Neural Network Retrieval of Inhomogeneous Cloud Parameters From Multispectral and Multiscale Radiance Data. Preliminary Results From MODIS Measurements.

Improvements of radiometric observational capabilities enable the development of new approaches for cloud parameter retrieval. In this study, we investigate the possibility of retrieving cloud parameter of inhomogeneous cloud and fractional clouds by using neural network techniques. The inverse model is defined, at the scale of retrieval, with 6 cloud parameters: the mean optical depth, the mean droplet effective radius, the fractional cloud cover, the optical thickness inhomogeneity, the effective radius inhomogeneity and the cloud top temperature. The retrieval procedure includes two separate steps: the first one is relative to the interpolation and the correction of radiance data (surface reflexion and thermal contribution effects). The second step concerns the cloud parameter retrieval. The input vector of the retrieval procedure uses multispectral information and different spatial resolution because we showed that the inclusion of sub-pixel radiance data as input vector components improved significantly the performance of the cloud parameter retrieval. The whole procedure was tested on different types of synthetic inhomogeneous and fractional clouds. All the cloud parameters of these types of clouds can be retrieved with reasonable accuracy. Following these studies, we applied this procedure to real measurements provided by MODIS on Terra.