Bird identification on 1290-MHz wind profiler data applying neural networks and neurofuzzy systems

Migrating birds can severely affect data from wind profilers operating in t he 1000 MHz range. Recent methods for removing bird contamination do not se em to solve the problem satisfactorily. Here, a new method, the Quantum Neu rofuzzy Bird Identification and Removal Deck (NEURO-BIRD) is presented. The algorithm has an overall classification rate of over 90 % for birds, clear air returns, and rain echoes for single, one-second wind profiler spectra. Even with very heavy migration, high quality hourly winds can be obtained. Because the source of contamination of the spectra is unambiguously identi fied, bird data can be supplied for ornithological research. NEURO-BIRD is very fast and well suited for real-time applications. (C) 2001 Elsevier Sci ence Ltd. All rights reserved.