32nd Conference on Broadcast Meteorology/31st Conference on Radar Meteorology/Fifth Conference on Coastal Atmospheric and Oceanic Prediction and Processes

Tuesday, 12 August 2003: 11:45 AM
Cloud Detection using Radar Wind Profiler Output
Paul Michael, Brookhaven National Laboratory, Upton, NY; and M. A. Miller and E. Clothiaux
Poster PDF (380.7 kB)
State of the art cloud sensing capabilities, including a Millimeter Cloud Radar (MMCR), have been installed at each of the Atmospheric Radiation Measurement (ARM) program sites at a single, central, location. Results from the MMCR along with ancillary data (e.g. ceilometers) are used in an algorithm named "Active Remote Sensing of Cloud Layers" (ARSCL), which has been shown to provide relatively accurate, high;resolution profiles of cloud location in the column. Knowledge of the cloud structure over a wider area, particularly at the Boundary Facilities (BF’s), would enable more rigorous testing of the parameterizations used in climate models.

A potential source of additional data is the National Oceanic and Atmospheric Administration 404 MHZ (75 cm wavelength) radar wind profilers (RWP), which are located close to the ARM Boundary Facilities. The RWP utilizes the Doppler shift of signals reflected from fluctuations in the refractive index of the atmosphere to determine mean radial velocities along three beam axes (vertical, tilted to North or East) which are converted to a 3 dimensional wind vector. The RWP based cloud detection algorithm postulates that clouds are characterized by anomalously large refractive gradients and tend to be significantly more turbulent than clear sky; thus increasing the reflectivity and spectral width signals relative to clear sky observations. To reduce the altitude and diurnal effects on the signal, and to bypass calibration issues, normalized reflectivity and spectral width variables are derived from monthly statistics (means and standard deviations) of the zeroth and second moments of the RWP signal as a function of time of day and altitude. The hypothesis is that clouds are indicated when a linear combination of normalized moments exceeds a threshold value. Ceilometer and satellite data that are available at BF’s can be included in an RWP algorithm. ARSCL/MMCR results are used as a standard in order to optimize the ratio of RWP moments used, the detection threshold and the inclusion of ceilometer and satellite data.

The quantitative measure of skill used to optimize and evaluate RWP algorithms is the “Receiver Operating Characteristic” (ROC); a tool appropriate for assessing the performance of a procedure that has a binary output. The ROC skill score is the fraction of altitude time points correctly identified as cloud minus the fraction of clear air points erroneously identified as cloud. Random procedures would yield a score of zero; perfect procedures, unity.

ARM Southern Great Plains Central Facility data for the year 2000 was used to test the algorithm. ROC skill scores for use of the RWP (with or without ceilometer data to define cloud base and GOES data to define cloud free periods) had a mean of about 0.5 (standard deviation of 0.2) when optimized daily. This is substantially better than chance but significantly less than perfect.

Supplementary URL: