Francine Marleau and I have been actively working on measuring galaxy structural parameters on the HDF NICMOS images. There are two datasets, Dickinson's and Thompson's. The Dickinson dataset is shallow, but it covers all of HDF-North. The Thompson dataset is very deep, but it only covers the WPFC2 Chip 4 region of the HDF. We chose to work on the Thompson set first. The Thompson set consists of 147 images with an exposure time of 896 seconds each.
NICMOS/NIC3 images are badly undersampled. Drizzling of undersampled images introduces artifacts in the PSF and in the noise that are extremely hard to model. For example, the residual image of a galaxy analyzed on a drizzled image will show a "ghost image" of the dithering pattern. Not good!! We got absolutely hideous looking model residuals when we ran GIM2D on the combined HDF-NICMOS image as we expected.
The approach we are taking is to simultaneously fit all the images without drizzling them. Those of you familiar with Peter Stetson's ALLFRAME stellar photometry package will know exactly what we are talking about. We do the same thing as ALLFRAME, but we deal with galaxies. So, we fit a model to 147 images simultaneously using a task called MGIMFIT2D. As the members of the GIM2D mailing list know by the number of updates on the list, MGIMFIT2D has evolved rapidly over the past few weeks. MGIMFIT2D features different modes of analysis to deal with different types of image stacks. In the case of the Thompson HDF dataset, we combined all 147 images, and ran SExtractor on it to make a deep object catalog. This catalog was then "imposed" on all 147 images in the image stack i. e. science and segmentation thumbnail images were extracted on a given image even though SExtractor would not have detected an object on that single image. We then generated TinyTim PSFs for all the objects on all 147 images (that's a lot of PSF's!). Finally, we used a GIM2D task called GGROUPER to group all thumbnail images together and to create all the required analysis scripts (a lot of tedious and awful image bookkeeping!). Once all the software was in place, it was very straightforward to run through the whole set-up.
As a "proof of concept" test, we applied our analysis method to a few
galaxies on a subset of 20 Thompson images. You can see the results by
downloading the two Postscript images below:
Combination of 20 F160W HDF-N NICMOS images used for test
Re-constructed GIM2D F160W residual image
IMPORTANT: This is a PARTIAL test i.e. not all galaxies have been analyzed yet.
ALSO IMPORTANT:
Those of you with good eyes will note that some galaxies around the
edges of the image appear to have been barely subtracted out by their
fits. Examples are the galaxies at (49, 246), (203, 256) and (254,
236). **** This is an artifact of the residual image reconstruction! ***
Take the object at (48, 245) for example. It was fitted on
16 of the 20 images in the test image stack because it was too close to
the edge of the frame on the remaining four images. So, when the residual
image is re-constructed, these four images contribute the full object flux
(and not residual object flux as they should have!) to the final residual
image. The 16 residual images for (48, 245) actually look beautiful
i.e. like pure noise!
Here is a first pass through the full set of 147 images (896 seconds each) with MGIMFIT2D. The images are :
Combination of 147 F160W HDF-N NICMOS images
Re-constructed GIM2D F160W residual image
Many galaxies show beautiful residual structures. This first pass does not include the NIC3 intrapixel response function in the PSF convolutions. The next version will in addition to further improvements to the GGROUPER/MGIMFIT2D pipeline (goodbye residual image reconstruction artifacts!!).