PSF Tool: FWHM values not constant?

[Terry Danks]

Hi All:

Basic misunderstanding here. I am looking for FWHM values to assess image and guiding quality.
Using the Dynamic PSF tool for the first time. It goes against my understanding of FWHM to have large values for bright (big) stars and small values for faint (small) stars? The range of FWHM values in my image are all over the clock, from 1.6 to 9+!

The whole concept of FWHM in an image is that the value should be near constant?

Please help me out here.

Terry

[naavis]

Hi All:

Basic misunderstanding here. I am looking for FWHM values to assess image and guiding quality.
Using the Dynamic PSF tool for the first time. It goes against my understanding of FWHM to have large values for bright (big) stars and small values for faint (small) stars? The range of FWHM values in my image are all over the clock, from 1.6 to 9+!

The whole concept of FWHM in an image is that the value should be near constant?

Please help me out here.

Terry

The FWHM numbers should stay fairly constant as long as you don't pick saturated stars or very very faint stars, and as long as your stars are not undersampled.

[georg.viehoever]

Are your images calibrated? Are they still linear when you measure FWHM?
Georg

[Terry Danks]

Not calibrated. Still linear. AFAICT, the FWHM values I am getting are really just some measure of star size on the image. 
My understanding is that, not only should FWHM values be pretty constant across an image, there really should be no correlation whatsoever between star brightness(size) and FWHM. None of these stars are anywhere near saturated.

[korborh]

Terry, can you post a screenshot of your dynamic PSF window showing the fit and values?
Are you selecting Auto for the fitting?
Do you have significant number of hot pixels near these stars?

[Geoff]

My understanding is that, not only should FWHM values be pretty constant across an image, there really should be no correlation whatsoever between star brightness(size) and FWHM. None of these stars are anywhere near saturated.

Your understanding is correct and that's the way it works for me. As one of the other posters suggested you should upload an image for others to look at.
Geoff

[Terry Danks]

Screenshot attached. I selected the stars by hand, clicking on/near them.
Arrows added in PS.

Terry

[korborh]

There are 3  very bright stars there.
How are you checking that these are not saturated?
These look like could be saturated and giving you the 3 values which are large.

[Geoff]

There are 3  very bright stars there.
How are you checking that these are not saturated?
These look like could be saturated and giving you the 3 values which are large.

I think we really need to see an original fits file.  My first thoughts were that the three bright stars were OK and the other two were too faint, giving overly small values.
Geoff

[Juan Conejero]

Impossible to say anything without the raw image.

[Terry Danks]

I certainly do not know what I am doing here. It seems the bright stars may well have been saturated. I determine that by using the mouse and looking at the K value.
Sat. for the STL is 65,000ADU. It appears though that one must be very choosy about the range of ADU values chosen to get representative FWHMs? It seems for this image to be from about 5K to 40K. Still there seems a definite tendency for the FWHM to increase with the ADU value? It seems as long as I stick with stars falling into a range of 15,000-37,000, I get FWHMs very close to 3.2.
I take this behaviour to be normal?

[Carlos Milovic]

If it is a non-blooming chip, then you may have entered the non-linear part of the well (above 40k), so FWHM measurements will no longer be valid. 5k to 40k seems a good range to me, to play it safe. Lower values might be too noisy.

[topboxman]

I have used Dynamic PSF tool to measure FWHM but it returns FWHMx and FWHMy. I also get inconsistent results between different stars and I make sure I select only unsaturated stars. At the bottom of DynamicPSF, change the "Scale mode" to Literal Value and fill in Image scale (arcsec/px) of your setup and it will display true FWHMx/y in arcsec. Hopefully your image scale is not larger than 1"/px because it will result even larger FWHM.

There's a script called FWHMEccentricity which measures the average FWHM of the whole image. It looks at unsaturated stars and I think it's pretty accurate. The script is under Image Analysis. FWHM is measured in pixels so you will have to manually multiply this pixel value with your image scale in arcsec/px to get true FWHM value in arcsec.

Peter

[Juan Conejero]

Some important things to keep in mind about DynamicPSF:

1. DynamicPSF is essentially a function fitting tool. Like any fitting process applied to real world data, PSF fitting has inherent uncertainty. This basically means that, unless you are working with a synthetic noise-free image, you should not expect to get the same results for very similar objects.

2. Meaningful PSF measurements are impossible with nonlinear data. One should always work either with calibrated linear frames, or integrated images.

3. Very dim stars cannot be fitted correctly because most of the PSF wings are below the noise floor.

4. Very bright stars cannot be fitted correctly because either the PSF is truncated (saturation), or the corresponding pixels have values outside the linear range of the sensor. Anti-blooming devices increase this problem.

5. Undersampled images are problematic. If the FWHM is 2 pixels or less, the function fitting process becomes very uncertain because it has to work with few discrete data points.

6. Noisy images are problematic. Noise is uncertainty and high noise means high dispersion, which doesn't help to fit a function accurately.

7. By default, DynamicPSF fits elliptic functions. However, meaningful elliptic function fitting requires well sampled data and relatively low noise. With low-SNR and/or undersampled data, one should always select circular fitting functions.

8. DynamicPSF allows you to sort the list of fitted stars by several criteria. One of these criteria is MAD (mean absolute difference) residuals. This allows you to filter bad or unreliable fittings very easily. You can also sort by FWHM, rotation angle, aspect ratio, etc. Sorting the data is the best and easiest way to remove outliers from the set of fitted stars.

9. The DynamicPSF tool has been thoroughly documented.

10. The excellent SubframeSelector and FWHMEccentricity scripts by Mike Schuster apply the DynamicPSF process on hundreds or thousands of automatically detected stars, using robust statistical analysis methods to provide very accurate results.

[Terry Danks]

Many thanks to all for the replies here. I now understand a lot better what is going on and how to use this tool. In actuality, the FWHMEccentricity script will quickly give me what I want very quickly and easily. i was not awre of it. Thanks, Peter, for pointing it out.