Biases in droplet radii and optical depths of marine stratocumulus retrieved from MODIS imagery

Abstract

The 2007 Intergovernmental Panel on Climate Change assessment established that the effect of clouds on climate contributes the largest uncertainty in predicting the future climate. Satellite observations provide an opportunity for learning about the behavior of clouds. This thesis seeks to assess the accuracy of cloud properties retrieved from multispectral satellite imagery and test the usefulness of satellite data in verifying conclusions based on aircraft observations that marine stratus appear to be formed through the nearly adiabatic ascent of moist air. Retrievals of cloud optical depth, a measure of cloud thickness, using 0.64-μm reflectances and droplet radii using separately 1.6, 2.1, and 3.7-μm reflectances were obtained with the MODerateresolution Imaging Spectroradiometer (MODIS). Owing to different amounts of absorption by liquid water in the near infrared, with the least at 1.6 μm and the most at 3.7 μm, the growth of droplet radius with cloud thickness should result in the largest droplet radii retrieved using the 3.7-μm reflectances and the smallest using 1.6-μm reflectances. Droplet radii retrieved using the 2.1-μm reflectances, however, are often the largest, and those retrieved using 3.7-μm reflectances are often the smallest. In addition aircraft observations indicate that the relationships between droplet radius, re, and optical depth, t, d ln r / d lnt e ,should be approximately equal to 0.2 for marine stratocumulus. While satellite observations in optically thick, overcast regions yielded d ln re/d ln t consistent with this result, retrievals for regions with broken clouds often yielded d ln r / d lnt e substantially smaller than 0.2. Clouds often exhibit large horizontal and vertical variability. In this thesis a simple radiative transfer model was used to predict reflectances at visible and near infrared wavelengths for clouds formed through the adiabatic ascent of moist air, and then a retrieval scheme based on vertically uniform clouds was used to determine if the departures from the behavior expected for adiabatic clouds might be caused by the assumption of spatial uniformity used in the retrievals. The simulations indicated that at all cloud thicknesses the progression of droplet size using 1.6, 2.1, and 3.7-μm reflectances followed that suggested by the absorptive properties of liquid water. The simulation also indicated, however, contrary to the observations, that when the clouds were optically thin, d ln r / d lnt e should be greater than 0.2, the value expected for the adiabatic ascent of moist air. The simulation was adapted to account for the effects of horizontal variations within clouds. Each 1-km pixel was given a subpixel distribution of optical depths based on a gamma distribution with mean taken from MODIS pixel-scale optical depths and variance given by Kato et al. (2006), obtained from Large Eddy Simulations of marine stratocumulus. Each subpixel was allowed to develop vertically following the adiabatic ascent of moist air. The average of the reflectances for the subpixels was used to retrieve the cloud properties for the pixels, and these properties were compared with the average properties of the subpixels. The retrievals obtained using 1.6-μm reflectances were most strongly affected by the addition of subpixel variations in optical depth and droplet radius. Mean droplet radius retrieved in the simulation was largest when using 1.6-μm reflectances, followed by that retrieved using 2.1 and 3.7-μm reflectances. The simulation of large droplet radii at the shorter wavelengths indicated that the large droplets observed in MODIS may result from horizontal variations within the 1-km MODIS pixel. The values of d ln r / d lnt e calculated for horizontally heterogeneous clouds were close to 0.2, but showed variations that extend both above and below this value, consistent with the MODIS observations. To illustrate the findings based on the simulations, visible optical depth and droplet radius were retrieved for MODIS 500-m pixels overcast by marine stratocumulus. The 500-m pixels were used to represent the subpixel variability within 2-km pixels constructed from the 500-m pixels. Differences between cloud properties retrieved using average radiances and averages of the subpixel properties were compared for overcast and broken-cloud regions using both MODIS and the Partly Cloudy Pixel Retrieval (PCPR) schemes for identifying the overcast 2-km pixels. In the regions overcast by optically thick marine stratocumulus, both methods of identifying overcast pixels led to small biases in the retrieved droplet radii and optical depths. Values of d ln r / d lnt e obtained in the overcast regions using the PCPR identifications were closer to the value of 0.2 expected for adiabatic clouds than those obtained using MODIS identifications. Additionally, values of d ln r / d lnt e calculated using the 500-m overcast pixels yielded better results than those calculated using the 2-km pixels. In regions containing broken clouds, the PCPR identification provided a smaller bias than the MODIS identification, however both methods showed greater biases than those calculated for regions overcast by optically thick marine stratocumulus. Values of d ln r / d lnt e for the regions containing broken clouds showed a positive value when using the PCPR scheme to identify overcast pixels, and a negative value when using the MODIS cloud mask to identify overcast pixels. Consistent with the results from the overcast regions, the d ln r / d lnt e obtained for the 500-m overcast pixels were in closer conformity with adiabatic clouds than the values obtained for the 2-km pixels using both the MODIS and PCPR identifications.