Clouds play an important role in the weather and climate system. Clouds dominate the vertical transport of energy and trace gases in the free atmosphere. For climate monitoring, information on the spatial distribution of cloud physical parameters, such as optical thickness and particle size, is of great importance. These parameters may be derived from spectral radiances measured from satellites in the solar spectrum. Radiative transfer models are used to simulate multiple scattering and absorption of light by cloud particles. This information is needed to relate the observed spectral reflectances to physical cloud properties.

This article presents the results a comparison study of five Radiative Transfer Models (RTM) for multiple scattering simulations on water clouds. The RTMs utilise different methods to approximate or solve radiative transfer. The radiances were simulated as a function of the optical depth, the viewing angle, the azimuth angle and the phase function for scattering. The calculations were performed at 0.63 and 1.6 micron. The observed radiances at these wavelengths provide information on respectively cloud optical thickness and cloud droplet effective radius. The models chosen were Doubling Adding KNMI (DAK), Monte Carlo, SHDOM, MODTRAN 4.1 and MODTRAN 4.2b. The MODTRAN 4.2b model is a beta version optimised for radiative transfer in a cloudy atmosphere, which was developed for this study. The study concentrated on plane parallel water clouds. The scattering phase functions of MODTRAN 4.1 are based on the Henyey and Greenstein equation. The scattering phase functions of the other models were generated with Mie code. The relationship between reflectivity and the view zenith angle was studied for 162 cases. The cases were selected on the basis of solar zenith angles, view zenith angles, microphysical properties, wavelength, and surface reflectivity. The Monte Carlo simulations were considered very accurate, and were used as reference of the true situation.

The comparison study showed that Monte Carlo, SHDOM and DAK model simulation results at 0.63 and 1.6 micron are close to each other. The differences may be explained by numerical noise of the different methods to model radiative transfer. The MODTRAN 4.1 results deviate greatest from the other models, with the biggest differences at large satellite viewing angles. MODTRAN 4.2 beta simulates the cloud reflectivities much better than 4.1, but at high viewing angles the difference are still large. Further research is going on to explain these differences.