Impact of Cloud Vertical and Horizontal Inhomogeneity on Multi-Spectral Retrieval of Liquid Water Cloud Properties: A Study for the GCOM-C/SGLI Cloud Product by Using A-Train and Landsat-8 Measurements

Abstract:

Cloud droplet effective radius (r_e) and optical thickness (τ_c) of liquid water cloud retrieved from multi-spectral measurement of satellite-borne sensors such as the Moderate Resolution Imaging Spectroradiometer (MODIS) are important parameters for understanding of cloud microphysics and droplet growth process. Japan Aerospace Exploration Agency (JAXA) is scheduling to launch a new earth observation satellite, the Global Change Observation Mission - Climate (GCOM-C), in Japan fiscal year 2016. The GCOC-C consists the Second Generation Global Imager (SGLI) instrument, which is a radiometer providing near-ultraviolet-to-thermal-infrared multi-spectral measurements at 250m-1km resolution with 1150-1400km swath. Motivated by the importance of r_e and τ_c retrievals, the GCOM-C/SGLI cloud product also provides them. However, it is pointed out by previous studies that multi-spectral-retrieved r_e and τ_c are impacted by cloud vertical and horizontal inhomogeneity. This study investigates the impacts by using A-Train and Landsat-8 data. First, we interpret three r_e retrievals from 1.6, 2.1, and 3.7µm-band measurements (r_e,1.6, r_e,2.1, and r_e,3.7) in terms of cloud vertical inhomogeneity by using synergistic measurements from the CloudSat/CPR and Aqua/MODIS. For this interpretation, we use an approach called Contoured Frequency by Optical Depth Diagram (CFODD), which is a joint frequency diagram of CloudSat/CPR radar reflectivity profile as a function of in-cloud optical depth profile as classified according to retrieved r_e (Nakajima et al., 2010; Suzuki et al., 2010). The CFODD approach visualizes the linkage of r_e,1.6, r_e,2.1, and r_e,3.7 to cloud droplet vertical profile. Second, we simulate the biases in r_e and τ_c retrievals at 1km resolution from the MODIS and SGLI induced by cloud horizontal inhomogeneity by using high-spatial resolution measurements of Landsat-8. And then we suggest two inversion models which estimate the biases in r_e and τ_c retrievals by using co-variance matrices of multi-spectral radiances: one model explains the bias due to clear-region-contamination based on a classical mixed pixel model which is parameterized by cloud fraction. The other explains the bias due to horizontal inhomogeneity of r_e and τ_c at subpixel scale in no clear-region-contamination case.