STScI Logo

flatfields noao.imred.ccdred



flatfields -- Discussion of CCD flat field calibrations


This topic describes the different types of CCD flat fields and the tasks available in the ccdred and spectroscopy packages for creating them. Flat field calibration is the most important operation performed on CCD data. This operation calibrates the relative response of the detector at each pixel. In some cases this is as simple as taking a special type of observation called a flat field. However, in many cases this calibration observation must be corrected for illumination, scanning, wavelength, and aperture effects.

The discussion is in three sections; direct imaging, scan mode, and spectroscopy. Though there are many similarities between these modes of operation there are important differences in how corrections are applied to the basic flat field observations. The application of the flat field calibrations to the observations using ccdproc is the same in all cases, however.

1. Direct Imaging

The starting point for determining the flat field calibration is an observation of something which should have uniform response at all points on the detector. In addition the color of the light falling at each pixel should be the same as that in an observation so the same filter must be used when determining the flat field (the issue of the matching the color of the objects observed at the appropriate pixels is ignored here). The best calibration observation is of a blank sky. If an accurate blank sky observation can be obtained then this is all that is needed for a flat field calibration. This type of flat field might be called a sky flat , though this term is more often used for a type of flat field described below. There are two difficulties with this type of calibration; finding a really blank sky and getting a sufficiently accurate measurement without using all the observing time.

It is usually not possible to get a blank sky observation accurate enough to calibrate the individual pixels without introducing undesirable noise. What is generally done is to use a lamp to either uniformly illuminate a part of the dome or directly illuminate the field of view. The first type of observation is called a dome flat and the second is called a projection flat . We shall call both of these types of observations lamp flat fields . If the illumination is truely uniform then these types of observations are sufficient for flat field calibration. To get a very accurate flat field many observations are made and then combined (see flatcombine ).

Unfortunately, it is sometimes the case that the lamp flat fields do not illuminate the telescope/detector in the same way as the actual observations. Calibrating with these flat fields will introduce a residual large scale illumination pattern, though it will correctly calibrate the relative pixel responses locally. There are two ways to correct for this effect. The first is to correct the flat field observation. The second is to apply the uncorrected flat field to the observations and then apply an illumination correction as a separate operation. The first is more efficient since it consists of a single correction applied to each observation but in some cases the approximate correction is desired immediately, the observation needed to make the correction has not been taken yet, or the residual illumination error is not discovered until later.

For the two methods there are two types of correction. One is to use a blank sky observation to correct for the residual illumination pattern. This is different than using the sky observation directly as a flat field calibration in that only the large scale pattern is needed. Determining the large scale illumination does not require high signal-to-noise at each pixel and faint objects in the image can be either eliminated or ignored. The second method is to remove the large scale shape from the lamp flat field. This is not as good as using a blank sky observation but, if there is no such observation and the illumination pattern is essentially only in the lamp flat field, this may be sufficient.

From the above two paragraphs one sees there are four options. There is a task in the ccdred package for each of these options. To correct a lamp flat field observation by a blank sky observation, called a sky flat , the task is mkskyflat . To correct the flat field for its own large scale gradients, called an illumination flat , the task is mkillumflat . To create a secondary correction to be applied to data processed with the lamp flat field image the tasks are mkskycor and mkillumcor which are, respectively, based on a blank sky observation and the lamp flat field illumination pattern.

With this introduction turn to the individual documentation for these four tasks for further details.

2. Scan Mode

There are two types of scan modes supported by the ccdred package; shortscan and longscan (see ccdproc for further details). They both affect the manner in which flat field calibrations are handled. The shortscan mode produces images which are the same as direct images except that the light recorded at each pixel was collected by a number of different pixels. This improves the flat field calibration. If the flat field images, of the same types described in the direct imaging section, are observed in the same way as all other observations, i.e. in scan mode, then there is no difference from direct imaging (except in the quality of the flat fields). There is a statistical advantage to observing the lamp or sky flat field without scanning and then numerically averaging to simulate the result of the scanning. This improves the accuracy of the flat fields and might possibly allow direct blank sky observations to be used for flat fields. The numerical scanning is done in ccdproc by setting the appropriate scanning parameters.

In longscan mode the CCD detector is read out in such a way that each output image pixel is the sum of the light falling on all pixels along the direction of the scan. This reduces the flat field calibration to one dimension, one response value for each point across the scan. The one dimensional calibration is obtained from a longscan observation by averaging all the readout lines. This is done automatically in ccdproc by setting the appropriate parameters. In this case very good flat fields can be obtained from one or more blank sky observations or an unscanned lamp observation. Other corrections are not generally used.

3. Spectroscopy

Spectroscopic flat fields differ from direct imaging in that the spectrum of the sky or lamp and transmission variations with wavelength are part of the observation. Application of such images will introduce the inverse of the spectrum and transmission into the observation. It also distorts the observed counts making signal-to-noise estimates invalid. This, and the low signal in the dispersed light, makes it difficult to use blank sky observations directly as flat fields. As with direct imaging, sky observation may be used to correct for illumination errors if necessary. At sufficiently high dispersion the continuous lamp spectrum may be flat enough that the spectral signature of the lamp is not a problem. Alternatively, flux calibrating the spectra will also remove the flat field spectral signature. The spectroscopic flat fields also have to be corrected for regions outside of the slit or apertures to avoid bad response effects when applying the flat field calibration to the observations.

The basic scheme for removing the spectral signature is to average all the lines or columns across the dispersion and within the aperture to form an estimate of the spectrum. In addition to the averaging, a smooth curve is fit to the lamp spectrum to remove noise. This smooth shape is then divided back into each line or column to eliminate the shape of the spectrum without changing the shape of the spectrum in the spatial direction or the small scale response variations. Regions outside of the apertures are replaced by unity. This method requires that the dispersion be aligned fairly close to either the CCD lines or columns.

This scheme is used in both longslit and multiaperture spectra. The latter includes echelle, slitlets, aperture masks, and fiber feeds. For narrow apertures which do not have wider slits for the lamp exposures there may be problems with flexure and defining a good composite spectrum. The algorithm for longslit spectra is simpler and is available in the task response in the longslit package. For multiaperture data there are problems of defining where the spectra lie and avoiding regions off of the aperture where there is no signal. The task which does this is apnormalize in the apextract package. Note that the lamp observations must first be processed explicitly for bias and dark count corrections.

Longslit spectra may also suffer the same types of illumination problems found in direct imaging. However, in this case the illumination pattern is determined from sky observations (or the flat field itself) by finding the large scale pattern across the dispersion and at a number of wavelengths while avoiding the effects of night sky spectrum. The task which makes this type of correction in the longslit package is illumination . This produces an illumination correction. To make sky flats or the other types of corrections image arithmetic is used. Note also that the sky observations must be explicitly processed through the flat field stage before computing the illumination.


ccdproc, guide, mkillumcor, mkillumflat, mkskycor, mkskyflat
apextract.apnormalize, longslit.response, longslit.illumination

Search Form · STSDAS