Rotational Raman scattering in the Earth's atmosphere ... - SRON
Rotational Raman scattering in the Earth's atmosphere ... - SRON
Rotational Raman scattering in the Earth's atmosphere ... - SRON
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Accurate model<strong>in</strong>g of spectral f<strong>in</strong>e-structure <strong>in</strong> Earth radiance spectra measured with GOME 89<br />
vectors v 1 ,...,v M describe <strong>the</strong> row space of <strong>the</strong> retrieval. The solution (4.14) is also known as <strong>the</strong><br />
solution of m<strong>in</strong>imum length (see e.g. Menke [1989]) and can be written as<br />
x est = K T ( KK T) −1<br />
y , (4.15)<br />
which is useful for any numerical implementation. For GOME measurements x est is strongly affected<br />
by <strong>the</strong> measurement error e sun <strong>in</strong> Eq. (4.12). This is caused by those terms <strong>in</strong> Eq. (4.14) that belong<br />
to small s<strong>in</strong>gular values. Here <strong>the</strong> noise on <strong>the</strong> measurement y gets amplified and <strong>the</strong>refore has a<br />
significant contribution to x est . However, ow<strong>in</strong>g to <strong>the</strong> spectral smooth<strong>in</strong>g that is part of <strong>the</strong> forward<br />
model of <strong>the</strong> Earth radiance spectrum <strong>in</strong> Eq. (4.12), <strong>the</strong>se error contributions are smoo<strong>the</strong>d out and<br />
thus have only a m<strong>in</strong>or effect on <strong>the</strong> correspond<strong>in</strong>g simulations of an Earth radiance spectrum.<br />
The m<strong>in</strong>imum length solution <strong>in</strong> Eq. (4.14) obviously does not conta<strong>in</strong> <strong>in</strong>formation of <strong>the</strong> solar<br />
spectrum <strong>in</strong> <strong>the</strong> null-space components. S<strong>in</strong>ce <strong>the</strong> null-space components describe <strong>the</strong> f<strong>in</strong>e spectral<br />
structures of <strong>the</strong> solar spectrum, <strong>the</strong> retrieved solar spectrum is a smooth version of <strong>the</strong> highresolution<br />
solar spectrum. This smooth<strong>in</strong>g is expressed by <strong>the</strong> so-called averag<strong>in</strong>g kernel A of <strong>the</strong><br />
retrieval [Rodgers, 2000]. For <strong>the</strong> m<strong>in</strong>imum length solution <strong>the</strong> averag<strong>in</strong>g kernel is given by<br />
A =<br />
M∑<br />
v i vi T . (4.16)<br />
i=1<br />
Thus, <strong>the</strong> relation between <strong>the</strong> true solar spectrum and <strong>the</strong> retrieved spectrum is<br />
x = Ax true + e x . (4.17)<br />
Here, e x is <strong>the</strong> error on <strong>the</strong> retrieved solar spectrum. The averag<strong>in</strong>g kernel A represents <strong>the</strong> projection<br />
of <strong>the</strong> state vector onto <strong>the</strong> row space (i.e. <strong>the</strong> non-null-space) of <strong>the</strong> <strong>in</strong>version problem. On <strong>the</strong> o<strong>the</strong>r<br />
hand, (1−A) represents <strong>the</strong> projection onto <strong>the</strong> null-space, where 1 is <strong>the</strong> identity matrix.<br />
Figure 4.4 shows <strong>the</strong> null-space component and <strong>the</strong> row space component of a solar spectrum<br />
retrieved from a simulated GOME solar measurement. Here we used <strong>the</strong> solar spectrum of Chance<br />
and Spurr [1997] as <strong>the</strong> true high-resolution solar spectrum x true . The figure shows that <strong>the</strong> null-space<br />
part (1−A)x true represents a significant component of <strong>the</strong> true solar spectrum. The rms of <strong>the</strong> nullspace<br />
component, related to <strong>the</strong> mean solar spectrum, is 23.4%. By def<strong>in</strong>ition, this null-space part is<br />
mapped to zero <strong>in</strong> <strong>the</strong> forward simulation of <strong>the</strong> GOME solar irradiance spectrum.<br />
4.4.2 Undersampl<strong>in</strong>g error<br />
If <strong>the</strong> solar spectrum is well sampled, <strong>the</strong> null-space contribution is mapped to zero for any sampl<strong>in</strong>g<br />
of <strong>the</strong> slit-averaged spectrum. This is also true for a sampl<strong>in</strong>g of <strong>the</strong> Earth radiance spectrum, which<br />
is subject to a similar but slightly stronger spectral smooth<strong>in</strong>g (see Fig. 4.1). However, it is known<br />
that <strong>the</strong> spectral sampl<strong>in</strong>g of GOME is too coarse, which results <strong>in</strong> a model bias for <strong>the</strong> simulation of<br />
<strong>the</strong> Earth’s radiance spectra of GOME. This error is called <strong>the</strong> undersampl<strong>in</strong>g error of GOME and has<br />
to be dist<strong>in</strong>guished from <strong>the</strong> <strong>in</strong>terpolation error described <strong>in</strong> Section 4.3.