Change page style:
Science Operations Statistics
Gemini Observatory tracks completion rates for queue programs, open shutter efficiency, acquisition times and weather losses for both telescopes. In addition, the Observatory has investigated the distribution of the real observing conditions. This information may be used in the future to better match the queue filling to the expected distribution of the observing conditions. Information on RA distributions as well as the demand for the various GMOS gratings in semesters 2006B-2008A is also available. The following summarizes current status of these Science Operations Statistics. The information on weather loss and delivered science nights, as well as the completion rates for queue programs is updated regularly. The remainder of the information originates from specific requests from committees or from internal analysis of operations questions. This information will only be updated as time allows or as the need for updated analysis arises. Last update: February 2013.
- Weather loss and delivered science nights
- Completion rates for queue programs
- Acquisition times
- Open shutter efficiency
- The effect of program length
- Distribution of observing conditions
- User demand for Target-of-Opportunity programs
- GMOS gratings - demand and execution
- Non-sidereal observations
- New users
- Long-term RA distributions
Each semester at the time of the Call for Proposals, the Observatory with input from the Operations Working Group decides on the number of offered science nights. The science queue (and classical nights) are filled using this number of science nights. The actual delivered number of science nights can either be larger (if planned engineering or commissioning tasks did not happen) or smaller (if engineering or commissioning tasks take longer than planned or if unforeseen events happen). In addition, the weather loss for a given semester will affect the completion rates. Figure 1 below shows the use of available time.
Gemini North suffered an unusually high amount of weather loss in the period from late December 2008 to late April 2009. Further details can be found here.
Figure 1: Use of available time for Gemini North and South during semesters 2005A to 2012B. On average, scheduled science nights make up 87.6% of all nights. The long term (2005A-2012B) weather loss amounts to 23.5%.
The completion rates of queue programs are closely tracked and queue planning is carried out to optimize the completion rates. Full multi-instrument queue planning was put in place at Gemini North starting in semester 2005A, with Gemini South following in semester 2005B. This change combined with better reliability of instruments and telescopes has led to a significant improvement in the completion rates across all scientific ranking bands. Further, in semester 2004A the sizes of the ranking bands were changed from equal size to roughly 20%, 30%, and 50% for band 1, 2, and 3, respectively. At the same time the national TACs were given the option of granting band 1 programs rollover status such that it would be active in the queue for a total of three semesters. Starting in 2007A, the ranking bands were adjusted to roughly 30%, 30%, and 40% for band 1, 2, and 3, respectively.
Figure 2 shows the queue completion rates for both sites. The completion rates for 2003A-2004B exclude ToO (Target of Opportunity) programs. For semesters 2005B and later, ToO programs are included in the completion rates. A ToO program's completion is defined as the executed time relative to the triggered time. If a ToO program has used all allocated time, it is counted as completed independent of the amount of triggered time. ToO programs that do not trigger any observations are excluded from the statistics. It should be kept in mind that there are still active band 1 programs with rollover status from the past two semesters, which can still reach 90-100% completion.
For band 3 programs the queue planning aims at getting a
significant amount of data for the programs that are in fact started,
rather than getting a little bit of data for many programs. The effect of this planning can be seen in Figure 2. The fraction of started band 3 programs that get at least 75% of the requested data is typically 70%.
Figure 2: Summary of the completion rates over a range of semesters from 2002B to 2012B and for both sites, as of Feb 8, 2013. Semesters 2012A and 2012B still have active programs with rollover status in band 1. Thus, the final completion rate for band 1 is expected to be higher than shown on this figure.
1 Band 1: 90% completion rate after rollover period
2 Band 2: 75% completion rate
3 Band 2: 85% of started programs should have 75% of data taken
4A Band 3: 80% of started programs should have 75% of PI defined minimum data taken
1 Band 1: 100% completion rate after rollover period
2 Band 2: 90% completion rate
3 Band 2: 100% of started programs should have 75% of data taken
4B Band 3: 80% of started programs should have 100% of PI defined minimum data taken
5 Band 3: 80% of started programs should have 75% of all data taken
Table 1: Completion rate requirements and goals, as endorsed by the Gemini Board and the Operations Working Group.
The acquisition times are tracked from the records in the observing database.
From an earlier study of this, it is known that the median time to slew and acquire
a guide star for a new target is about 6 minutes. The exact time of course depends
strongly on the length of the slew. In the following, this time is excluded from the
statistics. Figure 3 and tables 2 and 3 below summarize the acquisition times for
all the spectroscopic modes, and for the only imaging mode (NIRI+Altair) that has
significant acquisition time above the slew and guide star acquisition.
For the spectroscopic modes, the measured acquisition time is the time it takes to
image the target and align it in the spectroscopic aperture (slit, IFU or MOS mask).
For NIRI+Altair the measured acquisition time is the time it takes to center the
target on the NIRI array.
Comparison of the 2005B+2006A times, when the acquisition procedures for the various instruments were not homogeneous and not integrated with the rest of the software, and the 2008A times, after the procedures had been integrated, shows a marked improvement. The integrated acquisition software saves 3 minutes per acquisition, which adds up to 3 nights per semester per site.
Figure 3: Acquisition times for spectroscopic and imaging observing modes done in queue time for Gemini North and South. In all cases the improvement from semesters 2005B/2006A to 2008A due to the integrated acquisition procedures is clearly visible.
Table 2: Summary of the number of acquisitions as well as the acquisition times for spectroscopic modes and NIRI/Altair imaging, showing the improvement from semester 2005B/2006A to 2008A.
Table 3: Acquisition statistics for semester 2008A, showing a more detailed breakdown by instrument and observing mode.
The open shutter efficiency for Gemini instruments has been tracked since August 2004. For each night the open shutter time is extracted from the FITS headers of the obtained observations. Open shutter efficiency is defined as the sum of all science and calibration exposures obtained between evening and morning twilight, divided by the usable time available (hours between twilight less time lost to weather or technical faults).
Figure 4 summarizes the open shutter efficiency for two epochs, August 2004 - Feb 16, 2006, and semester 2008A. GMOS-N used to have about 5% higher open shutter efficiency than GMOS-S, which may have been due to the effect of consistent queue planning of all GMOS-N nights since its commissioning. However, in 2008A these differences are no longer present. There is a noticeable improvement in efficiency for the most frequently used instrument combinations. It can also be seen that multi-instrument night efficiency is a function of the mixture of demand for different instruments.
Figure 4: Open shutter efficiency for facility instruments at Gemini North and South for 2004-2005 and for semester 2008A. Open shutter efficiency is derived as the fraction of the usable time during the night, less any loss due to weather or technical faults.
An investigation was carried out to better understand how the requested program length affects the probability of a program getting the requested data. Only semesters 2005A-2009B were included in this investigation since these semesters closely reflect the current queue planning principles and methods, while this is not the case for the earlier semesters.
Figure 5: Comparison of the distribution of program lengths for scheduled programs (gray) and programs that got at least 75% of the requested data (orange). For band 1 and 2 the distributions are identical. Thus, the program length for these two ranking bands has no influence on whether the program gets data. For band 3 there is a tendency that more of the shorter programs get executed. This is as expected since the queue planning focuses on selecting programs from band 3 that have a fair chance of getting a significant faction of the requested data. The median program length in band 3 for scheduled programs is 9 hours, while the median program length for those that got at least 75% of the requested data is 6 hours.
In order to simplify the investigation of the effect of the observing conditions, six broad bins of observing conditions were defined as shown in table 4, based on the percentile bins used by the PIs to specify the required observing conditions.
Table 4: Observing condition bins, based on the percentile bins used by PIs to specify their required observing conditions.
All the science observations in the observing database for a number of semesters were mapped onto these six broad observing condition bins.
This was done both for the planned science observations and the executed
science observations. Figure 6 shows the distribution by bin.
Semesters 07B+08A are shown for both sites, and compared to the 18 weeks of winter at Gemini North in 2008/2009 - a prolonged period of poor weather, on which more details can be found here. The proportions predicted by the model currently used for filling queue time are plotted as well.
Figure 6: Fraction of useful time, sorted by observing condition bin, for three different time periods. The model fraction is shown in gray. The red bars correspond to the prolonged period of poor weather in winter 2008/09.
Figure 7 shows the percentage charged for Gemini's Target of Opportunity (ToO) programs, sorted by band. It is readily apparent that we are using 20-25% of band 1 time on ToOs. ToOs come in two varieties:
- Rapid ToOs: response times vary from a few minutes up to 24 hours. In 2010A, there were 10 triggers per month per site.
- Standard ToOs: response times greater than 24 hours. in 2010A, there were 20 triggers per month per site.
Figure 7: Fraction of time charged for Target of Opportunity programs, by band.
Because the two GMOSs can carry only three gratings simultaneously, it is of special interest to PIs with programs in band 3 to know which gratings are in highest demand and therefore will most often be in the instruments. Figure 8 shows this information for semesters 2006B-2008A.
Figure 8: GMOS gratings: Total planned time in science observations as well as the executed time are shown for the six gratings for GMOS-N and GMOS-S.
Figure 9 shows the fraction of time allocated to non-sidereal programs, from 2010A to 2011B. On average, Gemini North has about 9% of its time allocated to non-sidereal programs, with an even higher fraction in 2011B. Gemini South currently features less non-sidereal programs. The reason for this discrepancy is currently not well understood, although it should be noted that Gemini North's host, the University of Hawaii, is involved in many non-sidereal programs.
Figure 9: The fraction of time allocated to non-sidereal programs.
Table 5 presents the number of new Gemini users, where a "new" user is defined as a PI name with no programs on Gemini in any prior semester (starting with 2000). The numbers are relatively stable, with on average 50 new PIs each semester, representing 21.9% of all Gemini PIs.
Table 5: The number of new Gemini PIs over the past few semesters.
Last update: March 22, 2012 by Michael Hoenig