Offshore wind developments assessment - seabird collision risk, displacement and barrier effects: study

Appendix C

9.3 Full Workshop Report

Marine Scotland

Combining collision and displacement in ornithological offshore renewable energy assessments

Workshop

Victoria Quay 10 December 2019

Attendees

Marine Scotland

Tom Evans, Janelle Braithwaite, Elaine Douse

UK Centre for Ecology & Hydrology

Francis Daunt, Kate Searle

BioSS

Adam Butler

MacArthur Green

Mark Trinder

BTO

Aonghais Cook

RSPB

Aly McCluskie

HiDef Aerial Surveying

Ross McGregor

DMP Statistics

Bruno Caneco

SNH

Alex Robbins (phone)

JNCC

Julie Black, Lise Ruffino

Purpose of workshop

This workshop brought together experts from research, government and conservation bodies to discuss how best to implement combined modelling of displacement and collision risks from offshore wind farms (OWFs) for breeding seabirds. The aim was to discuss how current methods are used to estimate collision and displacement risks separately in assessments, with a following session on the extent to which the inputs, parameters and assumptions used in the different methods are consistent with each other. Finally, there was a session on how to combine collision and displacement risks into a single assessment of risk from both displacement and collision.

Agenda (Times/Details)

9:30 - 9:50 Reception & registration

9:50 - 10:00 Welcome and introductions

10:00 - 11:00 Presentations: estimating collision and displacement in assessments

• Methods of estimating collision effects (HiDef/DMP)
• Methods of estimating displacement effects (CEH)
• Combining effects: possible approaches (BioSS)

11:00 - 11:15 Tea & coffee

11:15 - 12:45 Combining effects: consistency of inputs (plenary)

• Avoidance vs displacement rates: detailed discussions
• flight speeds

12:45 - 13:15 Lunch

13:15 - 14:45 Combining effects: consistency of assumptions (plenary)

• time periods and life stages
• risk by behaviours
• variation and uncertainty

14:45 - 15:00 Tea & coffee

15:00 - 16:00 Final plenary: integration

Collision Methods Summary

Collision risk modelling (CRM) involves three broad types of data:

1. Wind farm data (turbine specs, latitude, site dimensions etc.);

2. Site specific seabird data (densities and flight heights);

3. Generic seabird data (biometrics, nocturnal activity, avoidance rates, flight speed and flight height).

More specifically, these include data on the following key parameters:

• Bird biometrics: length, wingspan, flight speed, flight type (flapping versus gliding);
• Aerial densities of birds in flight by month (from site surveys);
• Flight height distribution (proportion at collision risk height);
• Avoidance rates – see Cook et al. 2018 for further details;
• Wind farm characteristics (location, number of turbines);
• Turbine characteristics (blades, rotation, radius, hub height, blade width, pitch, % time operational for turbine);
• Hours of daylight (derived from latitude);
• Flux (derived within CRM - see below).

Flux (mean traffic rate, expressed as birds s^-1 m^-2) – based on bird flight speed and density (Band, 2012). This is then used to work out the number of bird flights through the rotor over any given time period, by scaling up based on the duration of the time period concerned and, the total area occupied by turbine rotor sweeps. More specifically, these calculations are used to estimate the proportion of the wind farm which is rotor swept, i.e.: Flux x prop rotor swept x prop at rotor height, which gives the number of at risk flights (rotor transits), which is then multiplied by p.collision ('PColl': wrong place wrong time) and then by avoidance rate.

Monthly total mortality (all birds derived from):

• Number of transits through the rotor (= flux x proportion at flight height);
• Collision probability (physical bird and turbine factors);
• Wind farm collisions (transits x collision probability x proportion of time operational)
• Overall avoidance rate = 1 – ((1 – macroAR) x (1 – mesoAR) x (1 – microAR))

Notes on equivalency of parameters within collision risk modelling and displacement/barrier effects modelling:

• Macro-avoidance although capturing displacement and barrier effects, refers only to birds in flight, not birds on the water;
• Area affected by displacement tends to include a border, but there is no border in collision risk modelling; but there is evidence for avoidance of up to 2km if not further from OWF perimeter;
• The matrix method does include buffer for displacement.

Displacement Methods Summary

The following parameters were identified as key inputs to displacement risk modelling when using the individual based model SeabORD:

• Bird Utilisation Distributions from GPS or at-sea survey data (site data typically not appropriate);
• Median prey availability in model region;
• Colony locations and sizes;
• Apportioning (if at-sea data; known if GPS);
• Bird behavioural parameters based on empirical data and in accordance with optimal foraging theory;
• Location, shape and size of OWF footprint(s);
• Displacement and barrier rate;
• 'buffer width (km) to be added for OWF footprints (within which displacement and barrier effects are assumed to occur; note this is called the 'border' within SeabORD); and
• 'displacement foraging buffer' width (km) to be added to OWF footprints (area into which birds are displaced during foraging).

Key questions:

• Q: Potential for prey redistribution around OWFs – can SeabORD account for this? A: this could be accounted for by altering the prey maps that are used within SeabORD between the baseline and the impact scenarios, but this has not been done before.
• Importance of thinking about birds that are displaced but not barriered – because there is evidence for birds flying through footprints but not foraging within them, so this category is probably quite important. Whereas there is no evidence for the converse (birds that are barriered but not displaced); however it may be that individuals are prepared to invest in a short detour to avoid turbines, but not to cease all access to a key foraging area, so this behavioural response is likely to be more complex than can be captured with simple rules.
• Q: Should this vary for an individual over time? Martin Perrow's work on terns shows changes in displacement propensity over time whereby birds fly through OWFs during chick rearing but not pre-breeding season and incubation – interpreted as birds being more constrained, and therefore less risk averse, during chick-rearing when energetics demands are higher? A: Yes, this could be included within the model with a re-working of some of the mechanisms and constraints assumed at different periods of the season, in order to develop a model capable of simulating over the full breeding season, not just the chick-rearing period.
• Also worth thinking about potential "shadow" effects – e.g. birds no longer using the area directly beyond a wind farm, thereby creating a loss of habitat – we (BTO) have some (as yet) unpublished data on this in Sandwich Terns. It shows flights between the breeding colony & OWF going right up to the OWF edge, with birds diverting round it. The area on the far side of the OWF from the colony is then not used by birds.

Summary of bird-related inputs and other processes required for model runs:

Inputs	sCRM	Matrix	SeabORD
Bird behaviour (flight and on water)	Flight only	Both	Both
Timescales	Monthly	Seasonal	Seasonal (chick-rearing)
Buffer	No buffer	Buffer	Buffer
Bird distribution & density	Site data	Site data	GPS or at-sea-based UDs
Displacement rate	Population level	Population level	Population level
Apportioning required?	No	No	To colony

Methodology for Integration Summary

• Habitats Regulations have to deal with the population in protected sites, not just colonies (unless the whole colony is within the SPA). So both the sCRM and SeabORD need to work at the SPA level, not only individual colonies.
• To inform the consenting process, the potential impacts of the key effects associated with developments are assessed through an Environmental Impact Assessment (EIA) in relation to baseline populations at site, local, regional and national levels. When preparing applications for Nationally Significant Infrastructure Projects (NSIPs) in England or Wales, or for equivalent national developments or major developments in Scotland and Northern Ireland, developers are legally required to consider if the project is likely to affect European sites by providing a Habitats Regulations Assessment or Appraisal (HRA). HRA is an iterative process and the emphasis is on understanding no Likely Significant Effects (LSEs) and demonstrating no Adverse Effects on Site Integrity (AESI) on relevant SPAs. If no LSEs on features of a European site, either alone or in combination with other plans or projects, cannot be ruled the HRA report provided with the application should enable the competent authority to then carry out an Appropriate Assessment (AA). The purpose of the AA is to ascertain whether there is no AESI on the relevant sites. Under the EC Birds Directive [2009/147/EC], sites are classified as Special Protection Areas (SPAs) based on the relative size of the population of a species, or suite of species, that they hold and must be maintained in a favourable condition.
• Currently, the expectation is that both the sCRM and SeabORD will be maintained as separate models, and this project will add the facility to use sCRM outputs within SeabORD.
• SeabORD applies to breeding adults from specific colonies, not all birds from everywhere; however, the input to the sCRM includes all birds seen flying in the footprint, including non-breeders, and potentially birds from colonies not modelled within SeabORD, when suing site-specific data from at-sea surveys.
• How to combine collision and displacement during the rest of the year needs to be considered because SeabORD currently only operates over the chick-rearing period.

Consistency of Inputs Summary

• Key is understanding the different types of avoidance assumed in the models: avoidance versus displacement rates, and the effects of the buffer;
• Important to consider the differences in input density information; e.g. all birds (sCRM) versus only breeding adults (SeabORD) or specific age classes; and all birds (displacement) versus only birds in flight (CRM);
• Need to consider flight behaviour – flapping and gliding (for collision probability the current guidance is assume everything is flapping), commuting and foraging, flight height;
• SeabORD could model meso-avoidance if each shapefile represented one turbine; however, this has not been attempted before; and
• The four most sensitive parameters in sCRM are flight height, flight speed, bird density, avoidance rate.

Avoidance and Displacement Rates Summary

• SeabORD models both displacement and barrier effects;
• Macro-, meso-, and micro-avoidance are all one parameter in the sCRM, although there is some empirical evidence for some species for differences between these. Crucially, overall avoidance estimates are partitioned into the three types for most species (Cook et al. 2018), however current values advised by SNCBs do not partition. We can then use the macro-avoidance rate in SeabORD; however, note that currently macro-avoidance rates and advised displacement rates may not be the same;
• This project should focus on black-legged kittiwakes (BLKI) and northern gannets (NOGA) because they are at collision and (in some jurisdictions) displacement risk. There are estimates for macro-avoidance from empirical work for both species;
• For other species would be good to lay out what sort of information you would need to estimate empirical macro-avoidance;
• Using Option 3 in the sCRM is better for integration because there are fewer things to correct – this is because it includes the relative distribution of flight height but still uses an adjustment to some extent (at present, Option 3 still uses the same flight speed adjustment as Options 1 and 2 - but there's the potential to update this given all the GPS tracking data available);
• Where and how do we take the adjustment part of the sCRM into account? This mostly comes into the micro and meso aspects of avoidance, so perhaps not relevant to the macro-avoidance rate that is required by SeabORD?
• A correction factor is applied in sCRM, but it can only be applied to meso- and micro-avoidance, so macro-avoidance is not affected by the correction. This means that the macro-avoidance rate can be used within SeabORD – the assumption is that macro-avoidance is the sum of displacement and barrier effects, and, therefore, appropriate for use within SeabORD which models both processes.

Input density information

• Differences between EIA and HRA were discussed (see notes above) – however, SeabORD is not always relevant for EIA because it is a model that operates at the scale of individual breeding colonies (although multiple colonies may be run at once). However, note that if the only source of birds is a single SPA colony then EIA is equal to HRA, for sites that are close to colonies during the breeding season. A single SPA may incorporate multiple breeding colonies, so this is relevant for HRA.
• Consideration needs to be given to 'All birds' versus 'breeding adults'.
• Including a buffer around a wind farm footprint within the collision risk modelling inputs is achievable because this is a specified input to the sCRM, therefore it can be adjusted to be the same as within SeabORD to ensure consistency between the two methods.
• sCRM only uses birds in flight at sea; however, the displacement matrix uses all birds seen at sea.
• The combined displacement rate and barrier rate must equal 'macro-avoidance' for the sCRM to be equivalent with both the matrix method and SeabORD – this is a critical assumption.
• SeabORD is only chick-rearing period so cannot be used in the non-breeding season.
• Consensus that we should be building the collision risk model into SeabORD, but need to convert data into flux in order to do that, estimate proportion of time in area and proportion of time in distance/area, needed in the collision calculations.
• The current sCRM uses the density of birds in flight – then 'PColl'(the probability of collision when a bird is in the rotor swept area) is multiplied by the number of birds passing through the rotor sweep over a given time period and the proportion of those birds at collision risk height (N.B. – applies to basic model only, not the extended model) – Q: can we plug 'PColl' into SeabORD? It is straightforward to extract 'PColl' from the Band (2012) model. A: Yes – this is straightforward, PColl is typically in the region of ~10% - but note that this applies to the Basic model only. For the extended model, need to use the collision integral, which accounts for the fact that, bird density, probability of interacting with rotor & probability of colliding all vary with flight altitude.
• SeabORD would ideally need a probability of collision per bird per time unit.
• Tagging data can estimate flux much better than at-sea survey data, we need a much better understanding of how birds move at sea. This could be achieved using radar and/or GPS. The current assumptions of a basically constant conveyer belt of birds moving through the wind farm are far too simplistic. This would reduce the need for the avoidance rate to be used as a correction factor. At present, estimates of the avoidance rate combine both avoidance behaviour and model error. Avoidance rates are estimated by comparing observed and predicted collision rates (Cook et al. 2018). The predicted collision rate will be derived from estimates of the total number of birds exposed to collision risk which are largely determined by the flux rate. Consequently, by being able to better estimate the flux rate, we will reduce its contribution to model error and, reduce the need for the avoidance rate to be seen as a correction factor (Masden et al. in prep.) A valuable future goal would be to document how GPS data can be used to estimate flux.
• If we can solve flux this gets at a lot of the bird behaviour issues as well, although not flight height.

Assumptions

• Are there differences in time periods used within the models? – for instance, nocturnal activity is incorporated in the sCRM (albeit coarsely), and nocturnal activity is also included in SeabORD but done in a different way (by setting the number of hours during a 24 hours period in which birds are assumed to be able to forage);
• Seasonal differences – SeabORD is chick-rearing but sCRM calculations are done monthly all year;
• SeabORD includes effects on chicks, but the sCRM does not – Q: would this make a big difference? A: Yes, over the lifespan of a OWF it could. Note that because the mortalities are typically plugged into a PVA, this would actually only be important if the collision occurred during the chick-rearing period: if the PVA is post-breeding census, adults must survive to breed and breeding happens instantaneously – so if a bird dies prior to breeding no offspring will survive, but if the bird survives the breeding period then it may have a successful breeding attempt (or may not, depending on the provisioning achieved by it and its partner);
• Note that SeabORD could be modified to cover the wider breeding season, for instance including incubation, but this would require a substantial amount of work to develop and parameterise different behavioural mechanisms during this time;
• sCRM can kill the same birds in different months over and over again;
• Assumption is usually that all age classes have same collision probability –however, it is likely that immatures differ quite strongly from adults in flight and foraging behaviour in many species;
• Foraging versus commuting flight could be included in the sCRM – e.g. by using different flight height and speed for different behavioural modes.
• The sCRM assumes 50% of birds are upwind and 50% are downwind, and that split affects collision risk;
• Flight speeds also different upwind/downwind – flux doubles with doubling of flight speed, but PColl only changes by about 2%; and
• Birds appear to show less strong avoidance of turbines that aren't spinning – this is implicitly included in the sCRM via the use of operational time in calculations; however, there is some evidence that birds may collide with static objects, and this potential is not currently included within the sCRM.

Variation and uncertainty

• The Matrix method doesn't include variation unless you vary the density of birds;
• sCRM: Typically there are data for 24 months of surveys, thus two surveys per calendar month, one from each of two years. Average density in each month is used, or if stochastic CRM then bootstrap of data from both months;
• Displacement matrix: average in each month (from the two surveys) but then for the season value, the highest of the individual month average values is used ('peak mean') – so if the period is Jan-Mar and you have average densities of 3, 7, 2 in those months (from raw values of 2+4, 5+9, 0+4 for the months) the seasonal displacement would be calculated using 7;and
• However, this means that the more you split the season (e.g. into more months) then the higher the risk is (because peak seasonal mean is used), which is illogical;
• Ideally would feed in the distribution of collision probabilities from the sCRM for SeabORD to use within the simulations; however, can only currently run SeabORD for 10s of simulations due to long processing times; it may be more appropriate to pick simulations and feed these into SeabORD rather than taking a mean and SD and using this to derive inputs because collision probabilities tend not to have unimodal distribution.

Integration

• Important to separate out what we want to achieve in this project, and what the larger objectives are for future work and recommendations;
• We should have a case study using black-legged kittiwake (BLKW) which integrates SeabORD and the sCRM. This should consider:

Behavioural split in SeabORD – SeabORD can partition out time in footprint into foraging and commuting, so should these different behaviours have different collision risks?

BTO project on commuting versus foraging flights using HMMs may be useful here?

It is likely that birds spend considerably more time foraging in footprints than flying through them, so potentially important to capture differences in collision risk?;

• There should be a northern gannet case study using the matrix method and the sCRM:

Important to explore available data for macro-avoidance rates to use in both approaches (e.g. Cook et al report)

Which mortality rate should be used in the matrix method? Same as in the SEANSE project?

Also should consider the Dierschke et al Biological Conservation paper on displacement.

Conclusions and recommendations

The workshop was extremely useful. By bringing together experts in collision risk and displacement modelling, we were able to summarise the key challenges in integrating collision risk and displacement, and agree an approach for integrating sCRM outputs into SeabORD in this project, using kittiwake and gannet as case study species;

Report should set out a vision for a combined model, covering the whole year;

Report should discuss what is needed to validation of individual based models, and give some advice on empirical data needed to do this in the future;

Report should discuss how better classification of uncertainty can be incorporated into IBMs, and that this should decrease as empirical data is used to better estimate key processes;

Report should discuss cumulative impacts (in-combination); and

Future needs were discussed:

• Need to parallelise sCRM to speed it up to the speeds that are obtained when using the underlying code;
• Need to integrate outputs from HMM models by BTO to partition by behaviour;
• Need to incorporate a distribution of flight speeds in sCRM (fixed rates currently used);
• Flight height should consider commuting and foraging flight separately;
• Need to incorporate 3D flight movements, based on the latest GPS tracking data (available for gannet and kittiwake, more planned for both species at more locations);
• Need further empirical data and validation of inputs and outputs of sCRM, matrix and SeabORD; and
• Further developments must consider the growing relevance of cumulative impacts. Some progress has been made but more will be needed.