MegaPrime/MegaCam Observing Statistics

Table of contents:

1. QSO validation, overheads, and weather

The bad weather over Mauna Kea has significantly affected the scientific operations of the MegaPrime imager over the past three years (since first light) during the winter season. The first year of operation (2003) also brought its fair share of engineering time and technical downtime while the entire imager (MegaPrime + MegaCam) was being tuned.

After three years of scientific operation under the QSO mode, it clearly appears that the bad weather is indeed the source of the observing efficiency issue. The following plot and table aim at providing a quick overview of the data gathering efficiency since the beginning of semester 2003B (the official beginning of science gathering with a stable instrument).

The CFHT Legacy Survey represents a significant fraction of the MegaPrime observing time (typically 55 to 60%). It is naturally pointed out here individually versus the entire share of observing time (the PI programs).

The statistics presented here are provided by the QSO team. A different representation (focused on a per semester basis) is proposed on the QSO site.

1.1 Graphical representation on a semester basis :

Description :
The bright green fraction of the plot represents the QSO validation level ( = exposure time + 40 seconds of overheads charge) for each observing run (hours per night). This is the fraction of time that produces data within the PH2 specifications. The CFHTLS fraction is represented with the dashed dark green line. The white section represents the fraction of time per night (hours per night) spent on data that do not get validated either because the 1) image quality, or/and 2) the sky background, or/and 3) the atmospheric transmission are out of specifications (mostly this is due to periods of variable seeing when it is hard to predict which scientific program to pick, resulting in an increase of unvalidated exposures). The orange area (hours per night) represents the average time lost per night because of the weather. The red area (hours per night) represents the combination of 1) observing overheads (readout time, filter change, telescope+dome moves,...), and 2) technical downtime (when either the instrument or the telescope are down), and 3) the engineering time (which indeed does not produce science). At last, the yellow line represents the average length of an observing night (hours). The semesters are marked by a white vertical dashed line.

Discussion :
Weather: the winter seasons clearly appear as the large orange blocks. The semesters "A" have been repeatedly doomed. Overheads, engineering, and downtime: clearly major progresses have been achieved over the first two years, with a clear decrease towards the end of semester 05A when the autofocus (in particular) became available. The observing efficiency has continuously increased since first light, and when pristine weather conditions meet with a flawless instrumental behavior, one reaches high peaks such as during the 05Bm05 run which brought nearly 8 hours of validation per night (with some nights peaking above 10 hours of validation per night!). Regarding the CFHTLS vs. PI programs, as the QSO paradigm dictates it, the effect of the bad weather is equally shared: the CFHTLS validation follows closely the PI programs validation.

1.2 Statistics per semester :

The following table proposes the overall statistics since mid-2003 (purple), and the individual statistics on a per semester basis: "A" semesters (winter/spring) appear in blue, the "B" semester (summer/fall) appear in green.

PeriodRunsNightsValidationCFHTLSNGVSUnvalidatedWeatherOverheadsNight Length
All 127 14.9 4.7 24% 6% 0.9 2.6 20% 9.9
03B 6 16.8 3.7 43% 0% 0.7 2.3 47% 10.5
04A 6 18.7 2.8 39% 0% 0.4 2.6 35% 9.4
04B 6 19.0 4.8 46% 0% 0.8 1.7 38% 10.5
05A 7 16.4 4.2 52% 0% 0.8 2.3 30% 9.3
05B 6 16.3 5.7 56% 0% 0.8 2.1 23% 10.5
06A 6 15.8 3.5 57% 0% 0.7 4.0 18% 9.4
06B 6 16.8 6.4 46% 0% 1.0 1.8 16% 10.5
07A 6 14.3 5.1 50% 0% 0.8 2.3 17% 9.4
07B 6 14.2 4.9 53% 0% 1.0 3.7 14% 10.5
08A 5 15.8 5.3 45% 0% 0.8 2.7 12% 9.5
08B 7 13.0 5.5 24% 4% 1.0 2.9 12% 10.2
09A 6 15.8 4.0 0% 29% 1.1 3.3 17% 9.4
09B 6 11.8 6.9 0% 6% 0.9 1.2 15% 10.5
10A 6 13.5 6.6 0% 29% 1.0 1.0 9% 9.4
10B 6 14.8 6.2 0% 8% 0.9 2.6 11% 10.5
11A 5 13.8 3.4 0% 32% 1.0 4.3 16% 9.5
11B 4 8.8 5.5 0% 10% 1.5 1.9 10% 9.7
12A 6 14.8 2.6 0% 30% 0.6 1.5 22% 9.4
12B 5 13.6 4.5 0% 0% 1.1 3.5 17% 10.4
13A 6 16.0 4.7 0% 0% 1.3 2.2 21% 9.8
13B 7 11.9 3.7 0% 0% 1.2 4.4 16% 10.2
14A 3 14.0 4.1 0% 0% 1.1 4.1 14% 10.1

Description :
The following columns apply for each given period:

Runs:number of runs
Nights: average number of nights per run
Validation: average QSO validation rate (hours per night)
CFHTLS: fraction of CFHTLS QSO validation
NGVS: fraction of NGVS QSO validation
Unvalidated: average QSO non-validated rate (hours per night)
Weather: average number of hours per night lost to weather
Overhead: fraction of overheads to observing (50% means equal time on each)
Night length: average night length (hours per night)

Discussion :
The validation rate for 05A is significantly higher than in 04A because the instrument behaved nicely and the impact of the decreased overheads was starting to show. 05B marks the first time MegaPrime/MegaCam was in cruising mode with reasonable weather statistics and a continuation of the instrument good behavior. The result was a validation rate higher than 5.5 hours per night over the semester. Regarding the reduction of the instrumental overheads and the increased reliability of the instrument, the improvements are clearly visible over the period 04B/05A/05B where they went from 34% to 25%, and then 19%, which is close to nominal considering the mode of operation of QSO. The semester 06A started with the worst weather conditions ever witnessed on Mauna Kea and the statistics for 06A appear already bleak. CFHT is still hopeful that at least 4 hours per night of validation for 06A will be reached by the end of July (this will be updated as 06A unfolds).

2. Image quality and seeing

The following diagrams were built using image quality measurements from all data collected through the first three years of QSO operation (March 2003 to February 2006). Only exposures longer than 60 seconds were considered in order to ensure that the natural seeing is well integrated. The data set contains only images taken through QSO, which started its scientific operation in March 2003: it does *not* include any engineering frames which were a dominant type in the first months of operation (see the data rate at the bottom of this page, specially in the r' band). The image quality per image is an average over the 10 central CCDs of the mosaic (Elixir statistics) such that the tunings of the wide field corrector do not "contaminate" this data set aimed at looking at possible atmosphere/dome seeing seasonal dependencies. In order not to bias the distribution towards better image quality, all exposures were included regardless of their QSO validation flag.

2.1 Image quality histogram per filter :

Description :
For each filter, the histogram gives the number of images obtained at the given image quality (all images over 60 seconds long, taken between Mar. 2003 and Feb. 2006). The total number of selected exposures is: u*= 2311 (3244), g'= 7232 (10008), r'= 7933 (10573), i'= 6978 (8684), z'= 3325 (4419) [The second number within the parenthesis is the total number of science and photometric calibration exposures acquired with the instrument in the given filter over that time period, they show that the statistics presented here are based on the majority of the data]. Each bar is 0.1 arcsec. wide and covers the range +/-0.5 of the given index (e.g. the 0.6" bar represents images with an image quality ranging from 0.56 to 0.65 arcsec.).

Discussion :
The "mode" seeing (the peak of the histogram) is: u*=1.0", g'=0.8", r'=0.8", i'=0.6", z'=0.6"; fairly standard values known for the CFHT atop Mauna Kea. The u* band being a lot more sensitive to atmospheric turbulence (and differential refraction also playing a non negligible role at high airmass), the diagram exhibits a wider spread of the image quality (histogram broader than in g' for example). The image quality in the i' and z' bands exhibit a similar behavior, and the same can be said for the g' and r' band. The average image quality (value integrated over the contiguous sections of the histogram with a number of occurrences higher than 10% of the peak value) is: u*=1.05", g'=0.88", r'=0.82", i'=0.78", z'=0.72". The following table summarizes the various numbers presented in this section:

Filter u* g' r' i' z'
Number of images 2311 7232 7933 6978 3325
Image quality (arcsec.) - Mode 1.00 0.80 0.80 0.60 0.60
Image quality (arcsec.) - Average 1.05 0.88 0.82 0.78 0.72
Difference to the r' band (average) +0.23 +0.06 +0.00 -0.04 -0.10

2.2 Average image quality per observing run :

Description :
The average image quality per filter is plotted for all the individual observing runs since first light. The idea is to outline time variations and/or correlations between filters in order to derive possible seasonal dependencies.

Discussion :
The first semester (03Am02 to 03Bm01) shows a global improvement of the image quality as the major tunings of the wide-field corrector took place during that period (to the point of affecting even IQ in the central area of the mosaic). Beyond that point, it is difficult to point out a clear trend one way or the other: the image quality is dominated by seeing variations from run to run, and no clear seasonal trend shows up. The u* band appears clearly a lot more sensitive to the atmosphere with regular surges above 1.2 arcsec. throughout the years. Note that bad weather (see the first plot of this page) tends to push the image quality values to some extremes (e.g. 05Am02).

2.3 Image quality versus airmass :

Description :
This diagram gives the average image quality for a given airmass domain:
  • 1.03 = 1.00 to 1.06 (00 to 20 deg. from zenith)
  • 1.11 = 1.06 to 1.15 (20 to 30 deg. from zenith)
  • 1.22 = 1.15 to 1.30 (30 to 40 deg. from zenith)
  • 1.40 = 1.30 to 1.50 (40 to 50 deg. from zenith)
  • 1.77 = 1.50 to 2.00 (50 to 60 deg. from zenith)
  • 2.50 = 2.00 to 3.00 (60 to 70 deg. from zenith)

Discussion :
As expected, the image quality in the u* band degrades very quickly above an airmass of 1.3 and shows sign of degradation as early as 1.15 airmass. the g', r', and i' filters remain fairly constant from 1.00 to 1.3, at which point they degrade at a fairly reasonable rate of approximately 0.2 arcsec per unity of airmass. Note that the irregularities in the i' data come from the time constraint of the CFHTLS SNLS which bring CFHT to sometimes observe the deep field on marginal conditions, creating a bias in the distribution. The z' band is fairly unaffected by the thickness of the atmosphere over the entire airmass range 1.0 to 3.0.

3. Sky transparency and brightness

3.1 Fraction of photometric conditions :

Description :
The fraction of exposures (all filters) taken under photometric conditions is plotted for all the individual observing runs since first light. The idea is to outline possible seasonal dependencies. Note that this is based exclusively on the light integration periods: the times when the Mauna Kea is weathered out (see first plot on this page) are not represented on this plot. The photometric flag is set conservatively by the QSO team for all exposure taken at a time the real-time sky transparency monitor SkyProbe did not show any sign of absorption (flat line) and the satellite IR map shows no cloud system moving above the Big Island (sometimes SkyProbe fails at detecting very faint uniform thin cirrus patches).

Discussion :
The sky transparency does not show any clear seasonal trend over the first three years. The average fraction of images obtained in photometric conditions is 75%.

3.2 Sky level :

Description :
The average sky background level (in ADU per second) is given for each filter for all the individual observing runs since first light. The idea is to outline possible seasonal, or solar cycles, dependencies.

Discussion :
There is a regular pattern in the i' and z' bands due to the fact that when the galactic plane is up in the sky (twice a year), CFHT seeks low galactic latitude fields by observing at higher airmass, hence causing higher background in both those filters. But overall, the sky background in all five filters has remained constant. The following table gives the average values in various units over these three years. Note that they do include all data taken a various airmass and Moon brightness, hence the average values are higher than the levels given in the technical specifications (zenith, and various levels of Moon brightness).

Filter u* g' r' i' z'
Sky level in ADU/sec 0.56 2.81 3.97 6.76 6.69
Sky level in mag / arcsec^2 22.2 21.7 20.8 20.0 19.1

4. Data rate

4.1 Image acquisition rate per night :

Description :
The average data rate per night since first light is given for each exposure type:
  • Bias: used to build the master bias (one per observing run).
  • Flat-fields: used to build the master flat-fields (one per filter per observing run).
  • Focus: used to derive the instrument focus by taking in out of focus images.
  • Objects: science, photometric calibration, snapshot, and engineering exposures.

Discussion :
The effect of bad weather during the past winters is clearly seen here. The most important feature though is the data rate increase when the autofocus became active in the middle of semester 05A: the number of focus sequences decreased from about seven night to none over the course of the next semesters. At 7 to 10 minutes per sequence, this freed a significant amount of time for science (about 50 minutes per night), and it shows on this diagram: the average number of frames acquired per night jumped from 68 to 80 (stats on 04B versus 05B), a 17% increase.

4.2 Accumulated data gathering :

Description :
This diagram shows the accumulation of the different frame types since first light:
  • u*,g',r',i',z': science, calibration, snapshot, and engineering frames in the given filter.
  • Other: other filters (CFH12K narrow band filters), open light, and pinhole mask frames.
  • Flats: flat-fields (all filters).
  • Bias: camera noise frames.

Discussion :
Interestingly, the pressure from the scientific communities on the various filters has been the same since first light, not so surprising since the CFHTLS with its clearly planned observing strategy accounts for about half of MegaPrime/MegaCam observing time. In three years of scientific operation, nearly 55,000 frames have been acquired, archived, processed and distributed. This represents approximately 40 terabytes of data (13 terabytes a year).