Predicting the Weather-Dependent Collision Risk for Birds at Wind Farms - Wind and Wildlife - page 13

Table 2 Collision risk estimates for wintering northern gannets at HR1 and HR2 offshore wind
farms, along with species-specific values of key input parameters

Horns Rev 1 Horns Rev 2

Northern gannet
Mean density of all wintering birds, individual/km 2 0.006 0.018
% of birds flying (estimated from ship surveys) 64 % 64 %
Mean density of flying birds in winter (Nov-Apr), individual/km 2 0.004 0.012
% of bird flying at rotor height 8.7 % 39.1 %
Collision risk (98 % avoidance), number of birds colliding 0 7
Collision risk (99.5 % avoidance), number of birds colliding 0 2
Collision risk calculated for pessimistic (98 % avoidance rate) and optimistic (99.5 % avoidance
rate) scenarios

Table 3 Significance and t- and F-values for the fixed parametric (wind directions, wind farm and
survey year) and smooth terms included in the GAMM for the common scoter
t-value p-value

Parametric Tail wind 4.825 <0.01

Side wind 5.481 <0.01

HR2 −7.043 <0.01

Smooth F-value p-value

Dist. to turbine 7.092 <0.01

Wind speed 3.068 0.08

Humidity 13.832 <0.01

R-sq. (adj) 0.10
Spearman's Rho 0.30
Sample size 2,374
The model was evaluated by fitting the model on 70 % and testing the predictive accuracy on 30 %
by estimating Spearman's rank correlation coefficient (Rho) between observed and predicted alti-
tudes. Adjusted R 2 value is given as an indication of variance explained by the model

leave land to cross the Fehmarn Belt in autumn, indicated that the birds fly higher in
lower wind speed and relative humidity. According to the model the birds further
gain altitude when leaving land and the red kites also flew higher in tail winds in
comparison to head winds (Fig. 8 , Table 5 ). The model had a good predictive ability
with a Spearman's correlation coefficient of 0.68, and the adjusted R 2 indicated that
the model explains 41 % of the variability in the data set. The model was also able
to account for most of the spatial autocorrelation in the residuals as we found
significant spatial autocorrelation only in lag 2 (1 lag = 250 m), the Moran's I value
was however very low, 0.06 indicating a weak autocorrelation.
We further used the model for predicting the flight altitudes in the autumn 2011
during mean weather conditions (in our model data set) in both tail (0°) and head
winds (180°). According to the predictions the red kites leaving Denmark cross the