Environmental Engineering Reference
In-Depth Information
meridional (V) wind velocity, air pressure (hPa, at 10 m), clearness (% at 10 m,
based on total cloudiness), relative humidity (% at 10 m) and air temperature
(°C at 10 m) from the weather model were integrated with the rangefinder and radar
track data. Clearness and humidity were considered to be proxies for visibility
(humidity is inversely correlated with visibility).
Processing of Track Data
The data collected using rangefinder and radars were processed (separately) before
use in the statistical analyses. Obvious outliers, wrongly located points within
tracks, were removed by visually inspecting the tracks. The track data were further
integrated with the weather data based on closest temporal (date and time) and spatial
(coordinates) match. The integration was made by linear interpolation between
time steps (1 h) in the weather time series data. The U and V wind velocities were
converted to wind speed (m/s) and wind directions (0-360°). A variable defining
the flight direction in relation to wind direction was also created for the Horns Rev
case study. The variable defined whether the bird was flying in head wind (within a
range of 90°), tail wind (within 90
o
) or side winds (within 90° from either side).
Prediction of Bird Movement and Migration Behaviour
Statistical models were developed to assess general patterns in bird migration
behaviour, its relationship to flight altitude, and how these related to wind and
weather conditions and distance to the nearest wind turbine. The general patterns of
flight altitude, considered here as representative of the altitude at which birds would
encounter turbines, were used in the estimation of the flight altitude relative to the
height of the rotors of turbines at the wind farms.
As the relationships between the response variable (altitude) and the predictor
variables in many cases were non-linear, the error structure of the data was non-
normally distributed and the track data was spatially and temporally autocorrelated, a
generalized additive mixed modelling (GAMM) framework was used (Zuur et al.
2009
). The autocorrelation was accounted for by using a correlation structure (corAR1).
The models were created using R version 2.13.0 (R Core Team
2004
) and the
“mgcv” package (Wood
2006
). The GAMMs were fitted with altitude (m) as the
dependent variable and the predictor variables mentioned above as smooth terms,
using thin plate regression splines (Wood
2003
). Flight direction in relation to wind
(head, tail or side winds), location (wind farm) and season were included as categori-
cal variables in the models. The most appropriate error distribution was used for model
fitting, either a gamma distribution with a log link, a Gaussian distribution or a
quasi-poisson distribution. The degree of smoothing was chosen by cross validation
using the “mgcv” package (Wood
2006
). The correlation structure called “corAR1”
was used for the random part of the GAMM (Zuur et al.
2009
). The “track ID” was
