course on inverse problems
play

Course on Inverse Problems Albert Tarantola Lesson XIII: - PowerPoint PPT Presentation

Institut de Physique du Globe de Paris & Universit Pierre et Marie Curie (Paris VI) Course on Inverse Problems Albert Tarantola Lesson XIII: Optimization and Linear Problems (Examples) 98 Linear Least Squares 4.3 The Gomos Instrument


  1. Institut de Physique du Globe de Paris & Université Pierre et Marie Curie (Paris VI) Course on Inverse Problems Albert Tarantola Lesson XIII: Optimization and Linear Problems (Examples)

  2. 98 Linear Least Squares 4.3 The Gomos Instrument in the Envisat Satellite As explained in Tamminen (2004) 4 , the GOMOS (Global Ozone Monitoring by Occultation on Stars) satellite is one of the 10 instruments onboard the ESA’s Envisat satellite (see fig- ure 4.21), which, in a polar sun-synchronous orbit at about 800 km above the Earth, is tar- geted on studying the Earth’s environment. The main objective of GOMOS is to measure the atmospheric composition and especially the ozone concentration in the stratosphere and mesosphere with high vertical resolution. In addition to ozone (O 3 ), the UV–visible spec- trometer (250–675 nm) can be used to detect also NO 2 , NO 3 , aerosols, and neutral density 5 . The GOMOS instrument uses the stellar occultation technique. The measurement principle is quite simple: the stellar spectrum measured through the atmosphere is compared with the reference spectrum measured above the atmosphere (see figure 4.22). Due to the absorption and scattering in the atmosphere the light measured through the atmosphere is attenuated and the attenuation is proportional to the amount of constituents in the atmosphere. The measurements are repeated at different tangential altitudes to obtain vertical profiles of the concentrations of different atmospheric constituents. Each occultation consists of about 70– 100 spectra measured at different tangential altitudes and each spectrum includes measure- ments at 1416 different wavelengths. 4.3.1 Simplified Theory Assume we only use one star, whose spectrum, measured far from Earth’s atmosphere, is S 0 ( λ ) . The spectrum S n ( λ ) of the n -th ray R n (a ray possibly passing through the atmo- sphere) can be computed as � � � d ℓ ∑ S n ( λ ) = S 0 ( λ ) exp α gas ( λ ) ρ gas ( z ( ℓ ) , θ ( ℓ ) , ϕ ( ℓ ) ) − , (4.20) R n gas where the coefficient α gas ( λ ) measured in the laboratory, represents the capacity of each gas in absorbing the signal in a given wavelength, and ρ ( z , θ , ϕ ) is the concentration (assumed small) of a given gas at point ( z , θ , ϕ ) in the atmosphere. The integral is along the ray path, and d ℓ represents the length element along the ray path. The path of each ray is, of course, refracted by the atmosphere, but, to simplify this numerical example we shall assume below that that rays are straight. One introduces the spectral ratio T n ( λ ) ≡ S n ( λ ) , (4.21) S 0 ( λ ) and, because we prefer Cartesian parameters 6 , the logarithmic spectral ratio t n ( λ ) ≡ log T n ( λ ) = log S n ( λ ) . (4.22) S 0 ( λ ) 4 I borrowed the idea of this exercise from a dissertation of Johanna Tamminen (2004), posted in the web. This short section (i.e., section ?? ) is obtained by “copy and paste” from her text. But I develop the formulas of the sections below in my way —and adapt them to a simplified version of the GOMOS experiment— and use the simple least-squares technique instead of the Monte Carlo Markov chain used by Tamminen. 5 Two infra-red channels are used to detect O 2 and H 2 O. 6 The concentration is not a Cartesian parameter, but let us avoid this complication here.

  3. 4.3 The Gomos Instrument in the Envisat Satellite 99 Figure 4.21: The ENVISAT satellite, with its ten instru- ments. Equation (4.20) then gives � d ℓ ∑ t n ( λ ) = − α gas ( λ ) ρ gas ( z ( ℓ ) , θ ( ℓ ) , ϕ ( ℓ ) ) , (4.23) R n gas which is our basic equation, solving the forward modeling problem. We shall see in the course, that it is possible to set the inverse problem taking as an unknown the function ρ ( z , θ , ϕ ) , but let us use here a simplified approach, where the Earth’s atmosphere is divided into a certain number of three-dimensional cells. We can then write α gas ( λ ν ) ρ gas t n ( λ ν ) = − ∑ ℓ n p ∑ q ∑ pqr ∑ , (4.24) pqr r gas where the indices { p , q , r } identify the cells (the index p may represent the altitude of a cell, the index q its latitude, and the index r its longitude), where ℓ n pqr is the length of the n -th ray inside the cell { p , q , r } , and where ρ gas pqr represents the concentration of each gas inside each cell (these quantities are now our basic unknowns). To prepare for subsequent notations, let us from now just write g instead of “gas”. Our last equation then becomes α g ( λ ν ) ρ g t n ( λ ν ) = − ∑ ℓ n p ∑ q ∑ pqr ∑ . (4.25) pqr r g Equivalently, we can write pqr α g ( λ ν ) ρ g t n ( λ ν ) = − ∑ ℓ n p ∑ q ∑ r ∑ , (4.26) pqr g i.e., K ng ν pqr ρ g t n ( λ ν ) = + ∑ p ∑ q ∑ r ∑ , (4.27) pqr g where I have introduced the coefficients K ng ν pqr = − ℓ n pqr α g ( λ ν ) . (4.28)

  4. 100 Linear Least Squares Equation (4.27) expresses a linear relation between the observable parameters { d i } ≡ { t n ( λ ν ) } and the model parameters { m µ } ≡ { ρ g pqr } , a relation that, below, shall be written using the two following forms: d i = ∑ G i µ m µ ; d = G m . (4.29) µ 4.3.2 Simplified Problem To simplify this numerical problem, let us disregard the variation of the gas concentration as a function of latitude and longitude, by assuming that it depends only on height (see figure 4.22). Equation (4.27) then becomes K ng ν p ρ g t n ( λ ν ) = + ∑ p ∑ , (4.30) p g with the coefficients K ng ν p = − ℓ n p α g ( λ ν ) . (4.31) The index p is the same as above (it represents the p -th layer of the atmosphere). satellite orbit S 0 ( λ ) Figure 4.22: As a first approximation, dis- regarding the variation of the concentration S n ( λ ) of different gases as a function of longitude and latitude, one may use the GOMOS in- strument to measure the variation as func- tion of height (figure adapted from Tammi- T n ( λ ) nen, 2004). 4.3.3 Numerical Application ❊①❡❝✉t❛❜❧❡ ♥♦t❡❜♦♦❦ ❛t ❤tt♣✿✴✴✇✇✇✳✐♣❣♣✳❥✉ss✐❡✉✳❢r✴⑦t❛r❛♥t♦❧❛✴❡①❡r❝✐❝❡s✴❝❤❛♣t❡r❴✵✹✴●❖▼❖❙✳♥❜ For the numerical application, we shall just consider two atmopheric layers and three gases in each layer (see figure 4.23), so we have six model parameters. We shall assume that we measure two rays, and, for each ray, the spectral amplitude at two wavelengths, so we have four observations. We have less observations than model parameters, but the explicit use of some a priori information makes this issue irrelevant. The linear system (4.30) now becomes, explicitly, ρ 1   1 t 1 ( λ 1 ) K 11 K 11 K 12 K 12 K 13 K 13 ρ 1       11 12 11 12 11 12 2   t 1 ( λ 2 ) K 11 K 11 K 12 K 12 K 13 K 13 ρ 2       21 22 21 22 21 22 1    = . (4.32)       t 2 ( λ 1 ) K 21 K 21 K 22 K 22 K 23 K 23 ρ 2       11 12 11 12 11 12 2      t 2 ( λ 2 ) K 21 K 21 K 22 K 22 K 23 K 23 ρ 3   21 22 21 22 21 22 1   ρ 3 2

  5. 4.3 The Gomos Instrument in the Envisat Satellite 101 n = 2 c λ = λ 1 , λ = λ 2 Figure 4.23: Artificial numerical ex- b a b λ = λ 1 , λ = λ 2 n = 1 ample, in an imaginary Earth (small globe and large athmosphere). Two rays available, with two wavelengths a = 10.5 m observed for each ray (four data val- b = 4.2 m ues). Two athmospheric layers, with c = 11.2 m three gases in each layer (six model pa- rameters). Let us introduce the vector d representing the four observable quantities, t 1 ( λ 1 )   t 1 ( λ 2 )   d = , (4.33)   t 2 ( λ 1 )     t 2 ( λ 2 ) the vector m representing the six model parameters, ρ 1   1 ρ 1   2   ρ 2   1   m = , (4.34)   ρ 2   2   ρ 3   1   ρ 3 2 and the “sensitivity” matrix K 11 K 11 K 12 K 12 K 13 K 13   11 12 11 12 11 12 K 11 K 11 K 12 K 12 K 13 K 13   21 22 21 22 21 22 G = , (4.35)   K 21 K 21 K 22 K 22 K 23 K 23   11 12 11 12 11 12   K 21 K 21 K 22 K 22 K 23 K 23 21 22 21 22 21 22 so we can write the forward modeling relation (4.32) as d = G m . (4.36) Using equation (4.31), we can explicitly write the sensitivity matrix: − a α 1 ( λ 1 ) − 2 b α 1 ( λ 1 ) − a α 2 ( λ 1 ) − 2 b α 2 ( λ 1 ) − a α 3 ( λ 1 ) − 2 b α 3 ( λ 1 )   − a α 1 ( λ 2 ) − 2 b α 1 ( λ 2 ) − a α 2 ( λ 2 ) − 2 b α 2 ( λ 2 ) − a α 3 ( λ 2 ) − 2 b α 3 ( λ 2 )   G = .  − c α 1 ( λ 1 ) − c α 2 ( λ 1 ) − c α 3 ( λ 1 )  0 0 0   − c α 1 ( λ 2 ) − c α 2 ( λ 2 ) − c α 3 ( λ 2 ) 0 0 0 (4.37)

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend