JP5.23
Cloud regime determined by reflectivities of Cloud Profiling Radar
Yong-Keun Lee, CIMSS/Univ. of Wisconsin, Madison, WI; and T. Greenwald and H. L. Huang
Cloud regime determined by reflectivities of Cloud Profiling Radar : 2-year of results
Clouds play a undisputingly important role in the climate system, and many efforts have been focused on better understanding of clouds with general circulation models, flight and satellite observations, and ground observations, etc.
Cloud Profileing Radar (CPR) operating on the satellite CloudSat since 2006 has opened a new era to determine the cloud features vertically, facing down 1.5 km across-track and 2.5 km along-track footprint at nadir at 94 GHz. The sample rate is 0.16 seconds and the vertical resolution is 250 m. Included are the regions between 82°S and 82°N. Two year CPR data is available since Jun 15 2006 which allows a capability to look at a year averaged and seasonal vertical cloud feature. Cloud regimes are investigated for two latitudinal region, tropical area ( 30°S – 30°N ), midlatitude area (60°S – 30°S, and 30°N – 60°N).
To determine cloud regimes, K-means clustering algorithm (Anderberg, 1973) is applied for all the reflectivity products since Jun 15 2006. Specifically, Cluster analysis code available on-line at the ISCCP Web site is used (http://isccp.giss.nasa.gov) as clustering algorithm.
K-means clustering algorithm has two issues: one of them is to determine the number of cluster and the other is to find the optimal solution out of huge amount of centroid histogram patterns.
To determine the number of cluster, several criteria have been applied following Rossow et al. (2005): 1 There must not be a significant change to the resulting centroid histogram patterns with random initial conditions and different subset of data, 2. the pattern correlation between the resulting centroids has to be low ( < 0.6), 3. the intra dispersion of members to each centroid is lower than the inter distance between the centroids.
Once the number of cluster is decided, a resulting centroid histogram patterns (RCHP) is selected from a subset of data and the pattern correlation is investigated between the RCHP and the local centroid histogram patterns (LCHP) out of several different subset of data. If the local centroid histogram patterns correspond to each RCHP with correlation coefficient greater than 0.6, the LCHP is considered as similar to RCHP.
Cloud regimes between tropical area and midlatitude area are similar with some exceptions. Some of the centroid histogram patterns are similar between two regions, such as thin cirrus cloud over low cloud, anvil cirrus cloud with high occurrence, and deep convection.
Joint Poster Session 5, Climate
Wednesday, 14 January 2009, 2:30 PM-4:00 PM, Hall 5
Previous paper Next paper