Real-Time CME Forecasting Using HMI Active Region Magnetograms

Real-Time CME Forecasting Using HMI Active. Region Magnetograms and Flare History David Falconer, Abdulnasser F. Barghouty, Igor Khazanov, and Ron Moore Forecasting X-class, M-class, CMEs, and SPEs from active region magnetograms using previous flare activity 2011 October MURI Meeting 1

More Less Relation between size and twist of an active region’s magnetic field with free energy: - + Twist Size Free Magnetic Energy 2 ~(Twist x Size)

MSFC Vector Magnetogram of a very Nonpotential Active Region: Observed Horizontal Field Potential Horizontal Field 25, 000 km An active-region field’s horizontal shear is concentrated along neutral lines where the field’s horizontal component is strong and the vertical component’s horizontal gradient is steep Observed-field upward (downward) vert. comp. is shown by solid contours or light shading (dashed contours or dark shading); red arrows show observed hor. comp. ; green arrows show hor. comp. of pot. field computed from obs. vert. comp. ; strong-observed-field (>150 G) intervals of neutral lines are blue. 3

Magnetic Measures Φ WLSG

Active-Region Magnetic Measures Vector measures (SDO/HMI) Magnetic Size Φ=∫|Bz| da (|Bz| >100 G) Free Magnetic Energy Proxy WLSG= ∫( Bz) dl (potential Bh>150 G) Where Bz is the vertical field and Bh is the horizontal field. In the line-of-sight approximation (SOHO/MDI) Bz is replaced by the line-of-sight field and Bh is replaced by the potential transverse field which is calculated from the line-of-sight field 5

Only Big Twisted Active Regions Produce CMEs A Productive Active Region has: • Large amount of free energy • Large magnetic size • Large magnetic twist • Similar tendencies for production of X and M flares. • • * Produced CMEs in 24 hours * No CMEs in 24 hours Measure of Active Region Magnetic Size Database 40, 000 Magnetograms From 1, 300 active regions 1996 -2004 Known flare/CME/SPE history Proxy of Active-Region Magnetic Free Energy

Correlation of AR Production of Major Solar Events with Free-Energy Proxy • Gray scale plot shows freeenergy/magnetic size distribution of 40, 000 magnetograms of 1, 300 active regions. Red contours are 0. 001, 0. 01, and 0. 5 event/day levels. • Free-energy proxy histogram of all active regions(black curve), and those that produce an X or M flare or CME in the next 24 hours.

Forecasting Events: Basic approach • From a vector magnetogram, in theory the free-energy of an active region can be measured. • Present vector magnetograms though are not capable of making these measurements. • Proxies of free-magnetic energy though do exist, and can be measured from line-of-sight magnetograms. • From a large database, we have determined empirical forecasting curves to convert an active region’s freeenergy proxy to its expected next-day event rates.

Empirical Forecasting Curves Forecasting an Active Region’s Next-Day Event Rates from Its Present Free. Energy Proxy • We have determined next-day event rates as a function of the active region’s free-energy proxy, for several types of events. • The sample is divided into 40 equally populated bins using the magnitude of our free-energy proxy. • For each bin, the average observed next. Event Rate 24 -hour event rate is determined, with uncertainties. • The event rates for bins with event rates of >0. 01 events/day are fitted with a power law. • This fit is used for converting our freeenergy proxy to expected event rate. • Event rate can be converted to the probability of an event. LWL SG(G) Proxy of Free Energy R=A(LWLSG)B P(t)=100%(1 -e-Rt)

2011 Advancements • Transitioned from MDI line-of-sight magnetograms to HMI line-of-sight magnetograms. Will discuss how. • With Near-Real-Time HMI magnetograms, we are now producing Near-Real-Time Forecasts. • The system was installed at NASA/SRAG in March 2011. NASA/SRAG do the radiation forecasts for the astronauts. • The six ground-based GONG sites serve as a backup. • Have identified how to improve event-chance forecasts by including previous flare activity.

Transitioning to HMI Line-of-Sight 1. Advantages of HMI over MDI HMI pixels 2” 0. 5” Cadence 96 minutes 45 sec LOS, 90 sec Vector Latency Approximately a day tens of minutes Magnetograph Type Line-of-sight Vector Operational 1996 -Jan 2011 May 2010 to present Now Turned off Transition Problem Now Operating LWL SG= ∫( Blos) dl The value of our free-energy proxy is resolution dependent. Increase resolution results in both longer neutral lines, and steeper gradients. The amount of increase varies from active region to active region. There might be something predictive in these different resolution dependent free-energy proxies, but until enough active regions are included in the sample, this will not be known.

Transitioning to HMI Line-of-Sight 2. Solution to Using HMI for Forecasting 1. Convert HMI to MDI magnetic field strength. 2. Then apply MDI point spread function. 3. Then cross-calibrate using overlap era. For MDI resolution • LWL (MDI)=1. 31*LWL (HMI) SG SG • Multiplicative uncertainty is 1. 22 HMI Smoothed to MDI Resolution HMI Unsmoothed HMI 12

Transitioning to HMI Line-of-Sight 3. Display of Recent Forecast from HMI (October 25, 2011) 13

Forecast based on Free-Energy Proxy and Flare History 1. Importance of Flare History Major flares in Last 24 hours Yes No Actual Number of X&M Flares 1092± 130 1387± 144 Number of X&M Flares Predicted from Free-Energy Proxy Alone Number of Active Region Magnetograms (% of sample) 750± 110 1780± 163 1780(4%) 38197(96%) Active regions that recently produced X&M class flares are ~50% more likely to produce X&M class flares in the near future, than free-energy proxy only would predict.

Forecast based on Free-Energy Proxy and Flare History 2. Recently Flaring and non-flaring forecast curves • Result: Active regions that have recently flared, show a significantly greater flare rate in the future. • Divide sample into flaring (have produced an X or M class flare in last 24 hours), and non-flaring (have not produced an X or M class flare in last 24 hours). • Develop separate forecast curves for both sets, by binning in a similar manner, but requiring at least 50 active-region magnetograms of the set in the bin. • Top Log-Log • Bottom Log-Linear • Dashed line Free-energy proxy forecast only.

Forecast based on Free-Energy Proxy and Flare History 3. Forecast Table • Forecast an event if the event rate is above 0. 5 events/day, and forecast no event if the rate is less than 0. 5 events/day. Free-Energy Proxy Only Free-Energy Proxy and Last 24 hour Flare History Contingency Table Forecast Yes Forecast No Actual Yes 544 1231 Contingency Table Actual No 540 37662 Forecast Yes Forecast No Decision Matrix Forecast Yes Forecast No Actual Yes Hit 50% Miss Rate 69% Actual Yes 689 1086 Actual No 563 37639 Decision Matrix Actual No False Alarm 1. 4% Correct Null 97% Forecast Yes Forecast No Actual Yes Hit 55% Miss Rate 61% Actual No False Alarm 1. 5% Correct Null 97%

Forecast based on Free-Energy Proxy and Flare History 4. Improvement to Forecast • The combination of flare history and free-energy proxy lead to improvements in probability of detection, Heidke skill score, and the false alarm rate, over free energy only. • • The Probability of Detection (The percent of Major flares, for which a yes forecast). The Heidke skill score ( skill score, 1 perfect forecast, 0 chance, -1 always wrong). • The false alarm rate (the percentage of forecast events, that turn out to be false). Skill Scores Free-energy proxy only Probability of Detection False Alarm % Heidke Skill Score Percent Correct 31% 50% 0. 36 95. 6% Free-energy proxy and last 24 hour flare history 39% 45% 0. 43 95. 9% Best 100% 0% 1 100%

Forecast based on Free-Energy Proxy and Flare History 5. To Do • Need to determine dependence of forecast accuracy on width of forward time window. • Determine if using last 24 hours flare history is optimal forecast accuracy. • Determine if more accurate forecast curves can be obtained for both sets by other curve-fitting methods. • Perform these studies for the other event types (X&M flares were studied first for best statistics). • Quantify how sample size affects results. • Incorporate the combination in the forecast tool.

Future Work: Tool upgrades to be done using HMI • Determine best way to implement combined flare and freeenergy forecast • Extend to CME, Fast CME, X-class Flare, and SPE. • Use deprojected vector magnetograms to measure WLSG – Waiting for automated ambiguity resolution – Will incorporate de-projecting of ambiguity-resolved activeregion tiles • Does HMI’s higher resolution give WLSG values that are more strongly correlated with AR major-event production than does the lower resolution of MDI? – This will need a large HMI database to determine.

Statistic backup slide Actual Yes Actual No Marginal Total Forecast Yes A B A+B Forecast No C D C+D A+C B+D A+B+C+D=N Marginal Total Decision Matrix Equations Forecast Yes Forecast No Skill Score Probability of Detection Heidke Skill Score False Alarm Rate Percent Correct Actual Yes Hit A/(A+B) Miss Rate C/(A+C) Actual No False Alarm B/(B+D) Correct Null D/(C+D) Formula A/(A+C)*100% 2(AD-BC)/[(A+C)*(C+D)+(A+B)*(B+D)] B/(A+B)*100% (A+D)/N Best 100% 100%

Transitioning to HMI Line-of-Sight 2. Solution to Using HMI for Forecasting 1. Convert HMI to MDI magnetic field strength. 2. Then apply MDI point spread function. 3. Then cross-calibrate using overlap era. Multiplicative uncertainty due to different instruments and their spatial resolutions Event Type MDI HMI-lowres HMI-full res X and M Flares 1. 07 1. 48 2. 71 X Flares 1. 29 1. 60 2. 86 CMEs 1. 10 1. 36 2. 16 Fast CMEs 1. 17 1. 41 2. 24 SPEs 1. 32 1. 51 2. 33 For MDI resolution • LWL (MDI)=1. 31*LWL (HMI) SG SG • Multiplicative uncertainty is 1. 22 HMI Smoothed to MDI Resolution HMI Unsmoothed HMI