Automatic Detection of Magnetic delta in Sunspot Groups
Sreejith Padinhatteeri, Paul A. Higgins, D. Shaun Bloomfield, Peter T. Gallagher
aa r X i v : . [ a s t r o - ph . S R ] O c t Solar PhysicsDOI: 10.1007/ ••••• - ••• - ••• - •••• - • Automatic Detection of Magnetic δ in SunspotGroups Sreejith Padinhatteeri , · PaulA. Higgins , · D. Shaun Bloomfield · PeterT. Gallagher c (cid:13) Springer ••••
Abstract
Large and magnetically complex sunspot groups are known to be associatedwith flares. To date, the Mount Wilson scheme has been used to classify sunspotgroups based on their morphological and magnetic properties. The most flareprolific class, the δ sunspot-group, is characterised by opposite polarity umbraewithin a common penumbra, separated by less than 2 ◦ . In this article, we presenta new system, called the Solar Monitor Active Region Tracker - Delta Finder(SMART-DF), that can be used to automatically detect and classify magnetic δ sin near-realtime. Using continuum images and magnetograms from the Helioseis-mic and Magnetic Imager (HMI) onboard NASA’s
Solar Dynamics Observatory(SDO) , we first estimate distances between opposite polarity umbrae. Oppositepolarity pairs having distances of less that 2 ◦ are then identified, and if thesepairs are found to share a common penumbra, they are identified as a magnetic δ configuration. The algorithm was compared to manual δ detections reportedby the Space Weather Prediction Center (SWPC), operated by the NationalOceanic and Atmospheric Administration (NOAA). SMART-DF detected 21out of 23 active regions (ARs) that were marked as δ spots by NOAA during2011 – 2012 (within ± ◦ longitude). SMART-DF in addition detected five ARswhich were not announced as δ spots by NOAA. The near-relatime operationof SMART-DF resulted in many δ s being identified in advance of NOAA’sdaily notification. SMART-DF will be integrated with SolarMonitor ∗ and thenear-realtime information will be available to the public. School of Physics, Trinity College Dublin, Dublin 2,Ireland email: [email protected] Manipal Centre for Natural Sciences, Manipal University,Manipal, Karnataka, India - 576104. Lockheed Martin Advanced Technology Center,Dept/A021S, B/252, 3251 Hanover Street, Palo Alto, CA94304, USA. ∗ SOLA: paper_final.tex; 14 October 2018; 15:45; p. 1 adinhatteeri et al.
1. Introduction
Solar flares are among the most energetic events in the solar system (energy up to ∼ J; Emslie et al. , 2012; Moore et al. , 2001; Aschwanden and Freeland, 2012)influencing a panorama of physical systems, from the solar surface through theheliosphere and onwards into geospace. Flares, along with coronal mass ejections(CMEs), are a major contributor to space weather – the interaction of magneticfields and particles accelerated on or near the Sun with the Earth’s magneto-sphere and upper atmosphere (Gopalswamy et al. , 2005; Messerotti et al. , 2009;Hapgood and Thomson, 2010). Although significant progress has been made inunderstanding the fundamental physics of solar flares and coronal mass ejections,accurate forecasting of these enigmatic events remains elusive (Gallagher, Moon,and Wang, 2002; Cui et al. , 2006; Leka and Barnes, 2007; Yuan et al. , 2010).Flares occur in the volume of atmospheric plasma above sunspot groups.Sunspot groups are formed by the convective action of sub-surface fluid motionspushing magnetic flux tubes through the Sun’s surface, the photosphere. Tur-bulent photospheric and subphotospheric motions jostle these flux tubes and,when the conditions are right, the sunspots produce a flare ( e.g. , Conlon et al. ,2008). Currently the exact conditions that lead to flaring are not known. Theanalysis of sunspot groups and their properties has allowed the most accurateflare prediction to date (Barnes and Leka, 2008; Bloomfield et al. , 2012; Ahmed et al. , 2013).Studying large volumes of data to identify sunspot groups and their variousproperties is generally done using feature-characterisation and tracking algo-rithms. The automatic detection of sunspots and their constituent structureshas been investigated using photospheric intensity images (Zharkov et al. , 2004;Curto, Blanca, and Mart´ınez, 2008) and magnetograms (McAteer et al. , 2005;LaBonte, Georgoulis, and Rust, 2007; Lefebvre and Rozelot, 2004). Variousfeature tracking algorithms were developed as part of the Heliophysics EventsKnowledgebase (HEK; Martens et al. , 2012). Verbeeck et al. (2013) provide areview of the robustness of four major algorithms used to automatically detectand characterise active regions (ARs) and sunspot groups. The algorithms thatthey discuss are: the Solar Monitor Active Region Tracker (SMART; Higgins et al. , 2011), an automated AR detection and characterisation algorithm thatuses magnetograms to detect magnetic features; the Automated Solar ActivityPrediction code (ASAP; Colak and Qahwaji, 2008, 2009), a set of algorithmsthat uses intensity images and machine learning to detect and predict flares; theSunspot Tracking and Recognition Algorithm (STARA; Watson et al. , 2009);and the Spatial Possibilistic Clustering Algorithm (SPOCA; Barra et al. , 2009).A comparative study of flare prediction between SMART-ASAP (a combinationof the feature detection of SMART and the machine learning of ASAP) andASAP was performed in Ahmed et al. (2013), which found SMART-ASAP to bethe more accurate.Sunspot groups have traditionally been classified using the McIntosh (McIn-tosh, 1990) and Mount Wilson (Hale et al. , 1919) classification schemes. The
SOLA: paper_final.tex; 14 October 2018; 15:45; p. 2 utomatic δ -spot detection McIntosh scheme classifies the complexity of sunspot groups from their white-light structure, whereas the Mount Wilson scheme uses magnetograms to classifythe spatial distribution of magnetic polarities.The original Mount Wilson classification scheme consisted of unipolar ( α ),bipolar ( β ) and mixed polarity ( γ ) designations, before being extended intothe current Mount Wilson scheme through the addition of the δ designation– opposite polarity umbrae being enclosed by a common penumbra (K¨unzel,1960, 1965). A statistical investigation of the connection between δ spots andmajor solar flares was published by Sammis, Tang, and Zirin (2000), findingthat the vast majority of X-class flares occur from sunspot groups that displaya δ configuration at some time. However, Mount Wilson classifications are onlyissued every 24 hours by the National Oceanic and Atmospheric Administration(NOAA) Space Weather Prediction Center (SWPC), so it remains unclear howlong it takes to produce a major flare after a δ spot forms. Automated detectionof δ spot formation from high-cadence data is necessary to answer this, with real-time implementation being beneficial for flare forecasting. Here, a new softwaremodule named SMART-Delta Finder (SMART-DF) is developed as an extensionof the SMART algorithm to automatically detect and characterise δ spots usingsimultaneous intensity and magnetogram data.In this paper, the SMART-DF algorithm is explained in Section 2. TheSMART-DF code is tested on several individual cases, with Section 3 presentingthe observation and data analysis. Section 4 includes the results and Section 5presents the conclusions and discussion of this work.
2. SMART-DF Algorithm
SMART-DF is a software package developed using Interactive Data Language(IDL) and SolarSoft (Freeland and Handy, 1998). The definition of the δ clas-sification as per K¨unzel (1965) is simply, a sunspot group with a “commonpenumbra enclosing opposite polarity umbrae”. Mt. Wilson Observatory uses amore specific definition for δ spots as “umbrae separated by less than 2 ◦ withinone penumbra have opposite polarity” . The only addition in this definition,compared to K¨unzel (1965), is the specific value for the maximum distancebetween two opposite polarity umbrae. SMART-DF is developed based on thisdefinition, and hence the distance threshold between the two opposite polarityumbrae is kept as 2 ◦ . This algorithm implements the following two conditionsfor δ detection. First, the region should have two opposite polarity umbrae withcentroids separated by less than 2 ◦ heliographic distance. Second, these twoumbrae should be surrounded by a common penumbra. These two conditionsmust be satisfied to classify a sunspot group as a δ spot. In SMART-DF thecandidates that satisfy the first condition are named δ candidates, while thosethat satisfy both conditions are labelled as δ spots.A simultaneously observed photospheric intensity image and line-of-sight (LOS)magnetogram are the necessary input to the code. The code also uses some SOLA: paper_final.tex; 14 October 2018; 15:45; p. 3 adinhatteeri et al.
X (arcsecs)-280-260-240-220-200-180-160 Y ( a r cs e cs ) (a) Umbra-penumbra identified (b) Umbrae centroids identified - distance estimated -50 0 50 100X (arcsecs)-280-260-240-220-200-180-160 Y ( a r cs e cs ) (d) δ - spot identified -50 0 50 100X (arcsecs)-280-260-240-220-200-180-160 Y ( a r cs e cs ) (c) Expected & observed penumbral area estimated Figure 1.
Summary of the SMART-DF algorithm. (a) Umbra-penumbra boundaries areidentified using continuum intensities and LOS magnetic field values. Blue (Red) contours markthe negative (positive) umbral border and green contours mark the penumbral boundary. (b)Each umbra is labeled based on area and the centroids are identified and marked. A plus-symbolis used to denote positive and an encircled cross is used to denote negative polarities. Alsothe distance between each positive-negative pair is estimated. (c) The boundary of expectedpenumbral regions around those positive (negative) umbrae which pass condition 1 (oppositepolarity umbrae with in 2 ◦ ) are marked with a red (blue) contour. (d) If all conditions in thealgorithm are satisfied, the identified δ spot is marked with a magenta circle. See Section 2 fora detailed explanation of the SMART-DF algorithm. parameters which must be provided as input. These include choosing a particularfield of view (FOV), instead of full disc, to detect δ spots. When a δ spot is found,SMART-DF will mark the location on an image, along with an output structureproviding details of magnetic and geometric properties of the identified features.The values of some input parameters are critical, since they decide the successof detection. The parameters and their optimised values used in this study arelisted in Table 1, and the reason for choosing these values for each parameterare explained in Section 3. Figure 1 shows the major steps of the SMART-DFalgorithm, which are described in the sections below. Regions of umbrae and penumbrae are identified using simultaneously observedcontinuum and magnetogram images. First, the intensity images are correctedfor limb-darkening using the Allen (1976) model. For each intensity image, asimultaneous magnetogram is also obtained and cosine-correction is applied tothe magnetogram (McAteer et al. , 2005). A histogram of continuum intensityfor a region with a mature sunspot generally shows a triple-peaked distribution
SOLA: paper_final.tex; 14 October 2018; 15:45; p. 4 utomatic δ -spot detection N u m be r o f p i x e l s Quiet SunPenumbraUmbra
Figure 2.
A histogram of the intensity distribution of a mature sunspot that is normalisedto the quiet-Sun intensity. The peak at 1.0 corresponds to quiet-Sun pixels, and the other twopeaks correspond to penumbrae and umbrae, as indicated in the figure. as shown in Figure 2. The Mean quiet-Sun intensity is defined as I QS , andintensities less than 0.65 I QS correspond to umbral pixels. Similarly, the peak at ≈ I QS corresponds to the penumbral intensity (Leka and Skumanich, 1998).In this study, all pixels with intensity less than 0.65 I QS and with a line-of-sightmagnetic-field intensity, | B LOS | ≥
500 G are defined as umbral pixels. Similarly,pixels within the intensity range of 0.65 – 0.9 I QS are defined as penumbral pixelsand above 0.9 I QS are defined as quiet Sun. These values are used globally forall images. Also, since our results are not dependent on flux within the umbralarea, the errors in the intensity levels used are not important. The above selec-tion procedure is verified manually by plotting contours around the umbra andpenumbra. An example can be seen in the top-right image of Figure 1. To avoid detecting very small pores and inter-granular lanes, only umbrae abovea certain minimum cut-off area ( A min ) are used to detect δ spots. The value usedfor A min in this study is given in Table 1. Each umbra with an area larger than A min in the FOV is labeled based on its size and polarity. Hence, the largestpositive-polarity umbra will be labeled U +1 , the next largest as U +2 , U +3 , ..., U + n and similarly, U − , U − , ..., U − m for the negative polarity. The distance betweeneach possible pairing of opposite polarity umbrae is calculated. The distance ismeasured from the center of the umbra, taken as the B LOS weighted centroid.Those pairs of umbrae with a distance of less than 2 ◦ in heliographic coordinatesare marked as first-level δ candidates, which satisfy the first condition of the SOLA: paper_final.tex; 14 October 2018; 15:45; p. 5 adinhatteeri et al. definition at the beginning of this section. Note that each candidate is a pair ofpositive and negative polarity umbrae, and only these candidates are consideredfor further analysis.
Penumbrae form around the umbrae, with or without azimuthal symmetry. Aregion around an umbral border with a certain width (the penumbral expectedlength, P el ) is selected and marked as the expected penumbral region ( P expected ).Actual penumbral filaments can be bigger or smaller in size than P el in differentcases. Hence, P expected is simply the area around umbrae where we expect tofind penumbrae. Based on observed intensity levels, each pixel in this regionis marked as either penumbral or non-penumbral. The area of such observedpenumbra ( P observed ) inside P expected is calculated. This is done for both the pairof umbrae in each δ candidate. Ideally, if penumbrae are formed with azimuthalsymmetry and fill the P expected region, then the ratio P frac = P expected /P observed should be equal to unity. This is not always the case. In most of the casesthe penumbra formation is not azimuthally symmetric and in some cases thepenumbral size can be less than the width of the P expected region, given by theparameter P el . Hence, expecting such a ratio to be equal to unity is not alwayspractical. The ratio also depends on the value of P el that we choose as input. Inthis study we have used different values of this ratio to qualify for second-level δ candidates, as explained in Section 4.The third level checks whether the penumbra around the pair of umbrae areconnected. This is done by determining whether P observed around each umbra inthe pair is connected (sharing some common area, P share ). The pairs of oppositepolarity spots that pass the three checks are marked as δ spots. This algorithmbecomes unreliable when the active region approaches the limb, due to projec-tion effects of both foreshortening and apparent ( i.e., false) LOS magnetic-fieldpolarity inversions. In this paper, we test SMART-DF only for ARs within ± ◦ longitude.
3. Observations and Data Analysis
SMART-DF was tested using data from the
Helioseismic and Magnetic Imager (HMI; Scherrer et al. , 2012). The HMI instrument was developed by the StanfordSolar Group and is a part of NASA’s
Solar Dynamic Observatory (SDO). HMIcan be used to obtain continuum intensity images as well as LOS magnetic-fieldstrength of the full solar-disc at high temporal cadence (every 45 seconds). HMIalso obtains magnetic-field vector information using polarimetry, but this is notused in this study. The primary input to SMART-DF is a pair of simultaneouscontinuum intensity images and magnetograms. All the ARs that appear on thesolar-disc during 2011 and 2012, with in ± ◦ longitude were analysed to testthe reliability of SMART-DF. A pair of intensity images and LOS magnetogramswere obtained with 12 hour cadence, i.e., at 00:00 UT and 12:00 UT every dayfrom the archived HMI data. SMART (Higgins et al. , 2011) is used to automat-ically identify ARs and define regions of interest (ROIs) on the solar-disc. All SOLA: paper_final.tex; 14 October 2018; 15:45; p. 6 utomatic δ -spot detection Table 1.
The parameters used by SMART-DF and their optimised values used for thisalgorithm testing study.Parameter Value Used Details A min Minimum/cut-off umbral area P el P frac
75% Ratio of actual penumbra with expected penumbral area P share
750 km Common shared area between penumbrae of two spots
ARs located within ± ◦ of longitude have been checked for δ configurationsusing SMART-DF as explained in Section 2. To calculate distances and areas onthe solar-disc, firstly helioprojective coordinates are converted to heliographiccoordinates using the world coordinate-system (WCS; Greisen and Calabretta,2002) programs (Thompson, 2006) available in the standard solar data analysissoftware (SolarSoft SSWIDL) and then spherical-trigonometry cosine-law is used(Smart, 1965).SWPC, operated by NOAA, releases a Solar Region Summary (SRS) textfile every day at 00:30 UT. SRS files contain the Hale classification of each ARon disc. NOAA classifies each active region by visually analysing the previousday data. Each day SRS between 1 January 2011 and 31 December 2012 wereobtained from the SWPC database and a list of ARs that formed δ configurationswas created. Detections of δ spots from SMART-DF were compared with thislist and a comparison test was performed using two methods.The first method tests whether all ARs detected by NOAA as having a δ configuration on a particular day were also detected by SMART-DF. For this,those days on which a new AR appeared on the solar-disc were noted down (basedon NOAA numbers in the SRS files). Each AR that formed a δ configurationaccording to the SRS files during 2011 and 2012 were listed and the date of itsfirst detection as a δ was noted. SRS is based on observations of the Sun theprevious day to the ‘release date’. Hence, SMART-DF was used to check for δ configurations in the data of that day (the previous day of the SRS release datethat has a δ spot). In this test, SMART-DF is limited to find less than or anequal number of δ spots compared to SRS. The results have been compared andare presented in Section 4 of this paper.The second method tests the relative detection rates of NOAA/SWPC andSMART-DF. On each day, the number of δ configurations identified by NOAAare counted and plotted as a histogram distribution. Full-disc images (simul-taneous continuum and magnetogram) are obtained with a 12-hour cadence,meaning twice per day for each day in 2011 and 2012. SMART-DF was used tolook for δ configurations and a similar histogram was created and over-plottedfor comparison. SMART-DF can detect less than, equal to or more instances of δ configurations NOAA/SWPC detections.There are four important input parameters needed to obtain reliable resultsfrom SMART-DF, as explained in the previous section (Section 2). They are:i) A min , the minimum umbral area to be counted as an umbra; SOLA: paper_final.tex; 14 October 2018; 15:45; p. 7 adinhatteeri et al. ii) P el , the expected length of a penumbra (or rudimentary penumbra, as it iscalled in the case of a forming penumbra);iii) P frac , the fraction of penumbral area observed with respect to the expectedpenumbral area.iv) P share , the common area shared by the penumbrae of two opposite polarityspots.The optimisation of these four parameters is critical for successful δ spotdetection. The optimised values used for this study are given in Table 1. The op-timisation was performed by varying these parameters to find the combinationswith the maximum successful detection rate of NOAA detected regions. Usersare free to change these values in the code to suit their needs or if they find anyanomalous AR going undetected ( e.g. , see the discussion on NOAA AR 12192 inSection 5). The minimum umbral area required to qualify as an umbra is fixedat 1.5 Mm , which is greater than the typical size-scale of granulation (knownto be ∼ P el could be aslarge as the minimum distance that is needed to pass condition 1 of a δ spot( i.e. , 2 ◦ in heliographic coordinate system). However, intergranular lanes andother intensity regions seen in complex active regions can be wrongly markedas penumbra when P el is too large. Hence, it is important to choose an optimalvalue that matches with typical penumbral length-scales. The third parameter isthe fraction of actual penumbral area with respect to expected penumbral area.Ideally this has to be equal to unity ( i.e. , 100%). However, the actual penumbrallength can be different from case to case, and in many cases penumbrae will notbe formed completely around both umbrae. Hence if we fix this value to 100%,even if there is a small region without penumbra, the algorithm will fail to countthem as a δ candidate. This fraction can be significantly different from caseto case, so SMART-DF was tested with different values of P frac on a known δ region. This resulted in an optimal value of 75% for P frac which was used in therest of this study. This value also depends on the value of P el chosen. The fourthparameter, P share , is fixed at 750 km (2 pixels ×
4. Results
The SMART-DF algorithm was tested on multiple ARs. One example can be seenin the bottom-right image in Figure 1. Three other examples of δ spot detectionare shown in Figure 3. In each row, the left side shows an HMI photosphericintensity image and the right side shows an HMI LOS magnetogram. The positive(north) and negative (south) polarity umbral areas are marked with red and bluecontours, respectively. Green contours mark penumbral areas. A circle is drawnto highlight the region of interest centred at the δ forming region.In the upper row in Figure 3, two main spots with opposite polarity areobserved, with the leading spot having negative polarity and the following spot SOLA: paper_final.tex; 14 October 2018; 15:45; p. 8 utomatic δ -spot detection -450 -400 -350 -300 -250X (arcsecs) Y ( a r cs e cs ) -450 -400 -350 -300 -250X (arcsecs)6080100120140160180200 Y ( a r cs e cs ) NOAA 11302 - 2011.09.27 00:01:30 -600 -550 -500 -450X (arcsecs) Y ( a r cs e cs ) -600 -550 -500 -450X (arcsecs)150200250300 Y ( a r cs e cs ) NOAA 11476 - 2012.03.09 01:01:30 -350 -300 -250 -200X (arcsecs) Y ( a r cs e cs ) -350 -300 -250 -200X (arcsecs)-320-300-280-260-240-220-200-180 Y ( a r cs e cs ) NOAA 11861 - 2013.10.11 09:37:30
Figure 3.
An example of a δ spot detected using the automatic code. The red (blue) contoursshow the positive (negative) umbrae and the green contours mark the penumbral border. Apink circle surrounds the location of a detected δ configuration. having positive polarity. The δ formation occurs in the region of flux emergencebetween the two main spots. It is also observed in this case that the δ spotis formed by multiple small umbrae with opposite polarity and having a rudi-mentary penumbra around the whole region. A similar configuration is observedin the emerging region, but with two distinct, large and stable bipolar spotsin the bottom right image of Figure 1. The central row of the Figure 3 showsanother case of two distinct δ spot formations in the same AR. One occurs inthe emerging-flux region, similar to the earlier case. Another δ is formed by asmall positive-polarity spot joining with a large negative leading-spot. Both spotsshare a common penumbra surrounding both polarities. The fourth example ofa δ spot detected by SMART-DF is given in the bottom row of Figure 3. Two δ forming pairs can be seen here. One of them is similar to the above example, SOLA: paper_final.tex; 14 October 2018; 15:45; p. 9 adinhatteeri et al. where a small positive-polarity spot joins with a large, stable negative-polarityspot. In some cases opposite polarities of a small bipolar region may approacheach other and form a rudimentary penumbra for a short time. SMART-DFindicates them as a δ since they satisfy both required conditions, and hence aremarked with a small circle in the figure (bottom row). Such cases are generallyseen at locations of flux emergence. A user can choose to avoid the detectionof such small-scale and transient δ spots by increasing the cut-off umbral-areathreshold ( A min ).The success rate of δ detection was also tested by comparing with the detec-tions by NOAA/SWPC. There are two methods of performing this comparison,as explained in Section 3. The first method of testing is to check whether eachAR is detected by both NOAA and SMART-DF on its first day of observation.During 2011 – 2012, NOAA’s SRS files reported 33 ARs as having a δ pair atsome part of their appearance on the solar-disc. Out of these, 23 ARs formeda δ within ± ◦ longitude. SMART-DF detected 21 out of these 23 NOAAdetections several hours before the SRS was released by NOAA. The algorithmfailed to detect any δ regions in NOAA AR 11164 and NOAA AR 11374 whencompared to SRS. Both were manually checked, on AR 11164 there was notenough penumbra at the beginning when the opposite polarities were within 2 ◦ ,and by the time the penumbra developed the opposite polarities drifted awayto more than the cut-off distance of 2 ◦ . In the case of AR 11374 the negativepolarity around the positive spot was spread like a plage and did not form asunspot with area more than A min .The second method of testing is to compare the number of instances of daily δ detections by NOAA and SMART-DF. The δ configuration in an AR may remainso for more than one day, and it will be counted again as another instance byboth SMART-DF and NOAA. To match with NOAA’s daily SRS, SMART-DFcounts each δ region only once per day (though SMART-DF uses a cadence oftwo observation per day). During the two years studied here, NOAA detected97 instances of δ formation within ± ◦ of longitude while the SMART-DFdetected 116 cases. Distributions of the two are given in Figure 4. In theseadditional detections by SMART-DF, five ARs were never reported by NOAA.It is possible that NOAA ground-based data did not show a δ spot, due to itsinferior spatial resolution or possible increased seeing. Since only two instances per day (at 00:00 UT and 12:00 UT) were used by SMART-DF for analysis,if a δ spot formed and disappeared between the two instances (with a gap of12 hrs), SMART-DF would have missed that case, while NOAA would havedetected them as they scan through out the day. A higher cadence data analysiswould show a better match in the distribution. At the same time, SMART-DFmay detect small, instantaneous δ s which may be missed, or ignored by manualanalysis by NOAA. When multiple δ configurations are formed in the same AR,both SMART-DF and NOAA list count them as one.
5. Discussion and Conclusions
In this article we report on the development of SMART-DF, an algorithm toautomatically detect δ spots in active regions. These forms of sunspot groups are SOLA: paper_final.tex; 14 October 2018; 15:45; p. 10 utomatic δ -spot detection N o . o f δ - s po t s de t e c t ed Jan Apr Jul Oct Jan Apr Jul Oct SMART-DFNOAA-SRS
Figure 4.
Number of δ spots detected by SMART-DF and NOAA every month of 2011 – 2012.Red bars represent the SMART-DF detection and blue bars represent the detections by NOAASRS. SMART-DF detected 116 instances of δ spots when compared to 97 by NOAA during2011 – 2012. known to be associated with major solar flares (Sammis, Tang, and Zirin, 2000).Current methods to identify δ spots are primarily based on manual checking ofground-based data ( e.g. , NOAA). Our near-realtime detection of the formationof δ spots can be used to flag an AR as a potential flaring region. SMART-DFwill be integrated into SolarMonitor , and hence will be available to the public.The results of comparing SMART-DF and NOAA/SWPC δ spot detectionrates are presented. It is found that during 2011 and 2012, 21 out of 23 ARs de-tected by NOAA as δ spots (within ± ◦ longitude) were identified by SMART-DF during the 24-hour period before the SRS release by NOAA (which occurs at00:30 UT every day). In some cases SMART-DF detected δ spots an entire day(during 24 – 48 hours period) before it was marked as a δ by NOAA. In additionSMART-DF detected five ARs which were never reported by NOAA as δ region.If daily instances of δ detections were to be counted, 97 instances were detectedby NOAA compared to SMART-DF 116 cases. SMART-DF might have detectedmore δ spots if the data had been analysed using higher cadence. In this study,data were obtained only once in 12 hrs, any instance of short term δ formationand disappearance during the interval are missed.SMART-DF determines that a spot is a δ configuration when two conditionsare met: i) opposite polarity umbrae within 2 ◦ and ii) common penumbra sur-rounding both polarity umbrae. However, the conventional definition of δ spotsis vague. While the Mt. Wilson Observatory uses the specification of within 2 ◦ separation, the initial definition by K¨unzel (1960, 1965) does not mention anyvalue. The recent appearance of the extremely large region NOAA AR 12192 wasone specific case where the distance between opposite polarity umbrae (both SOLA: paper_final.tex; 14 October 2018; 15:45; p. 11 adinhatteeri et al. Y ( a r cs e cs ) -350-300-250-200 Y ( a r cs e cs ) NOAA 12192 - 2014.10.20 22:00:30 -650 -600 -550 -500X (arcsecs) Y ( a r cs e cs ) -650 -600 -550 -500X (arcsecs)-350-300-250-200 Y ( a r cs e cs ) Figure 5.
HMI observation of AR12192 and it’s analysis by SMART-DF . The top andbottom rows of the figure represent two cases with different values for the maximum alloweddistance between opposite polarity umbrae (to be detected as a δ spot). The top row is withstandard definition value of 2 ◦ , and the bottom row is with a value of 5.2 ◦ . Red (blue) contoursshow the positive (negative) umbrae and green contours mark the penumbral border. A circlesurrounds the location of a detected δ configuration. between their centroids and between their closest edges) is more than 2 ◦ . Asshown in the top row of Figure 5, SMART-DF fails to find the correct δ spotpair ( i.e. , the two largest umbrae), instead selects a dark and very small negative-polarity region on the edge of the positive-polarity umbra. The correct pair ofumbrae was detected only when the restriction of distance was increased, withthe bottom row of Figure 5 the result when separations of up to 5.1 ◦ are allowed.Although large ARs like AR 12192 are not common, there may be several can-didates in which SMART-DF failed because the umbrae separation was slightlymore than 2 ◦ . In addition, the definition of penumbrae has changed significantlyduring the last decade with the advent of high-spatial resolution observations.In earlier observations, any shadow-like region around an umbra (with intensityless than the quiet Sun, but more than the umbra) was called penumbra. Morerecently, penumbrae are known to have distinct filamentary structure with a welldefined magnetic topology. SMART-DF, like most other algorithms, uses onlyintensity values around umbrae to identify penumbrae. When two pores are inclose proximity, a shadow-like region can form between them with an intensitybetween that of the quiet Sun and the pores ( i.e. , umbrae). Currently, SMART- SOLA: paper_final.tex; 14 October 2018; 15:45; p. 12 utomatic δ -spot detection DF will identify this as a penumbra and may use that to satisfy the second δ spot condition.The detection of a δ spot does not assure the occurrence of a flare, but for flareforecasting studies it will be important to determine how useful δ spot detectionis in identifying potentially flaring ARs. In a follow-on study, SMART-DF willbe used to investigate the evolution of different physical parameters in δ spots,including the observed changes in magnetic topology and plasma velocity. Acknowledgments
This work has received financial support from: EOARD (SP), IrishResearch Council- Enterprise partnership (PAH), European Space Agency Prodex programme(DSB). Courtesy of NASA/SDO and the HMI science teams for the data used in this paper.
References
Ahmed, O.W., Qahwaji, R., Colak, T., Higgins, P.A., Gallagher, P.T., Bloomfield, D.S.: 2013,Solar Flare Prediction Using Advanced Feature Extraction, Machine Learning, and FeatureSelection.
Solar Phys. , 157.
DOI . ADS .Allen, C.W.: 1976,
Astrophysical Quantities . Athlone Press. London.
ADS .Aschwanden, M.J., Freeland, S.L.: 2012, Automated Solar Flare Statistics in Soft X-Rays over37 Years of GOES Observations: The Invariance of Self-organized Criticality during ThreeSolar Cycles.
Astrophys. J. , 112.
DOI . ADS .Barnes, G., Leka, K.D.: 2008, Evaluating the Performance of Solar Flare Forecasting Methods.
Astrophys. J. Lett. , L107.
DOI . ADS .Barra, V., Delouille, V., Kretzschmar, M., Hochedez, J.-F.: 2009, Fast and robust segmentationof solar EUV images: algorithm and results for solar cycle 23.
Astron. Astrophys. , 361.
DOI . ADS .Bloomfield, D.S., Higgins, P.A., McAteer, R.T.J., Gallagher, P.T.: 2012, Toward ReliableBenchmarking of Solar Flare Forecasting Methods.
Astrophys. J. Lett. , L41.
DOI . ADS .Colak, T., Qahwaji, R.: 2008, Automated McIntosh-Based Classification of Sunspot GroupsUsing MDI Images.
Solar Phys. , 277.
DOI . ADS .Colak, T., Qahwaji, R.: 2009, Automated Solar Activity Prediction: A hybrid computer plat-form using machine learning and solar imaging for automated prediction of solar flares.
Space Weather , 6001. DOI . ADS .Conlon, P.A., Gallagher, P.T., McAteer, R.T.J., Ireland, J., Young, C.A., Kestener, P., Hewett,R.J., Maguire, K.: 2008, Multifractal Properties of Evolving Active Regions.
Solar Phys. , 297.
DOI . ADS .Cui, Y., Li, R., Zhang, L., He, Y., Wang, H.: 2006, Correlation Between Solar Flare Productiv-ity and Photospheric Magnetic Field Properties. 1. Maximum Horizontal Gradient, Lengthof Neutral Line, Number of Singular Points.
Solar Phys. , 45.
DOI . ADS .Curto, J.J., Blanca, M., Mart´ınez, E.: 2008, Automatic sunspots detection on full-disk solarimages using mathematical morphology.
Solar Phys. (2), 411.
DOI . ADS .Emslie, A.G., Dennis, B.R., Shih, A.Y., Chamberlin, P.C., Mewaldt, R.A., Moore, C.S., Share,G.H., Vourlidas, A., Welsch, B.T.: 2012, Global Energetics of Thirty-eight Large SolarEruptive Events.
Astrophys. J. , 71.
DOI . ADS .Freeland, S.L., Handy, B.N.: 1998, Data Analysis with the SolarSoft System.
Solar Phys. ,497.
DOI . ADS .Gallagher, P.T., Moon, Y.-J., Wang, H.: 2002, Active-Region Monitoring and Flare ForecastingI. Data Processing and First Results.
Solar Phys. , 171.
DOI . ADS .Gopalswamy, N., Barbieri, L., Lu, G., Plunkett, S.P., Skoug, R.M.: 2005, Introduction to thespecial section: Violent Sun-Earth connection events of October-November 2003.
Geophys.Res. Lett. , 3. DOI . ADS .Greisen, E.W., Calabretta, M.R.: 2002, Representations of world coordinates in FITS.
Astron.Astrophys. , 1061.
DOI . ADS .Hale, G.E., Ellerman, F., Nicholson, S.B., Joy, A.H.: 1919, The Magnetic Polarity of Sun-Spots.
Astrophys. J. , 153. DOI . ADS . SOLA: paper_final.tex; 14 October 2018; 15:45; p. 13 adinhatteeri et al.
Hapgood, M., Thomson, A.: 2010,
Space weather: Its impact on earth and implications forbusiness , Lloyd’s 360 Risk Insight, London.Higgins, P.A., Gallagher, P.T., McAteer, R.T.J., Bloomfield, D.S.: 2011, Solar magnetic featuredetection and tracking for space weather monitoring.
Adv. in Space Res. , 2105. DOI . ADS .K¨unzel, H.: 1960, Die Flare-H¨aufigkeit in Fleckengruppen unterschiedlicher Klasse undmagnetischer Struktur.
Astron. Nach. , 271.
ADS .K¨unzel, H.: 1965, Zur Klassifikation von Sonnenfleckengruppen.
Astron. Nach. , 177.
ADS .LaBonte, B.J., Georgoulis, M.K., Rust, D.M.: 2007, Survey of Magnetic Helicity Injection inRegions Producing X-Class Flares.
Astrophys. J. , 955.
DOI . ADS .Lefebvre, S., Rozelot, J.-P.: 2004, A new method to detect active features at the solar limb.
Solar Phys. , 25.
DOI . ADS .Leka, K.D., Barnes, G.: 2007, Photospheric Magnetic Field Properties of Flaring versus Flare-quiet Active Regions. IV. A Statistically Significant Sample.
Astrophys. J. , 1173.
DOI . ADS .Leka, K.D., Skumanich, A.: 1998, The Evolution of Pores and the Development of Penumbrae.
Astrophys. J. , 454.
DOI . ADS .Martens, P.C.H., Attrill, G.D.R., Davey, A.R., Engell, A., Farid, S., Grigis, P.C., Kasper, J.,Korreck, K., Saar, S.H., Savcheva, A., Su, Y., Testa, P., Wills-Davey, M., Bernasconi, P.N.,Raouafi, N.-E., Delouille, V.A., Hochedez, J.F., Cirtain, J.W., Deforest, C.E., Angryk, R.A.,de Moortel, I., Wiegelmann, T., Georgoulis, M.K., McAteer, R.T.J., Timmons, R.P.: 2012,Computer Vision for the Solar Dynamics Observatory (SDO).
Solar Phys. , 79.
DOI . ADS .McAteer, R.T.J., Gallagher, P.T., Ireland, J., Young, C.A.: 2005, Automated Boundary-extraction And Region-growing Techniques Applied To Solar Magnetograms.
Solar Phys. , 55.
DOI . ADS .McIntosh, P.S.: 1990, The classification of sunspot groups.
Solar Phys. , 251.
DOI . ADS .Messerotti, M., Zuccarello, F., Guglielmino, S.L., Bothmer, V., Lilensten, J., Noci, G., Storini,M., Lundstedt, H.: 2009, Solar Weather Event Modelling and Prediction.
Space Sci. Rev. , 121.
DOI . ADS .Moore, R.L., Sterling, A.C., Hudson, H.S., Lemen, J.R.: 2001, Onset of the Magnetic Explosionin Solar Flares and Coronal Mass Ejections.
Astrophys. J. , 833.
DOI . ADS .Nordlund, ˚A., Stein, R.F., Asplund, M.: 2009, Solar Surface Convection.
Living Reviews inSolar Phys. , 2. ADS .Sammis, I., Tang, F., Zirin, H.: 2000, The Dependence of Large Flare Occurrence on theMagnetic Structure of Sunspots.
Astrophys. J. , 583.
DOI . ADS .Scherrer, P.H., Schou, J., Bush, R.I., Kosovichev, A.G., Bogart, R.S., Hoeksema, J.T., Liu, Y.,Duvall, T.L., Zhao, J., Title, A.M., Schrijver, C.J., Tarbell, T.D., Tomczyk, S.: 2012, TheHelioseismic and Magnetic Imager (HMI) Investigation for the Solar Dynamics Observatory(SDO).
Solar Phys. , 207.
DOI . ADS .Smart, W.M.: 1965,
Text-book on spherical astronomy . Cambridge univ. press.
ADS .Solanki, S.K.: 2003, Sunspots: An overview.
Astron. Astrophys. Rev. , 153. DOI . ADS .Thompson, W.T.: 2006, Coordinate systems for solar image data.
Astron. Astrophys. ,791.
DOI . ADS .Verbeeck, C., Higgins, P.A., Colak, T., Watson, F.T., Delouille, V., Mampaey, B., Qahwaji,R.: 2013, A Multi-wavelength Analysis of Active Regions and Sunspots by Comparison ofAutomatic Detection Algorithms.
Solar Phys. , 67.
DOI . ADS .Watson, F., Fletcher, L., Dalla, S., Marshall, S.: 2009, Modelling the Longitudinal Asymmetryin Sunspot Emergence: The Role of the Wilson Depression.
Solar Phys. , 5.
DOI . ADS .Yuan, Y., Shih, F.Y., Jing, J., Wang, H.-M.: 2010, Automated flare forecasting using astatistical learning technique.
Research in Astronomy and Astrophysics , 785. DOI . ADS .Zharkov, S., Zharkova, V., Ipson, S., Benkhalil, A.: 2004, Automated recognition of sunspotson the soho/mdi white light solar images. In: Negoita, M., Howlett, R., Jain, L. (eds.)
Knowledge-Based Intelligent Information and Engineering Systems , Lect. Notes in Comp.Sci. , Springer Berlin Heidelberg, ISBN 978-3-540-23205-6.
DOI ..