A prototype auroral hyperspectral all-sky camera has been constructed and tested. It uses electro-optical tunable filters to image the night sky as a function of wavelength throughout the visible spectrum with no moving mechanical parts. The core optical system includes a new high power all-sky lens with F-number equal to f/1.1. The camera has been tested at the Kjell Henriksen Observatory (KHO) during the auroral season of 2011/2012. It detects all sub classes of aurora above ~½ of the sub visual 1kR green intensity threshold at an exposure time of only one second. Supervised classification of the hyperspectral data shows promise as a new method to process and identify auroral forms.
©2012 Optical Society of America
1. Short background
During the last decade Liquid Crystal Tunable Filters (LCTFs) have been developed by the company CRi (Cambridge Research & Instrumentation, Inc.). The ability to electronically tune the band pass wavelength of these filters throughout the visible electromagnetic spectrum makes them an ideal candidate for hyperspectral imaging [1, 2].
A high throughput all-sky lens has been constructed to match the optical design of a LCTF from CRi with an EMCCD (Electron Multiplying Charge Coupled Device) detector from the company Princeton Instruments (PI). This paper describes the optical design and performance of the assembled instrument as candidate for low light hyperspectral imaging of aurora and airglow.
2. Optical layout and design
Figure 1 shows the optical diagram and the lens mechanics of the all-sky lens, named NORUSCA II. The design includes a section or a chamber for optical elements such as interference filters or in our case a LCTF. See label (2) in Fig. 1. The circular fish-eye view of the lens is converted by the first part of the lens into a narrow beam of only ± 7° divergence. The effective aperture of optical elements that may be inserted into the filter chamber is 35 mm. The third part of the lens focuses the light from the filter chamber onto a circular image of 8 mm in diameter. The focal length f = 3.5 mm and the back focal length of the system is 17.5 mm, which is compatible with C-mount camera systems.
Note that there is no telecentric lens system in the design. This reduces the number of active optical elements needed to only 12 compared to a standard auroral telecentric all-sky lens system with 19 optical elements. As a consequence, the throughput is increased due to less transmission losses and high effective apertures. The net result is a high throughput all- sky lens with F-number close to one. Table 1 lists the key technical characteristics of the lens.
The filter chamber of the lens is matched in size to the LCTF filter. The acceptance angle of 7.5° and the effective aperture of 35 mm of the filter are matched to the input light beam divergence cone and its cross section diameter, respectively. The etendue is in other words conserved through the whole system.
Spectral tuning is obtained by using electronically controlled liquid crystal wave plates to a Lyot filter design . The wave plates behave as optical birefringent elements with an electrically variable retardance. The latter is controllable due the effect that the liquid crystal molecules are orientation sensitive to electric fields applied between the plates. Since the retardance is directly linked to wavelength, the filters are tunable.
The spectral range of the filter covers the visible part of the electromagnetic spectrum from 400 – 720 nm. The bandwidth is 7 nm at center wavelength 550 nm. The filter switches from one wavelength to another in just 50 ms. The transmission factors are ~10, 40 and 45% at wavelengths 430, 560 and 630 nm, respectively.
The low transmission of the LCTFs, especially in the blue part of the spectrum, is the main reason they have not already been effectively applied to auroral and airglow research. Note that standard interference filters have transmissions of ~80% at all wavelengths, which is 2-8 times better than the LCTFs. Nevertheless, our attempt is to try to compensate the low transmission with the use of the high throughput all-sky lens and an EMCCD detector from PI. The ProEM 512B image sensor is back-illuminated with peak quantum efficiency above 90%. The format of the sensor is 512x512 pixels on a square area of 8.2 x 8.2 mm2, which covers the circular lens image. The camera head is air cooled down to −70°C to reduce thermo electrical noise. Figure 2 shows the whole assembled system.
3. Basic system characteristics
3.1. Line spread function (LSF)
Figure 3 shows the polychromatic Line Spread Function (LSF) of the lens as a function of view angle to the optical axis calculated by the ZEMAX design software (http://zemax.com). The calculation includes diffraction. An artificial star or point source is according to Fig. 3 focused onto the image plane within a radius of 8 μm across the entire field of view of the lens. Since the size of the EMCCD pixels is 16 x 16 μm2, a star should theoretically be focused within one pixel. A comparison with real data of stars on the night sky is presented in Section 4.
The Modulation Transfer Function (MTF) is defined as the discrete Fourier transform of the LSF. If we assume optimal focus at 50% cut-off frequency of the MTF, the spatial frequency becomes 40-60 line pairs per millimeter.
3.2. Mapping function
The mapping function, R, or the radial center pixel distance of a point in the image plane is found by the use of a LED (Light Emitting Diode) illuminated pinhole. The pinhole is located at a distance of 1 m from the center of the front surface of the all-sky lens, on an arm that can be rotated an angle θ with respect to the optical axis. In this configuration, the pinhole acts as artificial star with an angle θ to the optical axis. A neutral density filter (SCHOTT NG9) in front of the pinhole is used to attenuate the intensity by a factor of 0.1. The arm is rotated in sub steps of 10° from θ = 0° to θ = 90°. In this experiment, the center wavelength was set to 555.7 nm with an exposure time of 100 ms and gain equal to 100. The results are shown in Fig. 4 .
It is clear that none of the standard mapping fisheye functions presented in Fig. 4 fit our data, especially for θ > 30°. Therefore, a 2nd order polynomial function is introduced asTable 2 .
R is for θ > 30 ° in-between the mapping functions of an equal area and an orthographic fisheye lens, and its maximum Rmax = 4.08 mm at θ = 90 ° matches the size of the EMCCD.
3.3. Spectral characteristics
The experimental setup for intensity calibration is described in detail . A diffuse re-emitting Lambertian screen is located a distance of z = 8 m from a standard 45W tungsten lamp (ORIEL SN7-1633). In our case, only the on-optical axis center pixel of the camera is used in order to make sure that its field of view is filled by the screen. Note that as long as the field of view of the center pixel is filled, then neither the distance of the camera or the angle to the screen matters.
The brightness of the screen as seen by the center pixel is given as
A mercury vapor tube supplied by Edmund Optics Ltd. (SN K60-908) is used for wavelength calibration of the camera. The mercury lamp is used as source to the screen. Note that the source for our calibration is the screen, not the lamp itself. Figure 5 shows the results of the calibration including the mercury lamp, the raw data counts and the calibration factors.
The wavelength calibration shows that the peak emissions of mercury, Hg 435.8 nm, Hg 546.1 nm and the 577/579.1 nm doublet are all identified within the measured Full Width Half Maximum (FWHM) of 5.03 nm, 6.96 nm and 8.51 nm, respectively. For Lyot filters FWHM varies with wavelength as5]. As seen in Fig. 5, the instrumental line profiles are close to triangular in shape.
Next is the sensitivity calibration. Figure 5 shows the spectral raw counts, Cλ, of the center pixel. The exposure time was kept fixed at 100 ms. Gain is set to 100. A 3rd order polynomial fit is applied to the spectrum. The overall counts are increasing with wavelength. This is mainly due to the 45W tungsten lamp, which peaks in the near infrared part of the spectrum at λ = 900 nm.
With focus on aurora, Mλ is converted to number of photons cm−2 s−1 sr−1, and multiplied by 4π x10−6 to express the screen brightness in Rayleigh (R) . The calibration factor is then given as
4. Experimental setup
One NORUSCA II camera was installed under a transparent dome at the Kjell Henriksen Observatory (KHO) on Svalbard, Longyearbyen, Norway (78°N, 16°E). Initial tests of the instrument started on 7th of November 2011 and ended on 29th of January 2012. KHO is in daytime located underneath the average dayside aurora oval. The two months of astronomical darkness at mid-winter makes the location one of the most ideal places for ground–based observations of both day- and nighttime auroras.
15 wavelength bands were selected to cover the most prominent auroral emissions in the visible spectrum. Table 3 shows the origin of the emissions and the calibrations factors based on the procedure in Section 3.3.
The emissions lines at wavelengths 486.1 and 656.3 nm are from proton impact excitation of hydrogen (Hβ and Hα), while the forbidden 557.7, 630 and 636.4 nm emissions are due to electron impact excitation of atomic oxygen [OI]. The N2+ band emissions at wavelengths 427.8 and 470.9 nm are produced by electron or proton impact ionization excitation. Correspondingly, the emissions at 500.2 and 568 nm are from ionized nitrogen (NII). The N2 bands at 662.4, 670.5 and 676.4 nm are excited by electron impact as well as cascading from higher levels. The 589 nm emission is from an excited layer of NaI atoms located at an altitude of 80 – 105 km, known as the Sodium layer. The emission line is also present in low pressure sodium vapor lamps used as outdoor lighting, which make it useful for quantifying light pollution.
Note that the exposure time is set to 1s for all channels. Each sequence of 15 exposures starts on the minute. There is a 1 s delay between each exposure. The instrument then waits 30 s for the next sequence to start. The main reason for the above setup was to reduce data storage. It can be optimized in speed by tuning to minimum filter swap time and increasing number of channels. Exposure time can also be reduced and the detector gain increased. A conventional multi-spectral imager with fixed wavelength interference filters will not be able to compete with the operational speed of the NORUSCA II camera. It should however be noted that standard interference filters have higher transmission and more narrow bandpass. In other words, the choice of filters depends on experimental requirements. The NORUSCA II lens can easily be used with a standard filter wheel.
5. Data samples
On 24th of January 2012 a Coronal Mass Ejection (CME) occurred at the Sun. The impact on Earth at ~15:00 UT produced magnificent aurora. Figure 6 shows a 1s exposure of the event at 15:15:03 UT from the NORUSCA II camera at KHO. Center wavelength is 557.7 nm, and gain is set to 100. The image is without any sky background or dark level subtraction. The raw data counts range from 0 to 30846 CTS/s. The brightest object in the image is Jupiter. A vertical scatter plot of a 64 x 64 sub frame centered on the planet reveals that the focus is within 3 pixels. Note that the center peak is ~3 times higher than its 2 neighbors. The planet is in other words focused within ~1 pixel, if atmospheric scintillation and other blurring effects are excluded.
Next step is to form color composite images. Channels 1, 2 and 3 are selected as blue, green and red components to the RGB color image, respectively. These channels are the first 3 exposures in the sampling sequence that starts every minute. Fig. 7 shows the RGB image of the sequence that started 15:15:00 UT on 24th of January 2012. A simple camera model is applied to the image that includes the mapping function in Eq. (1) together with translation and rotation. The optical center is shifted 4 pixels to the right and 2 pixels up from the center of the EMCCD. The radius of the compass rose is 260 pixels. Geographical North is tilted 28° clockwise from the vertical. The colored scaled bars represent the intensity scale of each color channel from 0 - 15 kCTS/s. The dark frame level is at 600 CTS/s with a standard deviation ofσ = 50 CTS/s. Note that the auroral green maximum intensity or the hotspot located ~25° from zenith to the South (S) has an intensity of ~70 kR, which classifies as bright nightside aurora.
The M8.7-class solar flare caused a CME that produced planetary index Kp = 5 geomagnetic storm. The impact on Earth occurred at 15:09 UT with intense auroral activity that lasted up to two hours. See the RGB animation in Fig. 7 (Media 1) of the whole event. Note that each frame is sampled every minute. The playback speed is 15 frames per second. The aurora was even detected through a layer of low altitude clouds that drifted across the field of view from 15:45 to 17:00 UT. Just after the auroral display close to ~17 UT, the rapid playback of the animation reveals a one hour dynamic weak intensity wave pattern of unknown origin. It resembles in shape and signature airglow and gravity wave interaction down in the mesosphere (80-90 km). See Fig. 7. It could just be a coincident or one could speculate that it is formed by auroral generated waves propagating downwards. Further studies are needed to reach a conclusion on this issue.
The opposite of intense green nightside aurora is weak red dayside aurora. Figure 8 shows a typical example of dayside aurora from KHO on 29th of December 2011 at 08:55 UT. The geomagnetic conditions are now classified as normal with low auroral activity level (Kp = 2). Magnetic local noon or the center of the cusp is then theoretically located in the zenith of KHO at 08:55UT. The threshold count range for each channel in the RGB image is now set to 0 – 3 kCTS/s. The auroral forms of Fig. 8 are ~5 kR for each channel. The red [OI] 630 nm emission fills almost the entire field of view due to its long lifetime of ~110s. The green [OI] 557.7 nm and blue N2+ 470.9 nm emissions are more discrete and narrow in shape with lifetimes of ~1s and < 10−7s, respectively. Scattered sun and moon light is seen in the lower South -West (SW) to East (E) horizon. The sun and moon are 12.9° and 4.3° below the horizon during the exposures, respectively.
The Milky Way and hundreds of stars are clearly seen as they rotate in time around Polaris (α UMi) in the rapid playback of the RGB animation of Fig. 8 (Media 2). The movie covers in time the pre-noon to cusp section of the auroral oval as observed from KHO (06:00- 09:40 UT). Again, each frame is sampled every minute, and the playback speed is 15 frames per second. The open closed field line boundary of the magnetosphere is identified as the gap between the red dayside auroral forms in zenith and the green belt above the horizon in the South (S).
The minimum detection threshold signal is assumed to be 3 times the dark noise level, or 3σ = 150 CTS/s. For the green [OI] 557.7 nm emission, the minimum detection limit then becomes 550 R. In other words, the NORUSCA II camera detects auroras above ~½ of the sub visual green 1kR intensity threshold at an exposure time of only one second.
The multi-channel or hyperspectral ability of the instrument opens the opportunity to apply image processing methods such as clustering and classification . In Fig. 8 (Media 3) all 15 bands of the dayside auroral image are used as input to a Supervised Maximum Likelihood classification. The result is obtained by the software MultiSpec developed by Larry Biel based on the research of the students of Prof. David Landgrebe, all at Purdue University in USA. 7 training areas were selected to represent different classes. 3 types of aurora were identified as discrete arcs, band structure and diffuse forms. 2 sky background regions were selected. In addition, two classes represent reflected light of the mountain side of Breinosa and scattered light from the sun and moon. As expected, low energetic diffuse red aurora is classified successfully. Energetic arcs and bands are also identified. Strong stars are classified as scattered solar light. It should be noted that the arcs moves fast during the sequence of band exposures, introducing classification errors. Nevertheless, the result shows promise of new methods to analyze and process auroral data.
6. Concluding remarks
A new all-sky camera is presented with hyperspectral capability in the visible range of the electromagnetic spectrum (430 – 720 nm). The camera uses a LCTF with a nominal FWHM of 7 nm at center wavelength 550 nm. A novel high throughput C-mount all-sky lens named NORUSCA II matches the size and design of the LCTF. The camera has been tested at the Kjell Henriksen Observatory (KHO). It is capable of detecting ~½ kR of green or red auroral emissions at only 1 second exposure time. The main advantage of the system is that it requires no moving mechanical parts to select the desired center wavelength within its spectral range. The camera uses only 50 ms to swap between 41 available center wavelengths. The major disadvantage of the system is the low transmission of the LCTF, especially in the blue part of the spectrum. The hyperspectral ability of the instrument opens for new processing methods such as classification of auroral forms.
This work was financially supported by The Research Council of Norway through the project named: Norwegian and Russian Upper Atmosphere Co-operation On Svalbard part 2 # 196173 / S30 (NORUSCA2).
References and links
1. W. L. Wolfe, Introduction to Imaging Spectrometers (SPIE Press, 1997).
3. P. J. Miller, “Use of tunable liquid crystal filters to link radiometric and photometric standards,” Metrologia 28(3), 145–149 (1991). [CrossRef]
4. F. Sigernes, J. M. Holmes, M. Dyrland, D. A. Lorentzen, S. A. Chernous, T. Svinyu, J. Moen, and C. S. Deehr, “Absolute calibration of optical devices with a small field of view,” J. Opt. Technol. 74(10), 669–674 (2007). [CrossRef]
5. R. Nakatsuji, PerkinElmer - Caliper Life Sciences, 68 Elm St., Hopkinton, MA 01748, USA (personal communication, 2012).
6. D. J. Baker and G. J. Romick, “The rayleigh: interpretation of the unit in terms of column emission rate or apparent radiance expressed in SI units,” Appl. Opt. 15(8), 1966–1968 (1976). [CrossRef] [PubMed]
7. D. A. Landgrebe, Signal Theory Methods in Multispectral Remote Sensing (Wiley, 2003).