Multilayer Retrieval of Cloud Top heights from MODIS over the Southern Ocean

 The Southern Ocean (SO) is one of the cloudiest places on Earth, with  distinct cloud properties including a high prevalence of multilayer clouds. Previous research has found that multilayer clouds contribute to net cloud radiative effect biases. In our previous paper,we compared and validated different LEO passive sensor retrievals (AVHRR-Patmosx, CMSAF, MODIS collection 6) over the SO against active sensor retrievals (CloudSAT- CALIOP). In the comparison of cloud top height, we found that a mean absolute bias of 0.65 km (AVHRR CMSAF), 1.03 km (MODIS), and 1.31 km (AVHRR PATMOS) was observed for single-layer cloud scenes cases. This mean bias increased to 1.86 km (AVHRR CMSAF), 3.22 km (MODIS), and 3.34 km (AVHRR PATMOS) for multilayered cloud scenes. One of the significant factors for the observed differences is the presence of  multilayer clouds.  

Given the results of the comparison and a need for more accurate cloud retrieval for multi layer clouds in particular, we developed a new multilayer retrieval algorithm for CTH from MODIS data over the SO region using an artificial neural network (NN) approach. The retrieval algorithm employs MODIS radiances and reanalysis datasets. The algorithm's performance for the topmost cloud layer demonstrates a significant improvement compared to the traditional retrieval approaches.  The MODIS CTHs mean bias error against the CloudSAT- CALIOP merged dataset was reduced to approx 0.02 km with an RMSE of 0.84 km. In multilayer scenarios,the CTHs of the top layer were retrieved with a MBE of 0.08km and RMSE of 0.98 km and the CTHs of the second layer with a MBE of 0.01km and RMSE of 1.54 km. The results were analysed to understand the influence on latitude, solar zenith angle, sensor zenith angle, cloud optical depth and surface temperature on the ANN algorithm. The research successfully demonstrated the usefulness of NN in retrieval algorithms.