1
Automatic detection of dynamic Ha dark
features from high-cadence FMT observations
Speaker: Liu Yu
@Kwasan and Hida Obs., Kyoto University
2
Introduction
• I present a fast program for automatic identification of dynamic chromospheric dark features from time series of full-disk solar images at three Ha wavelengths (center, and ±0.8Å)
• It is a natural requirement for making such an automatic program since daily data is very huge in quantity (at least 1fpm)
3
Introduction
• Usually, one dark feature revealed from the off-band Ha observations may be correspondent to a surge, or enhancement of chromospheric network, or activity of a filament
• The statistical work on them should facilitate the understanding of the triggers of small-scale reconnections, the coronal heating, and CMEs.
4
Introduction
• Particularly, the Ha surge activities have a very close relationship to both large-scale phenomena (SXR jets, flares), but also to small-scale bursts (Ellerman bombs, EUV blinkers)
• The large-sample statistical analyses on surges is a quite different way from many previous work
5
Hida FMT consists of six small telescopes, in which four ones obtain simultaneous full-disk data at different wavelengths.
Hida/FMT
6
Catalogs for FMT• http://www.kwasan.kyoto-u.ac.jp/observation/event/fmt/index_en.shtml
• Since 1992 May, nearly 20,000 dark events have been registered with detailed classifications. They are valuable resource for further statistical analysis of surges and filaments.
• However, making these catalogs is an intensive work for a staff because he/she not only has to search for so many dark features that are usually very weak, but also needs to make a best decision on their classifications and other important parameters.
• Therefore, a better way is needed
events with IB, L or M size marked
7
A program for automatic surge detection for FMT
Data including:• Hα• Hm ━ -0.8Å from Hαcenter• Hp ━ +0.8Å from Hαcenter• WL ━ white-light continuum
Data pro-processing:1. Deleted poor-seeing data (Hm, WL)2. Enhance the contrast (reduce limb darkening)3. Possess more information (Making into standard FITS fo
rmat)
8
1. Bad seeing judgment
• For Hm data
Thick lines: poly-fittings, p(t)
rule: I(t) < pI(t)*c, and
n(t) > pn(t)*(2-c), or l(t) < pI(t)*0.5,
or n(t) < pn(t)*0.1
I - t
Noise - t
Bad seeing
I(t)
p(t)
a noisy point is defined its intensity is lower than I(t) X c, where c=0.85
frame sequence
intensity
percentage of noisy points
9
•For WL data
rule: I(t) < pI (t), or
n(t) > pn (t)
I - t
Noise - t
10
2. Limb darkening correction
Before reduction
consists two steps:
(1) judge disk center and radius, using Sobel function for edge enhancement to obtain limb pixels.
(2) make limb darken profile using a radial median filter and then with a second order polynomial fitting to the inner ¼ core.
center-to-limb function: Limb ( r:r ) = median ( image ( r:r ) ) * a, where a = 0.6
11After reduction
this step helps to enhance the intensity difference between the dark features with their background
12
3. PGM→FITS
• After rotation and center correction, then transfer the original pgm data (Hm,Hp,Ha, WL) into fits type
• The fits files Contain useful information, e.g., B0, r¤, data spatial resolution, date and time etc, which makes it convenient for the following processes
speed: ~1s / frame (in my computer)
for a typical observation day (8hrs): ~30 mins
13
A flow chart for data pre-processing
Mean I(t), N(t), profiles
Fitting curves PI(t), PN(t)
Record this time
All hm / wl data
each hm / wl data
Produce limb darkening function and correct
Judge disc center, disc radius, B0
Cloudy?
1 yes
0
no
Standard FITS fileeach ha / hp data
Same time?
1 yes
Cancel this frame
0
no
B0 correction, center shift
14
Effective detection algorithm• The whole solar disc is evenly divided into 109 b
ox regions • In an Hm image, if one dark point group (cluster)
whose intensity is lower than 90% that of the located box region, then it is defined as a surge candidate
• For a series of Hm frames, if a same dark cluster appears frequently more than some threshold, then it is judged to be a real dynamic dark feature
• Time needed for an 8-hr observation: <90min, averagely
15
width: 39 pixels0 (out-disc)
250 (spot)
Darkpoint group
limb
adjustable
N
W
16
The choice of 90% intensity as an important parameter
01:15
90%
17
A simple example
The solar disk is evenly divide into 109 boxes. The sunspot pixels are valued with 250. The detection will be carried out in the inner circle for avoiding the noisy limb region.
18
First, the procedure searches for dark points from every box in an Hm frame
19
Checking if the darkpoints are in cluster, delete the isolated points.
Two clusters have been detected. The procedure records the median center and length of the dark clusters.
20
After then, the procedure identifies the real dark point clusters that appear for least two successive minutes according the recorded coordinates and times, and later coalesces the nearby clusters of the same time.
21
For 1999-Aug-03, a part of the event catalog is made like ---
22
An image containing four-wavelength observations
23
Main Parameters
• w : box size for dividing disc surface (19 pix)• di : intensity threshold (90%)• far : distance threshold for dark points from each other in a cluster
(15 pix)• dd : size of JPEG images (35 pix)• min : minimum dark points in a cluster (3)• rad : distance threshold to attribute a dark point to a cluster (15 pix)• dt : threshold of time difference to coalesce two events of the same
locations• theta : lat/long. confines for avoiding the noisy solar limb (65 deg)• leng : distance threshold for associating with sunspot and
filaments (20 pix)• varian : threshold for convergence of dark point clusters (100, 80)
24
Variance map Dark points Clusters in one frame
Clusters in all frames Clusters after coalescence
Basic processes in the procedure
25
A flow chart for surge detection
An Hm Fits image
All Hm files finished?
smooth
Weighted di ( ~ 90%±Δ)
Divide the disc into 109 boxes, value those pixels from spot umbrae and other dark features 250 and 10, respectively
Value in-disc darkpoint ¼ of median, out-disc 0
Obtain coordinates (x, y) of the10-value on-disk pixels
umbraeNearby WL image
save median (x,y) into xy_step0.dat
Possible Surges?YN
YN
Continued next page
START:
26
final xy_step0.dat
Check every record
xy_step2.dat
Cluster darkpoints with series number, Check group distance
Merge, obtain all clusters (x, y)
xy_step1.dat
keep successive events
Merge, ∆t≤15min
Delete isolated records
order, size estimate
Using Hm, Hp, time extension
xy_core.dat
Ha,Hp,Hm, Wl JPEG images
obtain (x0,y0), (t0,t1), size
Mpeg movies
Event catalog in time order
S/F/N association
Appear at least two successive minutes
e.g.,20031201event.txt
27
Contrast between personnel-made and machine-made catalogs
(1)80% personnel-detected surges are recognized by machine
(2)70% all dark events by personnel are also found by machine
(3)355% new events added by machine
(4)The events missed are due to…
28
Comparisons of starting and ending times for 28 surges from SurgeM+P
(a) t_personnel - t_machine (b) t_machine - t_personnel
From this figure, we can see that the starting times decided by machine are usually earlier than by eye; and most of the ending times decided by machine are later than by eye.
Ending times Starting times
29
Some negative factors in the detection
• Sunspots umbra region• Limb area• Data deleted for poor seeing• Resolution limitation• Some unexpected situation (airplane tracers, tree
branches, etc)
However, this program is designed to be able to overcome most of the shortages (except the resolution limitation) by carrying corresponding measures
30
On detection of bright Ha features
• Similar to FMT, SMART is also a multi-channel telescope, but it can supply high-resolution flare images. We wish with some efforts, the present procedure can be applied to SMART data for flares
• There have been several methods for flare detections: (1) Automatic image segmentation and features detection in solar full-
disk images, ESA SP-463,Veronig A. et al. (I_flare>2I_quiet) (2)Automatic solar flare detection using Neutral Network techniques,
Solar Physics,2002 Borda R.F. et al. (3)Automatic solar flare detection using MLP, RBF,and SVM, Solar
Physics, 2003, Qu M. et al. (4)Automatic solar flare tracking using image-processing techniques,
Solar Physics, 2004, Qu M. et al.
But they all use only Ha center data, SMART shall have the advantage to supply the important simultaneous off-band data.
31
Conclusions
• From the contrast to personnel-made catalogs, the results by the automatic program are proved to be robust in detecting dynamic dark Ha features with fast rate and more precise starting and ending times. 80%surges, 70%all dark events in the personnel catalogs are re-recognized, and 355% new events are added.
• With some modifications, the procedure may be applied for detection of flare ribbons observed by SMART
32
Three interesting surges in the seven daysA surge ejected vertically from one point
A surge ejected transversely from several points
33
A surge ejected abnormally from outside towards the sunspot