Clouds
exhibit fluctuations of their optical properties (local extinction,
effective radius, ice crystal shape) and geometrical characteristics (top
bumps, gap or fractional coverage, shape or structure) at different scales.
The way these spatial inhomogeneities affect the radiative transfer
(radiative fluxes, actinic fluxes, and radiances) is one of the major issues
of atmospheric radiation theory, for the direct problem (fast and accurate
computation of radiative properties when cloud parameters are known) as well
as for the inverse problem (estimation of cloud parameters from radiometric
data for example). The resolution of both problems is based on the
definition of cloud model, which is generally the model of plane-parallel
and homogeneous cloud. In order to investigate pertinent cloud parameters
for radiative transfer simulation, we developed a new stochastic cloud
model, the tree-driven Mass Accumulation Process or “tdMAP”, suitable to
generate realistic overcast and broken clouds in a unique framework.
Simulations of visible and thermal radiative properties of inhomogeneous
(tdMAP) clouds and of their homogenous equivalents are calculated with SHDOM
radiative transfer model. We present some results on radiative effects of
inhomogeneities cloud on visible and thermal properties for multi-layers
clouds (2 layers) and on actinic fluxes for one-layer clouds. We also
demonstrate that neural networks are suitable tools to develop fast and
accurate algorithm in order to compute inhomogeneous cloud fluxes or
radiance.