The Infrared Cloud Monitor for the MAGNUM Robotic Telescope at Haleakala
M. Suganuma, Y. Kobayashi, N. Okada, Y. Yoshii, T. Minezaki, T. Aoki, K. Enya, H. Tomita, S. Koshida
aa r X i v : . [ a s t r o - ph ] A p r THE INFRARED CLOUD MONITOR FOR THE MAGNUMROBOTIC TELESCOPE AT HALEAKALA
Masahiro Suganuma , , Yukiyasu Kobayashi , Norio Okada , Yuzuru Yoshii , , TakeoMinezaki , Tsutomu Aoki , Keigo Enya , Hiroyuki Tomita , and Shintaro Koshida , ABSTRACT
We present the most successful infrared cloud monitor for a robotic telescope.This system was originally developed for the MAGNUM 2-m telescope, whichhas been achieving unmanned and automated monitoring observation of activegalactic nuclei at Haleakala on the Hawaiian island of Maui since 2001. Using athermal imager and two aspherical mirrors, it at once sees almost the whole skyat a wavelength of λ ∼ µ m. Its outdoor part is weather-proof and is totallymaintenance-free. The images obtained every one or two minutes are analysedimmediately into several ranks of weather condition, from which our automatedobserving system not only decides to open or close the dome, but also selectswhat types of observations should be done. The whole-sky data accumulatedover four years show that 50 −
60 % of all nights are photometric, and about 75 %are observable with respect to cloud condition at Haleakala. Many copies of thissystem are now used all over the world such as Mauna Kea in Hawaii, Atacamain Chile, and Okayama and Kiso in Japan.
Subject headings: instrumentation: miscellaneous National Astronomical Observatory, 2-21-1 Osawa, Mitaka, Tokyo 181-8588, Japan e-mail: [email protected] Institute of Astronomy, School of Science, University of Tokyo, 2-21-1 Osawa, Mitaka, Tokyo 181-0015,Japan Research Center for the Early Universe, School of Science, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku,Tokyo 113-0033, Japan Kiso Observatory, Institute of Astronomy, School of Science, University of Tokyo, 10762-30 Mitake, Kiso,Nagano 397-0101, Japan Institute of Space and Astronomical Science, Japan Aerospace Exploration Agency, 3-1-1, Yoshinodai,Sagamihara, Kanagawa, 229-8510, Japan Department of Astronomy, School of Science, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo113-0013, Japan
1. INTRODUCTION
A cloud monitoring system, which watches the sky to detect clouds above an observatory,is a powerful apparatus for ground-based telescopes if we want to check the sky easily andto execute some remote or automated observations. The telescopes and their instrumentsare sure to be safe if we can monitor clouds on time and can close the dome slit before cloudcoverage becomes heavy and rain drops come. Other types of weather sensors, such as rainor humidity sensors, are sometimes late to alert us to close the dome. With a cloud monitor,we also can always be sure whether the data acquired by the telescope has been affected byclouds or not.One of the smartest methods of seeing clouds from the ground is to use some thermalinfrared wavebands in which clouds themselves emit thermal radiation or reflect radiationoriginating from ground or sea. A CCD camera with a fish-eye lens is cheap, but theappearance of clouds in optical is deceptive because their brightness depends strongly on theintensity of the moon and city lights that illuminate them. Using a similar system, Shamir& Nemiroff (2005) developed an algorithm to make a whole-sky opacity map by means ofmeasuring the extinctions for many stars. However, it does not give us a direct view of thecloud distribution in the sky.An uncooled thermal imager with panoramic optics suits a robotic telescope becauseit requires little maintenance. The Sloan Digital Sky Survey (SDSS) project has developeda scanning system using a single channel photometer cooled by liquid nitrogen (Hull et al.1994; Hogg et al. 2001). It has enough sensitivity and field of view, but their scanningmechanism is not easy to construct and its cooling system needs frequent hands. Recently,thermal infrared imagers without cooling parts sensitive enough to detect thin clouds havebecome available, and by combining these with some panoramic optics, now we can easily seethe whole sky in thermal infrared. This idea was presented by Mallama & Deganan (2002),and a similar one was developed by the Apache Point Observatory . However, it requiresstability and reliability for these systems to be put into practical use for a robotic telescopewhere no operator or engineer is onsite.We developed an infrared cloud monitor system that is successful for this use. Owingto this, we have achieved unmanned automated observation at the MAGNUM observatoryat Haleakala since 2001 (Kobayashi et al. 2003; 2004). The MAGNUM (Multicolor ActiveGalactic Nuclei Monitoring) project newly built a 2-m optical-infrared telescope at the Uni-versity of Hawaii’s Haleakala Observatory site on the Hawaiian Island of Maui, and has been http://irsc.apo.nmsu.edu/ §
2. In §
3, the analysis softwarethat detects clouds and evaluates the whole-sky condition is described. The performancein some respects in operating the MAGNUM observatory is presented and discussed in §
4. Finally, the weather trends seen in our accumulated whole-sky condition data over fouryears are discussed in §
2. System Overview
Figure 1 shows a schematic diagram of the MAGNUM Infrared Cloud Monitor. Theoutdoor hardware of the system is to the left of the figure. There are two aspherical con-vex mirrors with Cassegrain-like alignment. A blackbody reference plate for calibration isinstalled where the camera can see it near the edge of its field of view. All electrical de-vices, thermal imager, signal converter for the output signal of the imager, shutter, and 4 –thermometer circuit for the blackbody reference plate, are attached under the primary mir-ror and surrounded by an aluminum-pipe housing whose ceiling is the primary mirror. Onthe central hole of the primary mirror is installed a diamond window which is transparentto thermal infrared and prevents water from dripping into the housing. Photographs of theoutdoor hardware are shown in Figures 2 and 3. It is about 80 cm in height and about 35cm in diameter.The data observed by the outdoor hardware are acquired by a Linux PC, illustratedon the lower right of Figure 1. The PC controls the shutter with a digital I/O board,and also triggers data acquisition of images and temperatures. The controlling and data-acquiring software, as well as analysis software, are always working and getting sets of thedata including whole-sky raw images every one or two minutes throughout the night. Theoutput data are directed to various types of software through a LAN in the MAGNUMobservatory that includes the main manager of the observatory, the real-time scheduleror selector of target astronomical objects, and the information-collecting server for imageheaders of astronomical observations .The specifications of our infrared cloud monitor are listed in Table 1. Owing to theuncooled thermal imager and reflecting optics that widen the camera field of view, we canobtain almost whole-sky images in thermal infrared. This hardware alignment, a Cassegrain-like mirror system above an aluminum-pipe housing containing the thermal imager andelectronic parts, has a significant advantage in that the outdoor system becomes compactand is waterproof. The only movable component in the system is a shutter, the “PRONTERmagnetic E/100”, which works both as a shield against the sunlight during daytime and asa flat-fielding plate for image reduction.The main hardware components of our infrared cloud monitor are the thermal imager,the reflecting optics, and the data acquisition system, the details of which are describedindividually in the next subsections. The uncooled infrared imager that is sensitive in the 10 µ m waveband is one of the keycomponents of our cloud monitor. In order to see high-altitude clouds with good visibility,or to detect the thermal emission of clouds standing out against the dark background of cold GIF images of the whole-sky emissivity cloud maps can be found at http://banana.ifa.hawaii.edu/cloud/ ∼ µ m band, and the other isthe 10 µ m band. So far as we use only one waveband and cannot measure cloud temperature,the 10 µ m band is better to estimate cloud emissivity. This is because this wavelength isaround the flat-top part of the blackbody radiation of the clouds and ground, and flux fromthe clouds in this band is less dependent on temperatures than in the 3 ∼ µ m band.We use an Amber Sentinel Camera commercially produced by the Amber A. RaytheonCo. in 1997. This imager has an uncooled bolometer array of 320 ×
240 pixels for its detector,and is sensitive enough to detect thin cirrus clouds in thermal infrared. It outputs both ananalog signal of NTSC standard and digital signals of 12-bit depth with a frame rate of 30Hz. The specifications of the imager are tabulated in Table 2.The camera has an automatically offset flat-fielding function. It compensates for pix-to-pix scatter of bias and dark current signals that depend heavily on the temperature of theimager. This calibration is necessary for a quick look at raw images and is recommended everyseveral minutes; calibration can be triggered by PC through an RS232C interface. However,we do this calibration less frequently, and take integrated images of a closed shutter platefor more precise compensation (see § The Cassegrain-like reflecting optics is another key component of our infrared cloudmonitor. The camera’s field of view is not large enough to cover the whole sky, and mustbe expanded by other optics. Germanium crystals are generally used for the lenses of thesethermal imagers because it is well transparent in thermal infrared. Fish-eye lenses madeof this material, however, are not commercially available, and are also hard to develop orfabricate without a great deal of expenditure. Thus, it is reasonable to use some kind ofconvex mirror system.For the fundamental shape of the mirrors, we based ours on the particular asphericalones introduced by Chahl & Srinivasan (1997). When we look into these types of mirrorsalong with their optical axis, the appearance of the reflected field is not radially deformed;they preserve a linear relationship between the apparent angle from the image center andthe real radial angle from the field center. A simple spherical mirror produces a radiallycompressed image toward the image edge; the more distant from the image center we seethrough the mirror, the more radially compressed the objects appear.Now, we align the camera with the surface of a mirror using polar coordinates as shownin Figure 4, where the lens node of the imager is located at the origin, and the mirror surfaceshould be adjusted by the revolution of the function r ( θ ) around Z axis. Here, θ is an angleof line-of-sight with the Z axis in the camera field, and Θ is an angle of line-of-sight withthe Z axis in the negative direction in the real field that is seen through both the cameraand the mirror. According to Chahl & Srinivasan, Θ would proportionally correspond to θ if r ( θ ) is given by r ( θ ) = r (sin γ ) − /κ [sin( κθ + γ )] − /κ , (1)where r is the distance between the mirror and O along the Z axis, γ = tan − [ dr ( θ = 0) /dz ]is the initial angle of the mirror, in other words, a half of the vertex angle, and κ relates tothe proportionality constant α between θ and Θ, which means field widening power, as d Θ dθ = − − κ = α . (2)If we place the mirror with the convex side upwards and direct the camera to look down onthe mirror vertically, we can see the sky at zenith angle between Θ min and Θ max in degrees:Θ min = 2(90 ◦ − γ ) (3)and 7 –Θ max = θ max + 2 (cid:18) ◦ − tan − (cid:20) dr ( θ = θ max ) dz (cid:21)(cid:19) . (4)A circular image of the whole-sky is obtained by setting θ max as half the shorter angle of therectangular field-of-view of the imager.Chahl & Srinivasan (1997) also suggested a two-mirror system of Cassegrain-like align-ment, in which the aspherical shape introduced above is used as a primary mirror and acone-like shape is used as secondary, as shown in Figure 5. Here, the secondary with halfa vertex angle of β is placed with its surface facing the imager, which is equivalent to theconfiguration in Figure 4 with γ = β . The section of its surface is triangular, equivalent to α = 1, which means the surface has no field widening power. The section of primary mirrorsurface should be drawn similarly to Figure 4, but in an X ′ O ′ Z ′ coordinate system, in which O ′ is symmetrical with O about the section of the secondary surface. If β , r ′ , and γ ′ areproperly optimized, the light coming from the zenith ( θ = Θ = 0) can reach the imager,avoiding the secondary mirror by way of r ′ and the secondary vertex. Then, the entire skyincluding the zenith can be seen.However, though the idea is very attractive, we found a serious astigmatism aberrationin these optics which is mostly derived from the cone-shape of the secondary. When seenfrom above along with the optical axis, there is no curvature along with sagittal directions onthe completely cone-shaped surface, while its curvature exists tangentially. This aberrationbecomes extremely large if we use an imager with a large lens aperture. A combination ofour 71 mm-aperture lens imager and 240 mm-diameter primary mirror, being restricted byour manufacturing capacity, results in a point-spread-function (PSF) size over one millimeteron the detector, which corresponds to 10 degrees in the sky.We therefore did not completely follow the original two-mirror system scheme, andimproved it to upgrade image quality at the cost of view near zenith and some amount ofsensitivity. In detail, we significantly flattened the vertex angle of the cone-shaped secondarymirror to reduce its tangential curvature. Next, we introduced an aspherical shape similarto that of the secondary mirror in Figure 4, so that it also has a field-widening powerto some extent, similar to a primary mirror. In fact, O ′ projected by the section of thistype of secondary surface does not strictly converge at one point, but its effect on the fielddeformation was found to be ignored. Moreover, we adopted a smaller primary mirror hole toreduce the shadow area in the image center, and also to extend the focal depth. Because theprimary hole squeezes the imager’s lens aperture, the decline in sensitivity is compensatedby frame integration.As a result of these improvements, we adopted the parameters listed for the “Improved”model in Table 3. The parameters determined following the original idea of Figure 5 are 8 –also listed as a reference for the “Original” model. Our “Improved” model gives a shadowcircle with a radius of 11 degrees at the zenith. However, the area is not critical for usbecause the MAGNUM telescope rarely observes objects around the zenith because of thesparse distribution of celestial coordinates of our targets, and because of some operationalrestrictions of our telescope.Most remarkable is that the image resolution was dramatically increased in our “Im-proved” model. Figure 6 shows the spot-diagrams of ray-traced images for both “Original”and “Improved” models. The specifications of the two models are also tabulated in Ta-ble 4. The “Improved” model decreases the size of the point-spread-functions (PSFs) byan order of magnitude or more. There remain some astigmatism and field curvature in our“Improved” model, but the PSF size is within a few pixels over almost the entire field of view. The surfaces of the mirrors were shaped by diamond turning on brass that is easilyworked and goes well with gold coating. Using an ultra-precise, computerized, numericallycontrolled (CNC) turning machine, we obtained a surface roughness of about 20 nm in rms,which is fine enough as a mirror at a wavelength of 10 µ m. The surface was plated with solidgold, containing 5 % Cobalt, with a thickness of 2 µ m. Finally, a physical vapor sapphire wasdeposited on the surface with a thickness of 0.2 µ m for protection. The surface reflectivity isabout 95 % and no serious degradation has been seen in an outdoor operation of five years.For the first attempt to process the mirror surface, we tried an aluminum-based alloyplated by electroless nickel, and coated gold on the surface. But we found that the gold coatdegraded in a few months’ exposure to air. The pinholes in the nickel plating might makewater erode the aluminum base rapidly.On the central hole of the primary mirror was placed a chemical vapor-deposited (CVD)diamond plate with a thickness of 0.2 mm and transparency of about 80 % at λ ∼ µ mwith no coat. A germanium plate processed with anti-reflection coating on both top andbottom and protective coating on the top could also work in some environments, and hasbeen used in similar systems at several sites such as Mauna Kea in Hawaii and Atacamain Chile. However, during our test operation in Tokyo, Japan, the upper surface of thegermanium window degraded in a few months. We guess that rain in Tokyo is very acidic,which might enhance degradation of the protective coating. 9 – The thermal imager, the Amber Sentinel Camera, outputs images with analog signalsof NTSC standard as well as digital signals of 12-bit parallel channels at the rate of 30-frames per second. We use digital output because an analog signal of 8-bit depth data losesthe lower 4 bits of original signal that is much larger than the noise signal of one or twoanalog-to-digital units (ADUs), and is difficult to restore by frame integration afterwards.For each data bit, together with synchronizing clocks for frame acquisition, the digitalsignals are single-ended transistor-transistor-logic (TTL) standard. We convert these signalsto differential signals of RS-422 standard level by a hand-made digital electrical circuit withIC in order to transmit the signals to a PC about 15 m away from the outdoor system. Thisis because the TTL signals are too delicate in a noisy environment to send more than severaltens of centimeters at a high data rate.The signals are acquired by a Linux PC using a digital frame-grabber board, PC-DIGproduced by Coreco Inc. It can grab digital data of 12 parallel channels at the rate of4.6 mega-bytes per second for our imager, and its driver for Linux OS is supported by thecompany.We integrate the images for 5 seconds, corresponding to 150 frames, to increase thesignal-to-noise ratio. Integration for more than 10 seconds is not favorable because thewhole-sky images of clouds are often blurred by migrations of clouds.For each acquisition of the whole-sky image, we take shutter images immediately beforeand after it. In addition, the temperature of the blackbody reference plate is measured atthe same time.The sets of data are acquired about every one or two minutes while the elevation of thesun is below 15 degrees. We cease the operation almost entirely during the daytime, in casedirect rays of sunlight degrade the imager detector.
3. ANALYSIS SOFTWARE3.1. Data Reduction
Each whole-sky image is processed immediately after acquisition with the shutter im-ages, the temperature of the blackbody reference, and some calibration data measured inadvance. Figure 7 illustrates how we reduce a raw whole-sky image into the apparent emis- 10 –sivity map of the clouds.First, offset flat fielding for the pix-to-pix pattern is done by subtracting the averageof the two shutter images obtained before and after the incident exposure of the whole-skyimage. The automatic offset flat fielding by the camera is convenient for snapshot images,but not complete for our frame-integrated whole-sky images. We can do similar calibrationwith a higher signal-to-noise ratio using the frame-integrated shutter images.Next, background signals from the optics and the interior of the imager are subtracted.There are two types of components of background signal: flat-offset components and spatial-pattern components. They differ from each other with varying internal and environmentaltemperatures, and have to be subtracted separately.The flat-offset background component includes bias and dark current signals of thedetector along with thermal radiation from the interior of the imager and the optics, whichshould be compensated for each whole-sky image. Subtraction of the shutter image fromthe whole-sky image does not work well because the surface brightness of the shutter andthe thermal background from the optics vary independently. The flat-offset value of thebackground signal C off on the whole-sky image is calculated from the temperature T ref of theblackbody reference plate and its signal C ref on the incident whole-sky image as C off = C ref − Z B λ ( T ref ) dλ / g , (5)where B λ ( T ref ) is the Planck function for temperature T ref , and g is the signal-to-surfacebrightness ratio measured beforehand. The units of C off and C ref are ADU. We subtract thesingle value of C off from those of all pixels in the incident whole-sky image.There remains a spatial-pattern background component, which mainly originates in thebaffle of the reflecting optics and the atmosphere. The pattern of this type of backgroundis radially symmetrical and almost stable on clear nights. We therefore prepare a templatewhole-sky image for a clear night beforehand. The template image should be acquired whenthe sky is certain to be clear and reduced up to compensating the flat-offset background.We subtract the template image from all whole-sky images.Now that the background signals are subtracted, we calibrate the signals in the whole-sky image into a surface brightness value using the signal-to-surface brightness ratio g . Onecan determine the value of g in laboratory by exposing the blackbody targets of differenttemperatures, or can determine it at the observatory site by exposing both clear sky anda black object of ambient temperature at once. Note that the value of g is dependent onzenith angle, mainly because of vignetting on the aperture of the camera lens. We shouldtherefore measure g for several zenith angles and using a function fitted to them. Figure 8 11 –shows the data of g measured for our system at the MAGNUM observatory site.Finally, the surface brightness S in the image is converted to the apparent emissivity ǫ of the clouds at a 10 µ m waveband, which is related to S as S = ǫ · Z B λ ( T c ) dλ , (6)where B λ ( T c ) is the Plank function for the cloud temperature T c . According to the averageannual air temperature of 296 K at sea level in Maui Island and lapse rate of − . / kmfor standard atmosphere (COESA, U.S. Standard Atmosphere 1976), the expected ambienttemperature at 10,000-m altitude above Haleakala Observatory should be about 240 K.We therefore calculate ǫ assuming (hereafter fixing) the temperature of T c = 240 K asrepresentative of high-altitude clouds or cirrus.Note that ǫ includes reflection efficiency of a cloud as well as absorption efficiency, andin S , there is a significant amount of reflected emission by the cloud which originates in thesurface of the ground or sea. This means we cannot convert the apparent emissivity ǫ simplyto optical depth, which relates to actual absorption efficiency. However, particularly forhigh-altitude clouds, it is reported that a large amount of emission still originates thermallyin clouds themselves (Platt & Stephens 1980). Figure 9 shows the whole-sky cloud emissivitymaps obtained and processed under various sky conditions.It is very convenient for a remote watcher from Japan as well as at the Haleakala siteto see the whole-sky cloud emissivity maps on the Internet. However, our main objective inoperating the MAGNUM observatory is automated observation in real-time considerationof weather conditions. We therefore developed software to detect clouds from the whole-skycloud emissivity maps and evaluate observational conditions from them ( § § To determine whether clouds exist or not in a certain part of the sky, it is important tomeasure both the average and the standard deviation value of the emissivities in a small areain about that direction, rather than to refer to just one pixel value. Here, two elements limitsensitivity: one is variation in zero emissivity level caused by residual thermal backgroundsignal, and the other is pix-to-pix noise.The empirical value of the former for our system is about ǫ = 0 .
25, which is considerablylarge compared to those of thin clouds. This mainly comes from the residual pattern of the 12 –background radiation that is difficult to subtract completely from a single template whole-skyimage. Also, humidity has some correlation with residual background.The latter limit for the sensitivity is about ǫ = 0 .
015 as a noise equivalent signal of theimage, which is much less than the former. Thin clouds are easy to detect by their spatialfluctuations of emissivity rather than by emissivity values themselves.We therefore divide a whole-sky cloud emissivity map into 90 sub-areas, each being10 degrees in elevation and 20 degrees in azimuth. For each sub-area, we categorize thecloud condition into several levels using the average emissivity ǫ and the rms emissivity σ ( ǫ )calculated for the area.Figure 10 shows the σ versus ǫ diagram on which each sub-area can be evaluated. Asub-area is evaluated as “clear” only when both ǫ < .
25 and σ ( ǫ ) < .
05, otherwise regardedas being covered by some clouds. Except for “clear”, the sub-area is evaluated into “thin”,“thick”, or “rain” when ǫ < .
4, 0 . ≤ ǫ < .
5, and ǫ ≥ .
5, respectively. The conditionof “rain” means that the surface brightness is larger than that for blackbody of 300 K, andthe mirror system is possibly wet due to rainfall or moisture, though the direct detection ofrainfall should be done by rain sensors. Each sub-area is given a level of 0 for “clear”, 1 for“thin”, 2 for “thick”, or 3 for “rain” in order to calculate whole-sky cloud condition fromstatistics over all sub-areas (see next subsection).The main cause preventing detection of even thinner clouds is the residual pattern ofthe thermal background on the whole-sky emissivity maps, which increases the rms valueof emissivities even in a small sub-area. This could be improved if a relation between theradial pattern and the temperature of the reflecting optics is contributed or the temperatureof the reflecting optics is regulated. More fundamentally, we are soon going to improve thedesign of the reflecting optics so that there would be no vignetting objects in the optical pass.
To determine whether the sky allows observation or not and what type of observationis best to execute, we evaluate the whole-sky cloud condition using statistics over sub-areavalues calculated and labeled in §
4. PERFORMANCE4.1. Automated Observation with Infrared Cloud Monitor at MAGNUMObservatory
Our cloud monitor was located at the Haleakala site and began to give whole-sky cloudemissivity maps when MAGNUM observatory started its telescope operation in August 2000.Then, an automated monitoring observation of active galaxies with the cloud monitor wasput into practical use in early 2001. After refinements of several months, we achieved fullyautomated astronomical observation for an entire night. Now, we have continuous unmannedobservation, except for maintenance every several months (Kobayashi et al. 2003; 2004)According to the whole-sky cloud condition evaluated by the cloud monitor, our auto-mated observing system decides whether observation is possible or not. When the whole-skycloud condition is either “CLEAR” or “THINorPARTIAL”, the observing system opens thedome slit and commands observation. When the whole-sky cloud condition is “CLOUDY” or“RAINY”, the observing system closes the dome slit, and carries out no observation. Whenthe whole-sky cloud condition is “MEDIUM”, the observing system maintains the ongoingoperation.Moreover, according to the whole-sky cloud condition, we also determine what typeof observation should be executed. If the condition is “CLEAR”, which can be regardedas a photometric sky, all types of observations are possible. If the condition is “THINor-PARTIAL”, certain observations that are delicate under cloud extinction are restricted; 14 –observations such as standard star calibration, relative photometry between several sepa-rate fields, or imaging of faint objects are allowed only in “CLEAR” conditions. Instead,differential photometry between the bright objects in the same field of view is permitted in“THINorPARTIAL” conditions because it is barely affected by extinction fluctuations.The whole-sky cloud condition and status of the sub-area at which the telescope ispointing are recorded in the fits header of observed images. The whole-sky emissivity mapsare also archived so that we can check the quality of the observed astronomical data whenwe analyze them.The cloud monitor has mainly been working stably until now, except for several monthsof trouble with the frame grabber board. Regular maintenance includes wiping the duston the mirror and shipping whole-sky images to Japan when we visit the site every severalmonths.In the following two sections ( § § The most primitive function required of our cloud monitor is to determine whether thesky allows observation or not. When the sky becomes cloudy, the cloud monitor should closethe dome slit before rain falls. It also should stop meaningless and risky opening of the domewhen the sky is still covered by thick clouds.Table 5 shows the frequency distribution of various weather conditions over four yearsfrom two different weather-sensing systems including the whole-sky cloud monitor and therain sensor. Note that the rain sensor directly senses rain drops by means of changes in theresistivity of the electrical circuit, while the cloud monitor only inspects surface brightnessof the sky image. The percentages of rain-sensor output in the night are 85.8 % for “DRY”and 14.2 % for “RAIN”.This table indicates that a combination of “CLOUDY” and “RAINY” comprises 96 %of “RAIN”, i.e., rain drops can be avoided by this high probability from such a combinedcloud condition. The remaining 4 % probability corresponds to a situation in which therain sensor catches rain drops while the output from the cloud monitor is “CLEAR”, “THI-NorPARTIAL” or “MEDIUM”, and would decrease further if acquisition of whole-sky datawere carried out more frequently, because the approach of moisture is sometimes very rapid. 15 –The humidity sensor, however, usually helps to catch the moisture on its way up to theobservatory.
The next important function required of the cloud monitor is to determine whether thenight is photometric or not. The flux calibration of active galaxies using reference stars orstandard stars in different telescope directions often fails if we are uncertain whether thesky is entirely clear. Reliability of the whole-sky condition “CLEAR” can be estimated fromstatistics of accumulated standard star flux data, because these have been observed quicklywhile the whole-sky is “CLEAR”.Table 6 presents several statistical values of our standard star observations over twoyears while the instrumental throughput was relatively stable. Columns (1) and (2) are thewaveband and effective wavelength, respectively. Column (3) is the number of observations.The standard deviation of the fluxes over all observations σ all for each band is given incolumn (4), and the average over individual photometric errors < err > is given in column(5). Nominal extinction value for unit airmass Q atm , measured by intensive observations ofstandard stars on a few nights, are shown in column (6). The linear trend of flux decreasingwith time during the period, derived from changes in telescope throughput, is corrected.Airmass correction for elevation in each observation is done with a constant value in thetable.The photometric errors < err > are so small that they contribute little to σ all . Therefore,the scatter σ all mainly contains the day-to-day changes in extinction by the atmosphere orclouds.Converting σ all and Q atm to flux ratio, we show ∆ F/F against wavelength λ in Figure13. The vertical bars with inverted triangles on top are σ all , being corrected for < err > .Filled squares are Q atm . The solid line is linear fit to the filled squares except for the K -band which is particularly affected by water vapor. The line shows wavelength-dependenceof ∆ F/F ∝ λ − . , which is consistent with a typical trend of a mixture of Rayleigh-scatteringby molecules and Mie-scattering by small aerosols in the atmosphere (Cox 2000).A similar linear wavelength-dependence is seen in σ all in the optical, and should bedominated by daily or seasonal variation in the extinction by the atmosphere. On theother hand, σ all at longer wavelengths beyond the R -band is near constant, regardless ofwavelength-dependence of Q atm . We consider that this flat component of σ all could include 16 –variation in extinction by clouds missed by our cloud monitor, because the size of typicalcloud particles is on the order of ten microns and their wavelength-dependence of extinctionis white at a few microns or shorter. Therefore, photometric errors caused by extinction ofthe clouds are restricted to within a few percent.
5. Trend of whole-sky cloud condition at Haleakala
Bradley et al. (2006) overviewed meteorological characteristics at Haleakala with respectto many types of weather data, such as humidity, temperature, wind speed, and cloudcoverage. However, their analysis is based on compilation of various records with fairlylarge spatial and time resolution, including those taken by satellites. Our whole-sky cloudconditions are more straightforward and systematic, because our conditions are completelybased on direct measurements of clouds that are projected onto the sky above the observatory.Table 7 shows the proportions of whole-sky cloud conditions averaged monthly betweenJanuary 2001 and December 2005. Several conditions in early 2001 are combined becauseof test operation of analysis software. A total of 668,063 whole-sky images for 1,601 nightsgive the statistics in the table.Figure 14 presents the relative frequencies of the whole-sky cloud conditions combinedover the data in Table 7. The percentage of each condition is an average over the dataweighted by the number of nights in which the data were obtained. The monthly percentagefor combined conditions between January 2001 and July 2001 is divided into respectiveconditions, according to their average proportions after August 2001.It should be noted that despite “CLEAR” and “THINorPARTIAL” conditions, obser-vations were sometimes impossible due to other weather warnings such as high humidity andstrong wind. Moreover, observations were not carried out when the wet sensor warned thata dome was not dried out after rainfall or moisture. Concerning Haleakala, more than 50 −
60% of all night time is near photometric, and in about 75 %, it is feasible to execute particularobservations. Haleakala is therefore not worse than Mauna Kea where the observable skyrate is 60 −
80 % as one of the best locations for optical and near-infrared observations inthe northern hemisphere, along with good access.Next, Figure 15 shows the monthly average relative frequency of the whole-sky cloudcondition. Clear seasonal cycles over a year can be seen; There are high observable rates
17 –in summer and winter, and low rates in early spring and late autumn, in agreement withBradley et al. (2006). It has generally been said that there are a dry summer season andrainy winter season in Hawaii. Our data, however, demonstrates that midwinter is not verybad at Haleakala, as far as sky condition is concerned.
6. Conclusions
We developed an infrared cloud monitor weather system that has been most successfullysupporting an unmanned robotic telescope. It sees almost whole-sky in thermal infrared withno field deformation, sensitively detects thin high-altitude clouds, automatically evaluatessky conditions, and withstands outdoor environments for several months without mainte-nance. Owing to this system, the MAGNUM observatory has been achieving unmannedautomated observation at Haleakala for more than four years. Its evaluation of the whole-sky cloud condition being photometric, observable, or non-observable seems to be mainlysuccessful. It also proves that for optical and near-infrared observations, Haleakala is a sitecomparable to Mauna Kea. Copies of our cloud monitor are now used for many similarsystems at sites all around the world, including the Atacama region in the northern part ofChile.The development of our cloud monitor was supported by the Advanced Technology Cen-ter, National Astronomical Observatory of Japan. We are grateful to the Technical Centerof Nagoya University for our use of their ultra-precise turning machine. We thank H. Takamiand N. Takato for helpful discussion and advice on development. We also thank M. Doi, K.Motohara, and colleagues at the Haleakala Observatories for their help with facility mainte-nance. This research has been supported partly by the Grant-in-Aid of Scientific Research(10041110, 10304014, 11740120, 12640233, 14047206, 14253001, 14540223, 16740106, and17104002) and COE Research (07CE2002) of the Ministry of Education, Science, Cultureand Sports of Japan.
REFERENCES
Bradley, E. S., Roberts, L. C., Bradford, Jr., L. W., Skinner, M. A., Nahrstedt, D. A.,Waterson, M. F., and Kuhn, J. R. 2006, PASP, 118, 172Chahl, J. S., & Srinivasan, M. V., 1997,
Applied Optics , 36, 8275 (Eratta: 1999, 38, 1196) 18 –COESA, U.S. Standard Atmosphere 1976, (Government Printing Office, Washington DC)Cox, A. N., ed. 2000, Allen’s Astrophysical Quantities (4th ed.; New York: AIP)Hogg, D. W., Finkbeiner, D. P., Schlegel, D. J., & Gunn, J. E., 2001, AJ, 122, 2129Hull, C., Limmongkol, S., & Siegmund, W. A., 1994, Proc. SPIE, 2199, 852Kobayashi, Y. et al. 1998, Proc. SPIE, 3354, 769Kobayashi, Y. et al. 2003, Proc. SPIE, 4837, 954Kobayashi, Y., Yoshii, Y., & Minezaki, T. 2004, AN, 235, 537Mallama, A. & Degnan, J. J., 2002, PASP, 114, 913Platt, C. M. R. & Stephens, G. L., 1980,
J. Atmos. Sci. , 37, 2314Shamir, H. & Nemiroff, R. J., 2005, PASP, 117, 972Takato, N., Okada, N., Kosugi, G., Suganuma, M., Miyashita, A., & Uraguchi, F., 2002,Proc. SPIE, 4837, 872Yoshii, Y. 2002, in New Trends in Theoretical and Observational Cosmology, ed. K. Satoand T. Shiromizu (Tokyo: Universal Academy Press), 235Yoshii, Y., Kobayashi, Y., & Minezaki, T. 2003, AAS Meeting, 202, 38.03
This preprint was prepared with the AAS L A TEX macros v5.2.
19 –Fig. 1.— Schematic diagram of MAGNUM infrared cloud monitor system; The outdoorhardware is to the left of the figure and the indoor hardware is to the lower right. 20 –Fig. 2.— Outdoor part of MAGNUM infrared cloud monitor. Shown from top to bottomare a secondary mirror, a primary mirror, and an aluminum-pipe housing that contains athermal imager, a signal converter, a shutter controller, etc. The three rectangle platesextending from the upper edge of the housing are rain sensors that form a different systemfrom the cloud monitor, but share power and wires with it. 21 –Fig. 3.— Infrared cloud monitor at MAGNUM Observatory. The cloud monitor is seen nearthe center, on the roof of the container. 22 –Fig. 4.— Basic alignment of the imager and the section of the panoramic aspherical mirrorsurface introduced by Chahl & Srinivasan (1997).Fig. 5.— The basic alignment of the imager and the section of Cassegrain-like mirror systemthat was originally introduced. 23 –Fig. 6.—
Top:
Spot diagram through the focus for “Original” model of the mirror systemoptimized to our camera, following Chahl & Srinivasan (1997).
Bottom:
A similar diagramfor “Improved” model. Note that the unit of scale for the spots is micron, and the scale barin the top panel is ten times larger than that in the bottom. The pixel scale of the imagerdetector is 50 µ m. 24 –Fig. 7.— Block diagram of the reduction procedure from a raw whole-sky image into awhole-sky cloud emissivity map. 25 –Fig. 8.— The signal-to-surface brightness ratio g of MAGNUM infrared cloud monitorsystem, measured as a function of zenith angle. Dashed line, being fitted to the measuredpoints, is used for g in Figure 7 and equation (5).Table 1. Specifications of MAGNUM Infrared Cloud Monitor.Wavelength 8 − µ mOptics Thermal imager and two aspherical reflectivemirrors with Cassegrain-like alignmentSensor Micro-thermal bolometer array of 320 x 240pixField of View Circular field of 11 - 70 degrees at zenith anglePixel Scale 0.5 deg/pixelSampling rate one-or two-minute interval a Sensitivity ǫ ∼ a One image is integrated for 150 frames (total of 5 sec). 26 –Fig. 9.— Whole-sky cloud emissivity maps acquired and processed by MAGNUM infraredcloud monitor under various sky conditions. The sky condition is clear at top left, thin attop right, partially cloudy at bottom left, and entirely cloudy at bottom right. Note thatthe emissivity is apparent one, which is defined as eq.6, being assumed T c = 240 K. Thereare two shadow circles in each image: the small one at the image center is field vignettingby the hole of the primary mirror and the secondary mirror, and the other near the imageedge is a blackbody reference plate. 27 –Fig. 10.— Classifications of cloud status for each sub-area of 20(Az) × ǫ and standard deviation σ ( ǫ ). A numerical value in parentheses ineach zone indicates a level for calculating whole-sky cloud conditions (see § ∆ F / F λ [ µ m] U B V R I J H K ∝λ -2.4 Fig. 13.— σ all and Q atm in Table 6, plotted on wavelength vs. flux ratio plane. The invertedtriangles with vertical lines are σ all , being correlated with < err > . Filled squares are Q atm .The line fitted to the squares except for K -band shows wavelength dependence of λ − . . Thedotted line is one downwarded by a factor of three from the fitted line. 29 –Fig. 14.— Distribution of whole-sky cloud conditions at night at Haleakala over five yearsbetween Jan 2001 and Dec 2005. Percentages shown are weighed by the number of nightsper month. The percentage for combined conditions between January 2001 and July 2001 isdivided into respective conditions, according to their average proportions after August 2001.Fig. 15.— Mean monthly distribution of whole-sky cloud conditions at the Haleakala siteover five years, from 2001 to 2005. From the bottom to top, “CLEAR” (black), “THINor-PARTIAL” (gray), “MEDIUM” (light gray), “CLOUDY” (white), and “RAINY” (white).The same corrections as in Figure 14 were done for the data between 2001 January and 2001July. 30 –Table 2. Specifications of the Infrared Imager (Amber Sentinel Camera).Detector Uncooled micro bolometer arrayFormat 320 ×
240 pixelsWavelength 8 − µ mLens f = 50 mm, F /0.7 → ∼ F /1.4 a Field of View 18 ◦ × ◦ NEDT b < .
07 K for blackbody temperaturearound 25 ◦ CAnalog output NTSCDigital output 12 bit parallel (TTL standard signals)Remote control RS232C interfaceFrame rate 30 Hz a The lens was stopped down to about F /1.4 by a mask inorder to improve the image quality of the reflecting optics. b Noise Equivalent Differential Temperature.
Table 3. Design parameters of the original and improved Cassegrain-like two mirror systems optimized for AmberSentinel camera.Primary SecondaryModel O ′ ( X, Z ) a r ′ γ ′ α φ outb φ inc r γ α φ outb [mm] [mm] [degree] [mm] [mm] [mm] [degree] [mm]Original ....... (-52, 595) 530 85.0 11.0 240 80 300 85.0 1.0 80Improved ..... ( -2, 488) 360 89.6 4.4 240 38 340 89.6 2.3 100 a the position of the virtual node O ′ for primary in the coordinates XOZ b outer diameter c inner diameter, that is, the diameter of the center hole of the primary 32 –Table 4. Specifications of the original and the improved mirror system optimized forAmber Sentinel camera.Model Rms-size of PSF FOV a Focalratio[ µ m] [pixels] [degrees]Original ..... 600 − −
20 0 − F /0.7Improved ... 26 −
95 0 . − − F /1.4 a measured in zenith angleTable 5. Correlated classifications from two different weather systemsRain Cloud MonitorSensor CLR THN MED CDY RNY totalDRY 55.8 18.0 2.5 5.9 3.5 85.8RAIN 0.1 0.4 0.2 0.7 12.9 14.2total 55.9 18.4 2.7 6.6 16.4 100.0 %Note. — CLR, THN, MED, CDY, and RNY mean thewhole-sky cloud conditions, “CLEAR”, “THINorPARTIAL”,“MEDIUM”, “CLOUDY”, and “RAINY”, respectively. Notethat rain sensor directly senses rain drops or thick moisture,while the ”RAINY” evaluated by cloud monitor means thatthere are high emissivity regions at 10 µ m waveband in thefield of view. 33 –Table 6. Statistics of standard star observationBand Wavelength N-obs a σ allb < err > c Q atmd [ µ m] [mag] [mag] [mag](1) (2) (3) (4) (5) (6) U B V R I J H K Q atm . Linear trends of decreasing flux duringthe two-year period of observations are also corrected. a number of observations b standard deviations of flux over observations c average photometric error for each night d nominal extinction value per unit airmass measuredat our observatory 34 –Table 7. Monthly average of the whole-sky cloud conditions during nights.Year Month Nights Ndata CLR THN MED CDY RNY[%]2001 Jan 20 6981 90.4 a a b b b b b a combined with CLR, THN, and MED. bb