Automated Detection and Characterization of Solar Filaments and Sigmoids

K. Wagstaff, D. M. Rust,

B. J. LaBonte and P. N. BernasconiJohns Hopkins University Applied Physics Laboratory

Laurel, Maryland USA

Solar Image Recognition Workshop

Brussels

October 23-24, 2003

Objectives of Solar Filament Detection and Classification

• Report automatically on filament disappearances

• Provide warning of geomagnetic storms• Characterize magnetic flux rope chirality and

orientation of principal axis• Forecast pattern of Bz in magnetic clouds

Filaments observed in H on 1 January 2003 at 1708 UTC (BBSO image)

Filament Detection Method

• Identify filament pixels– Apply darkness threshold– Group dark pixels into contiguous regions– Prune out small dark regions and artifacts– Draw contours around filament boundary

• Find spines (filament centerlines)– Use simplified Kegl’s algorithm for finding the

principal curve defined by a set of points

Detected filaments with borders outlined.

Filaments with spines indicated.

Find Barbs (protrusions from filament)

• Identify points farthest from the spine• Follow boundary in each direction to find

bays, i.e. local minimum distances from spine

• Establish each barb centerline by connecting the farthest point to the midpoint of left and right bays

Barbs indicated by white lines.

Chirality (handedness) Classification

• Calculate angle between barb centerline and spine

• Classify barbs by obtuse and acute angles• Assign filament chirality based on majority

classification: right-handed for acute angles; left-handed for obtuse angles

Deducing filament chirality from barb counts.

The solar disk observed in H on 30 June 2002 at 1540 UTC (BBSO image). Ten filaments identified, five filaments classified.

Contoured filament with first approximation to spine.

Second approximation.

Fourth approximation.

Sixth approximation.

Eighth approximation.

Final approximation to spine and classification of filament.

Southern hemisphere filament rests in a right-handed flux rope.

Solar disk in H on 22 August 2002 at 1603 UTC (BBSO image)

Northern hemisphere filament rests in a left-handed flux rope.

Mirror image would be associated with right-handed flux rope.

Future Developments

• Make detection algorithm more robust

• Test against man-made lists

• Compare filament positions on successive images after correcting for solar rotation

• Set alarm bit if filament can’t be found

• Estimate geoeffectiveness from filament position on the disk and magnetic indices

Sigmoid Detection

• Sigmoid = elongate structure, S or inverse-S shape = signal of enhanced CME probability

• Present method: observers watching 24 hr/day

• Improved space weather forecasts require automatic, accurate sigmoid detection

X-ray Sigmoid

Algorithm developed for filament detection can be used on sigmoids.

Sigmoid Detection Problems

• Structure and intensity not well correlated

• Intensity dynamic range as high as 1000

• Internal structure makes detection dependent on spatial resolution

• Visibility varies with temperature. Visibility is best at 2 - 4 x 106 K, but often only 106 K images are available

Chart: Outline of a Sigmoid Recognition Program

contour the imageoptional: smooth/sharpen

trace out individual contoursstore contour data in a structure

compute k-curvature from data

extract curvature stats

compare curve stats to model stats

Sigmoid Decision

Curvature Stats and Measures

Curvature vs. Arc-length Plots

Structured Contour Data

Contour Map of Image

Solar Imagetaken from Yohkoh

Image Contouring

• Threshold image at different intensity levels• Lines of equal intensity create closed contours• Closed contours have distinct shapes

Characterizing a Shape with Curvature

• Curvature is change in tangent angle per change in arc length

• Counter-clockwise curving lines have positive curvature.

κ p =dθds

≈φ

q−p + p−o

κ p ≈σ ⋅

φv u 2+

v v 2

Interpreting the curvature-arclength plot

• Unique features: position of extrema and zeros; number of zeros; area under the curve; length of perimeter

κ ds=s2

s1

∫ dθ =s2

s1

∫ θ2 −θ1

Sample Case 1: Non-Sigmoid

• Number of Regions between zeros: 6• Extrema at: s = 0.05, 0.18, 0.35,

0.60, 0.70, 0.93• Area under the curve in each region:

-2.88, 0.09, -2.89, 0.08, -2.60, 1.03

Sample Case 2: Sigmoid

• Number of Regions between zeros: 4• Extrema at: s = 0.36, 0.50, 0.83, 0.94• Area under the curve in each region:

-4.36, 0.57, -4.53, 0.61

Successes and Problems

• 8 out of 10 Sigmoids Correctly Identified– 6 false detections in 4 different images

• Reasons for False Detections– Sigmoids are not yet precisely defined

– Sigmoids are often superposed on complicated background

• Recent Developments:– Algorithm refined and tested on SXI and EIT images

– Web-based implementation operates on real-time images

Conclusions

• Developed algorithm for automatic detection and classification of H filaments

• Developed algorithm for automatic detection of sigmoids

• Test results: sigmoid detector successfully flags periods of high activity