GEOS-5 - CloudSat Intercomparisons
Published September 12, 2008
A new study utilizes global analyses of the atmosphere obtained from the new GEOS-5 Data Assimilation System (DAS) developed for GMAO's Modern-Era Retrospective analysis for Research and Applications (MERRA) to examine the relationships between cloud parameters observed by the CloudSat satellite and predictors of convective cloud structure derived from MERRA reanalyses.
Background
Parameterization of clouds and convection remains a vexing problem for global models. Convective updrafts and downdrafts in the atmosphere have horizontal scales from 100s of meters to several kilometers. These motions are well below the horizontal resolution of current global atmospheric models and will probably not be resolvable in global models for a decade or more. Nevertheless, convection exerts a first order effect on the climate and atmospheric circulation of Earth. This effect is currently included in global models using simplified representations (or parameterizations) that are driven by the resolved flow features of the global model.
CloudSat is a satellite-borne cloud profiling radar (CPR) launched by NASA on April 28, 2006. CloudSat is designed to detect small cloud particles. The instrument provides profiles of cloud induced radar-reflectivity with a vertical resolution of 500 m. Profiles are measured approximately every 1-2 km along the satellite ground track. These profiles provide an unprecedented view of the vertical structure of clouds in the atmosphere. This study will examine the relationships between CloudSat derived cloud parameters and other atmospheric variables. We will utilize global analyses of atmosphere obtained from the new GEOS-5 Data assimilation system (DAS) developed for NASA's Modern Era Retrospective-analysis for Research and Applications (MERRA). Our focus here will be on convective systems which are poorly represented in models and which have so far eluded efforts to correlate simulations with CloudSat observations.
The purpose of this study is twofold. First we would like to place the huge CloudSat database in the correct meteorological setting in the hope of shedding light on the large scale dynamical controls exerted on convective systems by the atmosphere, using a state-of-the-art high resolution DAS. Second we would like to directly compare the output of the current convective and cloud parameterizations operating in the AGCM component of the GEOS-5 DAS with CloudSat observations.
Objectives and Techniques
Cloud or convective depth scale is a critical parameter for a number of atmospheric processes including tracer transport, radiative forcing, and diabatic heating. The depth of convection is typically underestimated in climate models. Our first task is to determine climatologies of cloud depth in CloudSat data and to relate them to predictors of cloud depth based on local atmospheric profiles of u, v, T and q. We use the following algorithm to automate estimates of cloud depth from CloudSat data. Raw GEOPROF-2B radar reflectivities are used as the starting point. Each granule of radar reflectivity consists of around 37000 profiles with 125 range bins (or levels) containing returns from different altitudes. Centers of the altitude range bins are spaced about 240 m apart although vertical feature resolution is around 500 m. For the analysis, each granule is subdivided into overlapping segments of 100 profiles, with centers spaced 50 profiles apart. (See Figure 1. Correlation coefficients are then calculated between each pair of vertical levels in the segment.Only those between level 105, the typical surface return level, and level 45 at around 15000 m are used. This results in a 60x60 correlation matrix for each segment (Figure 1c). A two-sided, one-parameter Gaussian fit is then made to the correlations from each level. This gives two correlation length scales (up and down) for each level in the CloudSat data (Figures 1d,e).
Figure 1.
Example of analysis procedure for GEOPROF-2B radar reflectivity granule (#01132, July 15, 2006, 01:52:00Z to 03:31:00Z). Panel A shows raw reflectivity from CloudSat for the entire granule, which consists of around 37,000 profiles with a vertical resolution of 240 m. Panel B zooms in on convective clouds observed in W Pacific (5N-5S, ~165W) near 02:40Z. The section bounded by the dashed lines contains 100 profiles. Panel C shows the level-to-level correlation matrix calculated for this section. We retain 75 levels of radar reflectivity, from near mean sea-level to around 18,000 m altitude, for the calculation. The resulting matrix is 75x75 and symmetric, with values along the diagonal of exactly 1.00. In panels D and E we illustrate the two-sided Gaussian fit that is performed to estimate an upward and downward correlation scale from each vertical level. The dashed vertical lines in (D) are placed at 3300 m an d14100 m. The corresponding correlation profiles are shown by the solid lines in (E). The dashed lines in (E) show best-fit one parameter Gaussians at these levels.We use atmospheric data from the 0.5x0.666 degree MERRA retrospective analysis interpolated onto CloudSat tracks in both space and time to derive a number of predictors for convective cloud depth. One of the simplest of these is the height to which an undiluted surface parcel would rise before losing buoyancy. This depends primarily on the moist static energy near the surface and the temperature profile aloft, as illustrated schematically in Figure 2. The undiluted surface parcel height is a simple estimate of the maximum possible convective cloud height for given atmospheric profiles of T and q.
Figure 2.
(Top) Sample temperature cross-section from MERRA retrospective analysis interpolated onto a CloudSat track; Granule 01132, July 15 2006, 01:52 Z to 03:31 Z. (Bottom) Schematic diagram illustrating how undiluted surface parcel height is determined from reanalysis fields. This will be used as a predictor for vertical correlation scales in CloudSat data.Results
Figure 3 shows a joint PDF of CloudSat vertical correlation scale from the surface upwards and the undiluted surface parcel height obtained from MERRA retrospective analysis for July 2006. The plot shows a clear statistical relationship between this simple predictor of convective cloud depth and observed cloud vertical correlation scales from CloudSat. Observed cloud vertical scales rarely exceed the undiluted parcel height estimated by MERRA. The shape of the PDF for a given value of the undiluted parcel height is approximately exponential, with many more observed correlation scales close to zero than to the maximum possible height predicted by the analysis. We believe the close match between the upper limit of the PDF in the horizontal and the y=x line may be fortuitous consequence of our analysis, but that the general character of the joint PDF is not.
The result in Figure 3 is encouraging in many respects. A simple analysis derived predictor for cloud height yields a good and physically-reasonable description of the statistics of vertical correlation scales in the CloudSat data. The lack of a good deterministic relationship between our simple analysis predictor and observed cloud vertical scales is almost certainly due in part to sampling issues related to CloudSat, but may also reflect a real underlying stochastic character in atmospheric convection. It is also possible that refinements in the analysis predictor could produce a tighter relationship with the CloudSat observations. However, we believe the first two possibilities, sampling and inherent stochasticity, are likely to dominate in comparisons of CloudSat convective cloud observations with large-scale models. Cloud resolving models should prove useful in interpreting results such as those in Figure 3.
Figure 3.
Joint PDF of vertical correlation scales obtained from CloudSat radar reflectivites (x-axis) and undilute surface parcel heights calculated from MERRA retrospective analysis profiles of T and q (y-axis). Results are shown for the month of July 2006.Future Work
We have aleady begun to examine results from the Goddard Cumulus Ensemble (GCE) cloud resolving model. The GCE will be forced with analysis profiles and tendencies from MERRA to extend the sample space of available GCE simulations to compare with CloudSat and other high resolution satellite data sets. In addition to addressing issues that may arise from satellite sampling, the GCE results will also be used to extrapolate knowledge gained about condensates and other satellite observables, to quantities such as vertical velocities and mass fluxes, which are central to parameterization development.
