Northern Sky Narrowband Survey: DR0.1

Data release DR0.1 contains a first complete set of Hα (656.3 nm) and low-resolution continuum images from instruments 1-3. The pre-processed data published on this site was calculated on 47 overlapping tiles with a size of 31° × 31°. Compressed 8-bit color images can be accessed in sections Overview images and Tile images. Full dynamic range images with linear intensity can be retrieved from an HiPS (Hierarchical Progressive Survey) repository. In order to assess the reliability of the data, sections Data acquisition and processing and Artifacts and limitations provide a brief description of the image processing and a detailed description of all known artifacts.

Overview images

The overview images shown below are 212° stereographic projections of the entire surveyed region. The projection type was used because it preserves shapes. Maximum image resolution in the JavaScript viewer (click on the images) is 30″ at the center (δ=90°) and about 11″ at boundary (δ=-16°). In the navigation mode of the Javascript viewer (press the N key or click the 'N' button) it is also possible to browse to other views on this site (including the tile images).

Click on the images to load a high-resolution (1.3 GP) versions using a JavaScript viewer.

Tile images

The tiles are arranged in 4 rows centered at declinations 76.125°, 50.375°, 24.625°, and -1.125°. The Size of each tile is 31° × 31°, with an overlap of at least 5.3°.

Click on the preview images below to load high-resolution (up to 10″) views in Hα and continuum, as described in the caption of the first overview image. Hα-only views like those in the second overview image, along with a legend, can be seen by clicking on the “[Hα]” and “[Hα legend]” links, respectively. The angle α below the preview images describes the right ascension of the image center. (The declination angle is found in the heading above the images. ICRS coordinate system is used.)

The tile images were processed individually with automatically chosen parameters. This might not always be optimal.

Declination δ=76.125°

Declination δ=50.375°

Declination δ=24.625°

Declination δ=-1.125°

HiPS / Aladin

Selected data with is made available as HiPS (Hierarchical Progressive Survey) and can be viewed with tools like Aladin. The HiPS's can be loaded in Aladin Lite by clicking on the HiPS ID's below.

Hα

Hα data are provided as FITS HiPS with linear intensity and full dynamic range (HiPS/IVOA ID: simg.de/P/NSNS/DR0_1/halpha) and as 8 bit compressed PNG HiPS for easy visualization (HiPS/IVOA ID: simg.de/P/NSNS/DR0_1/halpha8). The maximum resolution is 6.4″, usable resolution is about 10″.

Hα and continuum

A 8 bit compressed color HiPS with Hα (without continuum, mapped to red), blue continuum (including [OIII] and Hβ, mapped to green) and red continuum (without Hα but with [SII], mapped to blue) is available via the HiPS/IVOA ID simg.de/P/NSNS/DR0_1/hbr8. Emission nebulae appear reddish, while reflection nebulae range from green to blue. Stars are partially subtracted to make the faint nebulae visible. The maximum resolution is 12.9″.

Visual continuum

Star-subtracted visible continuum without Hα is available as FITS HiPS with linear intensity and full dynamic range via the HiPS/IVOA ID simg.de/P/NSNS/DR0_1/vc. This data is a combination of all three color channels (reg, green and blue). An 8 bit compressed (almost) true-color version without star subtraction can be found at simg.de/P/NSNS/DR0_1/tc8. The maximum resolution of both HiPS's is 12.9″ (usable resolution is about 20″).

Estimated variance for Hα and visual continuum

The estimated square noise (variance) for Hα and visual continuum can be obtained with the HiPS ID's “simg.de/P/NSNS/DR0_1/sn_halpha” and “simg.de/P/NSNS/DR0_1/sn_vc”. The maximum resolution is 12.9″.

The data are normalized to 30″ square pixels. This is approximately the smallest structure size that can be safely distinguished from stars in Hα. (The variance depends on the size of the area over which a signal is integrated. Therefore, such noise information is only usable if that size is given.)

Initially, the noise is estimated for each single exposure during stacking, in order to optimally weight them. The variance provided as HiPS is derived from this data. Channel mixing, e.g. due to continuum and Hα subtraction, is considered correctly. However, it should be mentioned that “noise” is not the same as “error”, i.e. artifacts are not taken into account. Nevertheless, the accuracy of the processed images is mainly determined by photon noise, whether from contamination or from the signal of interest. This noise is correctly reflected in this dataset.

Note that the FITS images may contain very large values (rather than infinity) at undefined pixels (usually due to sensor saturation at bright stars), which may make visualization difficult. On the other hand, variance allows for correct scaling by averaging (unlike reciprocal (square) noise, which would correctly handle undefined pixels).

Data description

Different datasets had been calculated from the raw data. Which one is used (for the images or the HiPS data), is described above. This section provides some additional details.

Hα

Hα data was captured with instruments 1 and 3. The usable resolution is about 10″

Hα was background-corrected and intensity-calibrated to Rayleighs using WHAM data; see Haffner et al., 2018 and HWAM-SS DR1. For continuum subtraction, the lower-resolution red and green channels from this data release have been used. To minimize artifacts around stars, continuum subtraction was applied after subtraction of (the same) stars, see the image processing section for details and the artifacts section consequences.

Continuum

Continuum data was captured with instrument 2. The usable resolution is about 20″

Unless otherwise stated, Hα was subtracted from the red channel with a factor that was determined empirically so that no underflow occur. Because Hα and SII emission strongly correlate, this also eliminates some — but no all — SII light. No attempt was made to subtract emission lines (like OIII and Hβ) from the green and blue channels. Thus, none of the continuum channels is totally free of emission lines.

Data acquisition and processing

Each region was recorded by more than one hundred short-time (up to 60s) exposures using multiple instruments. The camera pointing coordinates lie on a grid whose size is slightly smaller than 50% of the field height and 33% of the field width. This technique allows for detecting and rejecting outliers (e.g. satellite tracks and filter reflections) and minimizes the impact of varying field angles (e.g. filter transmission variations and optical distortions). These issues are discussed in detail in the section “Artifacts and limitations”.

Each single exposure is aligned (using an affine transformation with nonlinear distortion terms) and typically calibrated to more than 1000 stars from the PPMXL catalog, see Roeser et al., 2010, where position and projection type are given by tiles. The reference stars are also used for factor calibration so that the stacked result is automatically white-balanced to average stars. The stacking process is iterative, with results from previous iterations used to sort out outliers and help estimate the background. In order to maximize the signal-to-noise ratio, each single exposure is weighted according to the estimated noise. (This noise estimation is also responsible for underweighting hotpixels and badpixels. Photon noise of objects can be calculated from the results of the previous stacking iteration.)

The final stacking result of Hα has the highest resolution and is therefore used to extract point sources. Based on this source list (and assuming these point sources are stars), stars are subtracted from all stacking results. After that, continuum subtraction (from Hα) and Hα subtraction (from red continuum) are applied. (Performing continuum subtraction before star subtraction would cause too many artifacts around stars.)

The star-, continuum- and Hα-subtracted images are background-corrected, either using reference data (Hα uses WHAM data) or by suppressing frequency components that can't be detected by the cameras (all other channels; see section “Suppression of large structures due to background estimation”).

FITS HiPS's are calculated directly from these results. All 8-bit color images (“Overview images”, “Tile images” and 8 bit HiPS's) are dynamic range compressed (non-linearly high-pass filtered) and tonal curve corrected. Furthermore, the brightest stars are re-added in order to visualize the regions that suffer from contamination by star light.

The last overview image, without star, continuum, and Hα subtraction, was calculated directly form the final stacking results.

Artifacts and limitations

Suppression of large structures due to background estimation

The background for all channels except Hα has to be estimated from the darkest regions because no absolute reference data are available (for Hα WHAM data can be used). This method of background estimation also suppresses large homogeneous structures. Thus, the data should be considered as high-pass filtered. This effect is responsible for the ecliptic plane appearing as a double band in the continuum images rather than as a single broad band.

The upper detection limit for structures is about 3° in all directions. Objects that are smaller in at least one direction, like filaments, can be safely detected.

Background estimation from darkest regions is also responsible for the colored hue around dark nebulae, like here, because they are darker in shorter wavelengths. These hues are an indication that a larger structure has been suppressed and can be seen, for example, in the star fields near the galactic plane.

Reflections between filter and sensor

Reflections between interference filters, which are mounted in front of a lens, and imaging sensors cause artifacts, as depicted below.

These reflections occur at wavelengths where the interference filter is partially transmissive. Fully blocked wavelengths do not reach the sensor, and fully transmitted wavelengths are not reflected by the filter. The intensity of the artifacts, therefore, depends on the ratio of partially transmitted wavelengths to fully transmitted wavelengths. This is why such artifacts are only significant with narrowband filters.

Because each point is observed at different field angles (see Section “Data acquisition and processing”) these artifacts occur at different positions and can therefore be sorted out by the stacking software. Unfortunately, it has turned out that the outlier rejection thresholds were set too conservatively, resulting in spots in the Hα channel around bright stars, as shown in this example, where Hα is red. This will be enhanced in the next data release.

Satellite trails and other temporary effects

Similar to the filter reflections described in the previous section, such temporary effects can be removed by the outlier filter of the stacking software. How well this works depends on how many images the disturbances appear in:

• Satellite trails, airplanes, meteors and cosmics, which occur in only a few images (captured at the same time), should have been removed safely.
• Solar system objects, which occur in many images, are more difficult to distinguish from fixed objects and therefore may cause artifacts. (Bright planets are avoided by the capture software.)
• Geosynchronous satellite are so dense that they appear as a thin trail at δ=-7:25° in continuum images.

As mentioned, the software parameters were not chosen optimally in DR0.1. Because of this, and as more data is collected, these artifacts should be reduced in the future.

PSF variations

The star subtraction software has a limited ability to handle variable PSFs (point spread functions). PSF variance across the image field is controlled by a camera pointing grid, which is more than 3×2 times smaller then the field of view. There are two effects which cannot be fully compensated for and thus cause artifacts:

• Diffuse halos around bright stars caused by the 5 cm aperture Hα filters used on instruments 1. For example, such artifacts occur on the right boundary of this tile in the red channel (Hα): because the halos occur in only a small fraction of the image, PSF extraction did not work very well.
• Spikes around bright stars caused by two filters of the first set of 6 cm aperture Hα filters used on instruments 3, such as those visible around the star Mira. These filters have since been replaced.

Both kinds of artifacts should be reduced in the future as the fraction of images captured with the improved filters increases over time.

Starlight contamination and limits of the star subtraction

Star subtraction cannot perform miracles. Even if the PSF were known exactly, the signal-to-noise ratio near bright stars becomes very low due to photon noise caused by the starlight. In practice, in addition to these errors, inaccuracies in the PSF (see previous section) are also amplified at bright stars. Regions where the subtracted result becomes too uncertain are therefore interpolated. In general, artifacts around bright stars must be taken into account.

Two strategies are used to handle this situation:

1. Re-adding brightest stars: In some data sets (e.g. all color images), the brightest stars are re-added with a smaller scaling factor. This provides a natural visualization of the regions contaminated by starlight.
2. Providing variance data: Variance data is made available in the form of FITS HiPS's. This data accounts for photon noise and is therefore a reliable measure to assess contamination by starlight.

Elimination of small Hα sources and artifacts around M42

As described in image processing section, the Hα data are used for star extraction because this channel has the highest resolution. The drawback of this choice is that tiny Hα sources that appear as points are misinterpreted as stars and are eliminated too. In practice, this limits the spatial detection limit to about 30″.

Furthermore Hα data is saturated in brightest parts of M42 (Orion Nebula). This caused the a misinterpretation as a point source by the star subtraction software.

Variable stars

Because images are captured over a long period and input images are weighted based on (photon) noise, variable stars appear distorted in the stacking results. This distortions lead to tiny disc-shaped residuals after star subtraction, as shown in this example.

Remarks and discoveries

The results of the first data release were used to generate a revised list of uncataloged objects. Many of these objects are new discoveries. The most spectacular one is the 19 light-year-long jets from KX And.

The ecliptic plane appears as a wide double band in the continuum images (due to the high-pass filtering effect of the background estimation). Furthermore, the geosynchronous orbit is visible as a thin trail at a current epoch declination of δ=-7:25° (observed from a geographic latitude of 51:11°3; ICRS equatorial system is tilted about 0.15° relative to current epoch system). Due to continuum subtraction, neither effect is visible in Hα.

Statistics

More information about the camera array used for the data acquisition can be found on the instruments page.

Here is some additional statistical information:

	Hα	Continuum
Acquisition period	2018/11/10 to 2024/05/15	2018/11/10 to 2022/09/04
Total exposure time	4892 h	2084 h
Number of single exposures	2.95 × 105	1.68 × 105

Outlook

Data releases with the major version number 0 will contain pre-processed data only. Starting from DR1, it is planned that the stacking results, along with the scripts for deriving the processed data, will also be provided.

The first full coverage of OIII (500.7 nm, instrument 6a) and SII (671.7 nm and 673.0 nm, instrument 6b) is expected to be reached in the first half of 2025. These data (still in low quality) will be part of the next data release, DR0.2, which is not expected before summer of 2025.

A significant quality improvement is expected as soon as high-resolution continuum data is fully available (instruments 5a-5c). This will allow for a more accurate continuum subtraction and star detection. (Currently Hα is used for the latter purpose because it provides the highest resolution. The drawback is, that small emission sources may be recognized as stars.) These results, and thus the first release with data from all instruments, will be part of DR1, which is not expected before 2026.

Acknowledgements

I acknowledge the use of WHAM data for background correction and intensity calibration of the Hα data. The Wisconsin H-alpha Mapper (WHAM) and its H-alpha Sky Survey have been funded primarily by the National Science Foundation. The facility was designed and built with the help of the University of Wisconsin Graduate School, Physical Sciences Lab, and Space Astronomy Lab. NOAO staff at Kitt Peak and Cerro Tololo provided on-site support for its remote operation.

This work made use of PPMXL data for alignment and calibration.

References

 

L. M. Haffner, R. J. Reynolds, S. L. Tufte, G. J. Madsen, K. P. Jaehnig, and J. W. Percival. The wisconsin hα mapper northern sky survey. The Astrophysical Journal Supplement Series, 149(2):405, dec 2003. [ DOI | http ]

 

L. M. Haffner. WHAM SS Data Release, 2017. [ http ]

 

S. Roeser, M. Demleitner, and E. Schilbach. The PPMXL Catalog of Positions and Proper Motions on the ICRS. Combining USNO-B1.0 and the Two Micron All Sky Survey (2MASS). The Astronomical Journal, 139(6):2440–2447, June 2010. [ DOI | arXiv ]

Monitor calibration

For optimal display, the monitor should be calibrated so that as many shades of the grayscale shown below are distinguishable. At least the 4% steps should be separated by every monitor. The 2% steps are visible on better monitors. To distinguish the 1% steps, a HDR monitor is required.