7.10 IR Observing Strategies

7.10.1 Dithering Strategies

For imaging programs, STScI generally recommends that observers employ dithering patterns. Dithering refers to the procedure of moving the telescope by small angle offsets between individual exposures on a target. The resulting images are subsequently combined in the pipeline or by the observer using software such as AstroDrizzle. (See the DrizzlePac documentation.)

Dithering is used to improve image quality and resolution. By combining multiple images of a target at slightly different positions on the detector, one can compensate for detector artifacts (blemishes, dead pixels, hot pixels, transient bad pixels, and plate-scale irregularities) that may not be completely corrected by application of the calibration reference files. Combining images, whether dithered or not, can also remove any residual cosmic ray flux that has not been well removed by the up-the-ramp fitting procedure used to produce flt images (see Section 7.7.2 and Appendix E). Effective resolution can be improved by combining images made with sub-pixel offsets designed to better sample the PSF. This is especially important for WFC3/IR, because the PSF is undersampled by about a factor of 2 (see Table 7.5).

Larger offsets are used to mosaic a region of sky larger than the detector field of view. (Large offsets can also be used for “chopping” to sample the thermal background. This has been recommended for NICMOS exposures at wavelengths longer than 1.7 microns, where the telescope thermal background becomes increasingly dominant, but the thermal background is not a problem for WFC3/IR). In WFC3, all offsets must be accomplished by moving the telescope (whereas in NICMOS it was also possible to move the Field Offset Mirror).

Dithers must be contained within a diameter ~130 arcsec or less (depending on the availability of guide stars in the region) to use the same guide stars for all exposures. The rms pointing repeatability is significantly less accurate if different guide stars are used for some exposures. (See Appendix B of the DrizzlePac Handbook.). Mosaic steps and small dither steps are often combined to increase the sky coverage while also increasing resolution and removing artifacts. (See Section 6.12.1 for a discussion of the effect of geometric distortion on PSF sampling for mosaic steps).

The set of Pattern Parameters in the observing proposal provides a convenient means for specifying the desired pattern of offsets. The pre-defined mosaic and dither patterns that have been implemented in APT to meet many of the needs outlined above are described in detail in the Phase II Proposal Instructions. The WFC3 patterns in effect in APT at the time of publication of this Handbook are summarized in Appendix C. Observers can define their own patters to tailor them to the amount of allocated observing time and the desired science goals of the program. Alternatively, they can use POS TARGs to implement dither steps (Section 7.4.3). Observers should note that thermally driven drift of the image on the detector, occasionally larger than 0.15 pixels in two orbits, will limit the accuracy of execution of dither patterns (WFC3 ISR 2009-32). Additional information on dither strategies can be found in WFC3 ISR 2010-09, which provides a decision tree for selecting patterns and combining them with subpatterns. WFC3 ISR 2016-14 provides compact patterns with up to 9 steps (in the form of POS TARGs) designed to preserve sub-pixel sampling as much as possible over the face of the detector, given the scale changes introduced by geometric distortion.

7.10.2 Parallel Observations

Parallel observations, i.e., the simultaneous use of WFC3 with one or more other HST instruments, are the same for the IR channel as for the UVIS channel, previously described in Section 6.12.2.

7.10.3 Exposure Strategies

Given the variety of requirements of the scientific programs that are being executed with WFC3/IR, it is impossible to establish a single optimum observing strategy. In this section we therefore provide a few examples after guiding the reader through the main constraints that should be taken into account when designing an observation:

  • Integrate long enough to be limited by background emission and not read noise. Dark current is rarely the limiting factor.
  • Dither enough so that resolution can be restored to the diffraction limit and bad pixels and cosmic-ray impacts can be removed, while maintaining a homogeneous S/N ratio across the field.
  • Split the MULTIACCUM ramps into as many reads as possible for readout noise suppression.

In this regard, it is useful to consider Table 7.11, which summarizes the total background seen by a pixel, including sky, telescope, and nominal dark current, and the time needed to reach 400 e/pixel of accumulated signal, corresponding to 20 e/pixel of Poisson-distributed background noise. This last value, higher than the expected readout noise of ~12 electrons after 16 reads, is used here to set the threshold for background-limited performance. The passage from readout-limited performance to background-limited performance can be regarded as the optimal exposure time for that given filter, in the sense that it allows for the largest number of dithered images without significant loss of S/N ratio (for a given total exposure time, i.e., neglecting overheads). For faint sources, the optimal integration time strongly depends on the background (zodiacal, Earth-shine thermal, and dark current) in each filter, ranging from just 220 s for the F110W filter to 2700 s for some of the narrow-band filters.

These constraints put contradictory requirements on the ideal observing strategy. It is clear that, given a certain amount of total observing time, the requirement of long integrations for background limited performance is incompatible with a large number of dithering positions. Also, to split ramps for readout noise suppression decreases the observing efficiency, with a negative impact on the signal to noise ratio. Because the background seen by each pixel depends on the filter (Section 7.9.5), the optimal compromise must be determined on a case-by-case basis.

The optimal integration time needed to reach background-limited performance (see Table 7.11) can be compared with the integration times of the sampling sequences from Table 7.8Table 7.12 synthesizes the results, showing for each filter which ramp (SPARS, STEP) most closely matches the optimal integration times for NSAMP=15.

Table 7.11: Background (e–/pix/s) levels at the WFC3/IR detector.

Filter

Thermal

Zodiacal

Earth-
shine

Dark
Current

Total

Optimal
Integration
Time (sec)

F105W

0.051

0.774

0.238

0.048

1.111

360

F110W

0.052

1.313

0.391

0.048

1.804

222

F125W

0.052

0.786

0.226

0.048

1.112

360

F140W

0.070

0.968

0.267

0.048

1.353

296

F160W

0.134

0.601

0.159

0.048

0.942

425

F098M

0.051

0.444

0.140

0.048

0.683

586

F127M

0.051

0.183

0.052

0.048

0.334

1198

F139M

0.052

0.159

0.044

0.048

0.303

1320

F153M

0.060

0.153

0.041

0.048

0.302

1325

F126N

0.051

0.037

0.011

0.048

0.147

2721

F128N

0.051

0.040

0.011

0.048

0.150

2667

F130N

0.051

0.041

0.011

0.048

0.151

2649

F132N

0.051

0.039

0.011

0.048

0.149

2685

F164N

0.065

0.036

0.009

0.048

0.158

2532

F167N

0.071

0.035

0.009

0.048

0.163

2454

The columns show, from left to right: a) filter name; b) thermal background from the telescope and instrument; c) zodiacal background; d) earth-shine background; e) dark current; f) total background; g) integration time needed to reach background-limited performance, set at an equivalent readout noise of 20 electrons.


The selection of which sample sequence type (RAPID, SPARS, STEP; Section 7.7.3) must take into account the science goals and the restrictions placed on their use. Most observers have found that the SPARS ramps best meet the needs of their programs. Here are some factors to consider when selecting a sample sequence:

  • The RAPID ramp is a uniform sequence of short exposures. With its relatively short maximum exposure time, is suitable for a target consisting of bright objects that would saturate after a few reads in the other sequences. It is not appropriate for background-limited performance.
  • SPARS ramps, with their uniform sampling, provide the most robust rejection of cosmic-ray events, and can be trimmed by removing a few of the final reads to fine-tune the integration time with little degradation of the achieved readout noise. Thus they are considered the standard sampling mode.
  • STEP ramps are preferable where large dynamic range is needed; e.g., for photometry of stellar clusters. These ramps begin with a sequence of four uniform (RAPID) reads and end with a sequence of much longer uniform reads. The transition between the two uniform read rates is provided by a short sequence of logarithmically increasing read times. This design provides for correction of any non-linearities early in the exposure and allows for increased dynamic range for both bright and faint targets.

Finally, the selection of a given sample sequence type should also be made in conjunction with the number of samples (nsamp) that will be used to achieve the desired total exposure time for the observation. Long exposures should in general use a minimum of 5-6 samples in order to allow for reliable CR rejection and to allow for at least a few unsaturated samples of bright targets in the field. For very faint targets in read-noise limited exposures, a larger number of samples will result in greater reduction of the net read noise and a more reliable fit to sources with low signal. Short exposures of bright targets, on the other hand, can get by with fewer samples. This is especially true, for example, for the direct images that accompany grism observations. Because the purpose of the direct image is to simply measure the location of sources - as opposed to accurate photometry - they can reliably use an nsamp of only 2 or 3.

Table 7.12: Optimal exposure time needed to reach background-limited performance (see Table 7.11) for each WFC3/IR filter, along with the NSAMP=15 sequences that provide the closest match.

Filter

Optimal
exposure
time (sec)

SPARS

STEP

F105W

360

SPARS25

STEP50

F110W

222

SPARS25

STEP25

F125W

360

SPARS25

STEP50

F140W

296

SPARS25

STEP25

F160W

425

SPARS25

STEP50

F098M

586

SPARS50

STEP50

F127M

1198

SPARS100

STEP200

F139M

1320

SPARS100

STEP200

F153M

1325

SPARS100

STEP200

F126N

2721

SPARS200

STEP400

F128N

2667

SPARS200

STEP400

F130N

2649

SPARS200

STEP400

F132N

2685

SPARS200

STEP400

F164N

2532

SPARS200

STEP400

F167N

2454

SPARS200

STEP400

The benefits and disadvantages of each sequence type are discussed in the accompanying text.

7.10.4 Spatial Scans

Spatial scanning is available with either WFC3 detector, UVIS or IR. Conceptually producing star trails on the IR detector is the same as producing star trails on the UVIS detector, with a few differences discussed in WFC3 ISR 2012-08. This document is recommended to anyone preparing a phase II proposal that uses spatial scans for any purpose. Spatial scans are discussed more extensively in Section 6.12.5(UVIS imaging) and Section 8.6 (IR slitless spectroscopy). The former section describes star trails and the latter section describes spectra trailed perpendicular to the dispersion direction.

7.10.5 PSF Subtraction

IR imaging has been shown to be highly effective in detecting faint point sources near bright point sources (WFC3 ISR 2011-07). In this study, deep dithered exposures of a star were made at a variety of roll angles. Unsaturated exposures of a star, scaled down in flux to simulate faint companions of various magnitudes, were added to the deep exposures. The faintness of the companion that can be detected at a certain separation from the bright star depends on the degree of sophistication used to generate a reference image of the PSF to subtract from each set of dithered exposures. For a separation of 1.0 arcsec, five sigma detections could be made fairly easily for companions 8 or 9 magnitudes fainter than the bright star, and companions more than 12 magnitudes fainter than the bright star could be detected at separations of a few arcsec. Substantial improvements in detectability at separations less than about 2 arcsec could be made using the methodology described in the ISR to generate the reference PSF.
If observers want to use stellar images to subtract the PSF from a target comprised of a point source and an extended source to detect or measure the extended source, they should keep several points in mind:

  • IR pixels undersample the PSF (Section 7.6), so the stellar and target exposures should be dithered to produce good sampling of the PSF.
  • Position drift and reacquisition errors can broaden the PSF (WFC3 ISR 2009-32WFC3 ISR 2012-14).
  • If a single guide star is used for a visit, roll angle drift causes a rotation of the target around that star, which in turn introduces a small translational drift of the target on the detector. In recent years, as gyroscopes have failed and been replaced, the typical roll angle drift rate has increased from 1.5 mas/sec to ~17 mas/sec, producing a translation of up to 60 mas (0.45 IR pixel) in 1000 sec.
  • The characteristics of the PSF depend on the focus, which generally changes measurably during an orbit; its range in a particular orbit will not be known in advance (WFC3 ISR 2012-14).
  • The characteristics of the PSF vary with location on the detector (e.g., see ACS ISR 2003-06).
  • If exposures long enough to have good signal-to-noise to the desired radius will saturate the central pixels, one should consider making shorter exposures to avoid the effects of persistence. (Section 7.9.4Appendix D.)

While Tiny Tim modeling is available for the WFC3 IR detector, it has not been optimized to reproduce observed PSFs. Progress has been made in understanding the short-comings in the model implemented in version 7.4 (WFC3 ISR 2012-13WFC3 ISR 2014-10). See Section 7.6.4 for a discussion of on-going work to provide PSF models to observers.

7.10.6 The Drift and Shift (DASH) Observing Strategy

The term DASH (for “drift-and-shift”, Momcheva et al., 2016) has been adopted to describe the observing strategy of taking a series of WFC3/IR exposures of many targets within one orbit while the telescope is being guided under gyroscope control, thus avoiding the overhead cost of acquiring a new pair of guide stars for every slew between targets of greater than about 2 arcmin. A WFC3/IR sample sequence comprised of short exposure times is selected to limit image smearing within each time step, and the differential samples in one exposure are later aligned and combined to compensate for the greater drift due to gyroscope control. (See Figure 7.15.) The technique was designed to allow users to carry out shallow large-scale mosaic observations with the WFC3/IR camera, but has since been adapted to efficiently observe a collection of bright targets within a field ~1 deg across with WFC3/IR subarray apertures (which have shorter time steps for a given sample sequence) and WFC3/UVIS subarray apertures. More detailed discussions of HST guiding and considerations for program design are given below. See Section 6.12.5 for further discussion of the WFC3/UVIS observations.

A DASH orbit should begin with a guide star acquisition followed by exposure(s) using using fine guiding to ensure adequate pointing at the beginning of the orbit. Gyroscope performance in recent cycles has made it necessary to limit each DASH visit to one orbit and to institute a required minimum amount of time using fine guiding. The Orbit Planner in APT shows FGS Pause after the exposure(s) taken with fine guiding. The time between the end of the guide star acquisition (GS Acq) and the FGS Pause must be at least 5 minutes to provide time to update the gyroscope bias before proceeding with a slew or dropping to gyroscope mode. The orbit must therefore begin with more or longer guided exposures than in earlier cycles. The Program Coordinator will work with the PI to see if it is possible to plan orbits to achieve the proposed science that fit the scheduling windows.

All of the exposures should be grouped together into a non-interruptible sequence container in APT to confine them to one uninterrupted orbit. When placed in this container, the exposures will be observed without gaps due to Earth occultation or SAA passages. As a further precaution, if subarrays are used, the minimum size should be carefully chosen, and a series of increasingly large subarrays should be used over the course of the orbit as needed to allow for the accumulating drift. Observers should also consider using the offset pattern that minimizes the length of moves during an orbit. Offsets are made with a small angle maneuver (SAM), and the repositioning overheads increase with the offset size of the SAM. Overheads for SAMs of different sizes are shown in Table 10.1. More details on planning DASH observations in APT are presented in WFC3 STAN Issue 23.

Some of the earliest DASH obserations (described in Momcheva et al., 2016) were made in GO program 14114 (P.I. van Dokkum) with very stable gyro guiding. The exposures were taken with the full array using the SAMP-SEQ SPARS25 with NSAMP=10, 11, or 12, with 8 sample sequences per orbit. Each sample sequence had 9 to 11 differential exposures of 25 sec, and after dropping to gyro mode the drift in each 25 second interval was on average less than half of a WFC3/IR pixel. The pattern of SAMs resulted in overheads ~1 minute. Suitable step sizes in the WFC3/IR sample sequence depend on the expected accuracy of gyroscope guiding. This accuracy has changed with time. Observers should consult with their Contact Scientist when preparing the phase II proposal to determine step sizes short enough to produce an acceptable amount of smearing.

A new set of tools in a DASH pipeline was released in January 2021. These tools, along with a Jupyter notebook providing a walk through, are available through GitHub to assist users with calibrating DASH observations. Information regarding the pipeline and where to access it are reported in WFC3 ISR 2021-02 and in STAN issue 34. For users interested in receiving STAN publication announcements, please subscribe via the instructions at the bottom of our last issue.


Figure 7.15: Illustration of the “drift and shift” (DASH) method of restoring unguided WFC3/IR images.


(a) Shows the standard data product (the FLT file) of an unguided, gyro-controlled exposure. The objects are smeared due to the lack of fine guidance sensor corrections.
(b) Shows an example of the individual samples that comprise the final exposure. The smearing is small in each individual sample.
(c) Shows the reconstructed image after shifting the samples to a common frame and adding them.