Bird migration forecast: developing radar-based forecasts of bird migration over Switzerland

The Western Palearctic bird migration represents a massive flow of biomass that takes place twice a year between Europe and Africa. During the migratory seasons, the weather conditions induce strong variations in the migration intensity from one day to the other, with peak migration events involving hundreds of thousands of birds in flight. Prediction of these migration events allows to deliver warnings and minimize the risks of collision between birds and human infrastructures. Such collisions are thought to increase bird mortality during migration and are a source of danger and economic loss in aviation. Using six years of dedicated bird radar records in the Swiss Lowland and meteorological reanalysis, this study aims to understand the influence of weather conditions on the migratory fluxes over Switzerland and assess their predictability. Three different analysis are performed to answer these questions. First, a statistical analysis is carried out to gain insight into the role of weather conditions in modulating peak migration events. Second, boosted regression trees are trained to reproduce the migratory flux in the Swiss Lowland. Finally, a regularized linear regression is fitted to explore how the model interprets the spatial and temporal relationships between weather and migration.

The results show that specific weather patterns are associated with peak events. The statistical model indicates that bird migration is predictable from past days migration data and weather conditions. The addition of meteorological data from other locations in Europe proves to increase model performance. However, the quality of predictions varies importantly between years and the model tends to underestimate peak migration events. Finally, the linear model coefficients show that the model is able to detect spatial and temporal patterns for each meteorological variable. This study highlights the predictability of migration from weather predictors and emphasizes the importance of including spatial and temporal information on these predictors. Further analysis and adjustments are needed to provide a reliable forecasting system for the Swiss Lowland.