This thesis addresses two major problems in the field of radiative transfer (RT) in the
earth’s atmosphere. The first problem is linked with the need for significant computational
resources of RT in a three-dimensional (3D) atmospheric model. Although only highly
efficient one-dimensional (1D) RT models are employed for each pixel of the model domain
separately and independently, it is still not possible to utilize these models on a frequent
basis, compared to the rate at which meteorological variables are computed. That means
that the calculated radiative properties (RP) are held constant for a longer period of time,
while the prognostic meteorological variables are updated at a rapid rate. Even though
there is no detailed study about the consequences of this disproportion, an attempt was
made to develop an RT model which permits the fast computation of basic radiative transfer
properties which could be used in the future to update this information more frequently.
The developed model is based on the application of the radiative transfer perturbation
theory to realistic cloud fields column by column. It turned out that the application,
intended to replace the Independent Pixel Approximation (IPA), see below, is possible
and promising within the assumptions and constraints of the utilized methods. It could
be demonstrated that, depending on the actual case, errors in the pixel transmission and
reflection stay bounded to values of up to 10%−15%. In one case the achieved acceleration
could be investigated. It was about a factor of four compared to the direct application of
the usual forward variant of the model, although no numerical optimization was carried
out.
The second problem concerns the realistic treatment of the 3D interactions of clouds and
solar radiation. As implied in the above paragraph, 1D RT models are usually employed
column by column which suppresses the exchange of radiation between those columns.
Thus, fundamental 3D effects are neglected by this so-called Independent Pixel Approximation
(IPA). These comprise not only small scale contributions due to diffuse radiative
transport, but also large scale patterns like geometric effects of the inclined solar illumination.
Examples are blurred radiative structures due to radiative smoothing and the
shifted location of shadows and bright areas. To parameterize those effects strong efforts
have been undertaken during the last couple of years. However, no method has proven to
be completely satisfactory and ready for implementation. To carry this research one step
further two approaches have been adopted and extended. The first is the concept of the
Tilted Independent Pixel Approximation (TIPA). In contrast to the IPA, which ignores the
solar geometry, this method correctly accounts for the slant illumination due to the correct
tracking of the direct beam. As a result, the optical parameters in the slant columns are
arranged in a more realistic order and the attenuation and the positions of the RP are less
erroneous. To further improve this method a transformation has been developed which
yields 3D resolution of the RP in the original grid. Since the TIPA still does not include
any diffuse radiative exchange as another approach the Nonlocal Independent Pixel Approximation
(NIPA) has been explored. This technique uses 1D results and carries out
a convolution product to distribute RP across column boundaries. In order to arrive at
a fully independent treatment of this method a simplified derivation of the convolution
parameters was developed. Finally, TIPA and NIPA are combined to form NTIPA. These
approaches have proven to be superior to IPA with respect to several aspects. The improvement
ranges from several percent to 50% if maximum errors of the transmitted and
reflected light are considered. Criteria like the distribution of the errors or the vertical
profiles of the RP are also more preferable than their counterparts derived by IPA.