Detecting seabird displacement. A simulation-based geostatistical approach

(1)

Detecting seabird displacement

A simulation-based geostatistical approach

(2)

prof. dr. F. Eising University of Twente, chairman/secretary prof. dr. S. J. M. H. Hulscher University of Twente, promotor

prof. dr. ir. A. Stein University of Twente, promotor

dr. K. M. Wijnberg University of Twente, assistant-promotor

H. Skov DHI Group, Denmark

dr. ir. D. C. M. Augustijn University of Twente

prof. dr. J. C. Lovett University of Twente

prof. dr. C. J. F. ter Braak Wageningen University and Research Centre

prof. dr. J. van der Meer VU University Amsterdam, NIOZ

This research is supported by:

PhD@Sea, which is substantially funded under the BSIK-programme of the Dutch Government and supported by the consortium We@Sea.

ISBN 978-90-365-3185-6 Cover design by F. Huthoff

Copyright c 2011 by Blanca P´erez Lape˜na, Enschede, the Netherlands Typeset in LA_TEX

(3)

Detecting seabird displacement

A simulation-based geostatistical approach

DISSERTATION

to obtain

the degree of doctor at the University of Twente, on the authority of the rector magnificus,

prof.dr. H. Brinksma,

on account of the decision of the graduation committee, to be publicly defended

on Thursday the 28th of April at 16.45

by

Blanca P´erez Lape˜na born on the 26th of June 1977

(4)

prof. dr. S. J. M. H. Hulscher promotor

prof. dr. ir. A. Stein promotor

(5)

i

-¿Qu´e gigantes? -dijo Sancho Panza. -Aquellos que all´ı ves -respondi´o su amo-, de los brazos largos, que los suelen tener algunos de casi dos leguas.

(6)

(7)

Preface

This journey began with a farewell party. In 2006, after having been for several years in the Netherlands I thought I was leaving Enschede. It made me very happy that so many friends and colleagues came to my farewell party, but it made me even happier that a couple of days later Suzanne Hulscher offered me a position as promovendus at the Water Engineering and Management department of the University of Twente.

The offered opportunity was embedded in the We@Sea project, an initiative that focusses on opportunities and potential risks involved in offshore wind farm development. An earlier study within We@Sea that was carried out at the University of Twente by Henri¨et van der Veen involved impacts of such wind farms on the seabed morphology. My task now was to investigate potential impacts on marine fauna. Suzanne, thank you for this opportunity and for all your guidance along the way.

Without the expertise of several people, this thesis would not have been possible. Jan de Leeuw was the first to come on board to share his expertise. Jan, I am thankful for your large support at the beginning of this research. You left to Kenya but you left me in good hands. Shortly after, Alfred Stein became my promotor. Alfred, your involvement in the supervision team brought my work to another level. Thank you for pushing me the way you did and for the discussions we had. I hope you do not mind the long sentences in this preface. I am extremely thankful to my daily supervisor Kathelijne Wijnberg. It has been a pleasure working with you in all these years. I really appreciate that you were always available for discussions, which most of the time extended for long, and enjoyable, hours. I have learned a lot from you.

Thanks Robert Strobl, Gabriel Parodi, and Mindert de Vries for your com-ments on my very first version of the research plan. The user group meetings helped me in understanding the challenges in the project and which parts I should focus on. Henk Kouwenhoven, thanks for the useful insights from the developers point of view and for the data that I was allowed to use in this research. Jan Tjalling van der Wal, thank you for the valuable discussions. Mardik Leopold, your passion for research in ecological related problems in-spired me. Thank you also for your kindness and for keeping my attention to the relevant issues. Geert Aarts and Erik Meesters, thank you for the insights into the T1 seabird data set and the impact modelling.

Rodolphe Devillers, I am very thankful for the discussions at the ISSDQ 2007 Conference. If the research had not moved into another direction, I would have truly enjoyed working together with you. Rolf de By, thank you for the

(12)

guidance, the great times and for sharing your knowledge on the birds’ world with a “city rat”; Wim Bakker for boosting my python skills; Jacob Asjes and colleagues in We@Sea for reviewing the technical reports; David Rossiter for the discussions on geostatistics and impact assessments; Tom Hengl for the useful GEOSTAT summer school; Nick Hamm for the course on Model-based geostatistics; the R community for the useful forum discussions; Arta Dilo for the discussions on spatio-temporal analysis; Pieter Roos and Andries Paarlberg for making LA_{TEX run smoothly; Pieter van Oel for all the breaks we shared}

at UT and your help in organizing my ideas, and the Augustijns for all-round support! Freek Huthoff, for the time you spent in improving parts of this thesis. I still have a nice feeling remembering our Bayes’ moment. Also, Martin Poot, Johanna Saladas, Marieke Eleveld, Martine Graafland, and Rik Duijts for your efforts in supplying data and information.

Anke, Brigitte and Joke you always surprise me with your problem-solving skills. Thank you for all your help. René and Arthur, thank you for all the technical support and the nice chats during the installation processes. Arjan and Rianne, my roommates for most of the time at WEM. It was really enjoy-able to share the office with you as well as the conversations regarding work, photography, music, whistling in reverse and more. Leendert, my occasional roommate, thanks for your interest and your help in common aspects of both our research. Maite, gracias por nuestras conversacios acerca de la vida en in-vestigación, agua, ecolog´ıa y demás pajarillos. Marcela, por enseñarme a poner en práctica la teor´ıa de los l´ımites, creo que he llegado a entenderla. And all the former and current colleagues at WEM with whom I shared many great moments and conversations. I has been really enjoyable to work in such a nice atmosphere.

During these years I had the privilege of meeting many more amazing peo-ple, too many to mention all separately. To all of you: thank you, gracias, merci, danke, grazie, and of course, bedankt!

When Bertien and Mireia accepted to be my paranymphs I could not have been happier. Bertien, it is a pleasure to have you next to me at my defense, but mostly having you close during all these years. The Macarena will never be the same. To the other partner in crime Mireia, mil gracias por tu amistad y la energ´ıa positiva que siempre me has dado, te sales!

I want to deeply thank my Huthoff family for all the beautiful moments, and the many more to come!

Mis padres, Luis y Mercedes, Luis, Carolina y Andrea. Siempre a mi lado, en los buenos y ‘malos’ momentos. Gracias por todo, no se que har´ıa sin vosotros. . . os quiero con locura.

Freek, mi amor, cannot thank you enough for all these years. You always bring a smile to my face and you make me so happy, every single day.

Blanca P´erez Lape˜na. Enschede, 26 March 2011.

(13)

Summary

In recent years, offshore wind has emerged as a promising renewable energy resource. Apart from the technical challenges in the development of offshore wind technology, an important issue in putting this technology into practice is the tension between development of offshore wind farms and nature conser-vation interests. Therefore, it is important to apply sound impact assessment methods and to design adequate monitoring programs in order to protect and manage the marine environment.

With respect to ecological impacts, seabirds are likely to be affected by offshore wind farms as they may disrupt flight routes, and disturb feeding and resting grounds. Given the complexity of seabird behaviour, the variability in environmental factors that may affect it, and technical limitations in collecting seabird data at sea, detecting the impact of seabird displacement caused by wind farms is a challenge. In this thesis the ability to detect such impact is studied, focusing on the influence of seabird behaviour and survey characteris-tics.

The research approach consists of simulating realizations of seabird count surveys that could arise from the undisturbed situation, i.e. the situation with-out the influence of a wind farm. The simulated surveys are constructed by using parameter values that specify relationships between seabird abundance and environmental conditions, spatially autocorrelated random variation, and survey effort and design. For a given combination of parameter values and using hypothesis testing it is assessed whether the collected post-construction data is a realization of the undisturbed situation or, otherwise, different enough to conclude that the effect of a wind farm has been identified. By using differ-ent combinations of parameter values it is analysed how these variations affect the outcomes of a statistical test, and hence the outcome of an impact assess-ment. The ability to detect impact is investigated by analysing how often a displacement would be detected by the statistical test for various scenarios of reduction of bird abundance in the wind farm area.

In implementing this approach, seabird behaviour is described by a geo-statistical model, which relates seabird abundance to environmental conditions, includes spatially autocorrelated random variation, and accommodates zero-inflation in seabird count data. Survey characteristics are described in terms of the number of survey locations, their configuration, and the spatio-temporal variation of environmental conditions at the time of the post-construction sur-veys.

(14)

sit-uation is constructed by geostatistical simulation. The null hypothesis being tested is that there is no difference in seabird abundance in the wind farm area due to wind farm construction. The study shows that when taking species behaviour and survey characteristics into account, the probability of falsely re-jecting the null hypothesis (type I error) can be reduced. As a result, identified changes in seabird abundance can be better attributed to the presence of the wind farm.

The ability to detect impact is measured in terms of the probability of detecting an imposed impact: the statistical power. In the simulations, the im-posed impact is expressed as percent reductions (25%, 50%, 75%, and 100%) in the mean number of birds in the wind farm area after wind farm construction. The analysis of the simulations shows that different combinations of parameter values for seabird relations to environmental conditions, spatially autocorre-lated random variation, and survey effort and design can provide significantly different probabilities of detecting an imposed impact. In one of the investi-gated scenarios it is shown that, as the environmental conditions become more suitable for birds to be present in the study area, the probability of detecting a displacement increases. For example, when a 50% reduction in bird numbers has occurred, this probability increases from 0.54 to 0.86.

Temporal variability in environmental conditions also influences the ability to detect impact. This is of importance when surveys to collect seabird data span several days and the spatial pattern in environmental conditions varies between survey days. In impact studies it is common practice to aggregate these data over several days. This research shows that due to this data aggregation overestimation as well as underestimation of the ability to detect impact can occur, up to a factor of approximately two.

Once the ability to detect impact is quantified, Bayes’ theorem is applied to determine a suitable number of surveys for a particular study area. This way, a priori knowledge on the probability of drawing erroneous conclusions from survey data can be used to design a survey plan that distinguishes best between presence and absence of impact.

In conclusion, this thesis shows that seabird behaviour and survey char-acteristics can have a large effect on the ability to detect the displacement of seabirds due to offshore wind farm presence. Insight into how these com-ponents affect this ability can support environmental managers in preparing sound monitoring plans, which eventually lead to conclusions with acceptable confidence levels on whether impact occurs or not.

(15)

Samenvatting

Offshore windenergie is in de afgelopen jaren opgekomen als een veelbelovende bron voor duurzame energievoorziening. Naast de technische uitdagingen die de ontwikkeling van offshore wind technologie met zich meebrengen, is een be-langrijk punt bij het in praktijk brengen van deze technologie het spanningsveld tussen de ontwikkeling van windmolenparken op zee en natuurbeschermings-belangen. Daarom is het van belang om degelijke methoden toe te passen in effect studies en adequate monitoringsprogramma’s te ontwerpen om het mariene milieu te beschermen en beheren.

Met betrekking tot de ecologische effecten, is het waarschijnlijk dat zeevo-gels worden benvloed door de aanwezigheid van windmolenparken op zee, om-dat deze verstorend kunnen werken in foerageer- en rustgebieden en op vlieg-routes. Gegeven de complexiteit in het gedrag van zeevogels, de verscheidenheid aan omgevingsfactoren die daar invloed op hebben, en de technische beperkin-gen van vogeltellinbeperkin-gen op zee, is de verdrijving van zeevogels ten gevolge van de windmolenparken moeilijk te detecteren. In dit proefschrift wordt bestudeerd in hoeverre het mogelijk is om een dergelijk verdrijving vast te stellen aan de hand van zeevogeltellingen, waarbij gefocust wordt op het gedrag van zeevogels en technische kenmerken van de uitgevoerde monitoring.

De onderzoeksaanpak bestaat uit het simuleren van mogelijke realisaties van zeevogeltellingen die zouden kunnen worden verzameld in een ongestoorde situatie, d.w.z. een situatie zonder windmolenpark. De gesimuleerde tellin-gen worden uitgevoerd voor verschillende parameter waarden die de relaties tussen de omvang van zeevogelaanwezigheid en omgevingsfactoren specificeren, alsmede de ruimtelijk gecorreleerde willekeurige variaties in deze aantallen, en monitoringsinspanning en -ontwerp. Voor een gegeven combinatie van param-eterwaarden wordt met behulp van statistische hypothese toetsing vastgesteld of de daadwerkelijk verzamelde vogeltellingen na aanleg van het windmolen-park een realisatie kunnen zijn van de ongestoorde situatie, of zodanig daar van afwijken dat geconcludeerd kan worden dat een effect van het windmolen-park is gevonden. Door verschillende combinaties van parameterwaarden te gebruiken wordt geanalyseerd hoe deze variaties invloed hebben op de uitkom-sten van de statistische toetsen, en daarmee op de conclusies met betrekking tot het al of niet optreden van effecten van het windmolenpark. Het ver-mogen om verdrijving te detecteren wordt onderzocht door te analyseren hoe vaak, voor verschillende scenario’s van afname van zeevogelaanwezigheid in het windmolenpark gebied, deze verdrijving wordt gedetecteerd met behulp van statistische toetsing.

(16)

Voor de uitwerking van deze benadering wordt zeevogelgedrag beschreven met behulp van een geostatistisch model. Hierin wordt de omvang van de aan-wezigheid van zeevogels gerelateerd aan omgevingscondities. Daarnaast wordt ruimtelijk gecorreleerde willekeurige variatie meegenomen evenals het optreden van ’zero-inflation’ in zeevogeltellingen, d.w.z. het voorkomen van een over-matig aantal nulwaarnemingen. De technische kenmerken van de monitoring na aanleg van het windmolenpark worden beschreven in termen van het aantal observatiepunten, de locaties van deze punten en de ruimtelijke en temporele variatie van de omgevingscondities ten tijde van de uitvoering van de monitor-ing.

De zeevogeltellingen die zouden kunnen worden verzameld in een ongesto-orde situatie, wongesto-orden gegenereerd met behulp van geostatistische simulatie. De nulhypothese die getoetst wordt is dat er geen verandering in de zeevogelaan-wezigheid is ontstaan door aanleg van het windmolenpark. Deze studie laat zien dat wanneer zeevogelgedrag en monitoringskenmerken worden meegenomen, de kans om ten onrechte de nulhypothese te verwerpen (type I fout) kan worden verkleind. Hierdoor kunnen gedentificeerde veranderingen in de omvang van zeevogelaanwezigheid beter worden toegeschreven aan de aanwezigheid van het windmolenpark.

Het vermogen om een effect te detecteren wordt uitgedrukt in de kans om een bepaald opgelegd effect te detecteren: het onderscheidingsvermogen. In de simulaties wordt het opgelegde effect uitgedrukt in het afname percentage (25%, 50%, 75%, en 100%) van het gemiddeld aanwezige aantal vogels in het gebied waar het windmolenpark is aangelegd. De analyse van de simulaties laat zien dat verschillende combinaties van parameterwaarden voor de relatie tussen zeevogelaanwezigheid en de omgevingscondities, voor ruimtelijk gecorreleerde willekeurige variaties, en voor meetinspanning en -ontwerp, aanzienlijk verschil-lende kansen kunnen opleveren voor het detecteren van het opgelegde effect. Een van de onderzochte scenario’s laat zien dat wanneer de omgevingscondities meer geschikt worden voor aanwezigheid van vogels in het windmolenparkge-bied, de kans om verdrijving te detecteren toeneemt. Bijvoorbeeld, voor een geval waarin 50% afname in vogelaantallen was opgelegd, nam deze detectiekans toe van 0.54 naar 0.86.

Temporele variabiliteit in omgevingscondities heeft ook invloed op het ver-mogen om effecten vast te stellen. Dit is van belang wanneer vogeltellingen over meerdere dagen verspreid worden uitgevoerd en het ruimtelijke patroon in omgevingscondities verschilt tussen de opeenvolgende dagen. In milieu effect studies is het gebruikelijk om deze data over meerdere dagen samen te voegen. Dit onderzoek laat zien dat door dit samenvoegen zowel overschattingen als onderschattingen van het vermogen om effect te detecteren kunnen optreden, tot ongeveer een factor twee aan toe.

Wanneer het vermogen om effect te detecteren gekwantificeerd is, kan de stelling van Bayes worden toegepast om vast te stellen hoe vaak men de vo-geltellingen in een bepaald gebied bij voorkeur zou moeten herhalen. Op deze manier kan a priori kennis over de kans op het trekken van foute conclusies op basis van de tellingen worden gebruikt om een monitoringsplan op te stellen op

(17)

7 basis waarvan men het beste in staat zal zijn om onderscheid te maken tussen het wel of niet optreden van effecten.

Tenslotte, dit proefschrift laat zien dat zeevogelgedrag en technische ken-merken van de monitoring een groot effect kunnen hebben op het vermogen om verdrijving van zeevogels door windmolenparken op zee vast te stellen. Inzicht in hoe deze componenten dit vermogen benvloeden kunnen milieubeheerders ondersteunen bij het opzetten van goede monitoringsplannen, die uiteindelijk leiden tot acceptabele betrouwbaarheidsniveaus van conclusies over het al dan niet optreden van een dergelijk effect.

(18)

(19)

Resumen

Durante los últimos años la energ´ıa eólica marina se ha convertido en una prometedora fuente de energ´ıa renovable. Aparte de los desaf´ıos técnicos en el desarrollo de esta tecnolog´ıa, una cuestión importante para ponerla en práctica es el posible conflicto entre el desarrollo de parques eólicos marinos y los intere-ses de conservación de la naturaleza. Por lo tanto, es importante la aplicación de rigurosos métodos para la evaluación del impacto ambiental y el diseño de adecuados programas de monitorización para la protección y gestión del medio marino.

En referencia a los impactos ecológicos, las aves marinas pueden ser afec-tadas por los parques eólicos, ya que éstos pueden incidir en sus rutas de vuelo, y alterar las áreas de alimentación y descanso. La complejidad del compor-tamiento de las aves marinas, la variabilidad de los factores ambientales que pueden afectarlo, y las limitaciones de la recogida de datos sobre aves en el mar, constituyen un desaf´ıo para la detección del impacto en referencia a su desplazamiento. Esta tesis estudia la capacidad de detectar dicho impacto, centrándose en la influencia del comportamiento de las aves y las caracter´ısticas del muestreo en la toma de datos de contaje de aves.

La metodolog´ıa consiste en la simulación de realizaciones de muestras que podr´ıan derivarse de la situación sin perturbar, es decir, sin la influencia del parque eólico. Las muestras son simuladas utilizando una serie de valores de parámetros que especifican las relaciones entre la abundancia de aves mari-nas y las condiciones ambientales, autocorrelación espacial de los datos, y el tamaño y diseño del muestreo. Para una determinada combinación de valores de parámetros y utilizando una prueba de hipótesis, se evalúa si los datos recogi-dos, después de la construcción de un parque eólico, son bien los equivalentes a los de la situación sin perturbar, o de lo contrario, lo suficientemente diferentes como para concluir que el efecto de un parque eólico ha sido identificado. Uti-lizando diferentes combinaciones de valores de los parámetros se analiza cómo estas variaciones afectan a los resultados de una prueba estad´ıstica, y por tanto a los resultados de una evaluación de impacto. La capacidad de detectar el im-pacto se investiga mediante el análisis de la frecuencia con el que un desplaza-miento ser´ıa detectado por la prueba estad´ıstica considerando varios escenarios de reducción de la abundancia de aves en el área del parque eólico.

En la implementación de este método, el comportamiento de las aves mari-nas se describe por un modelo geoestad´ıstico que relaciona la abundancia de aves con las condiciones ambientales, incluye variación estocástica con autocor-relación espacial, y tiene en cuenta el exceso de ceros en los datos de contaje

(20)

de aves marinas. Las caracter´ısticas del muestreo son descritas por el número de sitios, su configuración, y la variación espacio-temporal de las condiciones ambientales en el momento del muestreo post-construcción.

El conjunto de muestras que podr´ıan derivarse de la situación sin perturbar es constru´ıdo utilizando simulación geoestad´ıstica. La hipótesis nula especifica que no ha habido un cambio en la abundancia de aves en la zona debido a la construcción del parque eólico. El estudio muestra que cuando se toman en cuenta el comportamiento de las aves y las caracter´ısticas del muestreo, la probabilidad de rechazar falsamente la hipótesis nula (error tipo I) puede ser reducida. Como resultado, los cambios identificados en el número de aves pueden atribuirse a la presencia del parque eólico.

La capacidad de detectar el impacto se mide en términos de la probabili-dad de detectar un impacto que ha sido impuesto: la potencia estad´ıstica. En las simulaciones, el impacto impuesto se expresa en porcentajes de reducción (25%, 50%, 75% y 100%) en la media del número de aves en el área del parque eólico después de su construcción. El análisis de las simulaciones muestra que diferentes combinaciones de valores de los parámetros para las relaciones entre la abundancia de aves marinas y las condiciones ambientales, autocorrelación espacial en los datos, y el tamaño y diseño del muestreo pueden generar im-portantes diferencias en las probabilidades de detectar un impacto que ha sido impuesto. En uno de los escenarios investigados se demuestra que cuando las condiciones ambientales son más favorables para la presencia de aves, la probabilidad de detectar un desplazamiento aumenta. Por ejemplo, cuando se produce una reducción del 50 % en el número de aves dicha probabilidad aumenta de 0,54 a 0,86.

La variabilidad temporal en las condiciones ambientales también influye la capacidad de detectar el impacto. Este hecho es especialmente importante cuando los muestreos de recogida de datos sobre aves marinas tienen una du-ración de varios d´ıas y el patrón espacial en las condiciones ambientales var´ıa entre los d´ıas del muestreo. En estudios de impacto es una práctica común el agregar estos datos. Esta investigación muestra que debido a la agregación de los datos pueden occurrir la sobreestimación y también la subestimación de la capacidad de detectar impacto, hasta un factor de aproximadamente dos.

Una vez que la capacidad de detectar el impacto ha sido cuantificada, se aplica el teorema de Bayes’ para determinar un número adecuado de muestreos para una determinada área de estudio. De esta manera, el conocimiento a priori de la probabilidad de llegar a conclusiones erróneas con los datos recogidos se puede utilizar para diseñar un plan de muestreo que mejor distinga entre la presencia y la ausencia de impacto.

En conclusión, esta tesis muestra que el comportamiento de las aves marinas y las caracter´ısticas de muestreo pueden tener un gran efecto en la capacidad de detectar el desplazamiento de aves marinas debido a la presencia de un parque eólico. El entendimiento de cómo estos componentes afectan dicha capacidad puede ser de gran ayuda para los responsables de toma de decisiones ambientales en la preparación de programas de monitorización que den lugar a conclusiones con niveles fiables sobre si se ha producido un impacto o no.

(21)

Chapter 1 Introduction

1.1 Motivation

The environment around us is constantly changing. But only in the last decades, humans have realized that their influence on the environment has, in some cases, led to its severe deterioration (Carson, 1962). Human activities such as logging, road construction, mining and agricultural development are causing tropical deforestation with an associated loss of biodiversity (Gardner et al., 2009), some agricultural practices have lead to water pollution affecting the growth of photosynthetic plants (Livingston et al., 1998), and high levels of industrialization have increased gas emissions, some of which have caused air pollution and damaged crops (Valle-Tascon and Carrasco-Rodriguez, 2004). In these times of rapid technological growth it is important to understand the con-sequences of human interactions with the environment (Vitousek et al., 1997) and therefore, more and more studies are emerging that focus on so-called impact assessments.

A particular system that undergoes increasing pressure due to human ac-tivities is the marine environment. The marine environment holds a vast range of economical functions. For example, in the North Sea these functions include oil and gas platforms, fisheries, cables (electricity and telecommunications) and pipelines, military activities, sand extraction and more recently offshore wind farms. Next to these economical uses, the North Sea also holds ecological func-tions (Figure 1.1). The area is a breeding ground for fish and important as a migratory route and wintering place for several bird species (van der Wal et al., 2009).

It is in this context that the impact of economical uses of the sea on the marine environment needs to be assessed if we want to maintain the marine environment in a healthy state for future generations.

Legislation at an international and national level has emerged to harmo-nize the relationship between offshore developments and the environment. For example, Environmental Impact Assessment (EIA) is a process of evaluating the expected impacts (both beneficial and adverse) of a proposed project that are likely to have significant environmental effects (Barker and Wood, 1999). As discussed in Glasson et al. (2005), in the context of EIA the direct and indirect impacts of a project need to be identified, described, predicted, and monitored (Figure 1.2). This requires knowledge on which elements of the marine environment may be affected by human activities, and which of these

(22)

Legend Platforms Other infrastructure Subsurface Surface N Legend Cables Electrical Telecom. Umbilical Other/Unkown N Legend Sand extraction Uncertain status Active Dredging Navigational N Legend Nature conservation areas N

Figure 1.1: Maps depicting examples of economical functions of the North Sea and marine nature conservation areas (modified after van der Wal et al., 2009).

elements should be considered as protected assets in the environmental impact assessment (K¨oller et al., 2006).

The monitoring of impacts after project execution is important for many reasons. From a management point of view, monitoring of impacts can as-sess the overall success of environmental management and protection (Dipper, 1998) and it can asses the success of mitigation measures in reducing impacts. Subsequently, decisions about future management actions can be made more effective (Wilson et al., 2010). Monitoring procedures can also serve to identify

(23)

1.2. Offshore wind energy and the environment 13 and correct (or mitigate) unanticipated impacts (Glasson et al., 2005). From a scientific point of view, monitoring can be used to check the accuracy of impact predictions (Wilson et al., 2010). Also, in complex environments such as the marine environment where the prediction of impacts is difficult, monitoring can be considered as a post-hoc impact assessment, where changes in the environ-ment are assessed to judge if a certain impact is acceptable (Herman and Heip, 1988). Identification of marine protected assets Description and prediction of potential impacts Decision about project execution Monitoring of actual impacts

Figure 1.2: Simplified Environmental Impact Assessment (EIA) process (modified after Glasson et al., 2005).

Although monitoring of impacts is of high relevance, it is one of the weak-est phases in the EIA process (Tomlinson and Atkinson, 1987; Dipper, 1998; Glasson et al., 2005; Ahammed and Nixon, 2006). Therefore, monitoring of impacts is the focus of this study.

In this study, the term impact assessment refers to the monitoring phase which involves the collection and analysis of data to assess changes in environ-mental parameters, in space and time, compared to what would have happened if a human activity had not been undertaken (Glasson et al., 2005). This is a challenging phase in an EIA given that the marine environment is a highly complex system with many processes interacting in an interrelated and often unpredictable manner (Crowder and Norse, 2008). This complexity together with the logistical problems of observing and studying the marine environment and the related high costs, lead to a poorly-understood undisturbed state of the environment (Jones, 2001).

Of the many human activities in the marine environment, offshore wind farm development is amongst the most recent and fast-growing activities. Since this technology is still in its infancy, knowledge of the impacts of offshore wind farm projects remains poor (Inger et al., 2009). A huge increase in installa-tion of offshore wind farms is expected in the near future (see Figure 1.3 for plans in the North Sea) and hence, it is imperative to be aware of the envi-ronmental side-effects of this new road that we are taking (Gill, 2005). This makes environmental impact assessment of offshore wind energy developments important.

1.2 Offshore wind energy and the environment

For centuries humans have relied on fossil fuels for energy production. Fossil fuels may one day run out (Kleiner, 2009), but more importantly, the decade-long utilization of fossil energy sources have caused adverse impacts on our environment (Inger et al., 2009; Witalec, 2009). It is therefore an assuring sign

(24)

United Kingdom Denmark Netherlands Germany Belgium Legend Approved area App. for approval Proposed area

Figure 1.3: EWEA’s 20 year offshore network development plan in the North Sea: an overview of approved wind farm areas, application for approved wind farm areas, and proposed wind farm areas (modified after map downloaded from www.ewea.org, see also Fichaux et al., 2009).

that more plans arise for alternative renewable energy sources. Amongst these renewable resources, offshore wind has been identified as a key resource. The vast amount of ‘wind space’ available in the coastal regions makes this resource attractive (Gill, 2005), also because wind farms constructed at sea can benefit from stronger winds (Pelc and Fujita, 2002; Esteban et al., 2011). In addition, offshore winds are less turbulent, thus decreasing the fatigue load and increasing the lifetime of the project (Punt et al., 2009). Harmful environmental impacts, however, should not only be identified when it is too late to reverse them.

EIA in the framework of offshore wind farms starts by identifying the pro-tected assets to be considered in an impact assessment procedure. In European countries, protected assets belong to two different categories, biotic and non-biotic.

Biotic Under this category, the focus is on impacts on marine fauna. Con-sidered protected assets include, amongst others, seabirds (local birds and migrating birds), marine mammals, fish, and benthos (K¨oller et al., 2006). The expected impacts are summarized in Table 1.1.

Non-biotic Under this category, the focus is on impacts on the visual seascape, sediment, hydrography and water quality in and around the wind farm area (Bruns and Steinhauer, 2005). The expected impacts are summa-rized in Table 1.2.

With a focus on biotic assets, seabirds have been identified as likely to be affected most by offshore wind farms (Exo et al., 2003; Garthe and H¨uppop,

(25)

1.2. Offshore wind energy and the environment 15

Table 1.1: Expected impacts of offshore wind farms for some identified biotic as-sets (K¨oller et al., 2006).

Assets Expected impacts

Local birds (resting and breeding birds)

Disturbance, displacement, barriers of movement and collision of seabirds due to construction and operation of turbines.

Migrating birds Collision or diversion of migrating birds due to

construction activities and operation of turbines.

Marine mammals Damage to or displacement of marine mammals due

to construction and operational noise.

Fish Damage to fish by sediment dispersion, vibration,

or electromagnetic fields and displacement of fish by introduction of new fish habitats (artificial hard substrates).

Benthos Damage and/or loss of benthic communities

through placement of turbines, changes of species composition due to introduction of artificial hard substrates, and/or change of sedimentation.

Table 1.2: Expected impacts of offshore wind farms for some identified non-biotic assets (Bruns and Steinhauer, 2005).

Assets Expected impacts

Visual seascape Change of seascape by wind turbines, disturbance

by intrusion of technical elements in a natural en-vironment, decrease of recreational qualities due to change in the appearance of the seascape.

Seabed and soil Seabed loss caused by foundation and scour protec-tion, changes in sedimentaprotec-tion, redistribution and re-suspension of sediments after wind farm cons-truction.

Hydrography and water quality

Foundations and scour protection of wind turbines might influence the current and/or the wave regime, spill of harmful substances during wind farm cons-truction, spill of oil during operation of the wind farm.

2004). Therefore, in this research the focus is on seabirds. The developed methods, however, can be applied as well to other marine fauna.

The following section expands on the description of potential impacts of offshore wind farms on seabirds.

(26)

1.3 Description of potential impacts on seabirds

Two types of impacts are considered with respect to offshore wind farms and seabirds, namely direct and indirect impacts.

Direct impacts

Direct impacts refer to impacts that take place during construction or during operation of the wind farm.

For example, disturbance can be due to the increased human activity in the wind farm vicinity or simply by the presence/noise of the wind turbines themselves (Drewitt and Langston, 2006). Such disturbances may have as a consequence the displacement of birds from suitable habitat areas, including their main feeding grounds (Fox et al., 2006). This displacement may produce a conflict if other alternative suitable areas are not able to accommodate the displaced birds (Langston and Pullan, 2003). Wind farms may also become barriers to seabird’s movement. Bird flocks may be forced to fly around the wind farm (Desholm and Kahlert, 2005), disrupting existing migratory flights, and local feeding and roosting flights. Whether a wind farm becomes a barrier depends on the species, the size of the wind farm, configuration of turbines, the amount of displacement of flying birds and their ability to compensate for their energy use (Fox et al., 2006). Bird collisions with offshore turbines is also an important issue (Exo et al., 2003). The associated increase in mortality may have high impacts at the population level (Langston and Pullan, 2003). When the location of the wind farm is in a migratory path or the area is used by specific seabird species, the collision risk will be higher (Drewitt and Langston, 2006). Moreover, weather conditions may influence the risk level for such collisions. For example, species under strong winds may be pushed towards the wind farm area. Also, in bad visibility conditions such as fog, birds may be attracted to the illuminated offshore structures, causing collisions with the hubs or rotors (H¨uppop et al., 2006).

Indirect impacts

Indirect impacts refer to changes made by a wind farm in marine processes or in other human activities that in turn affect seabird presence.

Wind turbines may change the marine conditions present in the wind farm location area. Examples include physical processes such as change in currents and sedimentation (Table 1.2). Furthermore, rotation of the blades creates a vortex, and may push birds towards the water with the danger of drowning (Fox et al., 2006). Vibration of the wind turbines may influence the distribution of fish, which in turn can influence the displacement of seabird species depending on this food source (Drewitt and Langston, 2006). Due to construction of the wind farm and the placement of cables, sedimentation processes may be altered. Turbidity levels may increase (Gill, 2005), which may in turn affect the feeding capabilities of visual hunting seabird species (Garthe, 1997; Markones, 2007).

Offshore wind farms will also produce an impact on other human activities. For example, certain types of fishery (e.g. beam trawling) may be forbidden

(27)

1.4. Monitoring of impacts, with focus on displacement 17 within the wind farm area (Wilson et al., 2010), which in turn may affect the behavior of seabirds in the area. Scavenging seabird species may be attracted to fishing vessels (Skov and Durinck, 2001) and may be displaced away from the wind farm area as their sources of food may no longer be available. Another indirect impact is the risk of seabird mortality whenever vessel accidents with the wind farm structures take place (Merck, 2006). Vessel accidents may have as a consequence oil spills which can cause seabird death, i.e. by affecting body parts such as feathers or by the ingestion of oil (Tseng, 1999). A further impact of fishery activities is the change in food resources within the wind farm area. As stated, beam trawling may no longer be allowed and this may cause an increase of fish numbers. As a result, a positive impact on seabird species that depend on this type of resources can be expected (Drewitt and Langston, 2006).

In this thesis the impact of wind farms regarding displacement of seabirds is considered. Displacement has been chosen as an impact because it is relevant as well to other marine fauna species (e.g. fish and marine mammals in Table 1.1). In addition, indirect impacts of offshore wind farm construction may also cause displacement of seabirds.

1.4 Monitoring of impacts, with focus on displacement

The term impact assessment is used in this work to refer to the analysis of changes in environmental parameters, both in space and time, compared to what would have happened if the project had not been undertaken. In particu-lar, the abundance of local seabirds is the environmental parameter of interest. Local seabirds refer to birds that reside for some time in and near the wind farm area (Leopold et al., 2010). The impact on migrating birds is excluded from this study and, therefore in the remainder of this thesis ‘seabirds’ only refers to local seabirds.

To perform an impact assessment, surveys to collect data on seabird abun-dance are undertaken prior to and after the construction of the wind farm. Survey methods for collecting data on species abundance consist of ship- and airborne-based methods. Ship-based techniques usually make use of a strip-transect methodology as described by Tasker et al. (1984). Within this method-ology birds are counted from both sides of the ship, over transects of 300 m wide, perpendicular to the ship track line, and at a recommended time interval of five minutes. Ship-based methods allow for a detailed collection of data in the surveyed area under different weather conditions, although the extension covered by these surveys tends to be small. Moreover, ship-based surveys allow the collection of seabird behavior information and the simultaneous registration of environmental characteristics governing the species observations, e.g. water salinity and temperature (Camphuysen et al., 2002). In airborne survey meth-ods, collection of seabird data is performed from an airplane. Two observers are located at each side of the plane and within a fixed distance from the plane track line (e.g. 100 m) the species, their numbers, and their behavior are

(28)

recorded. Airborne survey methods provide a higher coverage during the same amount of time as compared to ship-based methods but they are restricted to operate during good weather conditions (Camphuysen et al., 2002).

Assessing displacement of seabirds due to wind farm construction is usually carried out under the Before-After-Control-Impact (BACI) framework (Green, 1979; Stewart-Oaten et al., 1986; Murphy et al., 1997). Within this frame-work, changes are identified by comparing seabird abundance data before and after the construction in the wind farm area (impact area) and an area that is unaffected by the disturbance (control area). Such comparisons are made using different statistical analysis such as analyses of variance (ANOVA) and t-tests (McDonald et al., 2000; Smith, 2002; Stewart-Oaten and Bence, 2001). Investigations on the effect of offshore wind farms on displacement of sea-birds vary with respect to the survey methods employed and the type of impact analysis performed. In the remainder of this section, two examples of moni-toring studies in the North Sea are briefly presented together with their main findings on some common seabird species that have been investigated.

The Netherlands

In Dutch waters, there are currently two operational offshore wind farms, namely Offshore Wind farm Egmond aan Zee (OWEZ) and Princess Amalia Wind Farm (PAWF). Surveys with a focus on OWEZ were carried out from a ship to collect data on species abundance in the pre-construction and post-construction periods. The study area, of approximately 900 km2_{, comprised}

the wind farm area (of approximately 39 km2_{) and its vicinity. The method}

to detect changes in seabird abundance between pre- and post-construction periods consisted of fitting a generalized additive model or, if possible, a gen-eralized additive mixed model that includes as predictors the distance to the coast, the latitude and the OWEZ location. Two models were fitted, one for each of the pre- and post-construction periods. The OWEZ location parameter was tested for significance and it was assessed whether a significantly positive or negative effect of the wind farm area on bird numbers occurred between pre- and post-construction periods. This comparison was carried out using ex-pert judgement. Below, some results are presented of a two year monitoring program, described in detail in Leopold et al. (2010) and with focus on OWEZ. Statistical tests and expert judgement suggested that Divers (Red-throated Diver Gavia stellata and Black-throated Diver Gavia arctica), Terns (Arctic Tern Sterna paradisaea and Common Tern Sterna hirundo), and Auks (Com-mon Guillemot Uria aalge and Razorbill Alca torda) did not show avoidance for the wind farm area. For Northern Gannets (Morus bassanus), Common Scoters (Melanitta nigra), and Little Gulls (Larus minutus), there were insuf-ficient observations to conclude avoidance of the wind farm. Finally, Great Cormorants (Phalacrocorax carbo) showed preference for the wind farm area post-construction in some of the months investigated. In one month, however, this species showed avoidance for the wind farm area but the data and ex-pert judgement revealed that large numbers were resting on the structures of

(29)

1.5. Problem formulation 19 a gas-production platform only one kilometer north of the wind farm area.

Figure 1.4: Common Guillemot (Uria aalge) near IJmuiden, The Netherlands (pic-ture by Karel Mauer.)

Denmark

Two offshore wind farms (Horns Rev and Nysted) are subject to a monitoring program. Data on seabird abundance have been collected from an airplane. The method for monitoring displacement of seabirds is based on testing signif-icant differences between pre-construction seabird abundance data and post-construction data in the wind farm area, an area two km from the periphery of the wind farm area, an area between two and four km from the periphery of the wind farm area, and in a control area (more than four km from the outer-most turbines). Below, some results are presented of a three year monitoring program for the Horns Rev offshore wind farm, which are described in Petersen et al. (2006).

Statistical tests (e.g. t-tests) and expert judgement suggested that Divers (Red-throated and Black-throated) and Common Scoters showed significant avoidance of the wind farm during the three-years post-construction. Terns (Arctic/Common Tern) and Auks (Guillemot and Razorbill) were totally ab-sent from the wind farm area during the post-construction period for the se-lected months. No statistically significant attraction for any of the species was observed. Cormorants, however, occasionally occurred in large numbers in the wind farm area for social feeding during the post-construction period. For Northern Gannets there were insufficient data to test for statistically sig-nificant differences between pre- and post-construction periods. Little Gulls showed a preference for the wind farm area post-construction, but due to the low number of observations during the pre-construction period, no statistical differences were found between pre- and post-construction periods.

1.5 Problem formulation

The studies described in section 1.4 differed in their conclusions with respect to displacement of seabirds due to wind farm presence. Explanations may include i) the complexity in seabird species behavior and interaction with the marine environment, which may depend on location, ii) differences in data collection

(30)

and analysis methods, and iii) differences in the degree of expert judgement included in the assessments.

From a methodological point of view, monitoring the displacement of sea-birds is a challenging task. Seabird abundance data are often scarce due to logistic limitations in marine data collection. Seabird distribution patterns ex-hibit high spatial and temporal heterogeneity. Fluctuations in population sizes over time can explain increases or declines in the species numbers that are lo-cally observed (van Eerden and Gregersen, 1995). The patterns observed at the scale of impact studies will be influenced both by local time-varying con-ditions prevailing at the time and location of the survey, e.g. availability of prey (Hunt Jr, 1997), physical factors (Garthe, 1997; Markones, 2007), human activities (Camphuysen, 1995; Skov and Durinck, 2001), and by seabird distri-bution patterns occurring at coarser scales. For example, Common Guillemots breed on the cliff coasts in the Northern North Sea and Helgoland and arrive late summer and in autumn to forage in the Frisian Front (Lindeboom et al., 2005). When stratification in this area ends, usually mid-October (Weston et al., 2008), Common Guillemots disperse towards the Southern North Sea (e.g. off the coast of the Netherlands). Therefore, it is hypothesized that ad-vances or delays in the ‘breaking’ of the front may influence the number of Common Guillemots observed in, for example, Dutch waters (M.F. Leopold, personal communication, August 20, 2009). As a consequence, if an impact assessment is to be carried out locally in this area, environmental conditions occurring at a coarser scale may have an influence in the outcome of the as-sessment.

It is practically impossible to collect all related data in order to isolate a potential effect of an offshore wind farm on the displacement of birds. As a con-sequence, the conclusions regarding impact rely mostly on expert judgement, which may vary amongst assessments, and do not directly rely on the output of a statistical test. Moreover, the analysis of data collected following a given monitoring design may not provide evidence that displacement occurred when it in fact may have been present.

The problem addressed in this research is the ability to detect impact, knowing that issues such as species behavior and survey characteristics play a role. What is their influence on conclusions about impact?

Before this question can be answered, one needs to be able to include these issues in an impact assessment method. Seabird behavior can be translated into how seabirds respond to the environment. Two types of spatial variation in seabird numbers are identified. The first type is spatial variation induced by relationships between seabird abundance and environmental factors (un-derstood relationships). If data are available to model such relationships in a deterministic manner, seabird behavior can be included in the impact assess-ment. If data are not available, one may still want to somehow include seabird behavior in the impact assessment. The second type is spatial variation in sea-bird abundance which is not understood (non-understood relationships), i.e. due to the complexity in species behavior. Such variation should also have a place in an impact analysis. In addition, survey characteristics may play a role

(31)

1.6. Research objective 21 and should be explicitly accommodated in the analysis in order to assess their influence in conclusions about impact.

This research does not attempt to answer whether an impact has or has not occurred in a specific case or to enhance the knowledge on seabird behavior. It rather aims at supporting ecologists and environmental decision makers in their impact studies by investigating the effect of these issues on the ability to detect impact.

1.6 Research objective

The research presented in this thesis aims at providing support for impact studies on marine fauna by:

1. Developing an impact assessment tool that incorporates species behavior and survey characteristics.

2. Assessing the effects of species behavior and survey characteristics on the ability to detect impact on marine fauna displacement due to a spatial interference like a wind farm.

1.7 Research questions

In accordance to the aforementioned objectives, the following research questions (Q) are identified:

Q1. How can species behavior and survey characteristics be incorporated in an impact assessment method?

Q2. To what extent does inclusion of understood relationships between sea-bird abundance and the environment, affect the ability to detect impact? Q3. How is the ability to detect impact affected by patterns in seabird abun-dance that are not understood or can not be deterministically modelled? Q4. What are the effects of survey characteristics on the ability to detect

impact?

1.8 Research tools

In order to develop a method suitable to assess the influence of the specified issues on the ability to detect impact, three research tools were employed. Below, we describe these tools in more detail.

Hypothesis testing

Hypothesis testing is based on the truth of falsity of a proposition or hypothesis, on the basis of empirical evidence (Snijders, 2001). In the context of this study, a hypothesis may state that construction of an offshore wind farm will not cause changes in the abundance of a particular seabird species. Specifically,

(32)

the hypothesis may state that the mean number of a given species after the construction of the wind farm (µpost) does not differ from the mean number

pre-construction (µpre). Empirical evidence in this context may be summarized

by the average number of birds observed during the pre-construction and post-construction periods, or ˜µpreand ˜µpost, respectively.

Based on the work by Neyman and Pearson (1928), the first step in hypoth-esis testing is to define the null hypothhypoth-esis (H0) and the alternative hypothesis

(Ha). These hypotheses are stated in such a way that the researcher’s aim is to

falsify H0as evidence for Hausing sample data. In our example, the hypothesis

may be stated as follows:

H0: µpost= µpre

Ha: µpost6= µpre

Next, a test statistic is computed using the observed data. For example, the test statistic may be defined as the standardized difference between ˜µpost

and ˜µpre. Commonly, it is assumed that the test statistic follows a standard

reference distribution, e.g. the Student’s t -distribution or asymptotically the Normal distribution. If it is unclear whether this assumption is violated, the reference distribution can be empirically estimated through simulation by com-puting many realizations of the test statistic under the null hypothesis (Theiler and Prichard, 1996). The next step in hypothesis testing is to specify the level of significance α, i.e the probability that H0is rejected when it is in fact true.

For example, if α is set to 0.05, we determine the extreme regions that contain 5% of the area of the curve describing the reference distribution. This part of the curve is called the critical region (Davis, 2002). If the computed value of the test statistic lies in the critical region, H0is rejected.

Statistical power

Following the hypothesis testing discussion, failure to reject H0can be wrongly

interpreted as acceptance of H0 (Ludwig et al., 2001). To avoid such

misin-terpretation, researchers have chosen to endorse the ‘precautionary’ principle. Within the framework of this work, this principle recognizes that hypothesis tests, which aim to prevent type I errors (falsely rejecting H0) are frequently

in conflict with species conservation objectives (McGarvey, 2007), which aim to prevent type II errors (falsely failing to reject H0). The implementation

of the precautionary principle can be performed via a statistical power analy-sis (Sanderson and Petersen, 2002). Statistical power is defined as the proba-bility of rejecting H0 when it is in fact false and should be rejected (Mumby,

2002). A common approach to a power analysis is to specify a pre-defined environmental disturbance and to calculate the power to detect it (Maclean et al., 2007). Following the example in the previous section, a power analysis may begin with specifying a decrease in the mean number of birds using the wind farm area during the post-construction period. Then, the probability of rejecting H0, which in this case is false, is calculated providing the statistical

(33)

1.8. Research tools 23 Geostatistical simulation

The environment and its components interact in a complex manner. As men-tioned previously, species spatial variation may have physical causes. Some of these relationships between species abundance and the environment are under-stood and can be modelled deterministically. On the other hand, the complexity in species behavior may be such that spatial variation appears to be random. In this respect, one may need to look differently at spatial variation (Webster and Oliver, 2001).

Geostatistics provide a set of statistical tools suited for the analysis of data distributed in space (Goovaerts, 1997). In geostatistics, the assumption is that at each point in space, there is not just one value of a given property (e.g. number of birds) but a whole set of values (Webster and Oliver, 2001). The observed value y(xi) at sampling location xi is drawn from a probability

distribution with mean µ and variance σ2_{. In addition, observed values may}

have a certain degree of spatial dependence (Fortin and Dale, 2005). This characteristic is often described by what is known as the first law of geography: ‘Everything is related to everything else, but near things are more related than distant things’ (Tobler, 1970). This means that possible values of a property at nearby sampling locations are not independent and tend to be spatially autocorrelated. Such dependency is modelled in geostatistics by the spatial covariance function and the variogram, in which the variable Y (xi) at each

spatial location is a random variable (Webster and Oliver, 2001).

Under the assumption of second-order stationarity, the spatial covariance function is defined as:

C(h) = E[{Y (xi)} · {Y (xi+ h)} − µ2] (1.1)

where E indicates expectation value, Y (xi) is the variable at location xi, and

Y (xi+ h) is the variable at a location separated a distance h from xi.

Matheron (1965) defined the variogram as:

2γ(h) = E[{Y (xi) − Y (xi+ h)}2] (1.2)

For a set of data y(xi), i = 1,2,...,M (M = total number of sampling

loca-tions) the semivariance at lag h is calculated as:

ˆ γ(h) = 1 2N (h) N (h) X i=1 [y(xi) − y(xi+ h)]2 (1.3)

where N (h) is the number of point pairs separated by distance h, y(xi) is the

data value at location xi, and y(xi+ h) is the data value at a location separated

a distance h from xi.

The set of semivariances calculated for different values of h constitute the experimental semivariogram. For computational purposes, a function is fit-ted to the experimental semivariogram (hereafter referred as to the variogram model).

(34)

In geostatistics the term ‘simulation’ specifies the creation of values of Y (xi)

at locations xithat emulate the general characteristics of those observed

(Web-ster and Oliver, 2001). A realization is one possible outcome of such a simula-tion, where values at locations xi are drawn from the probability distributions

of Y (xi). The set of realizations have the same statistical properties in terms

of the mean µ and the variogram model.

In the context of this research, the samples that could have been collected in the undisturbed situation (without the presence of the wind farm) are charac-terized by the number of birds that are, on average, observed at survey locations plus stochastic variability described by a variogram model. The impact assess-ment method analyzes whether the post-construction survey is a realization of the undisturbed situation or, otherwise, different enough to conclude that the effect of the wind farm has been identified. To this end, the impact assessment method uses a geostatistical simulation procedure that is able to simulate many realizations of seabird count surveys arising from the undisturbed situation to be used as a reference to the observed post-construction data.

1.9 Outline of the thesis

This thesis comprises seven chapters of which a brief outline is given below (see also Figure 1.5).

Chapter 2 presents the method that allows to incorporate species behavior and survey characteristics in an impact assessment analysis. The aim of this Chapter is to construct a method that incorporates these aspects to further investigate their influence on the ability to detect displacement of seabirds due to offshore wind farm presence. Besides, it is demonstrated that taking these aspects into account influences the conclusions about a wind farm’s impact on seabird displacement. The objective of this Chapter is, therefore, to provide answer to Q1.

Chapter 3 expands on Chapter 2 with an application of the method in a real impact study using data collected on Common Guillemots at the Egmond aan Zee offshore wind farm (OWEZ). The aim of this Chapter is to demonstrate how the above mentioned aspects can be modelled in practice (Q1).

Chapter 4 extends the method presented in Chapter 2 to be suited for power analysis. In this manner, the amount of influence that aspects such as species behavior and survey characteristics have on the ability to detect impact can be analyzed (Q2 - Q4).

Chapter 5 focuses on situations where survey data on known relationships between seabirds and environmental conditions are not available. In other words, the focus is on the stochastic component of the geostatistical model. This Chapter explores to what extent the ability to detect im-pact is influenced by the assumption that characteristics of the stochastic component do not change over time (Q3).

(35)

1.9. Outline of the thesis 25 Chapter 6 provides a discussion on the achievements and limitations of the study, as well as on ways forward on the topic of this work. It also discusses the applicability of the developed methods and the management implications of the work.

Chapter 7 provides answers to research questions Q1 - Q4. Also, recommen-dations are provided in the context of impact assessment and monitoring of offshore wind farms on marine fauna.

Ability to detect impact of offshore wind farms on displacement of seabird species

Depends on Chapter 5 Spatial distrib. environmental conditions Configuration of observations Stochastic component Deterministic component Number of observations Temporal distrib. environmental conditions Stochastic component Deterministic component Chapter 3,4 Chapter 2

Survey characteristics Species behavior

Understood relationships Non-understood relationships Data available? yes no

(36)

(37)

Chapter 2 A simulation-based approach to impact

assessment

∗

Abstract

Assessing and monitoring the impact of offshore wind farms on marine fauna is vital if we want to achieve ecologically sustainable development of this re-newable energy resource. Given the complexity of the marine environment, a method capable of accommodating spatio-temporal behaviour of specific species and their interrelation with other marine phenomena is an essential prerequi-site for investigating whether or not there has been any measurable impact to date.

This paper presents a method based on geostatistical simulation to assess whether pre- and post-construction collected bird count data suggest displace-ment of birds due to the wind farm. The method takes into account spatial au-tocorrelation in species abundance at various scales, pre- and post-construction differences in environmental conditions and in survey effort and design.

We demonstrate that taking these factors into account influences the con-clusions about a wind farm’s impact on bird life. In particular, incorporating spatial autocorrelation in seabird numbers is an important factor in reducing the risk of wrongly identifying an effect of a wind farm on bird abundance.

Synthesis and applications. The development of offshore wind farms is of-ten in conflict with nature conservation interests. Environmental impact as-sessment and monitoring is essential to protect and manage the marine en-vironment. The method described here will allow scarce data to be utilized effectively as a basis for well-informed environmental decisions. In addition, the method will assist in the design of optimal monitoring procedures at a given site, balancing costs and effectiveness in detecting potentially harmful impacts.

Key-words: autocorrelation, impact assessment, geostatistics, offshore cons-truction, seabirds, spatial simulation, survey design.

∗_{This chapter has, in slightly modified form, been published as a separate paper: P´} erez-Lape˜na, B., K. M. Wijnberg, S. J. M. H. Hulscher, and A. Stein, 2010. Environmental impact assessment of offshore wind farms: a simulation-based approach. Journal of Applied Ecology 47(5)1110–1118.

(38)

2.1 Introduction

Marine waters have wide economical value including oil and gas reserves, sand extraction, fisheries and more recently offshore wind farms. The marine envi-ronment is also home to diverse marine fauna. Synergy between these economi-cal and ecologieconomi-cal functions is essential to ecologieconomi-cally sustainable development. The impact of human activities on the marine fauna must be assessed (Gill, 2005), particularly with respect to offshore wind farms since construction is a relatively recent development (Inger et al., 2009). Therefore, little is known about the impact on marine fauna (Merck, 2006; Garthe and H¨uppop, 2004), including seabirds (Drewitt and Langston, 2006).

Under the BACI (Before-After-Control-Impact) framework, the assessment of impact is based on comparisons between data (e.g. species counts) collected pre- and post-construction, in the potentially impacted area and in a control area (Stewart-Oaten et al., 1986; McDonald et al., 2000; Petersen et al., 2006). In the case of wind farms, the impact area is the location(s) where construction will take place. Ideally, the control area should have similar environmental conditions to the impact area but should be far enough away to be unaffected. Assessment of the impact of offshore wind farms through comparisons of pre-and post- construction species counts at impact pre-and control areas is generally not sufficient for three main reasons:

1. Spatial autocorrelation in seabird numbers.

2. The dynamic nature of environmental factors affecting the spatial distri-bution of seabirds.

3. Differences in survey effort and design between pre- and post-construction periods.

Spatial autocorrelation may occur as a consequence of seabirds respond-ing to variation over time in environmental factors such as water temperature, salinity, and water transparency (Garthe, 1997) that are in turn spatially au-tocorrelated. Spatial autocorrelation in species abundance data at such coarse scales can be addressed by building a deterministic model relating the expected (or mean) number of birds to environmental factors. However, the residuals of the fitted model may still be spatially autocorrelated (Pebesma et al., 2000). This may occur, for example, if other environmental factors affecting seabirds were omitted or due to the flock behaviour of birds. Where there is unex-plained positive spatial dependence in the data, pairs of species observations at given distances will tend to have similar values, hence cannot be consid-ered as independent observations. Traditional (non-spatial) statistical tests for calculating significant differences in the number of birds between the pre- and post-construction periods are not suitable, as the standard errors will be un-derestimated. This would result in inflation of Type I errors (Dormann et al., 2007), i.e. an increase in the risk of wrongly identifying an effect of the wind farm on the number of birds.

(39)

2.2. Materials and methods 29 The dynamic nature of environmental factors will affect the spatial distribu-tion of seabirds because condidistribu-tions during pre- and post-construcdistribu-tion surveys may not be the same. It will therefore be difficult to isolate the effect of the wind farm.

Finally, differences in survey effort and design (number and location of observations) between the pre- and post-construction periods may also generate differences in the number of birds observed that may be unrelated to wind farm construction.

The objective of this paper is to present a transparent and flexible method to detect differences in the number of a given species due to the presence of an offshore wind farm. In statistical terms, this means testing the null hypothesis of ‘no change’ in bird numbers in the wind farm area due to wind farm cons-truction. The method explicitly takes into account the effect of differences in the spatial dependence in species abundance at various scales as well as the effect of variation in post-construction survey effort and design.

2.2 Materials and methods

We test the null hypothesis that the construction of an offshore wind farm had no effect on the number of birds using the area with the wind turbines (hereafter referred to as wind farm area). A model relating species abundance to envi-ronmental factors is constructed to account for spatial autocorrelation in bird counts over large distances (hereafter referred to as coarse scale). Using this deterministic model, the number of birds that are expected at survey locations is characterized. To account for variation in environmental conditions between pre- and post-construction periods, the pre-construction situation is re-defined (hereafter referred to as ‘reference’ situation) to match the actual environmen-tal conditions of the post-construction survey. The deterministic model then predicts the expected (or mean) number of birds at post-construction survey locations.

To define the null hypothesis we use the difference between the mean number of birds in the wind farm (µwf) and control area (µc), hereafter referred to as

indicator. The null hypothesis and alternative hypothesis are stated as follows: H0: [µwf− µc]ref− [µwf− µc]post= 0

Ha: [µwf− µc]ref− [µwf− µc]post6= 0

where [µwf − µc]ref is the difference in the reference situation and [µwf−

µc]post is the difference in the post-construction situation. We refer to the

difference [˜µwf− ˜µc]ref− [xwf− xc]postas the test statistic. The values ˜µwfand

˜

µc are an estimate of µwf and µc for the reference situation, respectively, and

their difference is taken as a constant. The values xwf and xc are a sample

estimate of µwf and µc for the post-construction situation, respectively.

Due to stochastic processes, [xwf− xc]post will deviate from [˜µwf− ˜µc]ref.

Detecting seabird displacement. A simulation-based geostatistical approach

Detecting seabird displacement

A simulation-based geostatistical approach

Detecting seabird displacement

A simulation-based geostatistical approach

Contents

Preface

Summary

Samenvatting

Resumen

Chapter 1

Introduction

1.1

Motivation

1.2

Offshore wind energy and the environment

1.3

Description of potential impacts on seabirds

1.4

Monitoring of impacts, with focus on displacement

1.5

Problem formulation

1.6

Research objective

1.7

Research questions

1.8

Research tools

1.9

Outline of the thesis

Chapter 2

A simulation-based approach to impact

assessment

∗

2.1

Introduction

2.2

Materials and methods