linear Fitting

14. Reliability Plot (>C1.0)

15. McIntosh Evolution (corrected)
Results & Conclusions
Poor skill is a result of over-forecasting
Performs better than point-in-time method after correction
Correction of flaring rates leads both improved reliability and positive Brier Skill Scores across all flare classes
Brier Skill Score
Flare Class
McIntosh-Poisson
(Point-in-time)
McIntosh Evolution
McIntosh Evolution (corrected)
>C1.0
0.07
0.04
0.13
>M1.0
-0.19
-0.12
>X1.0
-0.11
0.02
BSS = 𝟏− 𝑴𝑺𝑬 𝑴𝑺𝑬 𝒄𝒍𝒊𝒎