Keywords: off shore, wind farm, monitoring, bird population, aerial survey

Transcription

1 ICES CM 2012/O:18 Not to be cited without prior reference to the author Aerial survey techniques for assessing offshore wind farms Stuart C. Clough 1, Stephanie McGovern 2, David Campbell 3, Mark M. Rehfisch 4 1 Director, APEM Ltd, Riverview, A17 Embankment Business Park, Heaton Mersey, Stockport, SK4 3GN. Tel: (s.clough@apemltd.co.uk) 2 Senior Ornithologist, APEM Ltd, Units 3 and 4a, Aviation Business Centre, Flint Road, Saltney Ferry, Chester, CH4 0GZ. Tel: (s.mcgovern@apemltd.co.uk) 3 Principal Remote Sensing Scientist, APEM Ltd, Riverview, A17 Embankment Business Park, Heaton Mersey, Stockport, SK4 3GN. Tel: (d.campbell@apemltd.co.uk) 4 Associate Director for Ornithology, APEM Ltd, Wellington House, East Road, Cambridge, CB1 1BH. Tel: (m.rehfisch@apemltd.co.uk) Keywords: off shore, wind farm, monitoring, bird population, aerial survey Abstract The offshore wind industry has grown rapidly, mirrored by an expansion and evolution of environmental survey techniques. Observer-based techniques have been supplemented, and sometimes superseded, by technology-based tools, each with strengths and weaknesses. It is imperative that developers choose the most appropriate survey tools, fully understanding the best methods for answering the relevant questions. Similarly, regulatory decisions must be informed by full comprehension of surrounding issues. Differences in bias and precision between survey methods and platforms make calibrations complicated, and best avoided if clear and far-sighted survey strategies are identified from the outset. Historically, bird and marine mammal surveys have relied upon observer-based visual transect methods. As technology has evolved, developers and regulators requiring new levels of scientific verification in their data have welcomed digital data acquisition. In this transition period, attempts have been made to splice data from different collection methods into an effective and usable form. This has led to desk-based and field-based comparisons between different methods, and an assessment of the ability of different methods to estimate populations with precision. Based on the experience gained over the last decade, we discuss the relative merits of the main survey techniques, and explore the challenges posed by inter-method calibration. We discuss recent developments in data collection and analysis methods, and present a vision of likely future developments. This will allow regulators and developers to make informed cost-benefit decisions based on specific requirements, and take advantage of current experience in this rapidly evolving field. Introduction As a result of the rapid growth in the offshore wind sector, the demand for spatial environmental data is constantly increasing (Wilkes et al., 2012). Traditional aerial survey methods suffer from reliability (Van Der Meer and Camphuysen, 1996) and safety (Thaxter and Burton, 2009) drawbacks which are overcome by new digital methods, and we discuss

2 application of these new technologies to offshore aerial bird surveys. Bespoke surveys collect a series of high resolution (HR) digital still images which are captured using state-ofthe-art, aircraft-mounted camera equipment, typically forming a grid of images across the survey area. This technique has been applied successfully to generate statistically robust population estimates and describe distribution for offshore wind farm sites around the UK. Obtaining accurate and precise estimates of bird abundance is crucial to determining and measuring effects of offshore wind farm construction, whilst exact geo-referenced locations of birds can inform decisions on turbine location. This technology can help answer essential questions about the impact of offshore development. The evolution of aerial survey methods Standards for boat based ornithological surveys were set out in the 1980 s and were developed into the 21 st century. Ornithological surveys have traditionally been carried out using observer-based methods, by boat or low-level aerial-based bird spotters; techniques that can often be costly and/or unreliable, along with posing greater risk to participants. Due to the height of wind turbines, it is also neither appropriate nor safe to carry out lowlevel visual surveys after the construction of a wind farm, and therefore alternative, safer methods would have to be employed post construction. The need for a complete survey package led to the first trials of digital data collection methods in 2007, with comparison and calibration surveys taking place from 2009 onwards. Calibration APEM has carried out several calibration flights to ensure continuity during the transition to digital aerial surveys and to assess the variation in population estimates resulting from the different survey methods listed below. Visual Boat-based The difference in raw bird counts recorded between aerial and boat-based methods may, in part, be due to boat association, namely an attraction to boats which may represent a source of food. Boat association with fishing vessels is typically displayed by certain seabird species, especially aerial foragers (Skov and Durinck, 2001; Gremillet et al., 2008). In comparison, digital aerial surveys are flown at an altitude that avoids disturbance to birds, consistent with COWRIE recommendations (Thaxter and Burton, 2009). For other species of seabirds, the presence of boats can have a displacement or avoidance effect. In the former case, birds flee the oncoming survey vessel. They are thus either not detected (under estimated) or are detected away from the vessel (affecting the recorded distance from the boat, which affects statistical estimates of population size). The distribution (and estimated abundance) of birds recorded in all of these cases does not represent a natural undisturbed scenario. Diver species in particular are sensitive to disturbance with specific conservation objectives to this end. Because digital aerial surveys operate at altitude, disturbance to target animals is kept to an absolute minimum, meaning accuracy of data is much greater.

3 Visual aerial While visual aerial surveys are useful for surveying large, remote areas in a short period of time, the low altitude generates large-scale disturbance amongst birds and Health & Safety concerns prevent post- (and sometimes pre-) construction monitoring. As the probability of an observer detecting an animal decreases with distance from the aircraft, the number of animals detected is not equal to the number of animals present, adding another layer of complexity to statistical analysis of the data. Digital aerial At present, there are two main digital aerial survey techniques. HR digital still imagery is based on a grid design, whereby a series of independent images with a randomised starting point are collected throughout the study area. HD Video methods, in contrast, typically collect a continuous stream of data along line transects which run in parallel across the survey region. Both methods allow the production of population estimates with a given level of precision; however statistical power is generally lower with continuous sampling methods due to a lower number of spatially independent samples collected during a survey (Thaxter and Burton 2009). Grid-based survey methods Grid-based aerial surveys are undertaken in twin-engine aircraft fitted with state-of-the-art cameras and bespoke flight management systems. These systems, controlled by camera technicians, produce digital still imagery at resolutions as high as 2cm (Figure 1). Such images are used to identify birds (and marine mammals) from a safe altitude where ecological disturbance is minimal. An aircraft can safely cover huge areas in a single day and collect large datasets quickly and efficiently. Data is collected in a grid pattern, each image being equidistant from its counterparts on the x and y axes. Survey design is defined prior to commencement of monitoring and therefore the amount of data collected is dependent on the size of the area and the spacing between image nodes. Analysis of imagery Whilst the collection of large datasets has clear advantages it has also presented the industry with a need to develop innovative ways to handle and analyse data. Bespoke bird identification software has been developed using novel object recognition algorithms. This has been linked to a tailored Graphical User Interface (GUI) which automates the geolocation of birds and enables the measurement of wing-span, length and flight height and direction, thus speeding up the bird identification process prior to analysis by APEM s specialist ornithologists. Automated analysis tools have also been developed in-house, to differentiate between species where similarities in colour and plumage are not distinguishable by eye. Developers often require a rapid turnaround between survey and data delivery, therefore maximising automation at this stage of the process without reduction in output quality - is key to meeting industry requirements. Such datasets can be stored permanently and re-visited when required.

4 Figure 1. Gannet Colony on Bass Rock (Firth of Forth) at c.2cm resolution Population estimates and statistical analysis APEM routinely use a systematic grid-based approach to data collection, collecting a representative sample of the birds present at the time of survey. Each sample can be considered an independent estimate of population. This survey approach is conducive to spatial statistical analysis which regulators deem increasingly reliable for bird population assessment. Non-parametric bootstrap methods are used to provide confidence limits and contribute to the high degree of statistical robustness. Precision calculations are used to obtain a coefficient of variation (CV) which can be used before or after survey to determine the ideal number of image samples to detect a change in the population of a particular species. For a species of interest, a CV of <0.16 is targeted, enough to be able to detect a doubling or halving of a population. Model-based spatial distribution estimates Using relevant covariate data, model-based statistical approaches such as General Additive Models (GAMs) can be used to produce density surface maps over the study area (Figure 2). These display areas of lesser and greater density, based on model predictions and including environmental parameters such as bathymetry, geological substrate, distance to shoreline and distance to relevant Special Protection Areas (SPAs). As an equal number of images are collected on the x and y axes in a gridded survey, coverage of the survey area is evenly spread, meaning that models including environmental variables are not biased by disproportionate coverage of relatively more (or less) important habitat. Pre and postconstruction geospatial comparisons can be performed to illustrate the changes in bird

5 densities across the survey area resulting from wind farm development. This analysis will detect and pinpoint where these changes are the most important, and will reveal whether these changes are statistically significant. 5 km Figure 2. Example GAMs model output The future of HR aerial photography The scientifically robust nature of the survey design, and the significant cost savings compared to other techniques, has led to a huge increase in the demand for HR aerial photographic surveys. As such, it is expected that this survey technique will quickly become an industry standard. Increasingly, developers are seeing the merits of HR still photography for other aspects of offshore wind farm development. For example, identifying and selecting the most appropriate route for the cable which brings the energy ashore is a complex process which can be informed by high resolution still photography. Remote sensing techniques are becoming increasingly mainstream for rapid and reliable environmental assessments. It is now possible for developers and regulators to debate the

6 potential impacts of proposed and constructed offshore wind farms with reference to scientifically robust data, without the need to employ the precautionary principle to account for the shortcomings and biases associated with the data collection method. Summary High resolution (HR) aerial photographic surveys provide a modern, viable and safer alternative method to traditional data collection and handling for the offshore wind industry. The high resolution data outputs obtained from HR aerial photographic surveys of proposed and operational offshore wind farm sites have generated robust datasets on which critical planning and implementation decisions can be based. The data outputs are of a very high standard, enabling detailed statistical analysis to examine avian population densities and abundance in the offshore environment. Confidence in the implementation of the data collection, handling and statistical analysis methodologies is ever increasing as methodologies are refined and uncertainty is reduced through improved survey design and modeling techniques. References Grémillet, D., Pichegru, L., Kuntz, G., Woakes, A.G., Wilkinson, S., Crawford, R.J.M. & Ryan, P.G. (2008) A junk-food hypthesis for gannets feeding on fishery waste. Proceedings of the Royal Society Series B, 275, Skov H., & Durinck J. (2001) Seabird attraction to fishing vessels is a local process. Marine Ecology Progress Series, 214, Thaxter, C.B. & Burton, N.H.K. (2009) High Definition Imagery for Surveying Seabirds and Marine Mammals: A Review of Recent Trials and Development of Protocols. British Trust for Ornithology Report Commissioned by Cowrie Ltd. Van Der Meer, J., and Camphuysen, C. J. (1996) Effect of observer differences on abundance estimates of seabirds from ship-based transect surveys, Ibis, 138 (3), Wilkes, J., Moccia, J., Arapogianni, A., Dragan, M., Plytas, N., Genacthe, A-B., Guillet, J., Wilczek, P. (2012) The European offshore wind industry key trends and statistics. The European Offshore Wind Energy Association.