Processing of Images Produced from Spectral Airglow Temperature Imager

Transcription

1 Processing of Images Produced from Spectral Airglow Temperature Imager Atanas Marinov Atanassov Abstract T he Spectral Airglow Temperature Imager is an instrument for ground-based spectroscopic measurements of the night-glow atmosphere emissions. The instrument registers airglow coming from the annular mesopause segment. The temperatures of various points of this segment are retrieved by sectors of the registered images containing spectral information. The first presented preprocessing procedure is aimed for traces of cosmic ray rejection. The second preprocessing procedure represented approach for dark image correction of signal images. A next stage of SATI spectrogram processing connected with deriving of spectral information from images and the determination of the rotational temperature of oxygen molecules emitting on altitude of the mesopause is considered. The results of the original and newly-proposed algorithms for filter parameter calculation and their importance for the final results for temperature determination on the basis of the О2 ( nm) emission measurements are presented. 1 Introduction The mesosphere - lower thermosphere region within the altitude range of ( km) is an area of large-scale and intensive energy and momentum transfer (Lindzen, 1981; Fritts, 1984; Fritts, 2003; Manson and Meek, 1988) in the course of which as a result of adiabatic and dissipative processes the thermodynamic condition of the environment is affected. The physical phenomena realizing this transport are the wind and the different wave processes - tides, planetary and gravity waves (Shepherd et al., 2006). The change of the thermodynamic condition of the environment influences on airglow emission mechanism. The investigation of this glow allows us to observe the state of the atmosphere in the area where it is generated. Ground-based optical instruments, equipped with CCD detectors, have been intensively developed lately (Taylor, 1997; Taylor et al., 2007; Shiokawa et al., 1999, 2007; Bageston et al., 2007). Among these instruments, the Spectral Airglow Temperature Imager (SATI) is distinguished by its capability to simultaneously measure signals, coming from different points of the investigated envinronment- annular sky segment (Wiens et al., 1997; Sargoytchev et al., 2004). The SATI images contain interference fringes. A radial image section contains spectral information; the azimuth dimension corresponds to the azimuth of the observed sky ring. That is why the SATI instrument is suitable mainly for gravity wave investigation (Zhang et al., 1993; Aushev et al., 2008; Gavrilov et al., 2002). Except for gravity wave investigations, SATI data have been applied for studying the seasonal variations of the rotational temperatures derived by O 2 and OH filters (López-González et al., 2004). The tidal variations and the seasonal trends in their parameters - amplitudes and phases, were studied by López-González et al., (2007). Planetary scale oscillations were also studied on the basis of SATI data (López-González et al., 2009). The responses of the mesosphere/lower thermosphere to stratospheric warming events were investigated by applying SATI data (Cho et al., 2004; Shepherd et al., 2010). Moreover, SATI data have been applied to the validation of the mesopause kinetic temperatures, measured by the Sounding of the Atmosphere using Broadband Emission Radiometry (SABER) instrument on the TIMED satellite (López-González et al., 2007; Remsberg et al., 2008). SATI temperatures have been used for validation of the Michelson Interferometer for

2 2 Passive Atmospheric Sounding (MIPAS), which measures onboard of EnviSat Satellite the atmospheric infrared emission from which profiles of temperature are derived (García-Comas et al., 2011). Therefore, the development of the approach for determination of the real temperature values of the mesopause is important. The basic ideas for the instrument and data processing were developed about two decades ago at the York University - Canada (Zhang 1991; Sargoytchev et al., 2004; Cho 2006). A version of SATI-3SZ (Petkov et al., 2008) was developed at the Stara Zagora Division of the Solar- Terrestrial Influences Institute in collaboration with the Space Instrumentation Laboratory at the Centre for Research in Earth and Space Science (CRESS) at York University - Canada. The objective of this paper is to present a part of the overall work, connected with the development of the SATI data processing algorithms developed in Space Research and Technology Institute, Bulgarian Academy of Science. 2 Scheme and Operation of the SATI Instrument The Spectral Airglow Temperature Imager (SATI) instrument was first described by Wiens et al. (1997) and a more detailed updated description was given by Sargoytchev et al. (2004). In brief, the instrument consists of a conical mirror, a Fresnel lens, an O 2 interference narrow band filter at nm wavelength, an optical lens and a CCD (charge-coupled device) camera (Fig. 1) Figure 1: General scheme of SATI instrument The operation of the SATI-3 instrument is automatic and PC-controlled. The measurements start after astronomical twilight in the beginning of the night and end in the morning before sunrise, when the Moon is under the horizon. Each spectrogram image is obtained with exposure time of 120 s. One dark image follows every eight spectrogram images. 3 Rotational Temperature Determination by Spectral Images. 3.1 Preliminary Image Processing Correction of High-value Pixels Figure 2a shows an arbitrary raw spectrogram image. The Dark current image looks like the corner of a spectrogram image without any particular structures. The preliminary processing of spectrogram I and dark D image sequences is associated with denoising of instrumental white noise and salt- piper noise, result from cosmic rays. Bright points are visible resulting from electric charged particles, which have passed through the detector. The correction of the values of the so-called hot pixels and charged particles-hit pixels is made after their identification. Figure 3 shows the histograms of the pixels distribution by analog digital units (ADU) of one dark current image (a) and one of a spectrogram (b). It is seen that the main part of the dark current histogram is a symmetric Gaussian type, however, the tail to the right shows the presence of pixels with increased values namely those which are searched for. The histogram of the spectrogram image is more complex due to the image structures the low hunch refers to the area, outside the histogram, while the high hunch is for the area, containing the spectrogram. Here again there is a tail, containing the high-value pixels. The histogram analysis provides the opportunity to find the pixels, whose ADU values are increased (impulse noise). The corrected values for dark image high value pixels D m,n, which are in the tail of histogram distribution could be expressed in different ways: averaging of neighbor pixels or their median, mean values of entire image as by the value correspondent to the maximum of the histogram. High value pixel from the histogram of spectrogram image could be corrected in analogical way. Figure 2b shows an image after the application of the respective processing procedure.

3 3 Figure 2: a) Original image; b) the image after denoising; and c) outlined crests after procedures for their finding Figure 3: a) Histogram of dark image; b) Histograms of spectrogram image Dark Image Correction All dark images are summed up and averaged in the original SATI instrument data processing algorithm for producing an average dark image D. After that every raw spectrogram image I(tim) is corrected in the same way by the following dark correction equation: S m,n (t im ) = r. [ I m,n (t im ) - Dm,n ]. (1) Coefficient r is defined by a relation between the averaged pixel sums of the four corners (20 20 pixels) of the signal and the mean dark images, tim spectrogram image registration time. The histogram analysis of the dark image allows determining the most probable ADU value for each image, hence the nocturnal course of the most probable value. The amplitude of this course varies sometimes within the range of dozens of units while, in other cases it is relatively constant. The time course analysis for each pixel displays individual peculiarities with certain boundaries, which are not negligible (Atanassov, 2010). This is shown after a suitable filtration of the highfrequency noise: D'm,n (t d ) = Φ tw ( Dm,n (t d )) = Φ tw ( D0m,n (t d )+ Ψ m,n (t d )), D'm,n (t d ) D0m,n (t d ), (2) D0 is slowly varying component of dark values and Ψ is a high frequency noise, td is dark image registration time and Φ- filtering operator. ADU values for each pixel taken from all pictures in the night are filtered (filtering in time). Both the amplitude and the absolute values, as well as the course of each mage pixels are different. After determining the dark image pixels courses for all image pixels we can apply the following correction approach referring to the time tim: td tim td+1 of the spectrogram image registering: D*m,n (t im ) = Dm,n (t d )+ ( t im - t d ), (t d+1 - t d )!"D'm,n (t d+1 ) - D'm,n (t d )#$ (3)

4 Dec Temperature[ K] time [hour] To Tn Do Dn Figure 4: Retrieved temperatures calculated according to original (T0) and proposed (Tn) dark correction approaches. D0 and Dn are filtered temperature series. * D m,n (t im ) is the dark image, calculated on the basis of linear interpolation and reflects better both the specific character of the individual pixels and the time changes. We obtain the corrected spectrogram image after subtracting the calculated dark image from the original signal image: * S m,n (t im ) = I m,n (t im ) D m,n (t im ). (4) The comparison shows (fig. 4) a certain difference in determining mesopause temperatures on the basis of the presented method for dark current image correction with regards to original one (the averaging of all dark images) Spectrogram Image Filtering The spectrogram image can be presented as a sum of the real signal S 0 m,n(t im), received by the instrument and random noise Ψ m,n(t im). The use of a 2D running average is an efficient high-frequency filter, applied for image processing (Gonzalez, 2002). Regarding the connection of the amplitude-frequency characteristics of this filter to the window length, on the one side, and the character of the structures in the signal 2-D field of application, on the other hand, a two pass filtering is selected with a window of 3x3 pixels. A special effective algorithm has been designed for simultaneous double filtration. 4 Filter Parameters Determination- Refractive Index and Central Wavelength. It is possible to extract information from each spectrogram, associated on the one hand with the maximums location and, on the other - with frequency energy distribution in the respective spectral part. The filter parameters µ- refractive index and λ 0- central lambda (which are dependent on filter temperature), are defined on the basis of the maxima position (interference rings radii), by solving the inverse problem. The following relation is used (Wiens, 1991): ( ) 2 r i r 2 i + f = µ # 2 1 λ 2 i % $ λ0 2 & (, i =1,6. (5) ' where r i is the radius of the i -th ring, f is the focal distance of the objective and λ i is the respective wavelength of spectral lines of O 2 spectrum. The unknown quantities (µ,λ 0) in (5) are determined on the basis of a linear regression. Three to six interference maxima (crests) are used (the minima (valleys) are not used). The deviation of the refraction index value, determined by the abovedescribed spectrogram analysis, from the measured laboratory values (µ=2.1551; 30 C) in some cases can be of the order of a few thousands. The values of λ 0 vary in the 5 -th sign, respectively.

5 5 The following approach is applied here to determine the respective radii. First, an initial approximation of the interference rings centre O 0 (m 0,n 0 ) is selected and on the basis of the image radial sections, the locations of the respective maxima are specified: ' R k,φ = S k.cos(ϕ ),k.sin(ϕ ) ; k =1 128; ϕ = ; Δϕ =1. (6) Initially, each image is subject to noise reduction by applying a two pass moving average filter (Smith, 1999). The problem for maximums searching cannot always be solved because of residual noise presence. An approach has been developed for searching and recognition of maxima, which are not sufficiently distinguished in the presence of residual noise. Every interference maximum possesses own individuality - the central ones are better expressed, while the outer ones sometimes and in some parts of the image might appear smeared and not very clear. The search is not on the basis of comparing values of pixels in image section- it is done on the basis of relationship between neighbor pixels. Every pattern P k({b i}i=1,l) is an arranged binary multitude {b 1, b 2,,b L}, as each element describes a relation between two adjacent points (increase-decrease ) from the section. Since the search procedure is not always productive, it is necessary to reduce the criteria and more or less recognize an interference maximum in the space of relations with a suitable pattern. This approach provides the opportunity to increase the number of the interference maxima found, especially for the external rings. This is of major importance for the interference image with weaker contrast obtained with not very strong useful signal. Some additional patterns which describe real possible relations between pixels values in radial sections are applied for separation of the outer 5 -th crest from the 6 -th one. After determining the coordinates of the interference maxima within the frames of the radial sections it is necessary to allocate them according to the interference rings. This is required because the maxima of all rings are not always found. In such cases the problem is solved on the basis of the analysis of the histogram for distribution of all determined maxima. The points for each ring with their coordinates are taken around the respective maxima in the histogram (Fig. 5). Having enough available points for each ring, the Least Square Fitting (LSF) can be applied to determine the centres and their radii. A multitude of points {(x k,y k)} k, l=1, for each k -th ring is obtained, which should satisfy the condition to lie as close as possible to a circle with radius r and centre coordinates (a,b), that is (x a) 2 + (y b) 2 = r 2 have to be fulfilled. Since the determined points do not lie exactly on the circle, in order to define that circle, it is necessary to minimize the following functional: E(a,b,r) = (L l r) 2, L l = (x l a) 2 + (y l b) 2. (7) l=1 The functional E(a,b,r) is minimized by an approach, which does not lead to solving a non-linear LSF problem. An iteration procedure is applied (Eberly, 2000), which is convergent and yields very good results: L = 1 L l=1 l a = x + LL a, b = y + LL b. (8a) x = 1, L a = 1 x l=1 l l=1, y = 1 y l. (8b) l=1 a x l L l, L b = 1 l=1 b y l L l. (8c) Before starting the iteration process, as an initial approximation of the centre coordinates it is assumed that a 0 = x, b 0 = y. After completing the iteration process and defining the centre coordinates its radius is determined as: r = 1 L l. (9) l=1 The entire process of defining the ring parameters is applied iterationally for their improvement. This process requires a good initial approximation of the image centre and it is quickly convergent. The proof for the process convergence is illustrated numerically. Figure 5 shows histograms of three sequential approximations. In the first histogram the last three internal rings are well separated, in contrast to the three internal rings. The second approximation of the spectrogram centre is selected to be center of one of the three internal rings. It can be seen that the third approximation does not improve substantially the maxima separation; the process for determination of the image centre and the rings radii is considered completed. The next step is the regression problem to determine the filter parameters (µ, λ 0). Figure 6 presents the nocturnal course of the filter parameters.

6 first second third pixels Radial distance from image centre [pixels] Figure 5: Histograms of peak distributions for sequential approximations Figure 6: Refractive index µ (a) and central wavelength (b) calculated by original approach C6 and by proposed one 6; the values calculated by proposed approach are more stable. 5 Rotational temperature determination by SATI images The original approach to determine the rotational temperature of O 2 from SATI images was developed basically by Zhang (1991) and later updated by Cho (2006). The SATI instrument registered light coming from annular mesopause segment area, where the maximum of the oxygen emission is located. SATI forms images, containing rotational spectra of the oxygen molecule corresponding to transitions O 2 (b 1 Σ + g X 3 g ). Example of an image registered by SATI-3SZ instrument with O 2 ( nm) filter is shown in Fig.2a. The values of the temperature in different points on the annular mesopauze segment are determined by the spectra formed on the base of the respective sectors of the image (fig. 8b). The used image sectors are determined by the angles γ and θ. The angle γ determines the sector width while the angle θ determines the orientation towards a selected direction. Every sector spectrum represents a series of values M p, each produced by aver-

7 7 Figure 7: Temperature courses for some sectors retrieved by filter parameters calculated by original C6 and proposed 6 approaches. aging the values of all the pixels of the sector, located at the distance p from the image center. The determination of the rotational temperature is achieved by comparing the measured spectrum M with synthetic spectra S Trot preliminary calculated for different temperature of the emitted gas and convolved with a filter passband function. The set of 190 synthetic spectra for interval of K is calculated with wavelength step 0.01 nm. The spectra S Trot are transformed to image space as spectra S Trot by using filter parameters µ and λ 0 before each comparison. The comparison between the measured spectrum M and the calculated and transformed synthetic spectrum S Trot is based on the regression equation: M = E. S Trot + B, (10) where E is an integral emission intensity for the filter transmittance interval and B is a mean background intensity. The solution of the regression equation (1) gives the values of E and B. Consecutively replacing with different synthetic spectra S Trot in (1) a minimum of functional is achieved:

8 8 δ Trot = p 2 1 p 2 p (M p E. S Trot,p B) 2.w p. (11) 1 p=p 1 The quantity S Trot,p = E. S Trot,p B in (2) is a spectrum similar to the respective synthetic spectrum for T rot. The weight coefficients w p are important for the minimization process and can be used as control parameters for calibration of the calculation algorithms. The temperature is found by iterative replacement of different spectra S Trot sequentially in (1) and (2) and determination of minimal Minkovsky distance between the measured M and S spectra. 6 Sector spectra derivation Temperatures in different points of annual sky segment must be determined for investigation of gravity waves propagation in mesosphere. Spectra from different image sector (fig. 8) must be determined to this aim. The image is divided into 12 sectors with angle g of 30 deg in the classic algorithm (Zhang, 1991). Figure 8. Sector of image and respective sector specter. 6.1 Moving sector approach with averaging Atanassov (2009) proposed a slightly different approach for sector spectra determination. Each sector is determined by an azimuthal angle θ and angle γ between the straight lines, determined by the following equations: an even distance from the image center. The values of angle g may vary in intervals of dozens of degrees. The variations of angle g determine the curve-linear calculation pattern (window) for averaging the pixel values. In order to choose the pixels restricted by the two segment radii, their indices are changed, depending on the following versions (Fig. 9): Version I: the sector is entirely in one of quadrants I, II, III or IV only and we can write down for index by Ox axis: i ( IT ( sing(1,cosϕ 1 )),IT ( R *MAX(cosϕ 1,cosϕ 2 ))), for cosϕ 1 > 0 i ( IT ( sing(1,cosϕ 1 )),IT ( R *MI(cosϕ 1,cosϕ 2 ))), for cosϕ 1 > 0 Version II: the sector falls into I and IV or II and III quadrants simultaneously. Then for the changing of the index by Ox axis: ( ( )). i IT ( sign(1,cosϕ 1 )),IT sign(r,cosϕ 2 ) We will note only that in the versions described here, sign(cos(ϕ 1 )) = sign(cos(ϕ 2 )). The function IT() is intrinsic function in the Fortran programming language which returns the nearest integer to the argument [8]. The functions MAX() and MI() return the maximum or minimum value respectively of the arguments. For the change of index by Oy axis in the above two versions we can write down j ( IT(k 1.i),IT(k 2.i)),k 1.i < k 2.i or j ( IT(k 2.i),IT(k 1.i)),k 2.i < k 1.i y = k 1 x, k 1 = tgθ y = k 2 x, k 2 = tg(θ +γ ) (12) Version III: the sector falls into I and II or III and IV quadrants simultaneously. Then: The determination of sector spectra for arbitrary azimuthal angle θ i+1 =θ i + Δθ is possible. All the values of sector pixels, located at the distance p from the image center are summed up and averaged. The determination of the sector spectra (Fig. 8) from sector with various angles g has an impact on the pattern used to average the values of the pixels located at i ( IT(R.cosϕ 1 ), IT(R.cosϕ 2 )),cosϕ 1 < cosϕ 2 or i ( IT(R.cosϕ 2 ), IT(R.cosϕ 1 )),cosϕ 1 > cosϕ 2 For the index change along the Oy axis in the last third version we have for cosϕ 1 < cosϕ 2 :

9 9 i (IT(R.cosφ 1,0) j (k 1.i, R 2 i 2 ) i [0,IT(R.cosφ 2 ) j (k 2.i, R 2 i 2 ) Analogously, the boundaries in which the indices change and by cosϕ 1 > cosϕ 2 can be determined. The distance of every pixel to the image centre, as above, determines the process of summing and averaging: S p = 1 I i,j,p = IT( i 2 + j 2 ) M p where S p is the p th element of the sector spectrum, M p is the number of all pixels at a distance of p pixels from the image centre. Actually, this approach for averaged sector spectra calculation like the original one with twelve sectors only, represents one-dimensional filtering with one of the most simple and widely used filter (Gonzalez & Woods, 2002), however, applied on a series of elements (pixels) disposed on curved lines- circles. What is special in our case is the formation of one-dimensional (curvilinear by circle) pattern on two-dimensional image for simultaneous calculation of the whole spectrum. Figure 10 shows temperature night courses calculated for sectors with different widths. Figure 9 Versions of samples for determination of sector spectra; 1- rejected pixels, 2- inner pixels and 3-pixel in the sector column. Figure 10. ight temperature trends for one sector retrieved by different sector angles (5, 15, 30, 45 and 60 deg). 6.2 Moving sector approach with median An algorithm for sector spectrum calculation is proposed by Atanassov (2010), where the sectors are determined by using an approach analogous to those presented in section 5.1. However, the intensities of the measured spectrum, instead of summing up and averaging, are determined analogously to the median filtering - the values of all the pixels located at the distance p from the image center are sorted and the value of the median element in the series of values is taken. Simultaneously with the determination of the distance for each pixel towards the image center, their values are stored in rows of a 2-dimensional area W q, p. Each column p of the area contains the read values for all pixels located at distance p towards the image center. The sizes of the area are determined so as all measured values of the pixels in the sector with radius R(~125p) and angle γ (5 40 ) to be stored. The application of a bidimensional structure presumes a non-efficient utilization of the storage. However, this approach is suitable because the number of pixels is small, additional pointers are not used, the realization is simple, and the sorting is effectively applied. Tentatively, dimension q of area W is determined as q = γ.r + Δ,

10 10 Figure 11. (a) octurnal trends of temperatures T o, T m and T cor for January 30, (b) Azimuthal distribution of the sector temperature T rot;γ,θ for one image (21,971[UT]). The differences between retrieved temperatures T m and T cor (a) are a result of countering sector temperatures with less sector background. (c) Azimuthal distribution of sector errors for the same image. (d) Azimuthal distributions of sector emission intensities (E) and backgrounds (B). where Δ is an additional quantity. The standard runtime subroutine sortqq() from the Visual Digital Fortran library is used to sort the values in the W columns. The results from this approach are very similar to the results produced by averaging. It may be applied without image pre-processing for cosmic ray rejection, which follows from properties of the median filtering (Gonzalez and Woods, 2002). 6.3 Determination of mean temperature In the classic algorithm (Zhang, 1991) the mean temperature T 0 and the intensity of emission are determined on the base of the mean spectrum M 0, produced by averaging the pixels located at the distance p (in pixels) from image center for the whole image. An approach for mean temperature T m determination based on averaging

11 11 of sector temperatures follows: T rot;γ, θ is hereby proposed as T m = Δθ 2π 2π Δθ T rot;γ,θ, where Dq is the step by azimuth. The size of Dq is selected so that the number of sectors is large enough (~100). The determination of the sector temperatures (fig. 11b) takes place simultaneously with the sector errors (fig 11c), sector emission and sector background (fig. 11d). The presence of special features (structures, asymmetry) in azimuthal distribution of the sector emissions and background may be applied by the evaluation of the usability of some parts of the images (some sector temperatures). The used sector background as a criterion for using respective sector temperature for mean temperature T cor calculation is illustrated in figure 11a. The night course of the mean temperature T 0, determined according to the original approach based on the mean spectrum and T m, determined by averaging of sector temperatures are shown. Calculated mean temperatures T cor based only on the temperatures of sectors for which the respective sector backgrounds are less than the average value of the whole image are shown too. Systematic enhancement of temperatures T cor in comparison with T 0 in range of 5 10K is available. θ=0 7 Conclusion An approach for SATI spectrogram processing was presented. It contains several stages - preliminary image processing (energy particles trace rejection, dark image correction of spectrogram image), filter parameters and sector spectra determination. The significance of algorithms for each stage of image processing for final results - mesopause temperatures determination is shown. The proposed dark image correction leads to a systematic reduction of calculated mean mesopause temperatures about 5K for a given datasets. The approach for filter parameters determination reflects the sector temperatures calculation. Systematic enhancement of retrieved temperatures up to 10 20K is apparent. Other distinguished characteristics of proposed approach are related to the absence of gross errors. The moving sector approach gives the possibilities to detail calculation of azimuthal distributions of sector temperatures, emission intensities and backgrounds. This allows estimation of the data quality and results in the course of data processing. Azimuthal distribution of sector temperatures, backgrounds, emission intensities and errors suggest ways to improve the calculated mean temperature. The development of SATI processing algorithms improves precession of mesopause temperature determination and will allow new more profound experiments about mesosphere dynamic to be conducted in the future. Acknowledgement The author would like to thank Dr. M.G. Shepherd from the Centre for Research in Earth and Space Science (CRESS) of York University in Toronto, Canada, for support and encouragement by algorithms development. The author thanks Mrs. M. ikolova for the technical assistance. Author Biography Atanas Marinov Atanassov At_M_Atanassov@yahoo.com Stara Zagora Branch Space Research and Technology Institute, Bulgaria References Atanassov, A. (2009). Algorithm for sector spectra calculation from images registered by the Spectral Airglow Temperature Imager, in: Proc. Int. Conf. Fundamental Space Research, pp Available from: <arxiv: v2> Atanassov, A., (2010) Dark image correction of spectrograms produced by SATI instrument. Compt. rend. Acad. bulg. Sci., 63, o 4, ,. Atanassov, A.(2010) Median Algorithm for Sector Spectra Calculation from Images Registered by the Spectral Airglow Temperature Imager, in: Proc. Int. Conf. Space, Ecology, Safety,, Available from: <arxiv: > Atanassov, A.. (2011). Comparison of Calculation Models for Determination of the Mesopause Temperature using SATI Images, Adv. Space Res. v. 47(11), Aushev, V. M., Lyahov, V. V., López-González, M. J., Shepherd, M. G., Dryn E. A. (2008). Solar eclipse of the 29 March 2006: Results of the optical measurements by MORTI over Almaty (43.031, E). J. Atmos. Sol. Terr. Phys. 70, Bageston, J.V., Gobbi, D., Takahashi, H., Wrasse, C. M. (2007). Development of airglow OH temperature imager for mesospheric study. Revista Brasileira de Geofísica, 25(Supl. 2), 27-34,

12 12 Cho, Y-M., Studies of the Polar MLT Region Using SATI Airglow Measurements, PhD Thesis, York University, Fritts, D. C., Gravity wave saturation in the middle atmosphere: A review of theory and observations, Rev. Geophys., 22(3), , Fritts, D.C. (1984). Gravity wave saturation in the middle atmosphere: A Review of Theory and Observations, Rev. Geophys., 22(3), García -Comas, M., Funke, B., López -Puertas, M., Bermejo- Pantaleón, D., Glatthor,., von Clarmann, T., Stiller, G., Grabowski, U., Boone, C.D., French, W.J.R., Leblanc, T., López-González, M.J. and Schwartz M.J. (2011). On the quality of MIPAS kinetic temperature in the middle atmosphere, Atmos. Chem. Phys. Discuss., 11, Gavrilov,,M., Shiokawa, K. & Ogawa T. (2002). Seasonal variations of medium-scale gravity wave parameters in the lower thermosphere obtained from spectral airglow temperature imager observations at Shigaraki, Japan, J. Geophys. Res., v. 107, O. D24, Glasbey C. A., R. Jones, Fast computation of moving average and in octagonal windows. Pattern Recognition Letters, 18, , moje da se mahne! Gonzalez R., R. Woods, (2002), Digital Image Processing, sec. ed. Prentice Hall, ew Jersey, 793. Eberly, D., (2000). "3D Game Engine Design", Morgan Kaufmann Publishers, San Francisco, CA , September Lindzen, R. S. (1981). Turbulence and stress owing to gravity wave and tidal breakdown. J. Geophys. Res., 86, López-González, M. J., Rodríguez, E., Wiens, R.H., Shepherd, G.G., Sargoytchev, S., Brown, S., Shepherd, M.G., Aushev, V. M., López-Moreno, J.J., Rodrigo, R., Cho, Y-M. Seasonal variations of O2 atmospheric and OH(6-2) airglow and temperature at mid-latitudes from SATI observations. Ann. Geoph., 22, , López-González, M. J., García-Comas, M., Rodríguez, E., López- Puertas, M., Shepherd, M.G., Shepherd, G.G., Sargoytchev, S., Aushev, V.M., Smithd, S.M., Mlynczak, M.G., Russell, J.M., Brown, S., Cho, Y.-M., Wiens R. H., Ground-based mesospheric temperatures at mid-latitude derived from O2 and OH airglow SATI data: Comparison with SABER measurements. J. Atmos. Sol. Terr. Phys. 69, , López-González, M. J., Rodríguez, E., García-Comas M., Costa, V., Shepherd M.G., Shepherd G.G., Aushev, V. M., Sargoytchev, S., Climatology of planetary wave type oscillations with periods of 2 20 days derived from O2 atmospheric and OH(6-2) airglow observations at mid-latitude with SATI. Ann. Geophys., 27, , Manson, A.H., Meek, C.E., Gravity wave propagation characteristics ( km) as determined by the Saskatoon MF radar (Gravnet) system: at 52, J. Atmos. Sci., 45(6), , Manson, A.H., Meek, C.E. (1988). Gravity wave propagation characteristics ( km) as determined by the Saskatoon MF radar (Gravnet) system: at 52, J. Atmos. Sci., 45(6), Petkov., Atanassov, A., Benev, B., Kanev, K., Hristov, G., Bankov, L., Sargoytchev, S., Shepherd, M. SATI-3SZ - a Spectral airglow temperature imager in Stara Zagora station: possibilities and first results, Proceedings of conference Fundamental Space Research, Sunny Beach, Bulgaria, Sep 2008, , Available from: <SATI-3 SZ/FSR> Remsberg, E.E., Marshall, B.T., Garcia-Comas, M., Krueger, D., Lingenfelser, G.S., Martin-Torres, J., Mlynczak, M.G., Russell III, J.M., Smith, A.K.,Zhao, Y., Brown, C., Gordley, L.L., Lopez-Gonzalez, M. J., Lopez-Puertas, M., She, C.-Y., Taylor, M.J. and Thompson, R.E. (2008). Assessment of the quality of the Version 1.07 temperatureversus- pressure profiles of the middle atmosphere from TIMED/SABER, J. Geophys. Res., 113, D17101, doi: /2008jd Sargoytchev, S., Brown, S., Solheim, B.H., Cho, Y.-M., Shepherd, G.G., López-González, M.J. (2004). Spectral airglow temperature imager (SATI) a ground based instrument for temperature monitoring of the mesosphere region. Appl. Opt. 43, Shepherd, G.G. Spectral Imaging of the Atmosphere. Academic Press, International Geophysics Series, v. 82, Shepherd G.G., Cho, Y-M., Liu, G., Shepherd, M. G., Roble, R.G. (2006). Airglow variability in the context of the global mesospheric circulation. J. Atmos. Sol. Terr. Phys. 68, Shiokawa, K., Katoh, Y., Satoh, M., Ejiri, M.K., Ogawa, T., akamura, T., Tsuda, T., Wiens, R.H. (1999). Development of optical mesosphere thermosphere imagers (OMTI). Earth Planets Space, 51, Shiokawa, K., Otsuka,Y., Suzuki, S., Katoh, T., Katoh, Y., Satoh, M., Ogawa, T., Takahashi, H., Gobbi, D., akamura, T., Williams, B.P., She, C.-Y., Taguchi, M., & Shimomai, T. (2007). Development of airglow temperature photometers with cooled-ccd detectors. Earth Planets Space, 59, Smith W.S. (1999). Digital Signal Processing, Second Edition, San Diego, California Technical Publishing. Taylor, M. J. ( 1997). A review of advances in imaging techniques for measuring short period gravity waves in the mesosphere and lower thermosphere, Adv. Space Res. Vol. 19(4), Taylor, M. J., Pendleton, W. R. Jr., Pautet, P.-D., Zhao, Y., Olsen, C., Babu, H. K. S., Medeiros, A. F., Takahashi, H. (2007). Recent progress in mesospheric gravity wave studies using nightglow imaging systems. Revista Brasileira de Geofísica, 25(Supl. 2),

13 Wiens R.H., Zhang, S.-P. Peterson R.. & Shepherd, G.G..(1991) MORTI: a mesopause oxygen rotational temperature imager, Planet Space Sci., 39, Wiens R. H., Moise, A., Brown, S., Sargoytchev, S., Peterson, R.., Shepherd, G.G., López-González, M.J, Lopez-Moreno J.J. & Rodrigo R. (1997). SATI: a spectral airglow temperature imager, Adv. Space Res., 19, Zhang S. P. (1991). Gravity Waves from O2 Airglow, PhD Thesis, York University. Zhang, S.P., Peterson, R.., Wiens, R. H., Shepherd G. G. (1993). Gravity waves from O2 nightglow during the AIDA 89 campaign I: emission rate/temperature observations. J. Atmos. Terr. Phys., 55 (3),