Abstract
During their migration, millions of birds cross the North and Baltic Sea each year, most of them during the night. With the rapid development of offshore wind farms in recent years collision risk of nocturnal migrants and the potential impairment of bird migration has increasingly come into focus. Yet, due to the difficulty of obtaining direct evidence of collisions or the avoidance behavior of nocturnal migrants at offshore wind farms, little is known with respect to their collision risk and associated number of fatalities. This is particularly true for the cumulative collision risk as no com prehensive data from multiple offshore sites was available yet.
Typically, bird migration shows high temporal variability at least part of which is driven by the prevailing weather conditions. Information on atmospheric variables can therefore be used to model and predict future migration intensities. Migration forecasts could then potentially be used to implement mitigation measures such as turbine curtailment in order to prevent bird collisions.
We used data on bird migration collected by marine surveillance radar over a period of nine years at 10 sites in the German Exclusive Economic Zone (EEZ) of the North and Baltic Sea to determine the relationship of migration intensity and flight height with atmospheric parameters and to con struct a forecast model predicting migration intensities in the German EEZ. Further, we applied a collision risk model (SOSS Band model) to estimate the cumulative number of collisions of nocturnal migrants at all German offshore wind farms that are operational or currently under construction.
Our models explained up to 80% of the variance in migration intensities in our multi-year, multisite dataset. Meteorological and time-related variables accounted for more than 70% of the variance alone. There was strong temporal variation in flux rates which varied across study years, within seasons and within the course of the night. Other factors such as study site, stage of the wind farm (baseline, construction, operation) and the company responsible for data collection were of minor importance.
Wind regime was the most important atmospheric driver of offshore migration. Flux rates increased with increasing tailwinds and, to a lesser degree, with seaward crosswinds indicating partial drift of birds to offshore locations. Our results also suggest that accumulation of migrants at the departure sites due to unfavorable wind conditions during previous nights affected migration activity.
Furthermore, nocturnal migrants generally preferred weather conditions characterized by low relative humidity and high barometric pressure which are usually associated with clear skies, moder ate winds and no precipitation. The relationship with ambient temperature contrasted between spring and fall. Spring migration intensities increased with increasing temperature while during fall migration intensities increased with decreasing temperature.
Variation in mean flight height was less well explained by meteorological variables with models accounting for approx. 50-60% of the variance. Migration intensity was the most important predictor of flight height indicating a strong positive correlation between the two variables. The relationships between atmospheric variables and flight height corresponded well with results from migration intensity with one noticeable exception. In contrast to flux rates, flight height increased with increasing headwinds. A possible explanation may be related to our meteorological data reflecting wind conditions at sea level with no information on variation of conditions with altitude.
Mean flight height also showed strong temporal patterns. It decreased in the course of the night presumably indicating an increasing proportion of birds preparing to land. Additionally, flight height increased during spring migration but decreased throughout fall. Systematic seasonal changes in weather (wind) conditions or seasonality in the species composition and variation in their preferred flight height may account for this pattern.
Predictive models were based on mean migration intensity per night for altitudes up to 200 m, the height range most relevant for offshore wind farms and explained about 40% of the variation in validation datasets. However, the accuracy of our models to predict nights in which a threshold of 250 MTR (migration traffic rate) was exceeded was relatively low (30-60%) due to a tendency of the model to underpredict high migration intensities. Even though optimization procedures increased accuracy of predictions to about 85%, the predictive performance seems insufficient for an application of the model to regulate potential turbine curtailment directly.
The cumulative number of collisions of nocturnal migrants at offshore wind turbines in the German EEZ during spring and fall migration was estimated at approx. 35,000; 16,000 and 8,000 birds assuming an avoidance rate of 0.956; 0.980 and 0.990, respectively. This illustrates the pivotal effect of the assumed avoidance rate on model outcome. Uncertainty and variation in other bird-related and turbine-related input parameters resulted in additional uncertainty of the model outcome though sensitivity analysis indicated that their potential impact was minor in comparison to avoidance rate.
The total number of collisions estimated for the North Sea was higher compared to the Baltic Sea reflecting the higher number of turbines in the North Sea, yet the number of collisions per turbine was on average about 50% higher in the Baltic Sea. With respect to seasonal variation, 36% of all collisions can be expected to occur in October alone. In relation to the total estimated number of migrants crossing the North and Baltic Sea, about 0.03% and 0.002% of these birds were estimated to collide each year, respectively.
Due to the high temporal variation of migration intensities, estimated collisions were also aggregated in time. By calculating the proportion of collisions that theoretically occur when migration intensity exceeds a certain threshold, we assessed the efficacy of potential turbine curtailment. For example, our data suggest that if turbine shutdown were implemented when flux rates exceeded 500 MTR, 27% of collisions could theoretically be prevented with turbine shutdown amounting to approx. 30 h per annum.
These considerations assume that collisions are strictly proportional to the number of birds migrating. However, it has been suggested that nocturnal migrants are particularly prone to collisions during unfavorable weather conditions and poor visibility. Occasions where high migration intensities concur with inclement weather may therefore have a strong impact on overall collision risk. Our data indicate that such events occur only rarely (0.5-8 h per year depending on the definition of poor weather and high migration intensity). If collisions were strongly aggregated by these events, turbine shutdown time could be further reduced to prevent a given number of collisions. However, better knowledge about the effect of weather on the avoidance behavior of nocturnal migrants at offshore wind farms is needed to gauge the effect of turbine curtailment on the number of collision fatalities.