In times of climate change, the pressure on biodiversity is high, there is a larger variability observed in seasonal biomass behavior, especially in bird migration. As part of mitigation, better anticipation of migration becomes key to preserving and understanding our ecosystem. At the same time, human activity has increased due to the installation of a larger number of wind farms and higher air traffic ( 80% of flight incidents were caused by collisions with birds), this makes it an essential component of citizen science to be able to detect, predict and adapt to evolving bird migration paths.

Radars dedicated to bird detection are scarce, few, and limited in resolution volume. It therefore becomes interesting to try and detect bird migration using existing networks of operational weather radars. However, sampling strategy and design are not necessarily well suited for such tasks. There are not enough standardized measures to compare the performance of an existing system.

The talk will present a bird detection algorithm developed for the MeteoFrance weather radar at NIMES where strong bird migration can be observed in the Rhone valley in the south of France and its validation using dedicated bird radars. This bird detection algorithm relies on a classical approach to distinguish meteorological echoes from biomass echoes. The results of this algorithm will be used to label radar PPI images with enough accuracy to build a training database consisting of radar polarization products. The labeled images are used during semi-supervised training of a U-net.

Prateek Gulati will present an opportunistic bird detection algorithm he developed for the MeteoFrance weather radar at NIMES, where strong bird migration can be observed. The objective is to build a training database of radar polarization products for the semi-supervised training of a U-net.

Lieu
LATMOS
Guyancourt
Salle 2202

Visio
https://uvsq-fr.zoom.us/j/91846173625?pwd=n5udQ1qfx97ThP5es7KpYurnJ7UhxF.1