The linear polarization of sunlight reflected by cloudy areas on Earth is s
ensitive to the cloud top pressure as a result of molecular scattering abov
e the clouds. We consider the derivation of cloud top pressures using polar
imetric data from satellites or aircraft. The inversion method used is base
d on adding/doubling calculations and a Newton-Raphson iteration scheme dev
eloped earlier to analyze the polarization of the planet Venus. A modificat
ion of the adding/doubling scheme is presented and used. This approach redu
ces the execution time for multiple scattering calculations with about an o
rder of magnitude. Two different atmospheric models were used. The first mo
del includes multiple scattering by a cloud layer of spherical water drops
and a higher layer of molecules and spherical aerosol particles, whereas in
the second model the reflection by the cloud layer is approximated by that
of a Lambertian surface and the aerosols in the higher layer are ignored,
thereby reducing the necessary computer time by about two orders of magnitu
de. Using simulated observations, the errors in derived cloud top pressures
due to the approximations made in the second model are compared with those
due to measurement errors. For a first application of our method to real m
easurements we used some photopolarimetric data obtained over the Atlantic
Ocean by the Global Ozone Monitoring Experiment (GOME). It is shown that fo
r the data considered the second model leads to errors in derived cloud top
pressures which are typically smaller than 80 mb. The measurement errors i
n the GOME polarization observations at about 355 and 490 nm lead to errors
in derived cloud top pressures which are typically smaller than 100 and 20
0 mb, respectively, so that the second model is sufficiently accurate to de
rive cloud top pressures from these observations. It also appears that the
measurements at about 355 nm are more suited to derive cloud top pressures
than those at about 490 nm. If more precise measurements are made, a more r
ealistic model, such as our first model, will be required. (C) 1999 Publish
ed by Elsevier Science Ltd. All rights reserved.