History

For several years, I have been exploring the feasibility of using a commercial CCD with a fisheye lens as a much cheaper alternative for detection of clouds in the moonless sky.  As well as being cheaper than the IR cameras, it offers higher resolution, no moving parts except the shutter and cooling fan, and no cryogens.   I supervised an engineering student, Mario Caceres, for several summers, who helped with this feasibility study and the presentation to ACTR in March 2000.  This clearly demonstrated the detection of clouds under starlight alone, and projected an extinction limit of a few percent with angular resolution of 0.18 degrees.   (Available in Powerpoint, on CD-ROM.)

On the basis of this presentation a backside illuminated 1Kx1K camera, the IMG1024S made by Finger Lakes Instruments, was approved for purchase.  It was discounted to $8.5K as a result of a clearance sale at Scientific Imaging Technologies (SITe).  A fast fisheye lens (Nikkor 8 mm, f/2.8) and data acquisition computer (Linux PC) were purchased with year end funds, at which time SOAR also elected to purchase a similar camera for Cerro Pachon as a site infrastructure contribution.  [To be more precise, CTIO bought the camera for SOAR with year end funds also, with a promise of reimbursement ...conditional on delivery.]

Various hemispherical windows were also ordered with year end funds, ranging from perspex domes from Edmund scientific to a custom optical quality MgF coated hemisphere.  The domes for large binnacle type compasses used in commercial shipping are one low cost source.  The one supplied by Ritchie Navigation for only $34, though polycarbonate and uncoated, is optically superior, and is often used for underwater cameras.   E&C Precision Optics have failed to produce an optical quality hemisphere, so the possibility of applying a hard AR coating to the polycarbonate window from Ritchie navigation will be investigated.   [2001 May 8: "Subal", an Austrian maker of underwater camera housings, claims to use optical quality glass domes which may be suiitable.]

The IMG1024S camera and fish eye lens have been operated in the lab and are about to be tested on the night sky.  I have not yet verified gain/noise/dark current performance.
 
 
 

Design Overview

Since the dark sky is almost invariant from night to night it is possible to subtract a reference frame formed from the median of previous (dark) nights so that a nominally flat image can be displayed at sufficient contrast to allow extinction and scattering to be perceived down to the limit imposed by the photon shot noise.   This has been demonstrated crudely by de-rotating clear frames taken on the same night.  Much better results will be obtained when the camera remains in a fixed position from night to night so that no de-rotation is needed.

Pixels can be binned together to improve the noise statistics, but only until the typical spatial scale of the clouds is reached.   Unfortunately a comparison of the angular scale of daytime Cirrus with that of the the moon or an outstretched thumb will quickly confirm that  the proposed resolution of 0.18 degrees (960 pixels across the sky) is not excessive, and that only slight binning can be used if at all.  This fine angular scale and the high winds found at altitude combine to require exposure times shorter than ~3 seconds to maintain acceptable contrast for Cirrus.

The unsurprising consequence is that camera sensitivity is crucial to detection of thin clouds.  To be specific, it was found that with fast optics (f/2.8), broad bandpass (no filter) and high quantum efficiency (backside illuminated CCD), one can detect extinction at about the 3% level.  Enough photons are collected so that moderate read noise can be tolerated.  Exposures short are enough that standard thermo-electric cooling is sufficient to keep dark current well below sky counts.

The smallest image circle for a fast fisheye lens (Nikkor 8mm, f/2.8, or Coastal Optics, 7.5 mm f2.8) is 22 mm requiring a moderately large CCD.  This is the primary cost driver since CCD prices scale roughly by area rather than by pixel count.   Fortunately the SITe 1024x1024 backside illuminated CCD with 24 um pixels was available at a bargain price ($8500 including camera electronics and software).  The 23 mm image circle for the Nikkor 8 mm maps into 960 pixels which can be displayed at full resolution with only very minor vertical truncation by most graphics cards.

Faster and shorter focal length lenses (e.g. GM23514C, 3.5 mm, f1.4), designed specifically for 1/2" CCDs, are available from Rodenstock Optics (Linos Photonics), but these are only quoted as providing 103.6° x 76.5° coverage.  These would be very attractive for a cheaper camera (smaller CCD) where horizon coverage was not required.  The faster lens would offset the loss of sensitivity due to the small CCD area.

Images will be acquired at 25 second intervals, flat fielded archived and scaled to 8 bits ready for distribution and display.  As a minimum, the most recent frames will be served up on the web, possibly binned 2x2.  An archive of previous images at full resolution could also be provided relatively easily.

A significant enhancement would be to provide animation the most recent hour or at 15 Hz frame rate.  Such smooth animation is needed to harnesses the eye's formidable pattern recognition capabilities which can extract correlated motions from the noise.  This considerably enhances the system's sensitivity to clouds and provides an intuitive way of showing weather patterns.  With a little practice, the eye can easily separate cloud and air glow .

Here are some sample Cloud Camera movies (*.avi, Windows Media Player), binned heavily to compress the movies to about 5 MB:

• direct
• with reference frame subtraction (cancellation is imperfect since de-rotated image was used as a reference instead of the median image from same sidereal time on previous nights)
• airglow (upon request ....to big to serve to web)

Comparison with ConCam

ConCam is a project at Michigan Technological University funded by NSF & NASA to monitor the whole sky continuously.  See www.concam.net  for  complete description.  They state that...

"Scientific objectives of the CONCAM project include the tracking of bright stars and highly variable phenomena such as novae, supernovae, optical counterparts to gamma-ray bursts. Astronomers might be interested in CONCAM images, however, for information they provide about weather and seeing conditions. CONCAM images will be uploaded soon after being acquired to a publicly accessible web page which can be inspected by astronomers in neighboring domes, astronomers attempting to observe from a remote location, or others generally interested in observing site conditions on any particular date and time."

We would be happy to build a CONCAM for the VLT.  We would ask you for about $10K to cover the cost of materials and ask that we not be charged internet transfer costs.  We ask in addition that all CONCAM data be made public domain as soon as it is taken.  We would then plan to transfer a copy of CONCAM.VLT data back to http://concam.net for display there under the link http://vlt.concam.net .

If you agree, we can ship you a CONCAM.  We will talk you though installation and set up http://vlt.concam.net from which VLT observers anywhere in the world can see the VLT sky.  You can mirror our software to make a local VLT mirror so that observers in the south can see the images with shorter download times.

By Products

• Light pollution monitoring:

Our light pollution measurements are made near zenith in clear weather.  The cloud camera will show how light light pollution varies over the whole sky weather and season and form year to year.   It can be used to monitor the variations in horizon glow under clear conditions, providing early warning of new trouble spots, and demonstrate light pollution in a form which is more emotive to the general public than spectroscopy.   The archive can also quantify the reduction of light pollution by low lying coastal clouds (i.e. how often and how much).
 
• Aircraft detection for Laser Guide Star safety:

Gemini are developing software in conjunction with Keck to to extract aircraft trajectory information.  The short exposures planned for the cloud camera are almost optimum for aircraft detection and the planned reference frame subtraction will eliminate source confusion.  The slaving of the cameras on Tololo and Pachon to Sidereal time will also ensure that exposures are synchronized.  This will allow full 3D trajectory information (position and velocity) to be extracted from a single exposure pair, maximizing the accuracy of flight path projections and minimizing latency.
 
• Air glow monitoring:

The cloud camera will build up records of how OH emission strength, time scales and spatial scales, and how these vary with time of day, season and solar cycle.
 
• Quantitative cloud cover statistics:

Estimates of percentage cloud cover each night could be logged automatically instead of relying on observer's nightly reports.
 
• Extinction and scattering measurements in clear sky:

Extinction varies even in the cloudless sky, for example when the  inversion layer rises in summer.  These variations will be evident in the cloud camera images.  It may be possible (someday) to calibrate the effects so that an absolute measure of extinction and scattering can be derived without need for a dedicated telescope.
 
• Outreach / publicity:

This camera will serve up all sky images of higher quality than are available elsewhere, promoting Tololo's reputation as an excellent observing site.   These could be used to draw public attention to our web links to long term weather and light pollution statistics.

STAGE  1  :  Basic deployment

• Construct weatherproof enclosure:

The design must provide complete weather protection while allowing air intake and exhaust.  At the same time the window and lens must be kept free from condensation.  My plan is to mount the camera at the top of a rectangular column (probably hardwood) through which intake air will be drawn upwards through an air filter to remove dust and water droplets.  To eliminate wind shake, the weather shield (topped by the hemispherical window) will cover but not contact this central column.  Warm air will be exhausted lower than the camera so that the weather shield (like an inverted cup) cannot fill with water to the level of the camera.   The exhaust air will pass between the weather shield and a concentric shell, then upwards and over the entrance window to aid demisting.   Additional demisting will be provided by waste heat from the power supply which will be located beside the lens in the dead air space just below the window.

The enclosure must be robust but does not require high precision and thus can be built by carpentry shop and a fiberglass contractor contractor in Coquimbo (or a one man week fill-in or overtime job in the instrument shop,).  Three days of drafting time are indicated below to make scale drawings.
 

• Construct reinforced concrete pillar on Tololo:                     ( Also required for ConCam.)

A 2 meter high obelisk on the Western side of the road near the 0.9m will provide a completely unobstructed view of the approaching weather to the South, West and North and minimize horizon obstruction to the West by aligning the YALO behind the 0.9m telescope and to the SW by aligning 4m and 1.5m.  The other telescopes hardly make it above the line of the Andes.  The height of the pillar is chosen to keep the camera slightly above flashlights, and curious visitors.   A greater ground clearance would be advantageous, but complicates window cleaning.
 
• Local archive and Web server:       ( Not supplied with ConCam.   .....Theirs are in Michigan)

The minimum software (scripting) requirement is to interface the camera to the Linux workstation using the Linux "Software Developers Kit" provided by the camera manufacturer,  to schedule the image acquisition over and appropriate range of times every day, then provide web access to the most recent image(s), and possibly to compressed images in the disk archive.   Images will have to be transferred to tape about quarterly.

STAGE  2  :  Enhanced visualization

• High speed animation:        (Not supplied with ConCam)

Experiments using the Linux freeware, ImageMagic, showed that the eye can extract patterns of motion from a noisy image very effectively, but only if replay is smooth and at sufficient frame rate.  ImageMagic delivers this performance by pre formatting images in memory prior to transfer to the display buffer.   It suffers from the drawbacks that loop length is limited by memory size, and to update (add new image delete oldest) it reloads all images from disk again.  Source code is available so it may be possible to overcome this second limitation.

In theory it is possible to pipe the data at sufficient speed from disk rather than RAM.  We will of course attempt to identify an existing product.  If this is not possible and a modification to ImageMagic proves too difficult, we could hold Mike Fitzpatrick's to his offer to write a simple front end to double buffer images into ximtool.   Rolando Cantarutti also has a test program written for the guide TV which is close to what we need.  The best solution is not clear so the estimate below is based on the scenario where the contract programmer double buffers data on disk into ximtool, with advice from Fitzpatrick and/or Cantarutti.

Software (a script) is needed to poll the Cloud Camera for data and  maintain N images in a video buffer.  More sophisticated synchronization than polling is possible but probably not necessary.
 

Resource Estimates

Roger Smith	0.5 mw
Contract Programmer	4 mw


 
• Reference frame subtraction:       (Not supplied with ConCam)

The brightness variations in the sky are too great to see the noise in darker regions without saturating the brighter areas.   This limits the visibility of clouds even with the very best animation.   However it is a simple matter to subtract out the known sky structure which is (nearly) invariant from night to night.  (Reference frame subtraction is only needed when the moon is down.)    The steps involved are
• Install NTP daemon in the cloud camera workstation, as we have done for other data acquisition systems, to guarantee accurate absolute time.
• Initiate exposures at the same Sidereal Time each night.
• During the day, build up an archive of clear dark sky frames by median filtering raw frames taken at the same Sidereal Time.  maintain a copy of the ephemeris to eliminate those frames occurring during twilight or moonlight.   This data base will accumulate automatically, and automatically track seasonal or solar cycle variations.   Then any difference between the current sky and the reference library represents weather or an event of interest.
• At night, select and subtract the reference frame.  Offering the astronomer either the flat fielded direct frames or the subtracted frames according to preference.
• Map to 8 bits using standard parameters.

 

Resource Estimates

Roger Smith	0.5 mw
Contract Programmer	4 mw

Occulting Ring :

Rather than attempting to track the moon, a simple and much more robust approach is to place a 3 meter diameter ring around the entrance window, inclined at 30 degrees from zenith so that it lies in the plane of the equator when the moon is at zero declination.  It is translated along the polar axis to accommodate the declination change of the moon (or sun).  It can either be made wider and be left in a fixed position all night or be driven during the night.  Automation is preferable since the same motion can be used to park the occulting ring below the horizon when the moon is down.    Click here to see the how the dimensions are calculated.

The rails and supporting structure could be built by a competent contract welder such as Benjamin Leiton.  Since the declination motion is moderate, the occulting ring can have considerable depth without changing the projected profile appreciably.   This allows it to be stiffened against high winds.  Fiberglass-foam sandwich (like a surfboard) will provide a good stiffness to weight ratio and can be out contracted.  The wheels from in-line skates will carry the weight easily while accommodating imperfections in the track built from welded angle iron.  Given that positioning requirements are very coarse, one could think of solutions requiring less capital outlay, but our favorite RS485 controlled servo motors (QuickSilver) would require least labor and is therefore preferred.

For operation in daylight an occulting ring is essential, and even if the camera is not used in the daytime the ring would protect the camera from heat and UV damage.
 

Switchable Filters :

The 8 mm Nikkor comes with a built in filter wheel.  Replacing the filter disk with one that has a toothed rim would allow a QuickSilver motor to be clamped to the lens so that remote filter changes could be made via the same RS485 port as the occulting ring.
 
 

Software refinements :

Roger Smith,   2001 April 20

2001 May 7:

Tests with the camera on a dark photometric night have demonstrated that the camera forms sharp images and operates reliably in continuous aquisition mode all night.  Images are sharp with minimal wings on bright stellar profiles.    Dark current and read noise were not aparent in images on the sky, and there are only a few small cosmetic defects.   The OH emission and/or scattering from aerosols at low angles brighten the sky near the horizon sufficiently that some long wavelength blocking may be desirable.  Images under a wider varaiety of weather conditions will be required to determine whether a filter changing motor will be desirable.

It was verified that direct sunlight on the lens will not produce anything like enough heat to damage the shutter.

2001 May 12 - Jun 8:

David has deciphered Skycalc and extracted the portions necessary to compute sun/moon rise & set times to allow the camera to startup and run autonomously.  This a general solution which can be applied wihtout modication at any site.

The workstation clock has been slaved to the GPS Time Server and exposures are now automatically taken at the same sidereal time each night to allow reference frame subtraction in future.  The frame rate need not produce an integer number of images per day.  We have devised an automatic image naming scheme which allows for easy file searches by either Date, UT or ST.

The SDK provided examples to read images from the camera into memory only.  David has incorporated the code to write the image buffer to a FITS file on disk.   Camera parameters, aquisition timing, and ephhemeris information have been added to the headers.

Roger has made a sketch of the enclosure concept.  Patricio Schurter is producing a 3D scale drawing as an exercise to learn his new CAD program.

Remaining Tasks:

Enclosure:

Finish data acquisition software:

Web interfaces:

Data processing:

Animation tool for local replay high resolution & frame rate:

actr@ctio.noao.edu