Estimating and correcting MODIS 1D cloud retrieval errors in marine boundary layer clouds

One significant source of uncertainty in the operational MODIS cloud product is the assumption of 1D radiative transfer in the retrieval algorithm. We report progress in an ongoing study designed to estimate and correct for these 1D errors in MODIS optical depth and effective radius retrievals in marine boundary layer clouds. The UCLA large eddy simulation (LES) model was used to generate realistic 3D cloud fields over the eastern north Pacific. The LES model was initialized with profiles from random columns of the ECMWF ERA-Interim reanalysis for 200 six hour simulations on a 160x160 grid with 50 m horizontal spacing. A wide variety of boundary layer cloud types and structure was generated. The new parallel-version of SHDOM was used to compute 3D radiative transfer in 1000 LES liquid water content fields. Water vapor absorption was included, though aerosol scattering was not, and a realistic ocean surface reflection model was used. Outgoing radiances were computed at 0.86, 2.13, and 11.0 um with solar/viewing geometries appropriate for MODIS. The high resolution radiances from SHDOM were averaged to simulate MODIS pixels.

The simulated MODIS pixels are input to a 1D lookup table, and the resulting optical depth and effective radius retrievals are compared to the true values averaged over each MODIS pixel. Bias and rms errors in optical depth and effective radius are computed and analyzed as a function of viewing and solar geometry and cloud type. A statistical 3D retrieval algorithm is made by training a neural network to predict the true optical depth and effective radius from the simulated MODIS radiances and textural information. The accuracy of the 3D retrieval algorithm will be reported.