STIS Instrument Science Report 95-002 STIS Design Reference Mission and Ground System Volume Requirements Stefi Baum, Phil Hodge, Ray Kutina June 1995 ABSTRACT In this ISR we describe a scenario for the expected useage of STIS. From that scenario and from our current expectations for the STIS pipeline calibration software we derive estimates of the expect rate of STIS exposures (100/day), datasets (50/day) downlink volume (2.3 Gbits/day), archive volume (3.3 Gbytes/day), CPU requirements (500 calwfpc2s/ day) and calibration reference file OPUS on-line storage volume (3 Gbytes). 1. Introduction In this ISR we describe how we expect STIS to be utilized for GO science. We provide an estimate of • the number of exposures generated by STIS, per orbit and per day • the number of unique configurations (slit plus MSM positions) utilized by STIS per orbit1 • the volume of data transmitted from the telescope per orbit and per day, which we refer to as the downlink volume • the volume of science data sent to the archive per orbit and per day (archive volume) • the number of CAL class datasets sent to the archive per orbit and per day • the estimated cpu time per day (in units of calwpc2) needed to calibrate STIS data in OPUS • the volume of calibration reference files OPUS must maintain in its operational environment at any one time, assuming they clear out unneeded files every 3 months. 1. This information is needed to compute the command volume, which is considered in a separate document. The information needed to determine the command volume is contained within Section 2 (See “STIS Useage Scenarios” on page 2.). 1 For the purposes of these estimates we assume that STIS will be operating as the prime instrument for 5 orbits per day, and has the potential to operate in parallel for 10 orbits a day. A summary of our results is provided in Table 1. Table 1. Conclusions of Estimate Study Entity Number STIS exposures per day 100 downline volume per day 2278 Mbits archive volume per day 2941 Megabytes archived CAL class datasets per day 50 CPU processing per day in units of calwpc2 CPU a 500 3-Month On-Line OPUS Reference file Volume 3 Gigabtyes a. equivalent number of wfpc2 datasets processed through OPUS The remainder of this ISR is organized as follows. In Section 2 (“STIS Useage Scenarios”, page 2), we provide a scientific scenario for the utilization of STIS, as the prime instrument, as the parallel instrument, and for calibration purposes. From this we derive the number of exposures the number of datasets and the downlink data volume expected from STIS per orbit/day (see Table 2 on page 4). In Section 3 (“Calibration in the OPUS Pipeline: Archive Volume and CPU Time Estimates”, page 9). we provide a brief description of the pipeline calibration of STIS data and determine the CPU calibration time estimates and the data volume generated per science dataset/per orbit/per day for STIS. In Section 4 “On-line OPUS Calibration File Requirements”, page 12, we provide a description of the on-line calibration file volume needs for OPUS. Section 5 (“APPENDIX A. Science Uses of STIS as Prime Instrument”, page 13), contains a detailed explanation of how we arrived at the scientific scenario (presented in Section 2) upon which the majority of the results presented in this document are predicated. All of the estimates presented in this ISR are based on our current (pre ground calibration and flight/ground software development) understanding of STIS, and our experiences with the existing HST spectrographic and imaging instruments. The estimates are as reliable as can be determined at this stage. 2. STIS Useage Scenarios STIS features large area detectors. The MAMA detectors have 1024 x 1024 pixels which when read out in high resolution mode can produce 2048 x 2048 pixel images. This is cur- June 1995 STIS Instrument Science Report 95-002 Page 2 rently expected to be the default option. Full frame MAMA exposures can be binned upon readout on board to produce 1024 x 1024 ‘lowres’ (16 Megabit) MAMA images, or 1024 x 2048 MAMA images, if binning is done in only one dimension. In the optical, the full STIS CCD readout is a 1060 x 1040 image (including the overscan regions). Given the large size of the STIS MAMA and CCD images, the upper envelope at which exposures can be generated for full STIS frames will be limited by the data volume allocation. Note that though the MAMA arrays are read out as 2048 x 2048 pixels, in order to ensure that a wavecal and a MAMA science frame occupy only 1/2 the data buffer volume, MAMA full resolution arrays will be saved in memory as 2018 x 2018 arrays. For the purposes of these calculations we assume MAMA science frames are 2018 x 2018 x 16 /1024 x 1024)= 62 Mbits and that a single MAMA science frame plus a single wavecal occupy 2048 x 2048 x 16 /(1024 x 1024) = 64 Mbits. For simplicity we refer to the full CCD images as 1024 x 1024 images of 16 Mbits. The target acquisitions will generate 3 small images within a single exposure, which totals 0.5 Mbits in size. All of the STIS instruments can also be utilized in subarray readout mode, in which case many more exposures (within the same data volume) can be generated per day. The maximum number of exposures possible in a single observation series is presently limited by ground software limitations to be less than 1296. In general (see Section 5, page 13), we do not expect users to utilize subarray readout mode, (with the exception of first order spectroscopy of single stars and of course target acquisitions and wavecals) unless forced to by data volume constraints, since it limits the spectral or spatial range of their observations. In the subsequent subsections we estimate the rate at which we expect exposures, datasets, and Mbits of data to be generated by STIS. We divide the discussion into three categoried • Prime GO/GTO Science • Parallel GO/GTO Science • non-Pointed Calibration programs. We summarize our results in advance in Table 2. June 1995 STIS Instrument Science Report 95-002 Page 3 Table 2: Volume Requirements Posed by Estimated STIS Observing Scenario Type of Science Percent of STIS Prime Time Exposures/Orbit Orbits /Day Exposures /Day Datasets /Day Downlink Vol/Day (Mbits) CAL Class Archive Vol/Day (Mbytes) TOTAL Archive Vol/Day (Mbytes) Faint Target Full Array UV Prime Science 60% 1 TA (3 subarray images) 2 Wavecal Exp 2 MAMA ((2018x2018) 3.0 15 9 386 579 676 FP SPLIT Full Array UV Prime Science 3% 1 TA Exp 1 Wavecal 5 MAMA (2018x2018) 0.15 1.05 0.3 47 70 82 Rapid Read, Subarrays, UV and Optical Prime Science 3% 1 TA Exp 1 Wavecal 50 MAMA (subarrays, full buffer) 0.15 7.8 0.3 19 29 34 Rapid Read, Full Array (lowres), UV and Optical Prime Science 3% 1 TA exp 1 wavecal 100 CCD or MAMA (1024 x 1024) 0.15 15.3 0.3 240 360 420 Faint Target, Full Array, Visible Prime Science 32% 1 TA 2 Wavecals 4 CCD 1.6 11.2 4.8 102 154 178 Parallel a Full Array (lowes) UV and Optical Science N/A 4 MAMA/CCD exp (1024 x 1024) 7.35 29.4 20 470 705 822 Non-Pointed Calibration (Internal) N/A 1 MAMA (2048x2048) 15 15 15 960 840 1080 TOTAL OVERALL N/A N/A N/A 95 50 2224 2737 3292 a. Assumes STIS ACCUM mode parallels of duration less than 5 minutes are disallowed and that all MAMA parallels are done in binned (lowres) mode to produce 1024 x 1024 images. Prime STIS GO/GTO Science To estimate the STIS volume from GO/GTO prime science we need to understand the ways in which STIS is likely to be utilized scientifically. As explained in detail below, we come to the following, conservative, conclusions with regard to STIS prime science useage: June 1995 STIS Instrument Science Report 95-002 Page 4 • ~60% of all STIS prime orbits will be devoted to long integration uv observations of ‘faint’ sources and will utilize the MAMAs in full frame readout. These orbits will consist of a target acquisition, two uv exposures at two MSM settings (but a single slit wheel position) and two wavecals. This will generate one target acquisition dataset containing a single exposure (but 3 images) of total size 0.5 Mbit, and two science datasets each with 2 exposures (1 science and 1 wavecal each). Each science dataset will be 64 Mbits downlink volume. • ~32% of all STIS prime orbits will be devoted to ‘faint’ optical observations and will utilize the CCD in full frame readout. These orbits will consist of a target acquistion and two crpslit exposure sequences at two MSM settings (but a single slit wheel position) with two accompanying wavecals. This will generate one target acquisition dataset of a single exposure of size 0.5 Mbit, and two science datasets, each containtaing 3 exposures (2 science and 1 wavecal). Each science dataset will be 32 Mbits downlink volume. • ~3% of all STIS prime orbits will utilize the FPSPLIT mode to obtain high dynamic range spectra in the UV. These orbits will consist of a target acquistion, and 5 full frame MAMA exposures, plus a single wavecal at a single MSM position. This will generate one target acquisition dataset of a single exposure with size 0.5 Mbit, and one science dataset, containing 6 exposures (5 science and 1 wavecal). The science dataset will be 312 Mbits downlink volume. Note that each of the 5 FPSPLIT exposures is at a unique slit wheel position. • ~3% of all STIS prime orbits will be devoted to full frame optical or 1024 x 1024 uv MAMA rapid readouts,, for example of planetary objects. These orbits will consist of a single target acquisition, 100 science exposures, and a single wavecal, all at a single MSM and slit wheel position. These orbits will generate one target acquisition dataset with one exposure of size 0.5 Mbit and one science dataset containing 101 exposures (100 science, 1 wavecal). The science dataset will be 1.7 Gbits downlink volume. • ~3% of all STIS prime orbits will be devoted to rapid readout uv or optical observations of bright stellar objects using subarrays, either for purposes of variability or due to the need to avoid saturation. These orbits will consist of a single target acquisition, 50 science exposures of approximate dimension 150 x 1024 and a single wavecal at a single MSM and single slit position. Here we have assumed the volume of data taken just fills the STIS science data buffer. These orbits will generate one target acquisition dataset with one exposure of size 0.5 Mbit, and one science dataset containing 51 exposures (50 science exposures and a single wavecal). The science dataset will be 128 Mbits downlink volume. This useage scenario is based on an understanding of the STIS instruments, a guess at the types of science they will be best suited for, a rough sketch of the STIS IDT science program, and an analysis of the to-date uses of the current HST spectroscopic instruments (FOS and GHRS). The logic used to arrive at this scenario is described in detail in “APPENDIX A. Science Uses of STIS as Prime Instrument”, page 13. For the purposes of estimating command volume (though not for any of the conclusions presented in this document) it is also important to estimate the nature of the target acquis- June 1995 STIS Instrument Science Report 95-002 Page 5 tions which wil be performed. Table 3 (condensed from memo from Steve Kraemer [IDT] dated Jun 21, 1995) summarizes types of acuisitions we expect to be performed. Roughly 1/2 of target acquisitions (those for science conducted through the narrow slits) are expected to be followed by peakups (acq/peak); the number of dwell points for different science modes of STIS is summarized in Table 4 (Steve Kraemer, 1995). Table 3: STIS Target Acquisitions Acquistion Type Percent of Acquisitions which are of this Type isolated point source 80% crowded field 5% diffuse source 10% planetary 5% Table 4: Peakup Acuisitions observation type dwell points per peakup echelle 15 first order 7 other (imaging) modes 0 Parallel STIS Observing Under normal conditions, assuming we disallow STIS parallel exposures shorter than 5 minutes in duration and restrict MAMA parallels to be binned (‘lowres’) 1024 x 1024 images, STIS parallel exposures will generate roughly 470 Megabits per day in data volume, within 30 exposures (translating into 20 datasets). If we were to take STIS parallel exposures (down to minute timescales) then the volume can go as high as 2070 Megabits, within 130 exposures (translating into roughly 100 datasets) per day! Note that such short MAMA exposures could be scientifically useful, so long as the total integration on any one patch of sky is of order 1 orbit, because the MAMAs do not suffer from read noise. Below we describe how we have arrived at these numbers and conclusions. Background To estimate the number and volume of STIS parallel exposures we need to know the distribution of exposure times for WFPC2 and NICMOS, the two primary instruments which STIS will be operating parallel to. For this purpose we use the Exposure Duration Estimates presented in Table 5. for WFPC2 (as derived from the HST Archive) and Table 6 for NICMOS (information provided by J. Mackenty). June 1995 STIS Instrument Science Report 95-002 Page 6 We assume that each individual primary exposure less than 5 minutes in duration is a separate pointing and will require STIS to read out. While this appears an extreme assumption, it is nevertheless reasonable because short integrations are typically either of bright sources or of planetary objects which are moving on short timescales. Primary observations of bright sources (times less that 5 minutes) will probably be dithered for both WFPC2 and NICMOS to assure flat field uniformity and/or allow for hyper resolution upon reconstruction. For integration times between 5 minutes and 20 minutes, where the sources are fainter, dithering may or may not be employed. Thus, for the longer duration exposures it is possible that a series of prime science exposures may be taken without moving the telescope. Note that presently, HST observes on average 6 targets per day, or equivalently, remains on a single source for roughly 2 orbits. Since much of the STIS parallel science probably requires multiple orbits per target to achieve the needed signal to noise, this suggests that STIS parallels will be scientifically useful. Parallel Volume Calculation If we disable STIS from taking exposures in parallel of duration less that 5 minutes, we will still be able to take parallel observations 90% of the time that WFPC2 is exposing as prime, and 60% of the time NICMOS is exposing as prime (see Table 5 and Table 6). The total volume of STIS data taken in parallel will then be roughly 470 Megabits. This data will be contained within 30 exposures, corresponding to 20 datasets as passed through OPUS into the archive. This calculation was done assuming • 4 exposures per orbit, (this is a conservative estimate to maximize volume, based on the fact that all CCD STIS parallels will be crsplit, allowing for spacecraft motion every 10 minutes for dithering, and allowing for multiple filters or gratings. A reasonable, less conservative and feasible estimate would be 2 exposures per orbit, cutting numbers in half.) • all STIS parallel exposures produce 16 Mbit 1024 x 1024 images (either CCD or binned MAMA exposures) • WFPC2 is prime for 5 orbits of which STIS is used in parallel 0.90*5 = 4.5orbits • NICMOS is prime for 5 orbits of which STIS is used in parallel 0.60 * 5 = 3 orbits If we allowed STIS to take parallel exposures of shorter duration (down to minute timescales) then this volume could go quite a bit higher. For instance, in the 10% of the time WFPC2 is taking short exposures (~0.5 orbits), and the 40% of time NICMOS is taking short exposures (~2 orbits), STIS could obtain 40 binned 1024 x 1024 exposures per orbit, leading to an additional volume per day of 1600 Megabits, or a total parallel STIS volume of 2070 Megabits per day! This data will be contained within 130 exposures, corresponding to roughly 100 datasets passing through OPUS into the archive. Note that short MAMA STIS exposures are in principal feasible and scientifically useful so long as the June 1995 STIS Instrument Science Report 95-002 Page 7 total integration time on one piece of the sky is large (~ 1 orbit). This is true even for observations of faint sources, because there is no noise associated with the reads and any spacecraft dithering between exposures of the prime can be removed on the ground subsequently. In practice, however, a very large effort would be involved in working with this data, the data volume costs are very high, and the scientific cost of disallowing such short exposures is minimal. For the purposes of all the volume calculations presented in this memo we assume that short (<5 minutes) parallel exposures with STIS are prohibited. Table 5: WFPC2 Exposure Duration Estimates Exposure Lenth (minutes) Number of Exposures Total Time (ksecs) Overhead corrected time (ksecs) fraction of time employed <1 1608 29 116 3% 1-2 400 30 75 3% 2-3 118 17.5 35 1% 3-4 216 44.5 78 3% 4-5 36 9.4 15 2% 5-10 765 304.3 396 10% 10-15 450 309.6 371 9% 15-20 770 756.9 870 33% >20 1016 1950 1950 50% Table 6: NICMOS Exposure Duration Estimates Exposure Duration (minutes) Fraction of Time Employed <1 40% ~5 25% ~10 35% Calibration Program Volume We estimate that the STIS calibration program will require us to take an internal unpointed exposure every orbit (e.g., in occulted time). We expect the calibration volume to be dominated by MAMA calibrations, because (1) there are 2 MAMA detectors for the single CCD detector, (2) two thirds of the STIS science is expected to be with the MAMAs, and (3) long total integration times will be required to build up sufficient signal to noise on calibrations due to the dynamic range limitations of the MAMAs. Thus we estimate a calibration program volume of one 64 Mbit exposure (dataset) per orbit. We can compare this estimate to the current situation for WFPC2. Roughly 1/3 of all WFPC2 exposures in the HST archive are internal calibration exposures, and WFPC2 currently takes roughly 25 June 1995 STIS Instrument Science Report 95-002 Page 8 exposures per day as prime instrument, suggesting that WFPC2 is taking roughly 1 unpointed internal calibration exposure per every 2 orbits. 3. Calibration in the OPUS Pipeline: Archive Volume and CPU Time Estimates The OPUS pipeline generically converts the science data and passes the data through the calstis calibration routine. This creates an increase in the data volume from transmitted to archived data. We can consider calstis (the calibration program for STIS data) to be effectively broken into two major parts, calstis-A which performs the 2-D Image Reduction and outputs a flat fielded image, and calstis-B, which performs the spectroscopic reduction and geometric correction and outputs the final spectroscopically and flux calibrated products. The vast majority of external science data passes through both calstis-A and calstis-B, however non-pointed calibration data taken under calibration programs will only pass through calstis-A. This means that the volume increase and CPU estimates will be different for non-pointed calibration and external science data. To summarize our results in advance • we conclude that pointed science data will undergo a volume increase by a factor of 14 from transmitted to archived bits, while non-pointed calibration data taken under calibration programs will undergo a volume increase by a factor of 9. The CAL class archived volume will be 12 times the downlink volume for science data and 7 times the downlink volume for non-pointed calibration data. • we derive estimate of the number of equivalent calwfpc2 CPUs (i.e., the total time to process STIS data through the OPUS pipeline relative to the time to process a wfpc2 dataset through the pipeline today). The calstis CPU time estimates range from 12.5 calwpc2 CPUs to process each MAMA full frame exposure, to 4 calwpc2 CPUS per CCD dataset, to 2 calwfpc2 CPUs for subarrays and non-pointed calibration exposures. Combining these estimates with the observing scenarios presented in Table 2, we conclude that roughly 500 calwfpc2 CPUs will be needed per day to process STIS data. Archive Volume Calculation Below we estimate the data volume increase from STIS downlinked science bits to archived science bits. To determine the data volume increase in OPUS, we need to consider two steps: (1) generic conversion and (2) calibration by calstis. Generic conversion receives the POD file containing the transmitted bits. From this it creates one copy of equal size (the EDT file set) and one raw CAL class dataset. For each POD bit, the raw CAL class dataset will have 2 bits. This is becase each CAL class science image will be a triplet; a science image, a data quality image, and a sigma image. We assume here that the raw sigma image will be just a header (place holder with the equation written in the header for deriving the sigma image. We further assume that the data quality June 1995 STIS Instrument Science Report 95-002 Page 9 file is fully populated and is a 16 bit per pixel image. In the case where the data quality image is blank (when no data quality bits are set for the raw data), this image will in fact be a zero size: thus we are again making a conservative estimate (maximizing expected volume).. With this assumptions, generic conversion will produce a 4-fold increase in data volume. Calstis will produce as intermediate or calibrated products, 32 bit per pixel science and sigma images and 16 bit per pixel data quality images. Thus each output calibration product is 5 times bigger than the transmitted science data. Calstis-A will produce one calibrated image product (the flat fielded image) for each science image input. All STIS data will be processed through Calstis-A. Calstis-B will produce a single calibrated image product for each cr-split CCD dataset and a single calibrated image for each input exposure MAMA image and each rapid-readout CCD exposure. Since the exposure and bit volumes are dominated by MAMA and rapid read CCD and MAMA exposures, for the purposes of roughly estimating the volume increase in OPUS it suffices to assume all science dataset experience a 10-fold increase in volume while passing through calibration. On the other hand, internal calibration data taken under calibration programs will be passed only through calstis-A and experience only a 5 fold increase in volume in calstis. CPU Requirements for Calibration of STIS data in OPUS Below we estimate the typical cpu time, in units of calwpc2, needed to calibrate STIS data in the pipeline. Since we have not yet built the pipeline, nor do we know the distribution of observing sequences with any accuracies, these are rough estimates. Adequate memory, to allow storage of multiple 2048 x 2048 images in memory at one time, will be crucial to the performance of calstis. There are at least three steps in the calibration of STIS images that are likely to be CPU intensive. These are geometric correction, convolving reference images with the Doppler smearing function, and combining CRSPLIT observations. Extraction of 2-D spectra are included as part of the geometric correction. Extraction of 1-D spectra from the 2-D spectra should be a small fraction of the total. Other steps should take significantly less time. The geometric correction would be applied to direct images to remove optical distortion from the OTA. For spectrographic images this step would remove distortion due to the spectrograph, and the output would have a linear wavelength scale. For time estimates the CALFOC algorithm (newgeom) and the IRAF task geotran were considered. The time required scales with the image area. When the dispersion is sufficiently high (i.e. echelle mode), the on-board software adjusts the pixel number of detected photons to account for the Doppler shift during the exposure. This means that the counts at a given image pixel came from various locations on the photocathode, so the appropriate flat field value to use for that pixel should include June 1995 STIS Instrument Science Report 95-002 Page 10 contributions from all those regions of the photocathode. If the darks show repeatable fine structure, then they will also need to be Doppler smeared before dark correction for the high dispersion modes. The size of the convolution kernel depends on the dispersion, the location of the target, and the locations of HST in its orbit during the exposure. For the high resolution band 1 echelle with 2048 pixels image width, the convolution kernel can be as large as 23x3 pixels. The time scales with the product of the image area and the width of the convolution kernel (assuming that the height of the kernel is fixed).The time required to combine images (CRSPLIT) using gcombine scales with the number of images to be combined. Timing tests were done on a Sun IPX. Calwpc2 took 2.5 minutes to run on a Sun IPX, so this time was used to normalize the times of the other tests. In other words, calwp2 is used below as the unit of time. For the values shown below, a 1024x1024 image was used, so times must be scaled depending on the input image size. Table 7: CPU estimates for Steps in calstis for 1024 x 1024 images calibration step time (calwpc2=1) geometric correction 2.0 Doppler convolution 0.32 combine four crsplit 0.50 combine two crsplit 0.23 Science data taken in different modes and different sequences will pass through a different series of calibration steps and the typical image sizes will also be different (affecting the CPU time estimates). In Table 8 below we give several examples. Table 8. CPU estimates By Mode, in units of calwpc2 observing mode image size calibration steps time in steps Total expected CPU time CCD crsplit=3 1024x1024 geo; crsplit=3 2.4 3.6 MAMA Echelle, single exposure 2048x2048 geo; Doppler 8.3 12.5 MAMA first order single exposure 2048x2048 geo 8.0 12.0 For non-pointed calibration datasets which pass only through calstis-A, we assume that the scaling time from wfpc2 to process is merely the image size: under that assumption each internal calibration (2048 x 2048) dataset requires roughly 2 calwpfc2 CPUs. We estimate the overall calwpfc2 CPUs per day needed to process STIS data in OPUS to be ~500 calwpcs CPUs (see Table 1)as follows. June 1995 STIS Instrument Science Report 95-002 Page 11 • For MAMA full frame highres science, we assume 12.5 calwfpc2 CPUs per exposure and from Table 2, we expect 15 such exposures per day (=188 calwpfc2 CPUs/ day). • For MAMA full frame lowres science, we assume 4 calwpc2 CPUs per exposure and we expect 40 such exposures per day (=160 calwpc2 CPUs/day). • For MAMA and CCD subarrays we assume 2 (calwpc2 CPUs per exposure and we expect 8 such exposures per day (=16 calwpc2 CPUs/day). • For typical CCD observations we assume 4 calwpc2 CPUs per dataset and we expect 10 such datasets (equivalent to 20 exposure) per day (=40 calwpc2 CPUs/ day). • For rapid read CCD observations we assume 3 calwpc2 CPUs per exposure and we expect 5 such exposures per day (=15 calwpc2 CPUS/day). • For Non-Pointed calibration exposures we assume 2 calwpc2 CPUs per exposure and we expect 15 such exposures per day (=30 calwpc2 CPUs/day) Note that there are major uncertainties in this estimate, including (1) the distribution of observation modes actually used, (2) the algorithm that will be used for geometric correction, (3) the complexity of the spectral extraction, (4) whether convolution for Doppler smearing should be applied to dark images as well as to flat fields, (5) how much additional time is needed to check the data quality flags and compute error estimates, especially for geometric correction, (6) how much processing is required for wavecals, and (7) the effect of using C instead of Fortran. Note in addition that the CPU time scales with image size; single small subarrays will process quickly. 4. On-line OPUS Calibration File Requirements From an estimate of the types, sizes, and selection dependencies of calibration reference files (details below), we estimate that OPUS will need (at a minimum) room to hold roughly 46 1024x1024 images and 59 2048 x 2084 files, for a total of 2.8 Gigabytes of disk space, over any 3 month period. This is a rough estimate. Justification In Table 9 below we list the calibration images needed by calstis, their sizes and an estimate of the rate at which they may be expected to change. The number listed is the total number of calibration science data images. Each data file will also have an accompanying data quality and error file of like dimension. The data quality file will be 16 bit rather than 32 bit; the science and error file will be 32 bit. Thus, a 2048 x 2048 calibration file has a size of 40 Megabytes and a 1024 x 1024 calibration file has a size of 10 Megabytes. June 1995 STIS Instrument Science Report 95-002 Page 12 Table 9: STIS Calibration File Requirements Name Selection Criteria Number Size Change rate bias CCD only, amplifier 4 1024x1024 bi-yearly shutter shading CCD only, amplifier 4 1024x1024 yearly dark detector, MAMA detector, CCD 2 4 2048x2048 1024x1024 monthly flat field observation mode, MAMA observation mode, CCD 51 25 2048x2048 1024x1024 bi-yearly spatial distortion detector, MAMA detector, CCD, amplifier 2 1 2048x2048 1024x1024 yearly The rate at which the calibration files will change with time is highly uncertain, as it will depend sensitively on the way in which the detectors and components respond to the radiation environment in space, however these estimates should be adequate for disk space resource determination. They are also consistent with the resources we are likely to have to analyse calibration data nad prepare and install new reference files. We also note that during and following SMOV a rapid rate of change of calibration images should be expected: the change rate numbers given are those expected for steady state. The volume estimates are dominated by the flat field images. It is possible that for ease of calibration, we will choose to divide the flat field calibration into a `low frequency’ component which is dependent on observation mode, and a `high frequency’ component (giving the pixel to pixel variations) which is dependent only on detector and broad spectral ranges. There would then be a much smaller number of high frequency flat field images, and a number of tables which modelled the low frequency flats. For the purposes of estimating disk volume capacity, however, it is safest to assume that there will be 76 flat field images. Here the 76 number is derived from an estimate of the number of STIS unique grating scan positions to just cover the spectral range. There will also be a large number of calibration tables. These should be small in size, and so should not affect the above volume estimates. 5. APPENDIX A. Science Uses of STIS as Prime Instrument In this Appendix, we describe how we expect STIS to be utilized for Prime Science. This provides the justfication for the assumptions presented in Section 2 which were used to calculate the ground system and operational requirements (as presented in Table 1 and Table 2). UV Versus Optical Science with STIS The majority of STIS science will most likely be carried out using the MAMAs as opposed to the CCDs. Looking at the uses of the existing HST spectrographs, one finds June 1995 STIS Instrument Science Report 95-002 Page 13 that roughly 1/3 of FOS observations are in the optical (would be conducted with the CCDs) and 2/3 in the UV (would utilize STIS MAMAs). Folding in also the GHRS observations (all UV and roughly equivalent in number to the FOS) would suggest that the bulk of the STIS observations would be carried out with the STIS MAMAs. Note however that the STIS CCDs are appreciably more sensitive in the optical than the existing HST spectrographs, which may shift the balance somewhat. The 2:1 ratio of MAMA to CCD use expected is also reflected in the existing rough scenarios for potential STIS GTO programs (see Table 10, below). Typical STIS Science: ‘Faint’ Source UV and Optical Observing The STIS IDTs have tentatively identified a number of major science goals and provided a rough view of the typical exposure lengths for a subset of these projects (Steve Kraemer, June 1994). This information is summarized in Table 10. Note that each spectroscopic observation of a new target will typically be preceded by a target acquisition and each change of grating element by a wavecal. Table 10: .Subset of Potential STIS IDT Observations (condensed from Kraemer Science Topic Detector Minutes per Exposure Number of Exposures Spatiall Resolved Spectra of AGN CCD 30 10 Echelle Spectroscopy of Quasars MAMA 1 MAMA 2 40 40 7 7 Trailed AH Hercules Stellar Spectrosocpy MAMA 1 30 4 QSO Off-Nuclear Spectra MAMA 1 MAMA 2 20 20 4 4 OB Stellar Winds in LMC MAMA 1 30 5 Emission Lines in SNR Filaments MAMA1 MAMA2 CCD 30 30 15 1 1 4 Long Slit Spectra of BL-Lacs MAMA 1 CCD 45 25 1 4 Spectra of Radio Source Counterparts MAMA 1 CCD 45 20 1 2 UV Spectroscopy of Faint Emission Line Galaxies MAMA 1 MAMA 2 45 45 1 1 Objective Priism Survey (white dwards/O stars/AGN) MAMA 1 MAMA 2 45 45 1 1 UV Imaging of AGN MAMA 1 45 1 1994) June 1995 STIS Instrument Science Report 95-002 Page 14 From the information presented in Table 10, one can conclude that a `typical’ STIS UV orbit will consist of 1 or 2 exposures of 20-45 minutes length. Such long uv exposure times are not unexpected. The MAMAs are photon counting detectors, and we expect them to predominantly be utilized to take relatively long exposures, since only with long integration times can high signal to noise be achieved while maintaining count rates in the linear regime. Conservatively then, we conclude that a typical STIS UV orbit will contain, 2 science exposures. As many of the STIS first order and some of the echelle gratings require ‘scanning’, we estimate 2 MSM positions per orbit for such a typical exposure, with 1 accompanying wavecal for each MSM position. We assume only a single slit position is used per orbit. Including the target acquisition, we thus predict 1 target acquisition dataset (a single exposure with 3 small images), and 2 science datasets (each with a science exposure and a wavecal). We assume that all such MAMA exposures are full frame, unbinned 2018 x 2018 images each with a volume of 62 Megabits. This is a reasonable if conservative assumption. It is true that first order grating MAMA observations of faint stars are likely to be read out as subarrays (since there is no loss of science in doing so) and so generate much less volume. However examination of the FOS and GHRS observations in the HST archive indicates that stellar observations at R ~ 10,000 or less (which would utilize the first order gratings for equivalent STIS observations) make up only ~5% of all uv observations taken with HST. Likewise, from the information presented in Table 10, we conclude (conservatively) that a typical STIS Optical CCD orbit will consist of 2 sets of crsplit exposures at different MSM positions but a single slit position, with an accompanying wavecal for each MSM position. This generates a single target acquisition dataset per orbit, and 2 science datasets per orbit, each with 2 science exposures and a wavecal exposure, and each having a volume of 32 Mbits. Exceptions: Instances where a large exposure/data volume is generated As described above, it is anticipated that the vast majority of STIS science will utilize fairly long exposure (> 15 minutes), generating small numbers of exposures/datasets per orbit and moderate data volumes. However, several scenarios can be envisioned where large numbers of exposures and in some instances large data volumes are created in short amounts of time. Below we describe the scientific scenarios in which this may occur. Note two things (1) the overall use of rapid reads (< 5 minutes) is expected to be small, occuring roughly 6% of the time, (2) in all cases the large number of exposures generated will be combined into a single dataset by OPUS for passage to the archive and to data quality evaluation. The scenarios described below in which we envision large numbers of exposres/data volume being generated quickly include: June 1995 STIS Instrument Science Report 95-002 Page 15 • Rapid Readouts for Time Resolved spectroscopy • Very Bright Object Observing • Fpslit observations. Repeatobs Mode for Time Resolved Spectroscopy For time resolved spectroscopy/imaging observers can use the `REPEATOBS’ (or nexp=many) option which will cause a series of identical exposures to be taken. Since there is a roughly 5 second interval between exposures, scientifically this mode would be used to monitor variability on minute timescales, generating typically a single exposure per minute. Monitoring of variability over the course of an orbit will thus generate roughly 50 exposures. OPUS will package these many associated exposures into a single datatset. We can expect that roughly half of the variability monitoring will be done on stellar objects with first order gratings, in which case subarrays can be used to limit the data volume without sacrificing science. In those cases, we assume a readout volume which fills the data buffer in a single orbit, or roughly 128 Megabits. The other half of the variability monitoring will likely be done on planets or in echelle format, in which case it will be desirable to readout the full array. In that case, the volume is limited by the rate at which data can be transferred from internal memory to the tape recorders (=1 full frame MAMA exposures every 60 seconds, or 1 full CCD exposure every 16 seconds) and the readout time for the CCD (27 seconds for full frame). We thus assume that we are reading out 1024 x 1024 arrays every 30 seconds in this mode, for both uv and optical observations. Thus a single orbit of CCD full frame or MAMA lowres full frame time resolved spectroscopy can then accumulate roughly 100 exposures or 1.7 Gbits within a single orbit. Such orbits will include a single target acquisition and a single wavecal exposure. One MSM position and one slit wheel position will be utilized for the science. Based on FOS and GHRS experience as derived from inspection of the HST Archive, time resolved spectrosocopy/imaging will constitute only a small fraction (<5%) of the GO observing programs,overall. Roughly half of these programs will probably be monitoring variability on second timescales and will be carried out in time tag mode with the MAMAs and PSD mode with the CCDs. Time tag and PSD mode will each generate just a single exposure (and dataset). These datasets can themselves be very large, with a gigabit produced in roughly 1000 seconds (data volume in fact limits the length of these exposures). For the purposes of the volume calculations we assume 6% of STIS science will be done in rapid readout mode for purposes of time resolved spectroscopy and very bright target observations (see below), inclusive. June 1995 STIS Instrument Science Report 95-002 Page 16 Very Bright Target Observations For very bright targets, observers may wish to break a long total integration time into many short exposures. This might be done to avoid saturation or the 65,536 counts per memory limit for the CCD (for gains less than 4, the memory limit will be the dominant factor) and to avoid the 65,536 counts per element limit for the MAMAs. This is expected to be an issue only for the brightest stars (e.g., O stars of magnitude brighter than 5 would need to be broken into subexposures of length multiple seconds with the medium gratings for the MAMAs). To give this some context, stars brighter than 10th magnitude (ubvr) constitute only roughly 5% of all FOS observations. (Note that the bright object protection for STIS will disallow observations of the brightest stars, see ISR under preparation [Mark Clampin]). This use of multiple exposures should be self limiting in the sense that the brightest sources should require the shortest total integration times, though astrophysical situations of faint lines on bright sources do exist. Perhaps more importantly, for the MAMAs nonlinearity sets in at roughly 200 counts/pixel/second, and users will typically want to choose a configuration which keeps them near the linear regime. At those count rates the memory limit is reached after 5 minutes. FPSPLIT Mode. The FPSPLIT option is expected to be used rarely, only to achieve the highest signal to noise on the brightest sources. Currently for GHRS, FPSPLIT mode is used 10% of the time. With STIS, FPSPLIT is expected to be used less (perhaps 3% of the time) because of the high overheads involved. The 5 FPSPLIT exposures generated from a single proposal instruction line will be combined into a single dataset by OPUS. June 1995 STIS Instrument Science Report 95-002 Page 17