STEP 2: Shapelet simulation information
STEP home Simulations Data Mailing List


Download from from The Original set, the Rotated set and the input shears and catalogues are now available

This directory contains simulated R-band images for the Shear TEsting Program (STEP). The images contain a known shear signal, and are intended as a blind test to calibrate various existing shear measurement methods, and as a useful tool to help develop new ones.

The images resemble ones that might be obtained with four ten-minute exposures in good conditions with Suprime-Cam on Subaru. Their pixel scale is 0.2" and the photometric zeropoint is 30.8. The depth, noise properties and input shear signal are constant across the image. In each case, image #00 contains zero shear, to be used if your shear measurment method requires an ensemble galaxy population without shear. The rest contain random shear values in both &gamma 1 and &gamma 2, but with |&gamma|<8%.

The PSF is also constant across the image - however, if you assume that is constant when interpolating to the positions of the galaxies, then please declare so in your analysis. The galaxy images contain a few stars, but I wanted to avoid having too many stars that all the galaxies become obscured. The starfield directory contains images with densely populated stellar fields. Obtain your PSF model from these, or co-add them with the image, if your method requires more. Versions of the starfields without background noise are available upon request for coadding. The PSF shapes (and the galaxy shapes, except for PSF B) are realised using shapelets, so the ellipticity of their isophotes is expected to vary as a function of radius: be careful! For more information about shapelets, the Shapelets webpage.

Six different PSFs are used. These have been chosen to test specific aspects of weak lensing measurement. They are:

PSF A - Typical Subaru PSF (~0.6"), shapelet galaxies.
PSF B - Typical Subaru PSF (~0.6"), pure exponential galaxies.
PSF C - Enlarged Subaru PSF (~0.8"), shapelets galaxies.
PSF D - Elliptical PSF aligned along +ve e1 direction, shapelets galaxies.
PSF E - Elliptical PSF aligned along +ve e2 direction, shapelets galaxies.
PSF F - Circularly symmetric PSF, shapelets galaxies.

For each PSF, there are 64 7'x7' images, each containing ~1200 useable galaxies (identical from image to image before shear). That should be enough to get 0.1% precision in each case. If necessary, I can make some more images, either with the same PSFs or with new ones (elliptical along -ve e directions? Suggestions welcome!). I'm also working on a set with a finer pixel scale for Doug.

Good luck!

Any comments/questions specific to these simulations should be directed to Richard Massey, rjm[at]

Notes on sky noise in STEP2

Dear STEPpers,

I have recently had a few queries about the noise model used for the STEP2 images. I hope the information below will clear up some confusion, without giving away any "advantage" that would be unfair on people who have already finished their analysis.

The STEP2 simulated images contain a relatively simple noise model. It is described to some level by a set of parameters stored in the image headers. However, they were really stored for my own reference; I wasn't anticipating that people would use them, and they are not arranged in the most transparent format. Here I describe the STEP2 noise model in a little more detail. The shapelet images created for STEP1 also contain this noise model.

There is a "photon counting" shot noise component added to the true flux in every pixel, proportional to the square root of the number of photons. The noise distribution should really have been Poisson, particularly for the very faint objects, but is drawn from an approximating Gaussian in practice. The rms of the Gaussian is 0.033.

There is also a noise component from sky background everywhere in each image. The mean background level has been subtracted, so the mean background level in the publicly available images is zero. Since it was high everywhere, it was definitely OK to model the fluctuations around this as Gaussian. For most analysis purposes, these are the most important element of the noise model. To make life a little more complicated, the noise model was built to mimic real Subaru data that I have, which was reduced using DRIZZLE, to help remove camera distortions. This procedure correlates the noise in adjacent pixels. As was used in the shapelet STEP1 images, and as we discussed while planning the STEP2 simulations, the simulated noise is smoothed (with a Gaussian kernel) to mimic the combined effects of DRIZZLE and image stacking. The rms of the sky noise is 1.65 after smoothing. A simulated image of a completely blank patch of sky is available from the STEP2 download page if you would like to further investigate the covariance between pixels. You should be particularly aware of this correlated noise when detecting objects (very faint, round sources are more likely to be fake) and when computing objects' weights from their detection S/N.


Last modified 5th Nov.
Shapelet simulations made available by Richard Massey, rjm[at]