Weather radar signal processing techniques

Automatic Detection and Mitigation of Wind Turbine Clutter
using Radar Data Products

I collaborate with Drs. Boon Leng Cheong and Robert Palmer to study the capability of WSR-88D radars to identify the weather radar interference generated by wind farms using readily available radar products. This project is funded by the NWS Radar Operations Center (ROC).

We developed a fundamental framework for automatic wind turbine detection using Level II data, i.e., reflectivity, Doppler velocity, and spectrum width, which focuses on areas where ground clutter filter has been applied. The goal is to indentify the residual wind turbine clutter signals that pass though the ground clutter filter, i.e., CMD (Clutter Mitigation Decision) flag has been marked. The fundamental idea is to mimic human’s visual identification capability to lock on stationary features when several consecutive images are inspected in a loop manner. The algorithm utilizes a series of consecutive radar images to derive six texture maps, which are essentially spatio-temporal statistical quantities for features on the radar images. The textures are subsequently ingested into a fuzzy-logic inference system for wind turbine clutter detection. The platform has been implemented and preliminary tests have been conducted with datasets from Dodge City, Kansas (KDDC), Dyess Air Force Base, Texas (KDYX) and Buffalo, New York (KBUF). Initial investigation has shown potentials in identifying wind farm regions with a window of 7-9 radar scans, which represent a time span of 35 to 45 minutes when the radar is operated at precipitation mode. A more thorough evaluation on the potentials of this platform is currently being finalized by the Radar Operations Center's Applications Branch.

The image on the left shows standard moments (left column) and six textures being derived in the algorithm (meddle and right columns). Also, the output of the fuzzy-logic detector is shown in the bottom panel, where the output is a value between 0 and 1 indicating certain degree of confidence. Yellow shades indicate detections (values > 0.5) of wind turbine clutter. Two wind farms, one located 40 km southwest while the other lies 25 km northeast of the KDDC radar site were successfully identified. Note the similar values in reflectivity, velocity and spectrum width for the precipitation at the southwestern region that would be nearly impossible to identify without a temporal history of data. The algorithm was able to separate the wind turbine clutter from the radar signals by using six temporal-spatial textures.

The most recent results are documented in this paper presented at the 2011 IEEE Radar Conference.

Back to WTC