A Cluster Analysis Approach to Comparing Atmospheric Radiation Measurement (ARM) Data with Global Climate Model (GCM) Results
| Forrest Hoffman | Oak Ridge National Laboratory |
| Salil Mahajan | Texas A&M University |
| William Hargrove | Oak Ridge National Laboratory |
| Richard Mills | Oak Ridge National Laboratory |
| Anthony Del Genio | NASA Goddard Institute for Space Studies |
Category: Atmospheric State and Surface
Mean tropospheric profiles of twelve distinct atmospheric states for ARM observations (left) and corresponding CCSM output (right) derived from cluster analysis of the combination of ARM observations and CCSM output. Shown here are temperature profiles (red, bottom axis), specific humidity profiles (blue, top upper axis), and wind speed profiles (green, top lower axis) against height (left vertical axis). Also shown is the surface pressure (asterisk, right vertical axis). The frequency plot (lower right) shows the population density of states from the ARM observations and CCSM results.
Cluster analysis was employed to compare ARM observational data at the Southern Great Plains (SGP) site with corresponding 6-hourly output from an integration of the Community Climate System Model (CCSM) run under the Intergovnmental Panel on Climate Change Special Report on Emissions Scenarios A2 for the current decade. Cluster analysis is a technique for classifying multivariate data into distinct regimes or states based on Euclidean distance in a phase space formed from the variables under consideration. A parallel clustering algorithm developed at Oak Ridge National Laboratory and designed for analyzing very large datasets was applied to obtain atmospheric column states from observations at the SGP site and, separately, from CCSM results based on vertical temperature, humidity, wind speed, tropospheric profiles, and surface pressure. A three-way process was implemented to compare ARM data with GCM output, where (1) CCSM output was projected onto states derived from ARM observations, (2) ARM observations were projected onto states derived from CCSM output, and (3) both ARM observations and CCSM output were projected onto states derived from the combination of the two datasets. Comparisons of 12 atmospheric states derived from the combination of ARM observations and CCSM output indicate that distinct singular states exist in each dataset. As shown in the figure, state 5 has no analog in the ARM observational data, while states 1, 3, 7, and 11 are never captured by CCSM. State 5 is characterized by very high humidity and temperature at the surface, and it has no analog in the observational data. States 1, 3, and 7 have very low frequency in the observations, so their absence from model predictions does not suggest a problem. However, state 11, which is characterized by high humidity and temperature with very low wind shear, is never predicted by CCSM. In addition, CCSM predicts an over-abundance of state 9 (low humidity and high temperature conditions) while under-representing state 4 (moderate humidity, temperature, and shear conditions). Misrepresentation of atmospheric states in CCSM over the SGP site could have impacts on predictions of cloud formation and hence the local radiation budget.
http://www.climatemodeling.org/arm
This poster will be displayed at ARM Science Team Meeting.
POSTER in PDF: for proper viewing, it should be viewed with Adobe Acrobat Reader. Download the latest version from the Adobe Reader website.
