Spatial and temporal analysis of cumulative environmental effects of offshore wind farms in the North Sea basin

The area of study (Fig. 6) was the Greater North Sea ecoregion, which includes the EEZs of six countries (England, Scotland, the Netherlands, Denmark, Norway and Germany). The Kattegat area, the English Channel, and the Belgium EEZ were omitted from the study area. The North Sea Marine Ecosystem is a large semi-closed continental sea situated on the continental shelf of North-western Europe, with a dominant physical division between the comparatively deep northern part (50–200 m, with the Norwegian Trench dropping to 700 m) and the shallower southern part (20–50 m)⁴⁸. The North Sea is one of the most varied coastal regions in the world, which is characterised by, among others, rocky, fjord and mountainous shores as well as sandy beaches with dunes⁴⁸. Apart from the marine seabirds feeding primarily in the coastal areas, under 5 km from the coast (e.g., terns, sea-ducks, grebes), the North Sea basin also hosts pelagic birds feeding further offshore, with some also diving for food (guillemot, razorbill, etc.). The North Sea basin is also a major habitat for four marine mammal species, of which the harbour porpoise and harbour seal are the most common. Moreover, fish ecology has been a widely studied topic, especially for commercial species, due to evidence of a decline in the fish stock, such as sprat, whiting, bib, and mackerel. Fish communities, and in particular the small pelagic fish group (such as European sprat, European pilchard), play also a key ecologic role, constituting the main pray for most piscivorous fishes, cetacean and seabirds⁴⁹, Based on early surveys, the predominant species divided by the three North Sea fish communities are: saithe (43.6% in the shelf edge), haddock (42.4% in the central North Sea, 11.6% in the shelf edge), whiting (21.6% in the eastern North Sea, 13.9% central North Sea), and dab (21.8% in the eastern North Sea)³⁴. More recent assessments of North Sea fish community are emphasizing the clear geographical distinction between the fish species living in the southern part of the North Sea, a shallow area with high primary production and pronounced seasonality, and northern part, a deeper area with lower primary production and lower seasonal variation in temperature and salinity. The southern North Sea fish community is represented by fish species such as lesser weever, while the northern North Sea fish community is represented by species such as saithe, with species like whitting, haddock representative for the North–West subdivision, and the European plaice having the highest abundance in the South–East community⁵⁰. The future fish stock and spatial distribution is however uncertain due to impacts of climate change related factors (e.g., growing temperatures)⁴⁹ and overexploitation.

The most prominent human activities in the North Sea basin are fishing, coastal construction, maritime transport, oil and gas exploration and production, tourism, military, and OWF construction³⁸. Within this list, the construction of OWFs has seen a rapid increase, aiming to reach a total cumulative installed capacity of 61.8–66.8 GW by 2030⁵¹. As indicated in Fig. 6, the new designated/search/scoping areas for the location of future OWFs will significantly increase the current space reserved for the offshore production of renewable energy in the North Sea basin.

Spatio-temporal database of OWF developments in the North Sea basin

For the input of the geo-spatial layers with the location of OWF areas we compiled a comprehensive spatial data repository in QGIS containing the shapefiles of analysed OWF, from 1999 to 2027 (last year of available official information on OWF development, Appendix D). The analysis was performed for the North Sea geographic area, referred here as the basin scale, taking into account the cumulative pressures from individual OWF projects (project scale). The main data sources for geospatial information for OWF, for the entire North Sea basin, are EMODnet (Human Activities data portal) and OSPAR, which were complemented by data on the country level, where needed; i.e. from Crown Estate Scotland (Energy infrastructure, Legal Agreements), Rijkswaterstaat for the Netherlands. From the available geo-spatial data for OWF, we selected the OWF in our area of study (Fig. 6) with the status of consent-authorised, authorised, pre-construction, under construction, or fully commissioned (operational). Therefore, planned OWF such as Vesterhavet Syd and Vesterhavet Nord, for which the start date of construction is still unknown, were not included in the analysis. Similarly, for the Horns Rev 3 OWF no geo-referenced spatial footprint was available in the open-access data sets, and therefore it was not included in the analysis.

The collected OWF geospatial data was aggregated to create a geospatial database, for the studied period of 1999–2050, composed by the following attributes: code name, country, name, production capacity (MW), area (({mathrm{km}}^{2})), number of turbines, start operation (year), installation time, and status in the period 1999–2050 (construction, operation, decommissioning). The created geospatial dataset was additionally cross-checked for integrity with the information provided through the online platform 4coffshore.com.

The lack of data regarding the construction time was complemented with the methodology proposed by Lacal-Arántegui et al.³⁶. Based on this research, we calculated the time required for OWF construction phase related activities multiplying 1.06 days by the known production capacity (total MW) for each analysed OWF.

The average time of operation is considered to be 20 years, probably profitably extendable to 25 years, as stated in a number of studies on the cycle of offshore wind farms⁵². For this case study, the operation time considered is 20 years (subject to change). Since there is little experience with the decommissioning of offshore wind farms (only a few OWFs have so far been decommissioned in the UK and Denmark), the decommissioning time is not yet clear. There are a number of parameters that influence the decommissioning time, which are: the number of turbines, the foundation type, the distance to port, etc. It is estimated that the time taken for decommissioning should be around 50–60% less than the installation time³⁷. Our study considers the decommissioning time as 50% of the construction time.

Time-aware cumulative effects assessment

In this study, Tools4MSP^53,54, a Python-based Free and Open Source Software (FOSS) for geospatial analysis in support of Maritime Spatial Planning and marine environmental management, was used for the assessment of the impacts of OWFs on the marine ecosystem, in the three development stages. We applied the Tools4MSP CEA module to the OWF of the North Sea basin for the period 1999–2050, taking into account the full life cycle of the OWF development, namely the construction, operation and decommissioning phases. The modified methodology from Menegon et al.³¹ and subsequent implementation⁵⁵, proposes to calculate the CEA score for each cell of analysis as follows (Eqs. 1, 2):

where eff is the effect of pressure P over the environmental component E and is defined as follows:

whereas,

• ({U}_{i}) defines the human activity, namely the OWF activity in the study area

• ({E}_{k}) defines the environmental components of the study area described in the Table 1

• ({d(E}_{k})) defines intensity or presence/absence of the k-th environmental component

• ({P}_{j}) defines the pressures exerted by human activities dependent on the three different OWF development phases (Annex B)

• ({w}_{i,j}) refers to the specific pressure weight according to the OWF phase

• ({s(P}_{j}, {E}_{k})) is the sensitivity of the k-th environmental component to the j-th pressure

• ({i({U}_{i, }M({U}_{i, }P}_{j}, {E}_{k}))) is the distance model propagating j-th pressure caused by i-th activity over the k-th environmental component

• ({M(U}_{i}, {P}_{j})) is the 2D Gaussian kernel function used for convolution, which considers buffer distances at 1 km, 5 km, 10 km, 20 km, and 50 km⁵⁶.

In Eq. (3), the CEA 1999–2050 describes the modelling over the time frame 1999–2050, whereas ({CEA}_{t}) is the cumulative effect of year t within the timeframe 1999–2050:

In this study, each final CEA score was normalised. To normalise the value of each initial CEA score obtained using the Eq. (1), we calculated its percentage of the sum of all CEA scores for all OWFs in the three development phases, period spanning the period 1999–2050 (({CEA}_{1999-2050})).

Environmental components

The selection of the environmental components (receptors) impacted by the identified pressures is an essential part of the scoping phase for OWF location, as monitoring the status (distribution, abundance) of different identified species represents a relevant indicator for the ecosystem status. For the evaluation of the habitats and species that can be affected by the cumulative ecological effects of OWF, we adapted the methodology of Meissl et al.¹⁴. Therefore, we selected the environmental components based first on their: (1) ecological value, supported by legal documents identifying species protected by law or through various national and international agreements (e.g. EU Habitats Directive, Wild Mammals (Protection) Act (UK), see Table 1 in Appendix E), to which we added species with (2) commercial value, but also with a (3) broad geographic-scale habitat occurrence of the species in the studied area, based on previous studies³⁵ and on 35 EIA studies for OWF in the North Sea basin.

Among the five fish species selected, sprat and sandeel play key roles in the marine food web (small pelagic fish), as prey source for piscivorous fish, cetacean and birds. The ecological value of sandeel, sprat, whiting and saither is also highlighted through EU or national protection agreements such as Priority Marine Features—PMF or Scottish/UK Biodiversity list (see Appendix E, Table 2). The list is completed by haddock, one of the fish species with commercial importance, highly dominant in the Central North Sea. With regards to the spatial occurrence at the basin level, the fish species selected are representative for both of the two distinct North Sea communities⁵⁰, the southern part of the North Sea (sprat), and the northern and north-west part (haddock, whiting, saithe).

The three selected seabird species are of ecological importance for the marine ecosystem, as indicated through the European, national and international protection agreements, such as the EU Birds Directive Migratory Species or the IUCN Red List (see Appendix E, Table 1). While razorbill and guillemot have similar feeding and flying patterns (low flight, catch pray underwater), there is evidence of different behaviors towards OWFs, with relatively more avoidance from razorbill compared to guillemot. In relation to the spatial distribution of the three selected species, there is a clear distinction between razorbill, highly present in the coastal areas of west North Sea basin, guillemot, with a relatively even distribution across the marine basin, and fulmar, one of the 4 most common seabirds in the studied area, in particular in the central and N–E parts.

In the marine mammals category we selected the harbor porpoise, indicated to be one of the most impacted species in this category⁵⁷, with a high occurrence in the North Sea basin. Its ecological value is emphasized by its presence in European and international lists for habitat protection, such as EU Habitats Directive⁵⁸, OSPAR List of Threatened and/or Declining Species⁵⁹, the Agreement on the Conservation of Small Cetaceans in the Baltic and the North Seas (ASCOBANS)⁶⁰. The harbor porpoise is the protected species in numerous Natura 2000 areas in the North Sea basin, such as the Spatial Area of Conservation Southern North Sea⁶¹ (British EEZ) or The Special area of Protection Kleverbank⁶² (Dutch EEZ).

Among the selected fish species, sandeel had the highest occurrence in EIA studies of OWF developments (23 out of 35), while guillemot had the highest occurrence among seabird species (25 out of 35). With an occurrence of 26 out of the 35 analysed EIA document, the harbour porpoise is the most studied mammal in relation to the impact of OWF.

As a result, we selected three EUNIS marine seabed habitat types (European Union Nature Information System)⁵⁸ (Appendix E, Table 2), three seabird species, one mammal species and five fish species (Appendix E, Table 1). The list can be extended; however, for this exercise we considered it sufficient.

The data sets used to represent the spatial distribution (presence/absence, intensity) of the environmental components in the studied area were obtained from multiple sources and were used in the Tools4MSP model either directly (EUNIS habitats, marine mammals, seabirds) or further processed using a predictive distribution model (fish species). In the case of EUNIS marine habitats, the data source was the online geo-portal EMODnet, through the Seabed Habitat service (Table 1), which provided GIS polygon layers for each habitat type and was further used to indicate presence/absence of a specific habitat.

For the distribution of the selected mammal species, the harbour porpoise, we used the modelling results of Waggit et al.¹⁶, translated into maps for the prediction of densities (nr. animals/({mathrm{km}}^{2})). The mapping approach starts with collating data from available surveys, which are further standardised with regards to transect length, number of platform sides, and the effective strip width. Finally, the standardised data sets were used in a binomial and a Poisson model, in association with environmental conditions (Table 1), in order to deliver a homogenous cover of species distribution maps, on 10 km × 10 km spatial resolution grid¹⁶.

For the distribution of the selected seabird species (razorbill, fulmar, guillemot), we used the results of the SEAPOP program (http://www.seapop.no/en/distribution-status/), through the open-source data portal (https://www2.nina.no/seapop/seapophtml/). The proposed methodology for creating the occurrence density prediction maps, on a 10 × 10 km spatial resolution grid, starts with the modelling of the presence/absence of birds using a binomial distribution and “logit link”. This was followed by the modelling of the number of birds using a Gamma distribution with a “log link” function, which also took into account geographically fixed explanatory variables (geographic position, water depth, and distance to coast).

The predictive model for the spatial distribution of fish species biomass (haddock, sandeel, whiting, saithe, sprat) was developed using AI4Blue software, an open-source, python-based library for Artificial Intelligence based geospatial analysis of Blue Growth settings (AI4Blue, 2021)⁶³. The model was based on two types of inputs: (1) the observation data on the presence of species and (2) data on the absence of species (absence data) for the period 2000–2019. Both data types were extracted by the ICES North Sea International Bottom Trawl Survey (NSI-IBTS, extracted survey year 2000–2019 including all available quarters) for commercial fish species, which was accessed on the online ICES-DATRAS database⁶⁴. Data was extracted using two DATRAS web service Application Programming Interfaces (APIs): (1) the HHData, that returns detailed haul-based meta-data of the survey (e.g. haul position, sampling method etc.) and (2) the CPUEPerLengthPerHaulPerHour for the catch/unit of effort per length of sampled species.

The presence data were represented by the catch/unit of effort (CPUE), expressed in kg of biomass of the specified species per one hour of hauling. The biomass was estimated by using the SAMLK (sex-maturity-age-length keys) dataset for ICES standard species. This approach is a viable alternative to presence-only data models, as it tackles the biased outcomes resulting from an non-uniform marine coverage of the data sets (mainly along the shipping routes)⁶⁵. The absence data were estimated using the methodology presented by Coro et al.⁶⁵, which detects absence location for the chosen species as the locations in which repeated surveys (with the selected species on the survey’s species target list) report information only on other species.

Additionally, the predictive model automatically correlates the presence/absence data with environmental conditions (Appendix E, Table 3) data to more accurately estimate the likelihood of species presence in the North Sea basin. Intersecting a large number of surveys containing observation data on the presence of selected species can return the true absence data locations, which represent a valuable indicator for geographical areas with unsuitable habitat (see methodology by Coro et al.⁶⁵). Those locations were estimated from abiotic and biotic parameters and differed to the sampling absences which were estimated from surveys without presence data⁶⁵. The environmental conditions (Appendix E, Table 2) data were accessed through direct queries using the MOTU Client option from the Marine Copernicus database. In order to input the layers to the CEA calculation, the input layer for the biomass was transformed using log[x + 1] to avoid an over-dominance of extreme values and all datasets rescaled from 0 to 1 in order to allow direct comparison on a single, unit-less scale⁵⁵.

The rescaled special distribution of biomass for the selected species are presented in Appendix F (Fig. a–j).

OWF pressures and relative weights

A systematic literature review was conducted to reach a first quantification of the OWF pressure weights (({w}_{i,j}),) in the construction, operation, and decommissioning phases (({U}_{i})). The OWF-related pressures specific to each of the phases of the OWF life cycle were based on the comprehensive analysis of all the existing Environmental Impact Assessment (EIA) methodologies used in the North Sea countries¹⁴. The review enabled the collection of 18 pressures that were subsequently compared and merged with the pressures established in the Marine Strategy Framework Directive, applied by the EU countries in the assessment of environmental impacts⁶⁶. Figure 7 illustrates the impact chain linking the three OWF development phases with the exerted 18 pressures and the 12 selected environmental components impacted.

Sensitivity in this research is defined as the likelihood of change when a pressure is applied to a receptor (environmental component) and is a function of the ability of the receptor to adapt, tolerate or resist change and its ability to recover from the impact⁶⁷. The criteria for assessing the sensitivities of environmental components is based on MarLIN (Marine Life Information Network) detailed criteria (https://www.marlin.ac.uk/sensitivity/sensitivity_rationale).

We validated the weights of pressures (({w}_{i,j}) from 0 to 5) and scores of environmental components sensitivities (({s(P}_{j}, {E}_{k})) from 0 to 5), as well as the distance of pressure propagation (≤1000 m to ≥ 25,000 m), through a series of 4 questionnaires for the marine mammals, seabirds, fish and seabed habitats. The compiled questionnaires were further validated through semi-interviews of 9 experts in the field of marine ecology, spatial planning, environmental impact assessment and offshore wind energy development. The expert-based questionnaires also included a confidence level for the proposed scores, which ranged between 0.2 (very low confidence: based on expert judgement; proxy assessment) and 1 (very high confidence: based on peer reviewed papers, report, assessment on the same receptor). The confidence level was used in determining the final scores for the pressure weights and species sensitivities. The final scores for weights and sensitivity scores were identified either by calculating the mean value (for cases where literature review scores and expert scores differed by > 2 units) or selecting the higher value—precautionary principle (for cases where scores from different sources differed by < 2 units). The definitions used, the values for the final environmental components sensitivity scores and the pressure weights are presented in Appendix G, Tables 9, 10, 11, 12, 13. Figure 8 illustrates the relative weights of the 18 analysed pressures in the three development phases. Here, a distinction can be made between localised pressures (e.g., Release of Sediment bound contaminants, sedimentation) and pressures with a higher spatial distribution (e.g., Underwater noise, Marine litter).

Representation of the spatio-temporal distribution of CEA scores

The cumulative impacts of OWFs over the 1999–2050 timeline (Fig. 2b) were represented in hexagonal choropleth maps using the outputs of the Tools4MSP model, over the analysed timeline 1999–2050. The original output of the Tools4MSP model was resampled over a hexagonal grid with a cell size of 10 km resolution (which corresponds with the distance (d) between the centroids of neighbouring cells). The hexagonal cells representing the CEA scores were then grouped in five classes, for the three selected years: 2020 (current state), 2030 (related to the EU energy targets benchmark year), and 2046 (final part of the analysed period, with higher impacts of the decommissioning phase). The final maps were realised using the open source software QGis 3.6.0⁶⁸.

The figures representing the temporal distribution of OWF areas, the CEA scores by phase (Fig. 1) and by country (Fig. 2a) were realised using Excel. The distribution of CEA scores by individual OWF, the link between pressures and environmental components, CEA scores per species (Fig. 4) and the pressure weights scatter plot (Fig. 8) were realised using R 4.0.2⁶⁹. The alluvial diagram representing the impact chain between OWF phases, pressures and species was realised using RawGraphs (https://rawgraphs.io/).

Sensitivity analysis

A sensitivity analysis was performed to understand the influence of the modeled parameters on the CEA results. For this purpose the Tools4MSP Software V3.0 (2021)⁷⁰ implements a novel module that enables a sensitivity analysis, applicable to CEA and to any other modules of Tools4MSP Modeling framework. In this study, three sensitivity analyses were performed, one for each OWF phase (construction, operation and decommissioning). For each phase two groups of uncertainty components were analyzed: the pressure propagation model (distance and weight) for each pressure generated within each OWF development phase and the effect functions, which models the effects of the generated pressures on the environmental components. For each phase 5600 randomized model runs were performed and CEA results were used to estimate the Total Order Index for each input variables (effect functions and pressure propagations). The confidence level included in the expert-based questionnaire was used to model the variability of input variables (uncertainty components) of the randomized model runs. The Total Order Index expresses the contribution of each variable in determining the output variance (uncertainty) including all variance caused by its interactions with any other uncertainty component. The advantage of the Total Order Index is that it can measure the effect of interactions in non-additive systems⁷¹.

Source: Ecology - nature.com