3. Methods
3.3. Modelling of habitat quality
Modelling of habitat suitability or quality of marine animals typically requires the computation of synoptic, dynamic variables (Doniol-Valcroze, 2005; Skov et al., 2005).
We constructed a local 3-dimensional model for the Horns Rev area covering the entire period 2002-2005. The model was set up using DHI’s model system MIKE 3, which is a fully dynamic, barotropic and baroclinic 3-D model. We used a finite difference grid, in which the hydrodynamic conditions are described in quadratic elements. The size of elements varied horizontally from 500 m in the core model area to 1500 m in the surrounding boundary area, which extended southwards to the north Frisian Islands (Figure 3.11). The vertical resolution was 2 m except for the surface layer, which has a depth of 5 m to take into account the tidal amplitude.
The hydrodynamic model, which was geo-referenced to WGS84, UTM zone 32, calculated the water levels (relative DVR90 datum), currents (3 components), temperature and salinity at half-hour intervals. The meteorologic forcings for the model have been delivered from Vejr2 in a resolution of 0.15º and a temporal resolution of 1 hour. The following data was interpolated at 1,500 m resolution and taken from the sea surface:
• Air pressure (hPa)
• Air temperature (ºC)
• Air speed and direction (10 minute means, m/s, radians)
The model has two open boundaries, which are forced with salinity, temperature and water levels derived from DHI operational waterforecast service for the North Sea (www.vandudsigten.dk). To obtain precise distributions of density differences time series data from seven sources of major freshwater discharges into the area were included (Figure 3.12). For the three German rivers Elbe, Weser and Ems, the actual daily discharge rates from Bundesanstalt für Gewässerkunde were used (http://www.bafg.de).
For the four Danish sources, climatic discharge data was used.
Figure 3.11. Hydrodynamic bathymetry model showing the core fine-scale area around Horns Rev nested into a larger-scale area.
Table 3.5. Specifications for the hydrodynamic model.
Area
Horizontal resolution
Origo and angle
East-west range North-south range
Vertikal range:
max depth, resolution and
layers West
coast
1500 m 6º 33’ 23’’ E 52º 56’ 60’’ N -1.951 º
232.5 km 0-154 points
364.5 km 0-242 points
48 m 2 m 1-25 layers Horns
Rev
500 m 6º 51’ 03’’ E 55º 05’ 10’’ N -1.763 º
120.5 km 0-240 points
81.5 km 0-162 points
36 m 2 m 1-25 layers
Figure 3.12. Sources of freshwater discharges, included in the model.
3.3.2. Analysis of environmental drivers and spatial modelling
The key environmental drivers behind the spatial dynamics of harbour porpoises at the Horns Rev 1 Offshore Wind Farm were analysed by carrying out a combined factorial and polynomial model design in PLS regression parallel to the analysis of drivers in acoustic activity (see Section 3.2.3 for details). Data from four surveys were analysed based on the criteria of large sample sizes (> 100 animals observed), temporal overlap with model data and different frontal positions of the large-scale density front: 28/7 2002, 8/8 2002, 6-7/8 2003 and 20-21/8 2005. During the four surveys the position of the density front changed between west, south, just east and east of the planned wind farm site. For each survey, the dynamic habitat variables were averaged for each tidal phase and collated in a raster GIS environment (Idrisi Version Kilimanjaro, ArcGIS Version 9.1) using UTM32 N Projection with WGS84 Datum at a spatial resolution of 500 m.
The initial version of the PLS model included more than 200 variables, the original variables and their cross-products and second- and third-order polynomials, while the final version was limited to the variables, which best reflected the distinct auto-correlation scale of the harbour porpoise survey data and showed the most significant regression coefficients.
A total of 23 potential original potential habitat variables were computed by post-processing the Horns Rev hydrodynamic model data and local bathymetry data:
1. Stability of the water column: Richardson Number, which is defined as
2
where g is the acceleration of gravity, β a representative vertical stability (commonly θ/ z, where θ is potential temperature), and u/ z is a characteristic vertical shear of the wind.
2. U = On-shore current vector at the surface (m/s) 3. V = Long-shore current vector at the surface (m/s)
4. Relative vorticity (or local eddy potential): dV/dx - dU/dy 5. W = Vertical (upward) current vector at the surface 6. D = Density at the surface (Kg / m3)
7. S = Salinity at the surface (psu)
8. T = Temperature at the surface (Celsius) 9. Water level (meters)
10. Gradient in U, measured as the slope of each grid cell based on the cell resolution and the values of the immediate neighbouring cells to the top, bottom, left and right of the cell in question using the following formula:
which measures the tangent of the angle that has the maximum downhill slope; left, right, top, bottom are the attributes of the neighbouring cells and res is the cell resolution
11. Gradient in V, same GIS method as 10 12. Gradient in W, same GIS method as 10
13. Gradient in surface density, same GIS method as 10 14. Gradient in surface salinity, same GIS method as 10 15. Gradient in surface temperature, same GIS method as 10 16. Bathymetry: negative values
17. Bottom relief: slope same GIS method as 10
18. Northern aspect of sea floor: Sine of the direction of the maximum slope values.
19. Eastern aspect of sea floor: Cosine of the direction of the maximum slope values.
20. Bottom complexity (F) calculated for 5x5 kernel: F = (n-1)/(c-1) Where n = number of different classes present in the kernel, c = number of cells
21. Distance to shallow areas (< 8 m water depth): Euclidean distance in m from each cell.
22. Distance to shallow area at Søren Jessens Sand (< 8 m water depth): Euclidean distance in m from each cell.
23. Distance to shallow area on Horns Rev (< 8 m water depth): Euclidean distance in m from each cell.
Spatial modeling techniques are increasingly recognized as important tools for extrapolating observations of marine animals to obtain spatial predictions of abundance or habitat suitability across large areas of ocean surface. We used the predictive presence-only model ENFA (Ecological Niche Factor Analysis) for developing models of habitat quality for harbour porpoises and harbour seals on Horns Rev. ENFA has been
( ) ( )
( ) ( ( )( ) )
(
− / •2 2 + − •2 2)
= right left res top bottom res Tangent
2005) and marine ecology (Leverette, 2004). The outputs of ENFA show two key aspects of the investigated species’ habitat: marginality and specialization. The principle of the analysis is the mathematical comparison between the environmental space represented by the species distribution and the global distribution in the Horns Rev area. Like the Principal Component Analysis, the ENFA summarizes environmental data into a few uncorrelated factors retaining most of the information. Habitat marginality can be defined as the direction on which the species habitat differs the most from the available conditions in the Horns Rev area. It is computed by drawing a straight line between the centroids of the ellipsoids of the global distribution and the species distribution. Habitat specialization is defined as the ratio of the standard deviation of the global distribution to that of the species distribution.
For harbour porpoises, ENFA was applied to the tidal phase scenarios of each of the four selected surveys using BioMapper Version 3 (University of Lausanne, 2005). The initial version of the habitat model included all 23 variables, yet in the final version we limited the variables to those with a clear impact on habitat marginality and specialization. For the visual surveys and satellite telemetry data for harbour seals, ENFA was applied to summarised sightings from all surveys and analysed only with the topographic variables.