In this section astrometric and photometric solutions are obtained, i.e. images are registered to sky coordinates and are (optionally) flux calibrated. Note that the determination of an absolute photometric zeropoint is optional. The relative zeropoint of exposures, i.e. transparency variations etc are always determined. Results are written to separate FITS headers which are later-on read during the coaddition process.
This little section controls the download of the astrometric reference catalogue. While programmes such as scamp can do that internally and thus invisible to the user, there are numerous cases where such an approach is problematic. The principle behind all astrometric solvers is to match a catalog of reference sources to catalogs containing sources in your images. In order to identify the correct solution, a certain minimum contrast (in the one or other statistical sense, depending on the method) is required. This can fail in several situations:
Astrometry in THELI usually works fine with the default settings, provided that the information in your FITS headers is halfway reliable. You can try and run it blindly, and if it fails then fine-tune some settings.
The enumeration above shows that in the end the success of the astrometric solution depends on the overlap between the reference catalog and the image catalogs. That is, you should not only think of the reference catalogue but also of the source catalog creation as the latter allows you to control how many sources are detected in your images.
You should aim for 100-1000 sources per chip, in the reference catalog as well as in the image catalogs. THELI will also work with a few 10s or with 10000+ per chip, but if you push it to the limits it can become difficult.
RA / DEC: This is only needed if your FITS header does not yet contain astrometric coordinates, or if these are significantly offset from the correct value. Otherwise the centre for the reference catalog will be determined from the FITS header.
Resolve target: Enter the name of your target. Once you click the button with the magnifying glass, THELI sends a query to the Simbad, NED and VizieR databases to retrieve the target’s coordinates (if not already contained in the FITS header).
After entering coordinates, THELI will prompt you with a notification window when you hit the Start button. Therein, you have three options:
Mag limit: Only sources brighter than this limit will be retrieved. This is your control over the density of the reference catalog.
Radius (optional; in arcminutes): Determined automatically, taking into account detector size and dither offsets. Objects within this radius of the nominal coordinates in the FITS header (or those entered manually) will be downloaded.
Once your selection is made, click on Get catalog to retrieve the reference catalog. The number of sources found will be displayed.
Choose from one of these astrometric reference catalogs:
|SDSS-DR9||23.0||Covers 14000 sq.deg around the northern galactic cap|
|PPMXL||21.0||Combines USNO-B1 with improved astrometry and 2MASS|
|USNO-B1||21.0||0.3”-0.4” rms uncertainty; can exhibit large systematics|
|2MASS||18.0||Near-infrared, low systematics|
|UCAC4||16.5||Good proper motions, all CCD based|
|TYC||11.0||The TYCHO catalog|
|ALLSKY||10.0||Based on GSC-2.3, filtered to mag<=10. Comes with THELI as downloading huge areas from CDS often resulted in incomplete sky coverage.|
TYC and ALLSKY should only be used for calibration of extremely wide fields (obtained e.g. with photo lenses) with very large pixel scale (much larger than 1.0”/pixel).
There are two ways to perform absolute photometry: an indirect approach using external standard star fields, and a direct approach using sources with known brightness in the field of view. THELI supports both methods, returning the photometric zeropoint ZP (for an image normalised to an exposure time of 1s).
Doing absolute photometry in THELI is very much simplified from a user’s point of view as compared to what actually happens in the background. The reason is that the absolute zeropoints for the images must be known BEFORE object catalogs are created and BEFORE they run through astrometry and relative photometry. Otherwise they cannot be propagated properly and an absolute zeropoint for the coadded image cannot be obtained.
THELI must do these steps to obtain a ZP:
Indirect approach: The standard star field has to be calibrated astrometrically in order to identify the standard stars unambiguously. This requires the download of a reference catalog, the creation of object catalogs, and a (simple) astrometric solution (no distortion fitting). Lastly, THELI will determine the ZP.
Direct approach: Astrometric and photometric reference catalogs must be downloaded for the target area, followed by catalog creation and an astrometric solution (such that targets in the field can be identified with those in the reference catalogs). Lastly, THELI will determine the ZP.
As you can see, obtaining the ZP requires catalog creation and astrometric solutions. The configuration parameters for these tasks will be taken from the configurations of the Create source cat task and from the Astro+photometry task described further below. You must make sure that these settings are sensible, otherwise the absolute photometry will fail. It is a good idea to familiarise yourself with these two tasks first before you dive into absolute photometry.
You must have observed at least one field with photometric standard stars in the same filter as your observations. Select the following standard star catalogs for observations in these filters (note that in particular in the near-IR the filter curves of these standard stars may not be the same as the ones used in your filter set. Check the pertinent publications for more details).
Landolt (Landolt 1992)
Stetson (Stetson 2000)
Stripe82 (Smith 2007)
Stripe82 (Smith 2007)
MKO (Leggett 2006)
HUNT (Hunt 1998, aka ARNICA)
PERSSON (Persson 1998)
PERSSON_RED (Persson 1998)
MKO_LM (Leggett 2003)
Your observations must have been conducted in one of the filters listed in the table above
The night should have have been photometrically stable, i.e. no change in transparency at a given airmass.
The standard star field(s) should have been visited several times at night at significantly different airmasses (for extinction calculation).
You may include standard star and target observations from many different nights. An independent solution will be calculated for each night.
Before you start reducing your target data, create a separate directory at the same level in the directory tree as the SCIENCE directory containing your main target. Include the name of your standard star directory in the Initialise section in the STANDARD field. When you reduce your target observations, the standard data will be processed too, including the same calibration files (master bias, master flat, if applicable also superflat and defringing) as the target observations.
You can collect different standard star fields (taken in the same filter) in the same STANDARD directory. THELI will automatically obtain a photometric solution for each night and present you with a dialog in which you can select if you prefer the 1-, 2- or 3-parameter fit, or if the solution is not acceptable.
Note that if your data set extends over several nights, it is sufficient for an absolute photometric calibration if only one night was calibrated successfully.
Once the task is started, it calculates three different photometric solutions for each night:
For each night, the three fits are graphically displayed as a line plotted over the corresponding data points. At the bottom of this dialogue, select which fit should be chosen, or if none of the fits is acceptable. Based on the plots you may fine-tune your manual selection of a fixed extinction and colour term coefficient, and then re-run the task. The zeropoints and coefficients will be written into FITS headers. If no ZP is found, the keyword is set to -1.
Note that if your data set extends over several nights, it is sufficient for an absolute photometric calibration if only one night was calibrated successfully.
The other possibility for absolute photometric calibration is to determine the zeropoint (ZP) directly from the data. You do not need observations of an external standard star field.
Direct photometric calibration in THELI does not take into account colour terms arising from different total throughput curves between your telescope/filter/camera combination and the one used for SDSS or 2MASS. If you need absolute zeropoints better than about 0.1 mag then you should consider fine-tuning the ZPs obtained with this method using a comparison of instrumental stellar tracks against those in the Pickels library.
Filename extension: After doing direct photometric calibration with FITTING METHOD = ZP for each chip, images have the character P appended to their filename extension, e.g.
In all other cases no chip-specific treatment is necessary and thus no indication has to be made in the file names.
In order to detect objects in the images for astrometry and photometry purposes, THELI uses SExtractor. We encourage you to make yourself more familiar with this tool (see also the other links on the SExtractor website).
The task creates the following sub-directories:
Maximum FLAG: Per default, THELI keeps only objects that were not flagged (FLAG=0) as problematic by SExtractor. However, there are cases when you need to include flagged objects as well in order to arrive at an astrometric solution. For example, for mildly saturated stars one can still determine a good centroid, which can be very helpful if the faint end of the astrometric reference catalog is close to saturation in your images. Similar considerations apply to objects close to image borders, etc.
Quoting from the SExtractor manual: The FLAGS parameter contains all the extraction flags as a sum of powers of 2:
For example, an object close to an image border may have FLAGS=16, and perhaps FLAGS=8+16+32=56.
Min. # of objects: If a catalog contains less objects than this number, then the catalog and the corresponding image are moved to lownum sub-directories and will not enter astrometry and coaddition.
Background level: SExtractor usually gets the background level correctly. However, it may fail if a large number of NAN pixels is present, and then the object detection will also fail. In this case you can manually set the background level, assuming it is the same for all images (I was forced to use this setting only once, for background-subtracted Pan-STARRS images).
Data is well sampled, but plagued with hot pixels: Activate this switch if your images have REALLY A LOT OF hot pixels and/or other strange small-scale features that might be mistaken for real objects. A lot of means comparable to or more than real objects. This probably makes only sense for HAWAII near-infrared arrays with very bad cosmetics. Optical astronomers can ignore this switch.
5 different methods are available for astrometry. The first two methods, Scamp and Astrometry.net, perform automatic mosaicing, whereas the two shift approaches do not base their solution on WCS coordinates. All but the last method (Header) perform relative photometry as well, i.e. measure transparency variations in a series of images.
Scamp: The preferred method, fast, creates meaningful check-plots. Can be difficult to find a match with the reference catalog if both the position as well as the position angle (and possibly a flip) are highly uncertain or unknown.
Astrometry.net: Matching is done differently as compared to Scamp. Relative zeropoints, and optionally distortion, is calculated afterwards with Scamp (automatically, using the preset Scamp parameters). Checkplots are provided by Scamp. Comparably fast as Scamp.
Shift (float/int): A very simple solver that calculates image shifts only. No rotations and no WCS parameters are involved. The shifts are computed in pixels with respect to the first image in the series. This is only useful for mid-IR data where only very few or only one object at all is visible. The astrometry of the coadded image will NOT be accurate as no comparison with reference catalogs is made. If you use the integer approach, only integer pixel shifts and no resampling will be performed during coaddition. This method cannot be used for mosaicing.
Header: This will simply copy the unchanged raw zero-order solution (CRVAL, CRPIX, CDi_j) present in the FITS headers, albeit without distortion terms. This can be chosen if the raw data already has a good astrometric solution in the headers (e.g. determined by other software) and you prefer to use this over the THELI solution.
No relative photometric calibration will be done when chosing this method.
All methods create separate FITS header files with the relevant astrometric (and, if applicable, photometric) information. The headers are stored in
Update header: Writes the zero-order astrometric solution (CRPIX, CRVAL, CD-matrix) into the FITS headers. This is usually not necessary unless you need accurate sky coordinates for some reason. The only such reason in THELI is when you have to extract a background estimate from a small sky section and you want this area to be exactly the same despite a dither pattern. See the sky subtraction for details.
Restore header: Restores the previous FITS headers if a zero-order solution was written into them using Update header.
We strongly recommend to have a look at the scamp manual for details as of how Scamp works.
The following parameters are available.
DISTORT_DEGREES: Distortion polynomial order. Distortion in the sense of Scamp is a variation of the pixel scale as a function of position. All other (linear) transformations are explicitly modelled as such and encoded in the CD matrix. The distortion degree entered in this field means:
Scamp can also make use of higher distortion polynomials, but they get increasingly more unstable and require a large number of sources.
FGROUP_RADIUS: The maximum angular separation between to pointings. If two images are further apart, then independent astrometric solutions will be calculated for them.
CROSSID_RADIUS: The search radius (in arcsec) for cross-identifications. This value can remain at its default setting of 2, unless the pixel becomes larger than 0.7 arcsec/pixel. In the latter case, THELI sets
ASTREF_WEIGHT: In Scamp the object catalogs and the reference catalogs have equal weights in the solution process. Some cases may require to put more weight on the reference catalog. If your reference catalog is very precise and accurate to within e.g. 1/5th of a pixel of your camera, then you may choose a very large weight such as 10 which would make the reference catalog control your solution. Usually, the resolution of the camera is equal to or better than the uncertainty of the reference catalog. The default value for this parameter is 1.
SN_THRESHOLDS (low|high): Minimum S/N of a source to be kept in Scamp (low), and the minimum S/N for some high-S/N statistic calculations (high).
Additional DISTORT_GROUPS | DEGREES | DISTORT_KEYS: In almost all cases one can ignore these three settings (i.e. leave them empty). One would enter here a comma-separated list of additional distortion groups, degrees and keywords. DISTORT_KEYS would also accept FITS keywords, which must be preceded with a ”:”.
Example: A set of images with arbitrarily distributed position angles is being stacked, and it turns out the distortion pattern is a linear function of the position angle. One would then define a new distortion group, DISTORT_GROUPS=2, with DISTORT_DEGREES=2 and DISTORT_KEYS= :POSANGLE” to reflect the dependence of the distortion of the position angle. For more details see the *Scamp manual.
ASTRINSTRU_KEY: The FITS header keyword used to identify exposures for which one astrometric solution should be calculated. The default value is FILTER, meaning that images with the same FILTER header keyword will have one solution calculated. Images with different values in those keywords will get their own solution. You can add several comma-separated keywords in order to break down the astrometric solution into smaller groups. If you want to calculate one solution for images taken in different filters, then just enter a string that does not exist as a keyword in the FITS header, such as “none”.
For more information about this tool visit www.astrometry.net.
The following parameters are available.
It happens that Scamp returns a wrong astrometric solution without recognising it. This usually happens when the images were not correctly matched to the sky. You do not need to go through a possibly lengthy coaddition to find out if the solution was good or not. In most cases it is sufficient to have a look at the various check-plots created by Scamp. You can find them in
In the same directory is also a scamp.xml file with further valuable information that can help you identify trouble-making images and other issues. In order to look at the XML file with firefox, you must first enter in the address field
and then locate the following parameter and set it to false:
security.fileuri.strict_origin_policy = false
Otherwise the XML file will not be displayed correctly.
If you can’t get it done with Scamp, you may equally well try Astrometry.net which uses a different matching algorithm that is superior if WCS parameters such as pointing and position angle and a possible flip are unknown.
Image distribution on the sky: fgroups
This should be the first plot you should look at. It displays the dither pattern together with unmatched (red) and matched (green) reference sources. You should recognise your dither pattern here.
Good fgroups plots
- Example 1: Megacam@CFHT
- Example 2: A mosaic of ACS/HST images. The overlap with the ground-based reference catalog is not very compelling, but sufficient.
- Example 3: WFC@INT (wide-field camera at the 2.5m Isaac Newton Telescope)
- Example 4: A large set of rotated GPC1@Pan-STARRS exposures
- Example 5: LBC_RED@LBT (the red camera at the Large Binocular Telescope, with a cubic distortion polynomial). Compare to example 4 below.
- Example 6: A mosaic of mosaics with WFC@INT
Bad fgroups plots
- Example 1: A wide field view obtained with a photo-lens. Only part of the image (green) is matched with the reference catalogue, due to distortions. Increase the CROSSID_RADIUS.
- Example 2: Total failure. Reason: very small field of view, insufficient sources in the image catalogs. Try lowering the detection thresholds and set distort=1 because not enough sources are available to determine many distortion coefficients.
- Example 3: Inaccurate CD matrix (wrong position angles and/or pixel scale). Try increasing POSANGLE_MAXERR and PIXSCALE_MAXERR. RA and DEC in the FITS header could also be way off. Try increasing POSITION_MAXERR.
- Example 4: LBC_RED@LBT (the red camera at the Large Binocular Telescope, with a quadratic distortion polynomial). Note that only parts of the detectors are matched properly to the sky. Compare to example 5 above.
- Example 5: A wide field of view of a milky way area. While the depth of the reference catalog was ok (about 200 sources returned) the object catalogs contained 10000+ sources per exposure. Increase the object detection threshold DT from e.g. 5 to 50 would bring the source density down to a reasonable level.
- Example 6: That’s a tricky one, based on MOSAIC-II@CTIO. Note that almost all frames are registered nicely. However, the rightmost exposure shows no green matches (only visible at the rightmost edge; the coadded image would be clearly corrupted, though). Reason here was that the WCS information in the FITS header of this single exposure was 7’ offset, whereas the others were all within about 1’. Increasing POSITION_MAXERR to 8’ solved the problem instantly, after many days of trouble shooting on the user side. Mostly problems with scamp have very easy solutions, directly in front of your nose. Lesson: Even modern telescopes mess with you every now and then (not the first time I have seen this; WFC@INT sometimes has had its headers off by 20’).
Image distortion: distort
Distortion can be displayed as a variation of the pixel scale across the detector area. Make sure these plots look as symmetric as possible.
Good distort examples
- Example 1: Megacam@CFHT. Note the slight irregularity in the upper half of the image centre. One can do better than that, however this distortion model did not lead to any visible ill effects in the coadded image.
- Example 2: Pan-STARRS. Due to the very short exposure time (30s) and the high galactic latitude only very few sources were visible in this field. Hence only a linear distortion model could be fit to the data. A quadratic fit looked smoother, but showed artifacts with similarly large deviations from reality.
- Example 3: WFC@INT. Note the slight mismatch between the two detectors at the lower left.
- Example 4: WFC@INT. Same as example 3, but using more objects in the reference and the source catalogs.
- Example 5: WFI@MPGESO. The wide-field imager at the 2.2m MPG/ESO telescope is one of the very few instruments where the largest pixel scale does not coincide with the optical axis. You can model it with distort=3, but a higher order model (distort=4..5) (shown: distort=5) yields better results as the curvature of the “ring” is better reproduced.
- Example 6: SuprimeCam@SUBARU
- Example 7: LBC_RED@LBT
Bad distort examples
- Example 1: A wide-field exposure through a photo-lens. Try increasing the number densities in the catalogs or CROSSID_RADIUS to obtain a more concentric pattern.
- Example 2: Pan-STARRS. Note how the chip at the upper right is completely off. It contained only very few sources due to excessive masking. The corresponding images from this chip were removed by hand, then astrometry was repeated.
- Example 3: Megacam@CFHT. Try e.g. to increase the overlap between object catalogs and the reference catalog.
- Example 4: HAWKI@VLT. This is based on a sparse field. Adding another, neighbouring, sparse field fixed the problem. Wide dither pattern patterns or independent data fields help.
External astrometric residuals: astr_referror1d
Astrometric residuals with respect to the reference catalogue after a solution was found. You’d expect a tight and featureless horizontal distribution whose width is solely determined by the precision of the reference catalog.
Good astr_referror1d examples
- Example 1: Megacam@CFHT, USNO-B1
- Example 2: Pan-STARRS. Note the slight deviations at the outermost edges. Those chips are largely masked and were difficult to match.
Bad astr_referror1d examples
- Example 1: Megacam@CFHT. Adjust the reference and object catalog settings, or try a different reference catalogue.
- Example 2: Pan-STARRS. The only trick that helped here was putting a lot more weight onto the reference catalogue, ASTREF_WEIGHT=10. Normally this should be avoided as the images have better resolution and higher object density than the reference catalogue. In this case the reference catalogue was SDSS and significantly deeper than these 30s exposures, hence increasing the weight for the reference catalog is justified.
Internal astrometric residuals: astr_interror1d
Astrometric residuals between the individual exposures after a solution was found. You’d expect a tight and featureless horizontal distribution whose width should be on the order of 1/5th-1/20th of a pixel.
Good astr_interror1d examples
- Example 1: Pan-STARRS
- Example 2: Good residuals for a one-chip camera (pixel scale 0.85 arcsec/pixel)
Bad astr_interror1d examples
- Example 1: Megacam@CFHT. Note the small bifurcation in the upper line. The rest is ok.
- Example 2: WFI@MPGESO. The magnitude limit for the reference reference catalog was not deep enough in this case, resulting in badly matched chips.
- Example 3: Note the huge range (10 arcsec) of the y-axis. Normally the range should be about one pixel. Reason here was that the reference catalog downloaded had no overlap at all with the data.
- Example 4: FORS2@VLT observations of a globular cluster. The match is generally good (very tight horizontal line), but there are obvious mis-identifications within the very crowded globular cluster. Repeat the object catalogs with e.g. DEBLEND_MINCONT = 0.01, and try 2MASS instead of USNO-B1. 2MASS has significantly better resolution in crowded objects than USNO-B1 which features many blended or otherwise inaccurate sources in such cases.