2.1 Day-Night Multilayer Cloud Detection/Retrieval System Using an Artificial Neural Network Approach

Wednesday, 10 January 2018: 10:30 AM
Ballroom G (ACC) (Austin, Texas)
Patrick Minnis, SSAI, Hampton, VA; and S. Sun-Mack, G. Hong, C. R. Yost, W. L. Smith Jr., Y. Chen, F. L. Chang, and R. Palikonda

Retrievals of cloud properties from satellites typically interpret the observed radiances in terms of a single, plane-parallel cloud layer. Several parameters, such as top height CTH, optical depth COD, and particle size CER are retrieved using a physical model. For retrievals of cloud systems consisting of two or more cloud layers, i.e., multilayered (ML) clouds, the resulting parameters often have significant errors. Overcoming those errors requires estimation of the cloud properties in each layer. Many multispectral techniques have been developed to detect ML systems with varying degrees of success. A few methods have attempted to retrieve the properties of the bottom and top layers resulting in some limited improvement of the cloud system characterization. In this presentation, artificial neural network (NN) algorithms are developed to detect ice clouds and estimate CTH and COD regardless of the underlying background, clear or cloudy, day or night. They are trained with CALIPSO and Cloudsat (CC) data matched to multispectral radiances and cloud properties from Aqua-MODIS. The trained NN algorithms are applied to various satellites, including GOES-16. The CC lidar and radar profiles provide the layering information and ice cloud optical depths that serve as output for the neural network NN algorithm. The trained NN results enable estimation of the height and optical properties of the upper-layer clouds. Ice CTH and COD are also estimated using the modified CO2-slicing technique used by the Clouds and Earth’s Radiant Energy System (CERES) to provide the best estimate of those parameters. Knowing the properties of the upper part of the cloud and its layering status, it is possible to determine the properties of the lower layer using a ML retrieval model. This paper furthers this approach to better define and monitor ML cloud systems for CERES and the near-real time NASA Satellite ClOud and Radiative Property retrieval System (SatCORPS), which use both polar-orbiting and geostationary satellite imagery to characterize clouds globally at high temporal resolution. In addition to facilitating more accurate estimates of the Earth radiation budget, the enhanced cloud products should also prove valuable for assimilation into numerical weather model analyses, aviation safety, and solar energy applications.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner