SAND project
Spectrophotometer for Aerosol Night Detection
Martin Aubé Ph.D.
Last update: 10/02/2009
Table des matières (Cacher)
- See the presentation made at 2005 Fall Meeting of the International Dark Sky Association
- Information on the 2005 intensive light pollution measurement experiment in California-Arizona-Utah
1. Objectives:
The final objective of this project is to develop a methodology to monitor aerosol optical properties during the night. This methodology benefits from the presence of aerosols in the atmosphere and their impact on the level of light pollution induced sky luminance (see project ILLUMINA). In the shorter term, we consider that the challenge of detecting light pollution represents by itself an important scientific problem.
2. SAND-1, SAND-2 and OBSAND-1
| 
|
| Figure 1: Sand-2 , Light pollution spectrophotometer Opto-mecanical design and realization: M. Aubé / M. Fréchette | Figure 2: OBSAND-1 , Light pollution spectrophotometer observatory |
In order to protect SAND-2 from possible bad weather conditions on astronomical sites, a engineering students team designed an automated observatory named OBSAND-1. ''
SAND-2 is a diffraction grating (600 lines per mm) long slit spectrometer. Spectral information is collected by a CCD camera.
3. SAND-3
We are now working on version 3 (SAND-3) of the spectrometer. In this new version, we replaced the optical head by the DSS-7 spectrometer manufactured by SBIG. Along with that modification, we designed a new portable observatory utilizing an acrylic transparent dome to protect the instrumentation from bad weather.
For each acquisition, a spectral image is produced (figure 3). On this image, spectral information is distributed on the horizontal axis (lines).
| 
|
| Figure 3: SAND-2 spectral image of light pollution above Sherbrooke , Canada as detected from the top of Mont-Mégantic (60 km flying distance). This image corresponds to an integration time of 3600 seconds. | Figure 4: Thermal noise image with an integration time of 3600 seconds and a CCD temperature of -10 o C. |
We must remove the thermal noise (dark frame). The dark frame is obtained by the acquisition of an image of the same integration time, while maintaining the obturator closed (figure 4).
| 
|
| Figure 5: Spectral image of light pollution above Sherbrooke (integration time = 3600s). This image was corrected for thermal noise. | Figure 6: Average of 509 lines of the spectral image of the light pollution from mount Palomar observatory toward Los Angeles (30 deg. elevation angle) as detetected during may 2005. A certain number of spectral lines were identified on this figure. The vertical axis represents the averaged calibrated spectral luminance whereas the horizontal axis represents the wavelength (which is related to the position in pixel on the CCD). |
The gray vertical lines on figure 5 represent emission lines of various chemical elements (mainly those contained in lamps). When the spectroscopic module is well oriented, these lines are parallel to the columns of the image. To ensured this parallelism, an image rotation algorithm is used. This last seeks the rotation angle in order to optimize the integrated columns signal.The spectrum present on each line of the image spread out from ultraviolet to near infrared. If we choose the period of observation to avoid the presence of the moon, the spectrum of each line is the sum of the spectrum of light pollution, of the spectrum of stars, the spectrum of the interstellar objects, the spectrum of the extragalactic objects and the aurorae located along the line of sight.Light pollution show a diffuse pattern, its angular variation in the sky is very weak (especially if the slit see a small fraction of the sky). On the other hand the stellar, interstellar and extragalactic contents are highly variable from one place to another along the slit.To highlight the spectrum of light pollution we are averaging several lines of the image . This operation increases the signal to noise ratio (S/N). Such an average indeed tends to emphasize the stable characteristics of the spectrum. We thus obtain a better contrast between the spectrum of light pollution and the remainder of the spectrum.
The following step consists in carrying out spectral calibration. The issue being to associate a wavelength value for each position in pixel on the image. To achieve that operation, we use a light source of which we already know a good number of lines (calibration spectrum; see figure 9). The calibration spectrum is analyzed by an multiple gaussian fitting algorithm based on a least squares optimization method (gfa.f). The algorithm generates, the position, standard deviation, the amplitude, and the integral of each Gaussian. We then carry out a statistical selection of the resulting set of lines to determine typical line width (which depends amongst other things on the width of the slit and the optics of the spectrometer). To select the lines, the programs tri.f excludes standard deviations which deviate by more than 1.5 times the standard deviation of the standard deviations. The average standard deviation is calculated from the resulting selection. This width is then imposed to all the lines of comparable width (standard deviation 0.5 standard deviation). Line width which do not enter these limits are left free. Another algorithm (ordorigin_type_lampe.f) search the theoretical spectral signature of the spectral calibration lamp according to the final lines selection. This recognition rely on the fact that we have an approximate value of the constant of proportionality between the position in pixel and the wavelength for the apparatus used. When the signature is recognized, a 2nd order polynomial is fitted to the pixel vs wavelength dataset. This polynomial is then applied to the sky spectrum.
| 
|
| Figure 7: Compact fluorescent lamp spectrum used for spectral calibration. | Figure 8: Multi-Gaussian fit (green curve) of the sky raw spectrum of the sky (not spectrally calibrated) (red curve). |
We apply also the multiple gaussian fitting algorithm to the spectrally calibrated sky spectrum in order to find the various lines. We use the previously computed standard deviation average (see figure 8).The polynomial previously found is also applied to the a filtered set (tri.f) of lines delivered by gfa.f.