2nd: Identified Bad Pixels
3rd: Chebychev Fit to Good Pixels
Bottom: Adjusted Field Strength
FTP site with 48 hours of Adjusted Data for 06 - 07 January 2014 every 12 minutes.
Adjusted Data for any 12-minute interval from 2014.01.03_00:00 to 2014.01.13_23:48 is available upon request.
Original Data for HARP 3563 are available in the SHARP data series using Lookdata or Exportdata
Original Data for HARP 3563 in remapped Cylindrical Equal Area coordinates centered on the HARP are available at Lookdata or Exportdata
Each time record contains five 1800x1200 images with the total field, inclination, azimuth, Doppler, and a mask. In the mask image, value=0 shows the original good SHARP pixels, value=1 indicates bad pixels that have been adjusted, and value=2 shows the addition area retrieved from the full disk vector data.
The original SHARP data are Rice compressed, i.e., scaled by a float (0.01 for field components and 0.5 for Doppler) and saved as integers. During post-processing, they were read in and written out as floats and then Rice compressed again using fpack using the same scaling. The maximum difference is half the scaling factor (0.005 for field components and 0.25 for Doppler). The compressed fits files have the extension .fz. They can be read using the standard fitsio_read_image module, or viewed using ds9 just like regular fits files. For other uses you may need funpack.
A single time step sample for 2014.01.07 at 18:36 TAI is just 16Mb:
The full compressed data set is about 9G for the entire 5-day interval. Data have been split into 10 tar balls for easier download. Links to the data in half-day chunks:
| Figure 1: Adjustment of HMI Vector Magnetic Field |
|---|
|
|
| Top: Inverted Field Strength
2nd: Identified Bad Pixels 3rd: Chebychev Fit to Good Pixels Bottom: Adjusted Field Strength |
HMI measures the vector magnetic field every 12 minutes over the full disk. Ref: Hoeksema et al., 2014
HMI Active Region Patches (HARPs) follow sunspots as they cross the solar disk. NOAA AR 11944 in January 2014 was part of HARP 3563.
A variety of observed quantities are computed and collected in the SHARP data series Ref: Bobra et al., 2014.
The HMI vector data are generally more reliable in strong-field regions. However, deep in sunspot umbrae and in some penumbral boundaries where the solar atmosphere is complex, the inversion algorithm used to determine the field parameters from the observations occasionally fails. The errors also depend on the relative Sun-Spacecraft velocity. Such errors can be problematic for modeling if not excluded or adjusted.
Here we describe a procedure that can be used to adjust the failed-pixel values.
Bad pixels are found in HMI vector magnetic field data for active region AR 11944 during its disk passage. Inversion failures most often appear as single pixels or patches in the sunspot umbrae. However, they are sometimes present outside of the umbrae, in which case they are almost always single, isolated pixels.
The reason for bad pixels is not fully understood; it appears to be the combined effect of low intensity, extremes in the orbital velocity, and limitations of the inversion technique. Often adjacent pixels with very similar observed polarization will give very different results. The reported field strength in the bad pixels are most often unacceptably high, sometimes hitting the hard-limit maximum of 5000 G. However, in some cases the reported field strength is unacceptably low. Bad values can usually be easily discerned visually by the apparent discontinuity relative to the surrounding good pixels.
Bad pixels can be automatically identified by setting thresholds for the reported error of the inverted field strength and on the reported chi-squared returned by the HMI inversion code. (Ref: Centeno et al., 2014). The thresholds are determined empirically as 350.0 G for field error and 180.0 for chi-square.
We adjust bad pixel values using a 2-D surface fit to the good pixels in the data. 49th-order Chebyshev polynomials of the first kind are used in the fit. The values at the bad pixels are then replaced by the values computed from the fit.
This adjustment scheme is applied to the observed HMI SHARP data segments: field strength (field.fits), azimuth (azimuth.fits), inclination (inclination.fits), line-of-sight magnetic field (magnetogram.fits), and Doppler velocity (Dopplergram.fits). The adjusted data are saved with different segment names: field.adjust.fits, azimuth.adjust.fits, inclination.adjusted.fits, magnetogram.adjust.fits, and Dopplergram.adjust.fits. An additional image segment, mask.adjust.fits, is also generated for each time step record that indicates which pixels have been adjusted (0 = pixel with original data; 1 = pixel with adjusted data). All data shown here are in the original CCD image coordinates.
Examples of adjusted data are shown in Figs. 1-4.
Shown in the top panel of Fig. 1 is the HMI vector magnetic field magnitude reported in the original data series hmi_sharp_720s. The bad pixels form a patch in the sunspot umbra. There is an obvious discontinuity visible between these bad pixels and the surrounding ones. The values are significantly higher than the surrounding pixels and it is quite clear that the inversion has failed. There are also two isolated, single bad pixels outside of the sunspot umbra. The values at those two pixels are anomalously lower that the surrounding pixels.
The second panel marks as dark spots the pixels identified as bad using the thresholds on the computed field error and chi-squared. All of the obviously bad pixels are successfully identified using the proposed scheme, at least in this image.
The third panel shows the fit to only the good pixels computed using the 2D 49th-order Chebychev polynomials. Most, but not quite all, of the original features are recovered, even the two valleys at the bottom of the image.
The bottom panel shows the adjusted data where the bad pixels have been replaced by the fit. The apparent discontinuities have been almost completely eliminated. Notice that a small arc of brighter pixels, probably still not quite properly adjusted, appears along the left and upper edge of the bad umbral patch.
| Figure 2: Adjusted Field Strength | |
|---|---|
| Figure 2. Curves in the bottom panel refer to field strength of the original data (dotted lines) and of the adjusted data (solid lines) along the three horizontal lines (gree, blue and red, respectively) that cross the image shown in the top panel. The green line also crosses a bad pixel outside of the sunspot umbra. The solid lines (adjusted data) eliminate the abnormal jumps shown in the dotted lines (original data). |
| Figure 3: Adjusted Inclination and Azimuth | |
|---|---|
| Figure 3 shows the inclination (top) and azimuth (bottom) of the original (dotted) and adjusted (solid) magnetic field componentns along the same horizontal lines shown in Fig. 2. Azimuth has been disambiguated. The difference between the original and adjusted data in the umbral patch is not very significant; however, the value at the single bad pixel at the left changes a great deal. |
| Figure 4: Adjusted Line-of-Sight Magnetogram and Dopplergram | |
|---|---|
| Figure 4 shows the line-of-sight magnetogram (top) and Doppleram (bottom) observations as originally report (dotted) and the adjusted values (solid). The adjusted data appear to be much smoother. Note that in strong-field regions, the line-of-sight magnetic field method begins to saturate sooner than the vector field method. |