TARPs: Tracked Active Region Patches from

TARPs: Tracked Active Region Patches from SOHO/MDI
SH23A
2087
Michael Turmon (JPL/Caltech); J. Todd Hoeksema, Monica Bobra (Stanford University)
Synoptic 2001 M-TARPs (lines mark month boundaries)
Summary
Methodology: Finding Active Regions
• 
We are developing a new data product for the MDI Resident Archive containing
tracked magnetic features on the scale of solar active regions, abbreviated MDITARPs (MDI Tracked Active Region Patches). This data product, derived from lineof-sight (LOS) magnetograms and continuum intensitygrams, is a companion to the
already-released HARP (HMI Active Region Patch) data product from HMI.
Together, the two data products cover May 1996 to the present, and should
eventually span two solar cycles.
• 
We first compute a full-disk activity mask given input magnetogram and
intensity images (Turmon et al. 2002) taking spherical geometry into
account (Turmon et al. 2010).
Large spatially-coherent regions are identified within the LOS magnetograms and
intensitygrams and tracked from image to image, accounting for merges as regions
grow. After the region disappears, the numbered track (“TARP”) is placed into a data
series by finding the smallest box of constant latitude/longitude extent that
encompasses all appearances of the region.
Mask:
2011/02/14 12:00
• 
Following Liu et al. 2012, MDI mask activity model
is obtained from HMI model by scaling by 1.4.
• 
Sample HMI mask (right) shows
typical mask appearance.
The MDI-TARP data series provides all geometric and heliographic information
needed to track active patches in MDI and other solar data sets. For each
numbered TARP, the data series defines at each time step a rectangular CCD cutout, and it provides a mask within the cut-out indicating the active pixels within a
regular, smoothly-evolving blob.
• 
Group pixel-scale activity from masks
into NOAA AR-scale regions
We used a matched filter approach
with an elongated Gaussian kernel of
FWHM ~50x25Mm (~40x20 MDI pixels)
at disk center.
MDI Mask
Summary keywords such as areas and integrated fluxes are included for each
appearance of the region. The data product described here is in draft form, with
release as a data series on Stanford University’s JSOC expected in June 2014.
• 
• 
After
M-TARPs are indexed by a number called HARPNUM, analogous to NOAA
AR number, and time step (T_REC).
• 
For a single M-TARP, the data series is a list of rectangular patches and
metadata containing the observed lifespan of the TARP. A 1-day pad is
appended on both ends of the TARP. This padding period is seen in the
empty, dotted boxes in the tracked frames (e.g., in the next poster panel).
• 
Each rectangular patch in the list is a cutout from the image plane. A patch
can be overlaid on a full-disk image (e.g., the LOS magnetograms) by a
simple coordinate shift. Patch WCS are included for other projections.
where P is a permutation matrix giving the B-to-A mapping.
Fast, exact solution by linear programming.
• 
The footprint covered by the patch is determined by the the smallest lat/lon
bounding box that encompasses all appearances of that TARP. See below.
• 
Chaining this association across many frames yields complete tracks.
• 
• 
Tracks can be finalized only after they are unseen for a long-enough time.
The animating idea behind the sizing of the lat/lon bounding box is that the
observer is hovering over the AR, staying at a constant latitude and moving
at a constant angular rate in longitude.
• 
Shown below are four patches of the 200 total making up M-TARP 7570.
The colored masks below are stored as bitmaps in a suitable integer
encoding.
• 
Compute the overlap area between extrapolated
track (using standard latitude-dependent motion
relationship) and new region.
• 
Overlap of patch a in A and patch b in B is D(a,b).
• 
Solve linear assignment problem to match A up to B:
A
B
Merging Tracks and Complex ARs
Kernel
at Limb
Kernel at
Disk Center
•  Tracks are first identified using past and current data. Thus, growing regions may
merge in later appearances.
•  Coping with the consequences of merges adds considerable complexity to the
implementation. This complexity is hidden in the final data product.
Convolved with Template
Identified Groups
•  Care is taken so that regions near the detection threshold are not cut into separate
temporal pieces as they exceed and then sink below the threshold.
2002.04.09
01:36
2002.04.11
17:36
2002.04.17
01:36
à
Time
2002.04.18
09:36
2002 Sep. 02, 11:11 TAI
Tracked Frames
• 
Before
Chain regions together to make a track: singlelink most likely tracker using overlap area
•  The MDI TARP is often not a contiguous region. For shrinking regions, the M-TARP
lat/lon bounding box can be much larger than the currently active area. It can even
contain other TARPs. Use the bitmap to determine what is part of the TARP.
This work was sponsored by NASA’s Heliophysics HDEE Program Element.
• 
The MDI TARP Data Product
• 
Mask:
zoom
Active Region Grouping
• 
• 
Bayesian approach trades off pixel-by-pixel agreement of the mask value to
the data against spatial coherence of labels (a prior).
Mask Model and Example
The MDI-TARP data series is intended for:
•  Subsetting individual active regions
•  Computation of space weather indices for individual active regions
•  Facilitating long-term or synoptic statistical studies of active regions
Methodology: Active Region Tracking
MDI TARP vs. HMI HARP Correspondence
MDI TARP vs. HMI HARP Region Boundaries
We checked for correspondence between MDI TARPs and HMI HARPs using the
~140 ARs in the May–October 2010 overlap period.
By checking location and shape, we determined matching regions (in green) and
misses (in gray). We find 130 matches and 11 misses of each type (TARP present but
no HARP, and vice versa).
The quick-looks here are sampled 1/day
from the 96m (15/day) full series.
M-TARPs are colored blobs. NOAA ARs
are yellow crosses.
The M-TARP/NOAA correspondence is
found and coded in keywords.
• 
For the 2010 overlap, we overlaid the nine largest MDI TARPs over the HMI
HARPs, projecting the MDI TARPs into the HMI coordinates by image WCS.
• 
Some HMI HARPs enclose more activity within one HARP (blue blobs below
correspond to differently-numbered MDI TARPs).
Color Key
HMI HARP 92 + MDI TARP 14113 at 2010.07.25 19:12:00 TAI
MDI TARP Pixels = 8601, MDI Active Pixels = 1320
HMI HARP 226 + MDI TARP 14242 at 2010.10.26 19:12:00 TAI
MDI TARP Pixels = 7192, MDI Active Pixels = 1193
50
100
Outside the HARP & M-TARP
150
Centroid of MDI-TARPs and HMI HARPs
Synoptic: May 2010 – October 2010
40
Example TARP merges: green (7570),
pink (7561). Expired: violet (7507),
red (7538), etc. New: tan (7586), etc.
In the M-TARP, outside HARP
50
200
250
100
300
100
200
300
400
500
600
700
800
900
1000
In the HARP, outside M-TARP
1100
150
HMI HARP 211 + MDI TARP 14226 at 2010.10.15 00:00:00 TAI
MDI TARP Pixels = 9358, MDI Active Pixels = 776
HARP & M-TARP coincide
200
30
50
100
250
Active in M-TARP, not in HARP
150
300
200
20
Active in HARP, not in M-TARP
Latitude (deg.)
250
350
300
350
10
450
MDI TARP
0
100
200
• 
HARP (HMI)
MDI−HMI Match OK
−10
Active in HARP & M-TARP
400
400
No MDI or HMI match
300
400
500
600
700
800
900
1000
100
1100
200
300
400
500
600
More typically, as below, the HARP and the M-TARP coincide well.
HMI HARP 115 + MDI TARP 14136 at 2010.08.09 20:48:00 TAI
MDI TARP Pixels = 10125, MDI Active Pixels = 810
HMI HARP 86 + MDI TARP 14105 at 2010.07.14 22:24:00 TAI
MDI TARP Pixels = 15672, MDI Active Pixels = 1646
HMI HARP 104 + MDI TARP 14127 at 2010.08.02 04:48:00 TAI
MDI TARP Pixels = 8138, MDI Active Pixels = 813
50
50
−20
50
100
100
100
150
150
200
150
200
−30
250
200
250
300
250
350
300
−40
300
2097
2098
2099
2100
2101
2102
450
350
400
2096
400
350
2103
500
450
400
500
450
100
Carrington Rotation Number
100
Some extra HARPs are found due to enhanced spatial/temporal resolution of HMI.
With current settings, some extra M-TARPs are found due to grouping (next panel).
These results are consistent with the good MDI/HMI agreement found by Liu et al.
(2012), especially for relatively high fields.
200
300
400
500
600
700
800
200
300
400
500
600
700
800
900
500
100
HMI HARP 185 + MDI TARP 14198 at 2010.09.23 20:48:00 TAI
MDI TARP Pixels = 8542, MDI Active Pixels = 762
200
300
400
500
600
HMI HARP 187 + MDI TARP 14205 at 2010.09.29 08:00:00 TAI
MDI TARP Pixels = 8866, MDI Active Pixels = 1329
700
HMI HARP 175 + MDI TARP 14187 at 2010.09.17 14:24:00 TAI
MDI TARP Pixels = 12522, MDI Active Pixels = 1412
50
50
100
50
100
150
100
150
150
200
200
200
250
250
250
300
300
350
300
350
400
350
100
200
300
400
500
600
700
800
100
200
300
400
500
150
Histogram of Track Lengths
MDI TARP Data Characteristics
• 
Tracks cover April 1996 – October 2010 (all usable MDI)
• 
15 years, 72100 masks, 6170 M-TARPs.
• 
Median M-TARP length = 4.1 days = 61 frames
459 are ≥ 12 days (180 frames)
• 
The ease of computing per-AR quantities should
enable new studies that would have been prohibitive.
Status and Plans
100
The draft data product here will be improved before an expected June 2014 release:
Excess of short−duration ARs
partly due to image artifacts
from energetic particles
1 day = 15 synoptic images
• 
Better suppression of extraneous patches due to energetic particles.
• 
Improved correspondence of MDI TARPs to HMI HARPs.
• 
We hope to identify sunspots within the activity mask.
50
0
0
1
2
3
4
5
6
7
8
9
10
11
Track Length (days) Excluding 1-day Padding
12
13
14
15
16
References
M. Turmon, J. T. Hoeksema, X. Sun, M. Bobra, “HARPs – Tracked active region patch data
product from SDO/HMI,” 2012 AGU fall meeting, abstract #SH13A-2246.
M. Turmon, H. Jones, J. Pap, O. Malanushenko, “Statistical feature recognition for
multidimensional solar imagery”, Solar Physics, 262(2), 2010.
Y. Liu, J. T. Hoeksema, P. H. Scherrer, et al., “Comparison of line-of-sight magnetograms taken
by SDO/HMI and SOHO/MDI,” Solar Physics, 279(1), 2012.
H. Jones, G. Chapman, K. Harvey, J. Pap, D. Preminger, M. Turmon, S. Walton, “A comparison
of feature classification...”, Solar Physics, 248(2), 2007.
M. Turmon, J. Pap, S. Mukhtar, “Statistical pattern recognition for labeling solar active regions:
Application to SoHO/MDI,” Astrophys. Jour., 568(1), 2002, 396-407.
600
700
800
900
1000
100
200
300
400
500
600
700
turmon@jpl.nasa.gov
todd@sun.stanford.edu
mbobra@sun.stanford.edu
National Aeronautics
and Space Administration
Jet Propulsion Laboratory
California Institute of Technology
Pasadena, California
Copyright 2012. All rights reserved.
800
900