Guidance Note 2: Guidance to support Offshore Wind Applications: Advice for Marine Ornithology Baseline Characterisation Surveys and Reporting
Version 1: January 2023
This document is part of a series of guidance notes available on Advice on marine renewables development - marine ornithology.
This guidance note sets out our advice on baseline site characterisation for marine ornithology. Please see Guidance Note 1: Guidance to support Offshore Wind Applications: Marine Ornithology - Overview which provides the context and description of our suite of guidance notes. Check the log of updates in guidance note 1 to ensure the most current version is being followed. We expect each and every developer to adhere to this guidance, including the recommended parameters to provide consistent and comparable results for all Scottish projects to be assessed as an individual project and cumulatively. This guidance note should be used in conjunction with the Cumulative Effects Framework (CEF).
Please note this guidance was written prior to the development of a NatureScot position on how to account for the ongoing Highly Pathogenic Avian Influenza (HPAI) mortality event within an impact assessment. We are currently developing this advice and will provide updates and guidance as they become available. In the interim please contact [email protected] should you have any specific queries.
Baseline characterisation of a proposed development area is a vital first step in the assessment process for offshore wind developments. It underpins the marine bird impact assessment and therefore the collection methods, presentation and analyses of the data informing the site characterisation need to be robust, transparent and provide the best available evidence.
This guidance note outlines the minimum standards we would expect for determining:
1) the use of pre-existing data;
2) the collection of new data,
3) the analysis of baseline characterisation data and
4) the presentation in both interim and final reporting stages.
While this note provides general guidance, detailed requirements for baseline characterisation and survey are development location and project-specific. We therefore advise that the approach to site characterisation and any survey design should be agreed with NatureScot prior to any surveys commencing.
2. Collection of new data vs use of existing data
The suitability of existing data for use in site characterisation for a development area is dependent on the quality and relevance, in particular the age and temporal and spatial coverage of previous surveys.
We currently accept the use of data up to 5 years old (prior to submission date). Beyond this the relevance of the dataset reduces (due to environmental variations and change). If data are to be used which will be 5 years or older at the time of application, or if there are any concerns regarding spatial or species coverage, please contact [email protected] for further advice before using such data to support any application.
Where existing data collected within the past 5 years are proposed for use, we would require justification on their relevance (e.g. spatial overlap and applicability) and a clear, robust methodology for use. We advise power analyses will be useful to determine whether sufficient data exist to enable adequate baseline characterisation and robust assessment of changes that may arise should the project be consented. Consideration should also be given to whether the age of the available offshore survey data is well-matched to that of those data underpinning the most recent population estimates at seabird colonies that could be impacted; if not this may require developers to undertake targeted colony counts to improve temporal alignment. We will also require early discussion and agreement on the suitability and proposed methods for analysis.
3. New data collection: survey design considerations
We advise that baseline characterisation should comprise two years of monthly surveys. Surveys should commence either at the start of the breeding season or the non-breeding season but not mid-way through a season, (i.e. surveys should commence in either March or October). Prior to any surveys being undertaken the survey methodology should be agreed with NatureScot.
We advise that consultation on any proposed survey design should consider the following points:
- Details of the proposed development area including potential cable routes, which should inform required survey area, survey aims and data collection methods. Survey design should also be informed by consideration of likely species to be encountered and specific characteristics (e.g. size, flight behaviour, density and distributions). Particular attention should be given to possible presence of cryptic species and nocturnally active species (e.g. shearwaters and petrels), which may not be recorded effectively using standard survey methods. Sources giving distributions of seabirds at sea can be used as a general guide to species that are likely to be present, from Stone et al. (1995) and Waggitt et al. (2019) An atlas of seabird distribution in North-West European waters.
- The following survey methodology information should be included:
- Planes and equipment used for digital aerial surveys – this should include flight elevation (ASL) of the observation platform, flight speed, GSD, survey design (grid/transect and spacing etc), percentage coverage, object detection and ID rates.
- Summary of vessels used for boat based surveys, vessel details should include length, height of observation platform, vessel type (e.g. fishing boat), vessel speed, and any additional relevant information.
- Vantage point (VP) details, such as, location, height, view shed (i.e. area observed from the VP), equipment used to detect animals and record distances, together with any relevant maps or figures. Note that vantage points methods are only viable for developments within 2 km of the coast.
- Survey team employed (for boat and shore-based methods): number of observers, split of observers (birds versus mammals), level of experience (confirm ESAS-trained bird surveyors with ESAS survey experience), time on survey/time off survey.
- Map of survey site showing planned transect routes and area to be covered by the surveys.
- Where flight heights are being collected then the method to measure flight height should be discussed and agreed with NatureScot and Marine Scotland.
- Consideration of the survey design area - we advise a minimum of a 4 km buffer for both digital aerial and boat surveys, with a wider buffer of 6km for commercial scale developments, to prevent influence of edge effects at 4 km when modelling marine bird distribution across a site. However, if sensitive species such as sea ducks and divers, in particular red-throated diver, are likely to be present a buffer of 10km will be required. When developers provide interim survey results, then potential issues with sensitive species will be considered and survey design adjusted if required. The reasons for designing a survey with a buffer of less than 6 km or with non-standard coverage or flight height must be included with any design statement issued prior to commencing surveys. If a previously agreed methodology is proposed then a reference to the agreed methodology should be provided and checked that it is still suitable to rely on if older than 12 months.
4. Data analyses
The impact assessment process for offshore wind requires three stages of analysis of marine bird survey data.
The first is the interim surveys which provide the initial results (we advise that interim results are provided on the basis of seasons i.e. the 1st breeding and 1st non breeding season of species present at the site, providing the temporal and spatial coverage being attained by the survey methodology as well as the type and umber of species.
The second is the baseline characterisation which provides the detailed spatial data for all species present and forms the basis of the main impact assessment (including density estimates which feed into collision, displacement and apportioning calculations). This is undertaken for all species present on the development site for which there are sufficient data.
The third analysis undertaken is at the post consent stage, which will be focussed on the key species for which impacts have been identified and require pre- and post- impact comparisons. This third analysis is not addressed further in this note.
3.1. Interim surveys
Reports should be submitted timeously to NatureScot and Marine Scotland as close to the end of each of the breeding and non-breeding season surveys from year 1 - to inform discussion and identify if any survey design modifications are required prior to starting the year 2 surveys. This should be submitted with sufficient time for review prior to starting the year 2 surveys. These interim survey reports should include:
- Information on data collection and analysis, and species observed within the survey area.
- Presentation of survey trip information, such as fidelity to planned transect route, number, dates and timings of surveys, proportion of surveys in each Beaufort sea-state, tidal state, weather conditions, visibility, how data is recorded, and an example of a completed field recording sheet.
- It should identify any issues with survey collection i.e. missed months, unexpected observations, high densities etc. It will not be necessary to present density surface modelling in interim reports. However, it would be helpful to discuss and agree how spatial heterogeneity in density might be analysed and which method(s) might be used.
- If actual survey routes differ from those planned then a map showing actual routes will also be required. Note that for boats, it is particularly important to include information on whether both sides of the boat were observed and whether this was the case for all species.
3.2 Baseline characterisation
Baseline site characterisation should include detailed spatial data for all species present and forms the basis of the main impact assessment. We recommend that the list of species for which model based population estimates would be generated should be agreed through scoping (and informed by the survey reports). All other species with sufficient data should have design based estimates.
The recommendations in the guidance document Mackenzie, et al. (2013) Statistical Modelling of Seabird and Cetacean data: Guidance Document. University of St. Andrews contract for Marine Scotland; SB9 (CR/2012/05) clearly sets out the key principles for modelling seabird and cetacean data – these have been summarised in appendix 1 to this note, however Mackenzie, et al (2013) provides context and detail for these.
We advise that the following points are taken into account while undertaking and reporting on data analyses:
- Present the methods used to extrapolate from the density sample to the population of the entire development site ± buffer. Consider how the heterogeneity in densities may affect the population estimate.
- Present information on the numbers of birds present in segments, with a description of segment definitions.
- MRSea should be used for density modelling approaches and users should check GitHub for the most up to date versions. Mackenzie, et al. (2013) should be followed for presentations of outputs and model fit. However, if the number of data points for a species is less than 10, or the species are present in a uniform distribution it may not be possible to run the spatial element of MRSea. If this applies we will require this explanation to be set out for any relevant species and design based estimates can be used.
- Density calculated from monthly surveys (either model or design based), which needs to be applied to half month periods because the seasonal definition splits the month, should be applied at the same value for each half month.
- Density estimates should be presented with uncertainty estimates, i.e. combined standard deviations and coefficient of variation, as defined by standard deviation divided by the mean. Density values should be derived using methods that account for spatial autocorrelation (e.g. blocked bootstrapping). Details of how spatial autocorrelation has been identified and subsequent approaches used should be provided.
We provide examples of possible covariates and sources that could be used in modelling in Appendix 2 – however, note this is not exhaustive.
Reports should be concise and clear; presenting key information from surveys and supported by maps, figures and tables that are easy to understand. The guidance below applies to the results from all survey types.
For interim and baseline site characterisation reports the following should be presented:
- Include appendices of raw counts in tabulated form.
- Presentation and discussion of species densities and distributions for the entire study site, broken down into the development site and buffer area. Consideration should be given to potential explanatory factors around estimated density and distribution, such as water depth, distance from the coast, and attraction to fishing boats.
- Densities should be presented for birds on the water and in flight, with a combination of the two to give the mean density and overall population.
- Details of any correction factors should be provided, e.g. how unidentified birds are apportioned to species and any potential biases in this; if availability bias been applied; or if distance methods used to account for surveying across ecological gradients.
- Where Distance software has been used, include information on model selection, plot of detectability function and density estimates with associated error bars.
- Presentation of seasonal/monthly population estimates to allow impact assessments to be conducted on the relevant population scale (i.e. breeding, over wintering etc.) The seasons should reflect our recommended periods.
For baseline characterisation reports to inform assessment we advise the following are presented:
Where model- based estimates are presented we require:
- Maps of mean density
- Maps of upper and lower confidence limits
- Distribution of Coefficient of Variation (CV as defined by standard deviation over mean)
- Scaled Pearsons residuals
- Predicted vs Observed Plot
- R-squared (R2) and root-mean-square error (RMSE)
- Cumulative residual plot
- Partial dependents plot
- Assessment of statistical significance of selected covariates
We advise Species Accounts include the following:
- Order first by taxonomic species groupings then by most abundant species within each grouping (e.g. gulls, auks). For unidentified species these should be described to the lowest taxon, with a certainty level reported for each. A description should be provided to specify what species the unidentified groupings are likely to encompass.
- Provide details on the conservation status of each species.
- Provide details on the population estimates for each species at the regional, national and biogeographical region, including age of these estimates and the population status, i.e. increasing, decreasing, or stable. Compare with the population figures derived from the site surveys to identify which are the key target species.
- Provide information on the species’ behaviour and site usage.
- Provide details on whether the species is considered vulnerable to climate change. Sources such as the MarPAMM seabird and climate change reports Study to examine the impact of climate change on seabird species off the east coast of Scotland and potential implications for environmental assessments and Searle et al. 2022.
- Provide any contextual information for the area together with references, e.g. other survey results, local knowledge on distribution and abundance, tagging work, and colony monitoring.
- If agreed during pre-application discussions, provide information on seabird flight direction and height.
- Maps should be at a scale to allow for sufficient contextual information to interpret a development location (for example, including the coastline, location of other proposed developments). Maps should include a compass rose and scale bar.
- Figures should have clear axes with explanatory labels. Gridlines should generally be avoided. Where appropriate, axes can be interrupted so that infrequent large data values do not swamp the information elsewhere in the plot. Figure legends should appear below figures and table legends above tables. Legends should allow the figure or table to be largely interpreted without having to refer to the body text. Column and/or row headings in tables should be explained if necessary, as should axes labels for figures.
- Identify the periods of peak abundance of key species and discuss how this may affect future impact assessment.
- Identify key potential impacts (e.g. collision risk or displacement) for the target species.
The following information should also be included in, or submitted at the same time as the survey reports:
- Data sources for other species of concern not addressed by the survey methodology (such as passage species).
- The results from any additional survey work, such as counts and tagging studies. Consider whether results from these surveys can be shown on maps alongside count data.
- Any other available surveys or contextual information that may be used in the Environmental Impact Assessment or any Habitats Regulations Appraisal – for example, ESAS surveys, Stone et al. (1995) and Waggitt et al. (2019).
- Clarification on how datasets from different surveys will be analysed, taking into account any constraints on data collection, and how these may be interpreted.
- Proposals for further survey work, including targeted survey work on key species.
- References for any published reports or papers cited.
Information from site characterisation surveys or other agreed data sources, along with species impact pathways and sensitivity to an impact will be used to further inform decisions on likely significant effects. Further information on the HRA process can be found on the NatureScot HRA guidance.
5. Tagging data
The use of tracking data from nearby colonies can provide valuable insights into the behaviour of birds around the development area and beyond. This can be used to confirm connectivity with an offshore wind development and understand the usage of n and around a development site. These data may also provide contextual information which may be used in aspects of the impacts assessment such as within SeabORD. If a developer is interested in undertaking tracking of seabird species or has tracking data they wish to use within their assessment, this should be discussed at the scoping workshops.
Mackenzie et al. 2013 reviewed existing modelling approaches used in offshore wind EIAs and made a number of recommendations for improvement. Mackenzie, et al. 2013 sets these out in more detail in section 9, however, the key principles to consider were highlighted as:
- What is the nature of the response variable? Is the response of interest continuous or discrete? Are there natural boundaries to the response variable? For example, are the response data counts bounded by zero? If the data are counts, are there large numbers of zeros?
- What is the observation process like? Are all the animals at the surface likely to be seen by the observer, or are they likely to be harder to see at distance from the boat/plane? Are the focal animals ever underwater and therefore unavailable to be seen at the surface by the observer?
- How were the data collected? Were the observations sampled randomly from a set of independent locations or is there some natural order to the response data in space and/or time? For example, were the counts sampled from a boat or plane travelling along transects across the ocean on a set of surveyed days?
- What is the broad nature of the relationships between the continuous covariates and the response data? Are the numbers of animals (or the presence/absence of animals) likely to rise and fall (or fall and rise) as the covariate increases in value? For example, animal numbers might tend to be low for shallow depths, most common at moderate depths, and then found in lower numbers again in the deepest waters (e.g. Figure 88 within Mackenzie et al. 2013).
- Is there a spatial element to the data? Is there a desire to include a spatial term in the model as a proxy for unmeasured covariates? For example, animal numbers (or the presence/absence of animals) at each spatial location will likely be due to a complex mix of covariates, and many of these covariates will not be available for inclusion in a model. This is likely to result in unmodelled patterns in the response data across the surface which can be approximated by a spatial smooth term.
- Exploratory Data Analysis
Exploration of the data is necessary to familiarise the user with the data available for modelling. Even simple plots can help the analyst identify any obvious errors in the data entry/download, any large gaps in the covariate range and any outlying values in the response data. The user should also ensure the candidate covariates considered are not too similar (‘collinear’) to avoid issues such as model instability.
- Choosing model covariates
Choosing which variables to include in a model is an important part of the modelling process; including too few variables makes for poor predictions and including too many variables can make predictions highly uncertain. Sometimes, the user has an interest in choosing between a select set of models chosen in advance (e.g. based on prior information/analysis) but often models are chosen using automated methods. If the automated approach is employed, efforts should be made to choose from the full set of models available; this enables the user to see how well all candidate models fit the data and either present the best model or some aggregate of similarly performing models. Whether a model is improved by including covariates or if it is meaningful to run the model is site and covariate specific, as if the covariates are low resolution they may not give much additional information. Using it in a predictive way, anything that helps describe distribution will improve the predictive power, and there are many studies that can help inform on what covariates may be used.
- Evaluating the performance of the fitted model
The performance of the chosen model can be assessed by examining how close the model predictions are to the response data. A good model should produce predictions which are close to the input data and high fit scores. Cross-validation is the preferred approach to model fitting but this takes a long time to run. We recommend using information criterion for choosing covariates and cross-validation choosing between models (e.g. month/season), and suggest the use of BIC/QBIC to avoid over-fitting.
- Diagnosing model problems
Modelling is an iterative process and statistical models must be updated, where possible, in light of any inadequacies uncovered. Specifically, it is important to interrogate the working model to see if model assumptions are reasonable and the associated results are defensible. For any particular model, the user needs to understand under what circumstances, and in what ways, does the model not describe the data well and convey these limitations to the reader along with model results.
- Interpreting model results
The uncertainty about the estimated parameters is crucial when drawing conclusions based on a model. For instance, while some model (impact-related) terms might be included in a model chosen using objective fit criteria (e.g. the QAIC/QIC) there might be considerable uncertainty about these estimates and it is important to view these coefficients in light of the associated uncertainty. For example, the magnitude of the estimated pre/post impact difference might raise considerable concerns. However, if ’no-change’ post-impact is also plausible in light of the data (as indicated by a 95% confidence interval about the impact parameter) then this will not likely provide compelling evidence for a pre/post impact difference. In this case, the 95% confidence intervals might better serve as ‘best’ and ‘worst’ case scenarios for any impact-related differences in animal numbers.
List of possible covariates
Below is a list of covariates that could be considered for modelling and some possible sources. Note that neither is exhaustive.
- Bathymetry – e.g. GEBCO, EMODnet
- Bathymetric slope – derived from bathymetry
- Sea surface temperature – satellite derived e.g. MEaSUREs-MUR (NASA Jet Propulsion Laboratory)
- Sea surface temperature gradient – derived from sea surface temperature
- Distance to sea surface temperature/ocean fronts – derived from sea surface temperature
- Mixed layer depth (if available at fine scale resolution)
- Chlorophyll a e.g. MODIS (NASA)
- Distance to the colony (Euclidean)
- Distance to anthropogenic activities, e.g. active shipping lanes and wind farms
- Distance to active fishing vessels (if collecting AIS data)
- Wind speed – e.g. ECMWF