Présentation de l’éditeur
Ecological sciences deal with the way organisms interact with one another and their environment. Using sensors to measure various physical and biological characteristics has been a common activity since long ago. However the advent of more accurate technologies and increas- ing computing capacities demand a better combination of information collected by sensors on multiple spatial, temporal and biological scales. This book provides an overview of current sensors for ecology and makes a strong case for deploying integrated sensor platforms. By covering technological challenges as well as the variety of practical ecological applications, this text is meant to be an invaluable resource for students, researchers and engineers in ecological sciences. This book benefited from the Centre National de la Recherche Scientifique (CNRS) funds, and includes 16 contributions by leading experts in french laboratories.
Key features • An overview of sensors in the field of animal behaviour and physiology, biodiversity and ecosystem. • Several case studies of integrated sensor platforms in terrestrial and aquatic environments for observational and experimental research. • Presentation of new applications and challenges in relation with remote sensing, acoustic sensors, animal-borne sensors, and chemical sensors. Sensors for ecology Towards integrated knowledge of ecosystems
Jean-François Le Galliard, Jean-Marc Guarini, Françoise Gaill
Sensors for ecology Towards integrated knowledge of ecosystems
Centre National de la recherche scientifique (CNRS) Institut Écologie et Environnement (INEE)
www.cnrs.fr Acr1932688304928-24242.pdf 1 20/03/12 16:12
Photographie de couverture / Cover Picture © CNRS Photothèque – AMICE Erwan UMR6539 – Laboratoire des sciences de l’environnement marin – LEMAR – PLOUZANE “A diver inspects a queen conch Strombus gigas during a scienti c expedition in Mexico. The queen conch is equipped with acoustic sensors, here nearby a receptor, in order to collect information on its behaviour and physiology in nature.”
© CNRS, Paris, 2012 ISBN : 978-2-9541683-0-2
sensors-001-344.indd 6 20/03/12 13:10 Contents
Foreword...... 11 I Ecophysiology and animal behaviour
Chapter 1 : Bio-logging: recording the ecophysiology and behaviour of animals moving freely in their environment Yan Ropert-Coudert, Akiko Kato, David Grémillet, Francis Crenner.... 17
Chapter 2 : Animal-borne sensors to study the demography and behaviour of small species Olivier Guillaume, Aurélie Coulon, Jean-François Le Galliard, and Jean Clobert...... 43
Chapter 3 : Passive hydro-acoustics for cetacean census and localisation Flore Samaran, Nadège Gandilhon, Rocio Prieto Gonzalez, Federica Pace, Amy Kennedy, and Olivier Adam...... 63
Chapter 4 : Bioacoustics approaches to locate and identify animals in terrestrial environments Chloé Huetz, Thierry Aubin...... 83 8 Contents II Biodiversity Chapter 1 : Global estimation of animal diversity using automatic acoustic sensors Jérôme Sueur, Amandine Gasc, Philippe Grandcolas, Sandrine Pavoine. 99
Chapter 2 : Assessing the spatial and temporal distributions of zooplankton and marine particles using the Underwater Vision Profiler Lars Stemmann, Marc Picheral, Lionel Guidi, Fabien Lombard, Franck Prejger, Hervé Claustre, Gabriel Gorsky...... 119
Chapter 3 : Assessment of three genetic methods for a faster and reliable monitoring of harmful algal blooms Jahir Orozco-Holguin, Kerstin Töbe, Linda K. Medlin...... 139
Chapter 4 : Automatic particle analysis as sensors for life history studies in experimental microcosms François Mallard, Vincent Le Bourlot, Thomas Tully...... 163 III Ecosystem properties Chapter 1 : In situ chemical sensors for benthic marine ecosystem studies Nadine Le Bris, Leonardo Contreira-Pereira, Mustafa Yücel...... 185
Chapter 2 : Advances in marine benthic ecology using in situ chemical sensors Nadine Le Bris, Leonardo Contreira-Pereira, Mustafa Yücel...... 209
Chapter 3 : Use of global satellite observations to collect information in marine ecology Séverine Alvain, Vincent Vantrepotte, Julia Uitz, Lucile Duforêt- Gaurier...... 227 Contents 9
Chapter 4 : Tracking canopy phenology and structure using ground-based remote sensed NDVI measurements Jean-Yves Pontailler, Kamel Soudani...... 243 IV Integrated studies
Chapter 1 : Integrated observation system for pelagic ecosystems and biogeochemical cycles in the oceans Lars Stemmann, Hervé Claustre, Fabrizio D’Ortenzio...... 261
Chapter 2 : Tropical rain forest environmental sensors at the Nouragues experimental station, French Guiana Jérôme Chave, Philippe Gaucher, Maël Dewynter...... 279
Chapter 3 : Use of sensors in marine mesocosm experiments to study the effect of environmental changes on planktonic food webs Behzad Mostajir, Jean Nouguier, Emilie Le Floc’h, Sébastien Mas, Romain Pete, David Parin, Francesca Vidussi...... 305
Synthesis and conclusion Jean-François Le Galliard, Jean-Marc Guarini and Françoise Gaill...... 331
NB: When cited in the text, a chapter of this book is identified accord- ing to the part it belongs to. For example, (III, 3) refers to the chapter 3 by Alvain et al. in the third (III) part of this book about Ecosystem properties.
Foreword
Altogether explorer, scientist, philosopher and one of the first world citi- zen, the German naturalist Alexander von Humboldt (1769-1859) is often considered as a founder of ecological sciences, though the word “ecology” was only coined several decades later by another German scientist, Ernst Haeckel (1834-1919). Equipped with the best sensors (thermometers, barometers, and so on) and familiar with advanced metrology techniques of its time, von Humboldt pioneered the field of plant biogeography, a discipline at the meeting point between botany, geography, climatology and geology. Von Humboldt major conceptual and methodological con- tributions consisted in collecting physical and geological data along with plant distribution maps to determine the physical and historical condi- tions favouring specific plant assemblages all over the world. With this approach, he ventured into previously unsuspected complex interactions between plants and their physical surroundings. Two centuries later, researchers are still striving to understand the ecological and evolutionary mechanisms that determine the distribution of plant and animal species.
Indeed, an accurate quantification of how organisms interact with each other and with their environment is at the heart of several grand chal- lenges in modern ecological sciences from the description of bio-geochem- ical balances to the prediction of ecosystem dynamics. However, contrary to von Humboldt and his followers, we can now explore thoroughly the natural world, thanks to major technological improvements in our ability to measure physical, chemical and biological quantities. Sensors are now part of the standard toolbox of most ecological studies, and play an impor- tant role in both exploratory studies of nature, experimental approaches, and the development of predictive ecological models. With the advent of more advanced technologies and the strong opportunities offered by nowadays available computing capacities, we are in a better position to integrate ecological information from sensors across multiple spatial, tem- poral and biological scales. This book, sponsored by the Centre National de la Recherche Scientifique (CNRS) in France, presents an up-to-date overview of the use sensors for ecology by some leading CNRS laborato- 12 Foreword ries. The book covers some of the main technological challenges in our field from bio-loggers attached on animals to remote sensing imaging capacities installed on satellites, and provides many examples of practi- cal applications chosen from ongoing CNRS programs. It is also tightly connected with the current frontiers in ecology and evolution throughout the world. We hope that the book will become an invaluable resource to students, researchers and engineers in ecological sciences.
Few books have reviewed methods and issues in the field of sensors for ecology. The reason is easy to understand: a great deal of techniques and sensor types do exist and are covered in specific reviews or journals. Keeping pace with the increasing number of sensors and technologies currently available is therefore a difficult task. Yet, this book provides in a synthetic way a balanced description of the new applications and challenges in ecological research that the use of remote, acoustic, animal- borne, chemical and genosensors represents. Here, we adopted a broad view on sensors – usually defined as a device that measures a physical quantity and converts it into a signal – to describe a quantity of tools including image and video analyses, biodiversity and life history sensors, and other less traditional methods. Furthermore, this book contains technical descriptions of some sensors even though it is not a handbook about sensor technologies. A techni- cal treatment is crucial to understand, design and integrate sensors for the purpose of ecological research, but this issue is already addressed in many handbooks. Instead, it was decided to focus this presentation on relevant applications and practical problems faced by ecologists during their research programs. Lastly, the book differs from a traditional presentation based on a stand- ard classification among sensor types and a discussion of the specific issues within each category of sensors (e.g. remote sensing versus chemical sensors). Ecological studies often need to integrate physical, chemical and biological data obtained with various types of sensors to get a comprehen- sive view of the state and dynamics of ecosystems. We therefore chose to make a strong case for the deployment of integrated, autonomous sensor platforms in the context of observational and experimental research infra- structures, and we present case studies where integrated data can inform predictive models. The integration of various types of sensors for ecologi- cal studies poses a series of complex problems, from the engineering and autonomy of the platform to the access and use of data for modelling. These difficulties go far beyond the design of a single sensor and are often specific of the scale at which sensors should be deployed. Foreword 13
The idea of this project started under the patronage of the CNRS Institut Écologie et Environnement in 2010 with the aim to produce a state-of- the-art review for sensors used in ecology in France, as well as to identify strengths and weaknesses in this field for the future. Under the supervision of Jean-Marc Guarini and Karine Heerah, a workshop was organised at the University Pierre and Marie Curie in Paris in January 2011. Twenty four contributions were made and the workshop’s program addressed remote sensing techniques, the use of sensors for in situ studies, and the use of sensors for experimental studies. From these contributions and the dis- cussions that followed, a few were selected to be published in the four sections of this book. The first section deals with animal behaviour and physiology, a very active field of research that raises both strong technical constraints and ethical issues. In a second section, we discuss the use of sensors in intra-specific and inter-specific biodiversity studies, and provide key examples ranging from the use of acoustic sensors or genetic methods to image analysis. We then present a series of ecosystem studies relying on advanced remote sensing and chemical sensors; those studies focus on measuring feedbacks between organisms and their geo-chemical envi- ronment. Our last section groups three integrated ecological studies. Two discuss observation platforms of ecosystems and one describes an experi- mental marine ecology infrastructure. The latter demonstrates how sen- sors can be used to manipulate environmental conditions and study the effect of environmental changes on ecosystems. Each chapter was organ- ised so that it reviews existing methods and sensors, and discusses current difficulties and requirements for future technological development.
We would like to thank the authors for their participation and their kind patience during the editorial process. Angeline Perrot assisted us during the two last months of this project and was extremely efficient at organ- ising reviews, editing all text and making the iconography tidy from an heterogeneous pool of figures and photographs. The editors would like to thank the CNRS and the Institut Écologie et Environnement (INEE) for their financial support, as well as the University Pierre and Marie Curie for hosting the workshop. Jean-François Le Galliard acknowledges the sup- port of the TGIR Ecotrons program and from the UMS 3194 CEREEP- Ecotron IleDeFrance as well as the financial support from ANR Equipex PLANAQUA coordinated by École normale superieure and CNRS (con- tract ANR-10-EQPX-13).
Jean-François Le Galliard, Jean-Marc Guarini and Françoise Gaill January 16, 2012 Paris, France
I Ehi cop ys ology and animal behaviour
Chapter 1
Bio-logging: recording the ecophysiology and behaviour of animals moving freely in their environment
Yan Ropert-Coudert, Akiko Kato, David Grémillet, Francis Crenner
1. Setting the scene
1.1 Sensing with bio-loggers Bio-logging refers to the fastening of autonomous devices onto (mainly) free-ranging animals to collect physical and biological information (Naito, 2004; 2010; Cooke et al., 2004; Ropert-Coudert et al., 2005; see also Costa, 1988, although the term bio-logging was not used in these times). It should be noted here that bio-logging is sometimes referred to as biologging. However, as Naito (2010) pointed out, the latter term is misleading as it is used in molecular biology. As such bio-logging differs from telemetry in the sense that data are stored locally in the memory of the devices and not transferred via radio waves or other transmitting means. This move from biotelemetry to bio-logging was done in order to address practical difficulties related to data transmission. Thus, this comes, as no surprise that bio-logging was firstly used in ecological and physiological studies to investigate marine, far-ranging, diving species, as water represents a barrier to radio signals. Bio-logging studies were initially conducted on species with a body mass large enough to accommodate the large size of the very first loggers: seals and whales. As miniaturisation progressed, smaller species of seals and seabirds became target species for bio-logging approaches. Among seabirds, penguins (Sphenscidae) represent an intensively studied family because of their adaptation to aquatic life 18 Ecophysiology and animal behaviour and their consequently denser, larger and more robust body. Nowadays, bio-logging can be applied onto an impressive range of species, terres- trial or aquatic, whether these are mammals, birds or reptiles (see I, 2). Bio-logging developments are one step away from moving into the insect realm as radio-telemetry is already available to study terrestrial and flying insects (Vinatier et al., 2010; Wikelski et al., 2006). Immediate consequences of local storage are the necessity to retrieve the device to access the data and develop appropriate sensors to gather data about physical and biological information on relevant time scales. It there- fore clearly appears that bio-logging primarily refers to a methodological approach and has generated research to improve existing technologies. Yet, bio-logging is more than a mere catalogue of tools and techniques. The possibility to obtain an uninterrupted flow of information pertaining to both the activity and physiology of animal and its immediate, physical surroundings revolutionised the way we consider several fields in biology. We could draw a parallel with the field of genetics and how it evolved from Gregor Mendel crossing variety of peas to the advanced technologies of molecular sequencing. Similarly, the ecologist with its notebook possesses now a suite of approaches to examine animals living freely in their envi- ronment. In this context, bio-logging applications ranges from physiologi- cal investigations to the comprehension of the functioning of ecosystems, by relating a change in physical parameters of the environment to a change in the behaviour of both a predator and its prey, at the same spatial and temporal scales (Ropert-Coudert et al., 2009a).
1.2. Bio-logging in the scientific community The word bio-logging was coined at the occasion of the first symposium about the topic held in 2003 in Tokyo, Japan. Over the past decade, three additional symposia took place: Saint Andrews (Scotland) in 2005, Pacific Grove (USA) in 2008, and Hobart (Tasmania) in 2011. The next bio-log- ging symposium will be organized in France and is tentatively scheduled for Strasbourg in September 2014. The number of manufacturers has steadily increased since the inception of Wildlife Computers (USA) in 1986, the first – to the best of our knowledge – bio-logging company ever. Nowadays, the core of the bio-logging production is concentrated in the North America and Japan (Ropert-Coudert et al., 2009b), but emerging companies in the UK (CTL), Iceland (Star-Oddi) or Italy (Technosmart) are gaining worldwide momentum (Table 1). A non-negligible propor- tion (a rough estimate of 20%) of bio-logging devices is still produced in research institutions, the so-called custom-made bio-loggers, and is thus accessible only through collaborations between researchers. In Europe, for example, research-driven developments are found in the Sea Mammal Part I – Chapter 1 19
Table 1: A non exhaustive list of the most-used bio-loggers together with their weights and the sensors they include, as well as the name of the manufacturers. UK
USA
Survey,
USA
USA Computers,
Antarctic Manufacturers
Lee,
ritish B Cefas technology Ltd, UK Star-Oddi, Iceland Little Leonardo, Japan TechnoSmart, Italy Mr. Oceangraphic Hole Woods Institution, Sea Mammal Research Unit, UK Swansea University, UK Wildlife transmission)
transmission)
(data
transmission)
(data
(data Argos
Argos
GPS, GMS
Sensors light, speed, GPS,
temperature, temperature, temperature,
Depth, GPS GPS Depth, Depth, Depth, temperature, light, speed, acceleration, magnetometer 370 2.71.7 Depth, temperature Temperature, light 130 Depth, temperature, speed, acceleration, magnetometer 22 22 225 300545 Depth, audio, pitch, roll, heading 370 Depth, temperature, conductivity 2.5 Light, wet or dry status 30 Depth, temperature, light, acceleration, magnetometer tag
Model (g) Weight I
Phone
SRDL tag Cefas G5 DST bird DST magneticORI400-D3GTW1000-3MPD3GT 19W400-ECG 9DSL400-VDT IIGiPSy Depth, temperature, magnetometer, tilt 82 57CatTraQ Depth, temperature, acceleration SPLASH10-F-400 Depth, temperature, image ECG DTAG CTD tag GPS Daily Diary 42 Mk 15 Mk9 20 Ecophysiology and animal behaviour
Research Unit of the University of St. Andrews, which organized the 2nd bio-logging symposium. In France, the only openly declared bio-logging development team is found at the Institut Pluridisciplinaire Hubert Curien in Strasbourg. The next big step for the bio-logging community will be to form a society so as to reach an official status and help structuring the community. Bio-logging is especially expected to play an important role in the forthcoming decade regarding conservation issues and will repre- sent a crucial tool to assess large vertebrate species distribution and links between the physical environment and the biological response of animals to its variation (see Cooke, 2008).
2. Overview of bio-logging applications
2.1. Reconstructing the movement and feeding behaviour The ancestors of all bio-loggers are probably time-depth-recorders, com- monly referred to as TDR in several instances. These devices record hydrostatic pressure according to time so as to reconstruct diving activity of sea animals. Oddly, the very first incarnation of a TDR, which was attached to a freely-diving Weddell seal Leptonychotes weddelli, consisted in coupling a kitchen timer with a pressure transducer (Kooyman, 1965; 1966). Subsequent devices also functioned on a mechanical basis, such as miniature pencils that were animated by pressure changes and drew the profiles of dives onto a miniature paper (e.g. Naito et al., 1990). The emer- gence of solid-state memories put an end to this era of clever handcrafting. Nowadays, TDR can weigh as less as 2.7g and are able to capture depth and temperature data every second for around 10 days. When associated with GPS, they provide localisation onto both the horizontal and vertical dimensions, on a large range of species. Originally, TDR delivered only a 2D view of the diving activity (depth according to time) but progresses in behaviour reconstruction came from the utilisation of accelerometers. Accelerometers record gravity-related and dynamic acceleration signals and can be used to provide specific information about the movements of the body, such as walking gait (e.g. Halsey et al., 2008) or head-jerking (e.g. Viviant et al., 2010). The poten- tial of accelerometers to reconstruct time budget activity was demon- strated in several instances (e.g. Yoda et al., 1999; Ropert-Coudert et al., 2004a; Watanabe et al., 2005). The addition of gyroscopes and magne- tometers makes it possible to reconstruct the precise path of animals in the three dimensions. This approach, called “dead reckoning” (Wilson et al., 1991), is very prone to making substantial errors. For example, a Weddell seal diving for ca. 17mn would accumulate an error in its posi- Part I – Chapter 1 21 tion calculated via dead reckoning of nearly 100m over this period (see figure 5a in Mitani et al., 2003). While methods exist to take this error into account (Mitani et al., 2003), dead reckoning is yet to be imple- mented at time scales longer than a few days. Anyway, the precision of tracking techniques thanks to GPS development makes it unlikely that dead reckoning will become a major approach. Small body movements, such as limb movements (Wilson and Liebsch, 2003), can also be finely reconstructed using Hall sensors, i.e. sensors measuring the intensity of the magnetic field. In this case, a magnet placed on one mandibular plate facing a Hall sensor glued onto the other mandibular plate (figure 1) allows researchers to determine when a prey has been swallowed and, following a proper calibration, the size and type of prey (Wilson et al., 2002; Ropert-Coudert et al., 2004b).
Figure 1: Schematic representation of the jaw movement recorder on a gentoo penguin Pygoscelis papua (top) and a young wild boar Sus scrofa (bottom). A magnet and a Hall sensor, sensitive to the strength of the magnetic field are placed on the two mandibles, facing each other. When the mouth opens the Hall sensor senses a reduction in the intensity of the magnetic field and sends this information via a cable to the bio-logger attached on the body.
Finally, in the context of assessing animal movements on a world-wide scale, the two major developments of the recent decades feature the 22 Ecophysiology and animal behaviour advent of GLS (global location sensors) and of GPS (global positioning system). Global location sensors are miniaturised units that store light measurements at regular intervals, from which position can be estimated (using day length and noon time). Initially described by Wilson et al. (1992a), this method revolutionised migration studies because devices are particularly small (around 1g), cheap, and are able to record data up to several years. They can therefore be deployed year-round on a wide range of individuals and species (Fort et al., in press). Miniaturised GPS (the smallest ones currently weigh 5g or less) usually have shorter recording times yet far higher spatial resolution than GLS (a few meters versus a few tens of km). Their generalised use triggered a quantum leap in the spatial ecology of free-ranging animals (Ryan et al., 2004)
2.2. Reconstructing the internal temperature and heat flux Animal-borne bio-loggers also benefited physiological studies as these bio-loggers allowed researchers to investigate internal adjustments to the constraints of, for example, experiencing extremely low temperatures (Gilbert et al., 2008; Eichhorn et al., 2011). These feats cannot be real- ized in the confines of a laboratory. Reduced core temperature in the body of deep divers like the king penguins Aptenodytes patagonicus shed a new light on the physiological mechanisms involved in energy savings at great depths (e.g. Handrich et al., 1997). In parallel to the externally- attached bio-loggers that recorded mandibular activity (see above), mea- surements of temperature in the stomach (Wilson et al., 1992b; Grémillet and Plos, 1994) or the oesophagus of endotherms (Ancel et al., 1997; Ropert-Coudert et al., 2001) also permitted to explore when these animals fed onto their exothermic prey as their swallowing induced a drop in the temperature (see figure 2 and additional discussions around the principle and the limitations of this method in Hedd et al., 1995; Ropert-Coudert et al., 2006a). Heat flux measurement bio-loggers may also be used to study homeothermy in animals swimming in cold waters (e.g. Willis and Horning, 2005).
2.3. Reconstructing the heart effort: ECG vs. heart rate One challenge in ecophysiology is to determine energy expenditures of free-ranging animals. Field methods based on doubly-labelled water exist but these are long-term methods that integrate the energy expended over a period of few days (Speakman, 1997). Further to the point, these methods require multiple capture and handling, which are not always easy to imple- ment in the field, especially for shy and sensitive species. Cormorants, for example, respond to handling with intense overheating. Last but not least, Part I – Chapter 1 23
Figure 2: Two temperature signals (°C) recorded by sensors placed in the upper part of the oesophagous (top) and in the stomach (bottom) of an Adélie penguin Pygoscelis adeliae fed with cold food items. Each ingestion is visualised as a sudden drop in the temperature, followed by a slow recovery. the doubly-labelled water method is expensive and implies a laboratory specifically equipped with isotope analysis facilities. In contrast, the mea- surement of heart rate can give an idea of the energy expended, as heart rates are linked with metabolic rates (Nolet et al., 1992; Green et al., 2001; Weimerskirch et al., 2002). Although the shape of the relationship is often unclear (Froget et al., 2002; Ward et al., 2002; McPhee et al., 2003), mea- suring heart rate still enables the estimation of the effort allocated to basal versus non-basal (e.g. locomotor) activities. Among the bio-logging approaches for measuring heart rate, two tech- niques have emerged: i) heart rate recorders (HRR) that detect the heart beat and store in their memory the interval between each heartbeat or the number of heart beats per certain time period; ii) electrocardiogram recorders (ECGR) that monitor and store the complete electric signals allowing to access the complete PQRS profile of a heartbeat. both sys- tems measure the electrical activity of the heart transmitted via 2 or 3 electrodes placed in different parts of an animal’s body. HRR have an extended autonomy since they only count intervals (Grémillet et al., 2005) but are prone to error because the ability to distinguish heartbeats from electric noise due to muscular activity depends solely on an on- board algorithm. In contrast, ECGR requires a processing of the signal but this ensures that only heartbeats are counted (Ropert-Coudert et al., 2006b; 2009c). However, commercially available ECGR have limited autonomy. 24 Ecophysiology and animal behaviour
Figure 3: Recordings of heart rate on a captive mandrill Mandrillus sphinx. A. Photograph of the collar where devices are attached and the electrodes protruding from it. B. Collar mounted on the mandrill with the electrodes plugged on the skin and secured by bolts. C. A comparison of the heart rate recorded by two different devices: a heart rate counter (Polar Watch, blue) and an electrocardiogram (ECG recorder, red). The latter allows the user to visualize each heart beat as a PQRS complex and is thus much more reliable than heart rate directly given by the counter (calculated via an internal algorythm to which the user generally cannot access). The heart rate given by the counter shows large variation that are absent on the signals derived from the ECG. © Jacques-Olivier Fortrat.
The comparison of the heart rate signals of a sleeping mandrill Mandrillus sphinx directly derived from a commercially-available heart rate monitor (© Polar Electro, France) and the one calculated from an ECGR (Little Leonardo, Japan) illustrates well the risk of applying tools that are devel- oped for a specific use (here, the Polar Watch is intended for measuring heart rate during human exercise) onto an animal model without prior Part I – Chapter 1 25 calibration work (figure 3). The need to reduce the risk of storing electro- myograms generally leads researchers to implant the HRR in the body, while ECGR can either be implanted or externally attached. Implantation is not trivial as it involves anaesthesia and surgery, with all the associated risks, and is not always easy to perform in the field (see Green et al., 2004; Beaulieu et al., 2010).
2.4. Viewing the environment: image data logger Data contained in bio-loggers are used to reconstruct the activity and, in some cases, the environment in which the animals move. But the dream of all users is to be able to visualise directly what the animals are seeing. Images, if they do not give access to physiological information per se, are a smart and informative way of studying behaviour. Images are also attrac- tive to a large audience as they do not always require specific knowledge to be interpreted. As communication towards the public becomes paramount to Science, this is a non negligible asset for bio-logging approaches that use digital-still picture recorders or even video recorders. The National Geographic Crittercam project was a pioneer in merging the scientific community with common people. However, their usefulness to answer scientific questions was often questioned. Digital-still cameras take images following a definite sampling interval which is not always adequate for short time events like prey capture. Yet, these techniques can provide unravelled insights into prey identifi- cation (figure 4, see also Davis et al., 1999; Watanabe et al., 2006), prey density (Watanabe et al., 2003), group behaviour (Takahashi et al., 2004a; Rutz et al., 2007) or the biomechanics of flight (Gillies et al., 2011). Video recording systems have limited autonomy and are still rather bulky to be used without the risk of impairing the performances and health of some animal models (see the bulkiness of a video recorder mounted on an emperor penguin in the figure 1 from Ponganis et al., 2000). Recent advances in miniaturisation allowed for these devices to be placed on the head of a flying seabird (Sakamoto et al., 2009). In an applied context, it has been recently proposed to use newly-developed, highly miniaturised digital-still picture recorders mounted on seabirds to monitor pirates fish- ing boat (Grémillet et al., 2010).
2.5. Reconstructing the environment: animals as bio-platforms The pioneers of bio-logging soon realised that this technology not only allowed the study of animals in their natural surroundings, but also to access their biotic and abiotic environment. Especially in the oceans, where sampling through the water column is impossible from satellites 26 Ecophysiology and animal behaviour and expensive from research vessels, this approach led to remarkable advances. As soon as time-depth-recorders were coupled with positioning devices and temperature sensors, the thermal structure of water masses could be assessed. This was first conducted in Antarctica by Wilson et al. (1994), which used penguins equipped with data loggers to map thermal gradients across the 100m of the Maxwell Bay. Not only did they assess this abiotic parameter, but they also cross-checked this information with an estimation of krill biomass in this water mass, which was based upon the predatory performance of the birds. This approach was revolutionary. Yet, temperature measurements were too coarse to be adequate for proper oceanography work. It is only a decade later that seabirds were equipped with loggers measuring ocean temperature to 0.005K and depth to 0.06m, values accurate enough to track the vertical movements of the thermo- cline off Scotland in the North Sea (Daunt et al., 2003). However, this approach was then only used to investigate areas that had been already studied, and had been sampled using conventional, ship-based surveys.
Figure 4: Image data loggers. A. A digitial-still-picture logger (Little Leonardo, Japan) mounted on a great cormorant Phalacrocorax carbo in Greenland (left, © David Grémillet) together with a view of the logger itself (B) and four examples of pictures taken by the logger (C). The examples show fish prey caught in the beak of the cormorant. Part I – Chapter 1 27
The next step consisted in using free-ranging marine animals fitted with bio-loggers to sample unknown areas. For instance, Charrassin et al. (2002) used temperature data collected by diving king penguins to identify a previously-unknown water mass off Kerguelen in the Southern Ocean. However, operational oceanography requires real-time assessments of biotic and abiotic parameters, for instance to parameterise models of ocean circulation and climatic processes (IV, 1). This was not possible using ancient archival tags fitted to marine predators, since those had to be recovered to download the data, sometimes weeks or months after the actual measurement. Such problem was solved by the use of a sys- tem integrating bio-physical sensors of the environment (e.g. water colour, temperature, salinity) and sensors of the animal’s movements (3D accel- eration, depth and speed) with the Argos positioning and transmission system. Such tools are large, require substantial battery power, and can only be deployed on large marine mammals for the time being, in particu- lar elephant seals (Mirounga leonina). However, they allowed a major step forward because elephant seals cruise the Southern Ocean in areas that are beyond the reach of satellite or vessel-based oceanography, especially in the marginal ice zone off Antarctica and at depths of more than 1000m (Charrassin et al., 2008). From these areas, devices fitted to these large, record-breaking divers can send new data which are now being routinely integrated into ocean physics models (Roquet et al., 2011).
2.6. Multi-information sensors: the special case of accelerometry A single parameter may not always be sufficient to address a scientific question, such as in the case of the dead reckoning technique that we mentioned earlier (section 2.1). However, the use of multiple sensors is not always possible since it generally leads to an increase in the bulkiness of the devices. Fortunately, accelerometry can be used to derive more infor- mation than only the posture or the activity of animals. For example, with sensitive accelerometers, it is possible to detect the faint signal of the heart rate in the movements of the cloacae of a bird and thus address physiological questions without the need for electrodes and/or implanted materials (Wilson et al., 2004). In addition, since a rough 70% estimate of the energy is expended through movements, overall dynamic body accel- eration (ODBA) or partial dynamic body acceleration (PDBA), derived from 3-axes or 2-axes accelerometers, respectively, was proposed as an index of energy expenditures (Wilson et al., 2006). ODBA and PDBA are indeed significantly related to oxygen consumption in a variety of species, and both offer a good proxy of metabolic activity when combined with heart rate loggers (Halsey et al., 2008). Apart from accessing physiological parameters, these sensors can also be used to infer prey availability in the 28 Ecophysiology and animal behaviour environment. Changes in wing beat frequency and amplitude are increas- ingly used to infer prey encounter in birds (Ropert-Coudert et al., 2006b), while detection of head jerking movement are related to prey capture in marine mammals (Suzuki et al., 2009; Viviant et al., 2010).
3. The road to bio-logging is paved with good intentions but…
3.1. The standard bio-logging trade-off Increasing the life-time of a bio-logger while keeping the same level of per- formances leads to the following paradox. On the one hand, the amount of information stored is increased, and consequently the memory capac- ity has to increase too; on the other hand, the energy required to power the electronic circuit is increased, and so should be the battery size and weight in order to address this extra demand. Based on the power con- sumption of a unit, it is possible to adapt batteries of different capacities to the devices in order to adjust the working-time to the specific needs of a study. However, a longer working-time means larger and heavier batteries and bio-loggers, which may have an impact on the health of the species targeted or even become inappropriate (I, 2). This balance between small units with a lesser impact on the animal but reduced life time, and larger devices with enhanced functionalities but restrictions on their applicabil- ity, is a major problem seriously dealt with by the bio-logging community i) for ethical reasons, and ii) to ensure that the data collected are reliable and are as close to the norm as possible (Ropert-Coudert et al., 2007). Regarding the impact of bio-logger, one must be aware that animals are generally shaped to optimise their movement through a medium. Swimmers are hydrodynamically featured, while flying animals present a specific adaptations to reduce their body mass. Thus, any externally- attached item may impair these features and lead to an increase in energy expended or a change in behaviour. In parallel, we already mentioned the negative consequences of implanting bio-loggers. Guidelines are regularly produced to reduce the negative impact of bio-loggers (Casper, 2009). Bio- loggers, for example, should weigh less than 3% of the body mass of flying birds (Phillips et al., 2003) and less than 4-5% of the cross-section of the animal (Bannasch et al., 1994). Despite these guidelines, we believe that the scientific community should move forward to adopt a common code of conduct. Indeed, the bio-logging community is very mindful about the need to reduce the impact of devices, but newcomers may not always be aware of guidelines specifically designed for bio-logger deployments (see above). In some instances, referees are not aware of them and accept papers that present ethical concerns or which Part I – Chapter 1 29 results are questionable due to the negative influence of a bulky device on the performances of the animals. Which institution could be in charge of ensuring that the appropriate guidelines are followed? Some scientific jour- nals have taken the lead in addressing this problem: for example, Animal Behaviour has very strict ethics regulations and ask the authors to address them before submission to peer review. The pressure to produce attrac- tive results could, however, hinder these efforts as it sometimes pushes researchers to emphasise outputs against rigor (see Ropert-Coudert et al., 2007). Conversely, enforcements of strict rules would also be detrimental without consideration of the benefits that overstepping them could bring in terms of new scientific results.
3.2. Beyond sensors and devices: homogenising analyses and sharing data Originally, each research group using bio-logging approaches developed its own method for analysing the data generated by bio-loggers. This led to the emergence of several analytical programming codes that tackled the same question and therefore, to a divergence in the way bio-logging data were processed. For example, the bottom phase of a dive can be defined in several different manners, leading to values that are not comparable from one study to another. The trend of diversifying the analytical methods is also enhanced by the presence of free software like R that allows users to create and disseminate their own codes and thus their own definitions for various parameters. In addition, the possibility offered by most bio- loggers of selecting the frequency at which the sampling is done also leads to diversification and renders comparisons across data sets difficult. In physics, the “sampling theorem” states that the sampling frequency must be at least twice that of the signal’s highest component frequency (for a periodic signal) to avoid aliasing. Similarly, biologists suggested that the sampling interval should not represent more than 10% of the duration of the biological event that one wishes to measure (e.g. the lowest sampling frequency to measure a 600sec dive of a Weddell seal is 60sec, Boyd et al., 1993; Wilson et al., 1995). Not adopting a proper sampling protocol may lead to misinterpretation of the data and false biological conclusions (Ropert-Coudert and Wilson, 2004). Recently, the question has become a topic of reflexion on the occasion of various workshops. Can we (and should we) homogenise bio-logging data analysis? The difficulty to define the best practice in that case is twofold. First, devices always evolve and become more efficient or collect new types of data. Consequently new analytical methods are required to handle these novelties. Secondly, the analytical method depends upon the questions sought. In that sense, the currently best practice would not stay best for very long. Yet, we need to be able to compare datasets taken in 30 Ecophysiology and animal behaviour different locations, time and using different means, especially if we are to tackle large-scale questions. Methods like down-sampling, although nec- essarily frustrating, are keys to address such issues. We strongly advocate for working groups to explore paths for the homogenisation of analytical procedures within the framework of, for example, the Expert Group in Birds and Marine Mammals of the Scientific Committee for Antarctic Research (SCAR), or the newly-formed group of experts in accelerometry that was constituted on the last bio-logging symposium in Hobart. In addition to this issue, the use and share of data from bio-logging must be optimised. A whole book could be filled with the issue of data sharing, but only the surface will be scratched here. The million of data points that are now routinely recorded by data loggers and the multiplicity of the research teams using such an approach make it necessary to centralise, archive, and ultimately share the data. Some researchers had been collect- ing bio-logging information over several decades and onto a large range of individuals and species. Upon retirement, their data would be lost if no system stores them. This is only recently that specific data repository have emerged. The tendecy is now to multiply storage points, each scientific society recognizing the need for a database on their specific topic. For example, marine researchers studying the localisation and diving activity of polar top predators can store their data into the database managed by the SCAR (SCAR-marBIN and Antabif) that are themselves linked to marine databases at a larger scale (OBIS, SeaWiFS, etc.). This multiplica- tion and cross-sharing of datasets among databases, while duplicating the work, guarantee the permanence of a dataset as it will still be available even if one database is closed. An incentive to sharing the data is found in the recent effort to consider data sharing as a genuine publication, asso- ciating a DOI to a data set. As such, institutions evaluating a researcher’s output can value his/her effort towards the scientific community through this marker.
3.3. Bio-logging: an academic and commercial endeavour Efficient bio-logging equipment is generally achieved through a close collaboration between engineers and users. However, research institu- tions able to combine both expertises under the same roof are scarce. In some privileged situations, an academic collaboration can be developed between universities so as to link a department of biology and an engi- neering department for example. The highest technical sophistication can then be attained and complex and specific questions be answered. Once a prototype is created, engineers face more practical duties that may be less intellectually satisfying. Among those, the issue of proper conditioning and packaging of the device is critical. Most dysfunctions of bio-loggers Part I – Chapter 1 31 are due to practical packaging problems. Solving these problems requires a multidisciplinary and complex engineering approach. Once the equip- ment has finally been validated, biologists would request a large number of units and this is precisely when academic systems reach their limits. Indeed, academic bodies are (and probably should) not be involved into mass production as this would mean adopting an industrial approach to bio-loggers production. Industrial production implies that electronics hardware, software, connectic systems and batteries, circuit design and protection, casing and packaging, tests and validation, are all included at once in the reflexion process. Additionally at each stage of develop- ment, costs are balanced and they influence decisions at the next stage. Industries usually aim at producing the best device according to the cost it represents for them; and this is generally decided with consideration of the market, the number of potential customers and the most reasonable price per unit. Real and viable situations generally lay between these two positions. Subcontracting industrial fabrication could be an alternative for academic developers. Academic engineers and/or researchers could also create a start-up company based on what they developed to initially address their scientific needs. However, this involves an optimal knowl- edge of the scientific and technical need, as well as of the practical pro blems that may be encountered in the field while using the equipment. In a nutshell, everything reverts to the following question: is the demand originating from users asking for specific developments (greater perfor- mance, new sensors…) or from the engineers anticipating the application of new technologies? Both stimulations are probably necessary to draw an ambitious but realistic product specification.
4. Where do we go from here?
4.1. Going toward large-scale deployment For decades, the paucity of manufacturers, the expensive price of bio- loggers, their restricted memory or battery capacity, as well as the lack of adapted analytical tools precluded the deployment of numerous units at a time. Thanks to technological advances, such as those taking place in the mobile phone industry, some cheap, low consumption and consequently small bio-loggers have started to appear on the market. With these, large- scale deployments have become achievable. While occasionally dozen of devices had been deployed simultaneously to explore cooperative div- ing (Takahashi et al., 2004b), the first large-scale deployments, in both space and time, originated through programs like the Tagging of Pacific Pelagics (Topp, Block et al., 2003, see also http://www.topp.org/). Since 32 Ecophysiology and animal behaviour the inception of the Topp programs, thousands of tags have been attached to 22 top predator species in the Pacific, including whales, sharks, sea turtles, seabirds, pinnipeds and even squids. Mass production of devices is now a reality: it allows researchers to work at unprecedented spatial scales and on entire populations of studied animals. In this field, the United Kingdom has taken a huge step forward. For example, the long-life, min- ute geolocators developed by the British Antarctic Survey are deployed on a worldwide scale (e.g. Conklin et al. 2010). Recently, mass-production of GPS for mobile phone also created an alternative market where cheap GPS can be purchased by researchers who can re-conditioned them specifi- cally to their needs. As an illustration of this, the IPHC bio-logging unit is modifying commercially-available GPS units (Cat Traq from Perthold Inc., http://www.mr-lee-catcam.de/ct_index_en.htm) to make them suit- able for use on wild animals. However, there is a negative side to this large-scale enthusiasm: cheap devices do not always meet the usual sci- entific criteria. Lesser reliability or lower degree of technical information must be balanced with the benefits that can arise from the use of these mass-production bio-loggers. In other words, caution in the use of cheap devices must be taken to avoid impacting scientific excellence. Thorough calibration must be a premise to large-scale deployments.
4.2. Importance of multiple sensors As evoked briefly earlier in this chapter, the use of multiple sensors – when applicable – offers an added value by providing a much complete picture of the behaviour and physiology of the animals in their envi- ronment. The combination of simple sensors (e.g. pressure sensor and temperature sensor) became a standard in even the simplest data loggers, but genuinely multi-sensor loggers are still few. Among those, it is worth mentioning the “daily diary” unit developed by Prof. Rory Wilson at the University of Swansea. Despite their relatively small size ranging between 21 and 90g according to the size of animal, these bio-loggers can contain up to 14 different channels of both slow and fast sampling sensors work- ing simultaneously (Wilson et al., 2008). Apart from the daily diary unit, multi-sensing devices, either developed by research teams or commercially available (Wildlife Computers, Little Leonardo, Greeneridge Science, etc.), are used in large body sized models, e.g. fin whales Balaenoptera physalus (Goldboegen et al., 2006). To extend the applicability of multi- sensing devices to smaller animals, special developments are needed (I, 2); for example, a drastic reduction in the consumption is a pre-requisite to a generalisation of multi-sensing to species smaller than a 1-2 kg ani- mal. In addition, new chemical sensors to detect for example the level of oxygen in the water or the blood will pave the way for new generations of Part I – Chapter 1 33 multi-sensing bio-loggers with new requirements and constraints for the developers. Here, a distinction must be made depending on the acquisi- tion rates of these new sensors. The deployment of sensors for quasi-static parameters for which the sampling interval is equal or longer than 1s (e.g. temperature, light, pressure…) would not cause any trouble as transducers use low power and the volume of data is small. However, the use of sensors for medium speed parameters sampled typically between 10 and 100Hz (e.g. accelerometers, gyroscopes, etc.) requires larger memory volume and greater energy to store the data. Even stronger difficulties are faced for sensors that acquire high speed parameters (more than 100Hz) like elec- trocardiograms, electromyograms, or electroencephalograms. Numerous technical problems occur, and a special electronic architecture is needed to manage the high volume of memory, high speed communication for data transfer, and so on. With the million of data points that the daily diary units can generate, the next challenge will be to develop a software able to handle, display and summarise the complex information delivered by the next generation of bio-loggers. Prof. Wilson thus invested an important amount of energy, resources and time in developing such a tool and did it in such a way that its utilisation can reach a larger public than the scientific community alone (Wilson et al., 2008, http://www.swan.ac.uk/biosci/research/smart/ smartsoftware/). The software allows users to interpret the data from the bio-loggers so as to truly reconstruct behaviour and visualise it. For exam- ple, data points from the magnetometer, gyroscope, accelerometer and altitude sensors are combined and the result on the screen is an albatross (a computer graphic one, of course) flying in three dimensions following the exact paths that the original albatross flew. Beyond the example of the daily diary, visualisation software to accommodate complex and large datasets and display them in a pleasant and efficient manner is becoming increasingly available. The statistical free software R is of course power- ful and readily accessible but its lack of user-friendliness may sometimes limit its popularity for complex analyses and representations. Alternatives to R are numerous and we can only mention Igor Wavemetrics, which was recurrently presented at the last bio-logging symposium (http://www. wavemetrics.com/).
4.3. Combining the best of biotelemetry and bio-logging Biotelemetry – at least in theory – clearly has its advantages, especially as long as securing data is concerned. However, real-time data transmission is practically hampered by numerous factors leading to a temporary inter- ruption in communication, which in turn means a definite loss of mea- surements (Vincent et al., 2002; Costa et al., 2010). These blank periods 34 Ecophysiology and animal behaviour are generally due to technical limitations (e.g. signal attenuation, wave’s absorption by the environment, electromagnetic interferences, etc.) and to the behaviour of the animal to which the transmitter is attached (e.g. relative position of body and antenna, immersion in water or in a bur- row, etc.). In comparison, bio-logging seems the perfect solution. Yet it suffers from an important drawback: the bio-logger has to be connected back to a computer at the end of the experiment to retrieve the data, which means that the animal has to be still alive, re-localised, re-captured and should still be carrying a bio-logger that is still functioning! In other word, deploying a bio-logger represents a binary game: if only one point goes wrong in the chain, no data are collected. Obviously, combining the capabilities of the two methods seems to be the solution. An ideal device would record permanently the data in an embedded memory, and would then transmit them regularly to a base station. Of course, this basic principle needs to be adjusted to each experi- mental situation. Data may be transferred following a fixed schedule, for instance when an animal returns to a fixed location in space and time. As a consequence this would only require a single base station installed within radio range of such a site where the animal is known to be found at regular interval, and with a bidirectional connection between the logger and the station. The base station would be filled gradually with data from the logger, and be downloaded by the user when needed. If the animal dis- appears only the data collected after the last transfer with the base station are lost. Alternatively, the base station can interrogate the environment at fixed schedules or be triggered manually to search for a telemetric logger within its reception range (see the approach developed by the University of Amsterdam, Shamoun-Baranes et al., 2011). The reverse strategy con- sists in asking the telemetric logger to regularly scan the radio-frequency environment, in order to search for a base station. In this case, scatter- ing numerous base stations in a given experimental area would enhance the success rate of data transfer. These stations can also communicate between each other to optimise data organization and synchronisation. Additionally, each base station can communicate with a large number of telemetric loggers. The last step in this concept consists in bio-loggers able to communicate not only with base stations, but also among themselves, leading to a genuine network of communicating devices. Data would then be shared with all the loggers coming within communication range and then transferred to a base station when one logger is close to it. A non- negligible side aspect of such an approach is the possibility to investi- gate proximity between animals, including time, duration and possibly distance of encounters. While theoretically attractive, a fair amount of development has to be done to reach this grand challenge. Both advances in electronics and data communication protocols are required. Progresses Part I – Chapter 1 35 in theoretical studies over software that could be able to manage such complex sets of interactions are paramount to the future success of these bio-logging networks and cannot involve only one type of institutions.
5. Conclusion
Bio-logging has gone through several steps from mechanical to digital, and from bulkiness to miniaturisation. The field is now moving towards globalisation and large scale coverage. In the marine realm, bio-logging coupled to automatic identification and weighing systems such as those that exist in the Antarctic could serve as a basis for long term monitor- ing programs. Such observatories would thus act in parallel with weather or oceanographic stations to deliver data on Antarctic biodiversity. This concept can be extended to the terrestrial realm with a network of sensing nodes monitoring the state of terrestrial ecosystems over time. With the rapid modifications affecting all ecosystems on Earth, monitoring pro- grams such as these are urgently needed. The diversification of the data collected, the increase in the temporal coverage and accessibility of bio- logged data, and the possibility for large number of units to be deployed in a given environment concur to promote bio-logging as the key approach for ecological sciences in the future.
Authors’ references
Yna Ropert-Coudert, Akiko Kato, Francis Crenner: Université de Strasbourg, Institut pluridisciplinaire Hubert Curien, UMR 7178, Strasbourg, France
David Grémillet: Centre d’écologie fonctionnelle et évolutive, UMR 5175, Montpellier, France University of Cape Town, FitzPatrick Institute, DST/NRF Excellence Centre, Rondebosch, South Africa
Corresponding author: Yan Ropert-Coudert, yan.ropert-coudert@iphc. cnrs.fr 36 Ecophysiology and animal behaviour
Aknowledgement
We thank Prof. Y. Naito for his dedication to promote bio-logging even after his retirement. We also thank Prof. Rory Wilson for sharing his pas- sion for the development and application of novel bio-loggers to a wide range of species.
References
Ancel A., Horning M., Kooyman G. L., 1997. Prey ingestion revealed by oesophagus and stomach temperature recordings in cormorants. Journal of Experimental Biology, 200, pp. 149-154. Bannasch R., Wilson R.P., Culik B., 1994. Hydrodynamics aspects of design and attachment of a back-mounted device in penguins. Journal of Experimental Biology, 194, pp. 83-96. Beaulieu M., Ropert-Coudert Y., Le Maho Y., Ancel A., 2010. Is abdominal implantation of devices a good alternative to external attachment? A comparative study in Adelie penguins. Journal of ornithology, 151, pp. 579-586. Boyd I. L., 1993. Selecting sampling frequency for measuring diving behaviour. Marine Mammal Science, 9, pp. 424-430. Casper R. M., 2009. Guidelines for the instrumentation of wild birds and mammals. Animal Behaviour, 78, pp. 1477-1483 Charrassin J.-B., Park y. H., Le Maho y., bost C.-A., 2002. Penguins as oceanographers unravel hidden mechanisms of marine productivity. Ecology Letters, 5, pp. 317-319. Charrassin J.-B., Hindell M. A., Rintoul S., Roquet F., Sokolov S., Biuw M., Costa D. P., Boehme L., Lovell P., Coleman R., Timmermann R., Meijers A., Meredith M., Park Y.-H., Bailleul F., Tremblay Y., Bost C.-A., McMahon C. R., Field I. C., Fedak M. A., Guinet C., 2008. Southern Ocean frontal structure and sea-ice formation rates revealed by elephant seals. Proceedings of the National academy of sciences of the USA, 105, pp. 11634-11639. Conklin J. R., Battley P. F., Potter M. A., Fox J. W., 2010. Breeding latitude drives individual schedules in a trans-hemispheric migrant bird. Nature Communication, 1, pp. 67. Cooke S. J., 2008. Biotelemetry and biologging in endangered species research and animal conservation: relevance to regional, national, and IUCN Red List threat assessments. Endangered Species Research, 4, pp. 165-185. Cooke S. J., Hinch S. G., Wikelski M., Andrews R. D., Kuchel L. J., Wolcott T. G., Butler P. J., 2004. Biotelemetry: a mechanistic approach to ecology. Trends in Ecology and Evolution, 19, pp. 334-343. Costa D. P., 1988. Methods for studying the energetics of freely diving animals. Canadian Journal of Zoology, 66, pp. 45-52. Part I – Chapter 1 37
Costa. D P., Robinson P. W., Arnould J. P. Y., Harrison A.-L., Simmons S. E., Hassrick J. L., Hoskins A. J., Kirkman S. P., oosthuizen H., Villegas- Amtmann S., Crocker D. E., 2010. Accuracy of Argos locations of pinnipeds at-sea estimated using fastloc GPS. PLoS One, 5, e8677. Daunt F., Peters G., Scott B., Grémillet D., Wanless S., 2003. Rapid-response recorders reveal interplay between marine physics and seabird behaviour. Marine Ecology Progress Series, 255, pp. 282-288. Davis R. W., Fuiman L. A., Williams T. M., Collier S. O., Hagey W. P., Kanatous S. B., Kohin S., Horning M., 1999. Hunting behavior of a marine mammal beneath the Antarctic fast ice. Science, 283, pp. 993-996. Eichhorn G., Groscolas R., Le Glaunec G., Parisel C., Arnold L., Medina P., Handrich Y., 2011. Heterothermy in growing king penguins. Nature Communications, 2, pp. 435. Froget G., Handrich Y., Le Maho Y., Rouanet J. L., Woakes A. J., Butler P. J., 2002. The heart rate/oxygen consumption relationship during cold exposure of the king penguin: a comparison with that during exercise. Journal of Experimental Biology, 205, pp. 2511-2517. Fort J., Pettex E., Tremblay Y., Lorentsen S.-H., Garthe S., Votier S., Pons J.-B., Siorat F., Furness R. W., Grecian,W. J., Bearhop S., Montevecchi W. A., Grémillet D., in press. Meta-population evidence of oriented chain- migration in northern gannets. Frontiers in Ecology and Environment. doi:http://dx.doi.org/10.1890/110194. Gilbert C., blanc S., Le Maho y., Ancel A., 2008. Energy saving processes in huddling emperor penguins: from experiments to theory. Journal of Experimental Biology, 211, pp. 1-8. Gillies J. A., Thomas A. L., Taylor G. K., 2011. Soaring and manoeuvring flight of a steppe eagle Aquila nipalensis. Journal of Avian Biology, 42, pp. 377-386. Goldbogen J. A., Calambokidis J., Shadwick R. E., Oleson E. M., McDonald M. A., Hildebrand J. A., 2006. Kinematics of foraging dives and lunge-feeding in fin whales. Journal of Experimental Biology, 209, pp. 1231-1244. Green J. A., Butler P. J., Woakes A. J., Boyd I. L., Holder R. L., 2001. Heart rate and rate of oxygen consumption of exercising macaroni penguins. Journal of Experimental Biology, 204, pp. 673-684. Green J. A., Tanton J. L., Woakes A. J., Boyd I. L., Butler P. J., 2004. Effects of long-term implanted data loggers on macaroni penguins Eudyptes chrysolophus. Journal of Avian Biology, 35, pp. 370-376. Grémillet D., Kuntz G., Woakes A. J., Gilbert C., Robin J.-P., Le Maho y., Butler P. J., 2005. Year-round recordings of behavioural and physiological parameters reveal the survival strategy of a poorly insulated diving endotherm during the Arctic winter. Journal of Experimental biology, 208, pp. 4231-4241. Grémillet D., Müllers R. H., Moseley C., Pichegru L., Coetzee J. C., Sabarros P. S., van der Lingen C., Ryan P. G., Kato A., Ropert-Coudert Y., 2010. Seabirds, fisheries, and cameras. Frontiers in Ecology and Environment, 8, pp. 401-402. 38 Ecophysiology and animal behaviour
Grémillet D., Plös A. L., 1994. The use of stomach temperature records for the calculation of daily food intake in cormorants. Journal of Experimental Biology, 189, pp. 105-115. Halsey L. G., Shepard E. L., Hulston C. J., Venables M. C., White C. R., Jeukendrup A. E., Wilson R. P., 2008. Acceleration versus heart rate for estimating energy expenditure and speed during locomotion in animals: tests with an easy model species, Homo sapiens. Zoology, 111, pp. 231-241. Handrich Y., Bevan R. M., Charrassin J.-B., Butler P. J., Pütz K., Woakes A. J., Lage J., Le Maho Y., 1997. Hypothermia in foraging king penguins. Nature, 388, pp. 64-67. Hedd A., Gales R., Renouf D., 1995. Use of temperature telemetry to monitor ingestion by a harbour seal mother and her pup throughout lactation. Polar Biology, 15, pp. 155-160. Kooyman G. L., 1965. Techniques used in measuring diving capacities of Weddell seals. Polar Record, 12, pp. 391–394. Kooyman G. L., 1966. Maximum diving capacities of the Weddell Seal, Leptonychotes weddelli. Science, 151, pp. 1553-1554. McPhee J. M., Rosen D. A. S., Andrews R. D., Trites A. W., 2003. Predicting metabolic rate from heart rate in juvenile Steller sea lions Eumetopias jubatus. Journal of Experimental Biology, 206, pp. 1941-1951. Mitani Y., Sato K., Ito S., Cameron M. F., Siniff D. B., Naito Y., 2003. A method for reconstructing three-dimensional dive profiles of marine mammals using geomagnetic intensity data: results from two lactating Weddell seals. Polar Biology, 26, pp. 311-317. Naito Y., 2004. New steps in bio-logging science. Memoirs of the National Institute of Polar Research, Special Issue, 58, pp. 50-54. Naito Y., 2010. What Is “Bio-Logging”? Aquatic Mammals, 36, pp. 307-322. Naito y., Asaga T., ohyama y., 1990. Diving behavior of Adelie penguins determined by Time-Depth-Recorder. The Condor, 92, pp. 582-586. Nolet B. A., Butler P. J., Masman D., Woakes A. J., 1992. Estimation of daily energy expenditure from heart rate and doubly labelled water in exercising geese. Physiological Zoology, 65, pp. 1188-1216. Phillips R. A., Xavier J. C., Croxall J. P., 2003. Effects of satellite transmitters on albatrosses and petrels. The Auk, 120, pp. 1082-1090. Ponganis P. J., Van Dam R. P., Marshall G., Knower T., Levenson D. H., 2000. Sub-ice foraging behavior of emperor penguins. Journal of Experimental Biology, 203, pp. 3275-3278. Ropert-Coudert Y., Beaulieu M., Hanuise N., Kato A., 2009b. Diving into the world of biologging. Endangered Species Research, 10, pp. 21-27. Ropert-Coudert Y., Brooks L., Yamamoto M., Kato A., 2009c. ECG Response of koalas to tourists proximity: a preliminary study. PLoS One, 4, e7378. Ropert-Coudert Y., Grémillet D., Kato A., Ryan P. G., Naito Y., Le Maho Y., 2004a. A fine-scale time budget of Cape gannets provides insights into the foraging strategies of coastal seabirds. Animal Behaviour, 67, pp. 985-992. Part I – Chapter 1 39
Ropert-Coudert Y., Kato A., 2006a. Are stomach temperature recorders a useful tool for determining feeding activity? Polar Bioscience, 20, pp. 63-72. Ropert-Coudert Y., Kato A., Baudat J., Bost C.-A., Le Maho Y., Naito Y., 2001. Feeding strategies of free-ranging Adélie penguins, Pygoscelis adeliae, analysed by multiple data recording. Polar Biology, 24, pp. 460- 466. Ropert-Coudert y., Kato A., Chiaradia A., 2009a. Impact of small-scale environmental perturbations on local marine food resources: a case study of a predator, the little penguin. Proceedings of the Royal Society of London B, 276, pp. 4105-4109. Ropert-Coudert Y., Kato A., Liebsch N., Wilson R. P., Muller G., Baubet E., 2004b. Monitoring jaw movements: A cue to feeding activity. Game and Wildlife Science, 21, pp. 1-19. Ropert-Coudert Y., Kato A., Wilson R. P., Cannell B., 2006b. Foraging strategies and prey encounter rate of free-ranging Little Penguins. Marine Biology, 149, pp. 139-148. Ropert-Coudert Y., Wilson R. P., 2004. Subjectivity in bio-logging science: do logged data mislead? Memoirs of the National Institute of polar research, 58, pp. 23-33. Ropert-Coudert y., Wilson R. P., 2005. Trends and perspectives in animal- attached remote sensing. Frontiers in Ecology and Environment, 3, pp. 437-444. Ropert-Coudert Y., Wilson R. P., Grémillet D., Kato A., Lewis S., Ryan, P. G., 2006b. Electrocardiogram recordings in free-ranging gannets reveal minimum difference in heart rate during flapping versus gliding flight. Marine Ecology Progress Series, 328, pp. 275-284. Ropert-Coudert Y., Wilson R. P., Yoda K., Kato A., 2007. Assessing performance constraints in penguins with externally-attached devices. Marine Ecolgy Progress Series, 333, pp. 281-289. Roquet F., Charrassin J.-B., Marchand S., Boehme L., Fedak M., Reverdin G., Guinet C., 2011. Delayed-mode calibration of hydrographic data obtained from animal-borne satellite relay data loggers. Journal of Atmospheric and Oceanic Technologies, 28, pp. 787-801. Rutz C., Bluff L. A., Weir A. A. S., Kacelnik A., 2007. Video cameras on wild birds. Science, 318, pp. 765. Ryan P. G., Petersen S., Peters G., Grémillet D., 2004. GPS tracking a marine predator: the effects of precision, resolution and sampling rate on foraging tracks of African Penguins. Marine Biology, 145, pp. 215-223. Sakamoto K.Q., Takahashi A., Iwata T., Trathan P. N., 2009. From the eye of the albatrosses: a bird-borne camera shows an association between albatrosses and a Killer Whale in the Southern Ocean. PLoS One, 4, e7322. Shamoun-Baranes J., Bouten W., Camphuysen C. J., Baaij E., 2011. Riding the tide: intriguing observations of gulls resting at sea during breeding. Ibis, 153, pp. 411-415. 40 Ecophysiology and animal behaviour
Speakman, J.R., 1997. Doubly labelled water: theory and practice, Chapman & Hall, London. Suzuki I., Naito Y., Folkow L. P., Miyazaki N., Blix A. S., 2009. Validation of a device for accurate timing of feeding events in marine animals. Polar Biology, 32, pp. 667-671. Takahashi A., Sato K., Naito y., Dunn M., Trathan P. N., Croxall J. P., 2004a. Penguin-mounted cameras glimpse underwater group behaviour. Proceedings of the Royal Society of London B, 271, pp. 281-282. Takahashi A., Sato K., Nishikawa J., Watanuki Y., Naito Y., 2004b. Synchronous diving behavior of Adelie penguins. Journal of Ethology, 22, pp. 5-11. Vinatier F., Chailleux A., Duyck P.-F., Salmon F., Lescourret F., Tixier P., 2010. Radiotelemetry unravels movements of a walking insect species in heterogeneous environments. Animal Behaviour, 80, pp. 221-229. Vincent C., McConnell B. J., Ridoux V., Fedak M. A., 2002. Assessment of Argos location accuracy from satellite tags deployed on captive gray seals. Marine Mammal Science, 18, pp. 156-166. Viviant M., Trites A. W., Rosen D. A. S., Monestiez P., Guinet C., 2010. Prey capture attempts can be detected in Steller sea lions and other marine predators using accelerometers. Polar Biology, 33, pp. 713-719. Ward S., Bishop C. M., Woakes A. J., Butler P. J., 2002. Heart rate and the rate of oxygen consumption of flying and walking barnacle geese (Branta leucopsis) and bar-headed geese (Anser indicus). Journal of Experimental Biology, 205, pp. 3347-3356. Watanabe S., Izawa M., Kato A., Ropert-Coudert Y., Naito Y., 2005. A new technique for monitoring the detailed behaviour of terrestrial animals: A case study with the domestic cat. Applied Animal Behaviour Science, 94, pp. 117-131. Watanabe Y., Bornemann H., Liebsch N., Plötz J., Sato K., Naito Y., Miyazaki N., 2006. Seal-mounted cameras detect invertebrate fauna on the underside of an Antarctic ice shelf. Marine Ecology Progress Series, 309, pp. 297-300. Watanabe Y., Mitani Y., Sato K., Cameron M. F., Naito Y., 2003. Dive depths of Weddell seals in relation to vertical prey distribution as estimated by image data. Marine Ecology Progress Series, 252, pp. 283–288. Weimerskirch H., Shaffer S. A., Mabille G., Martin J., Boutard O., Rouanet J.-L., 2002. Heart rate and energy expenditure of incubating wandering albatrosses: basal levels, natural variation, and the effects of human disturbance. Journal of Experimental Biology, 205, pp. 475-483. Wikelski M., Moskowitz D., Adelman J. S., Cochran J., Wilcove D. S., May M. L., 2006. Simple rules guide dragonfly migration. Biology Letters, 2, pp. 325-329 Willis K., Horning M., 2005. A novel approach to measuring heat flux in swimming animals. Journal of experimental marine biology and ecology, 315, pp. 147-162. Part I – Chapter 1 41
Wilson. R P., Cooper J., Plötz J., 1992b. Can we determine when marine endotherms feed? A case study with seabirds. Journal of Experimental Biology, 167, pp. 267-275. Wilson R. P., Culik B. M., Bannasch R., Lage J., 1994. Monitoring Antarctic environmental variables using penguins. Marine Ecology Progress Series, 106, pp. 199-202. Wilson R. P., Ducamp J. J., Ress W. G., Culik b. M., Niekamp K., 1992a. Estimation of location: global coverage using light intensity, in: Priede, I.G. and Swift S. M. (Eds). Wildlife telemetry: remote monitoring and tracking of animals. Ellis Horwood, Chichester, pp. 131-134. Wilson R. P., Liebsch N., 2003. Up-beat motion in swinging limbs: new insights into assessing movement in free-living aquatic vertebrates. Marine Biology, 142, pp. 537-547. Wilson R. P., Pütz K., Grémillet D., Culik B. M., Kierspel M., Regel J., Bost C.-A., Lage J., Cooper J., 1995. Reliability of stomach temperature changes in determining feeding characteristics of seabirds. Journal of Experimental Biology, 198, pp. 1115-1135. Wilson R. P., Scolaro A., Quintana F., Siebert U., Straten M. T., Mills K., Zimmer I., Liebsch N., Steinfurth A., Spindler G., Müller G., 2004. To the bottom of the heart: cloacal movement as an index of cardiac frequency, respiration and digestive evacuation in penguins. Marine Biology, 144, pp. 813-827. Wilson R. P., Shepard E. L. C., Liebsch N., 2008. Prying into the intimate details of animal lives: use of a daily diary on animals. Endangered Species Research, 4, pp. 123-137. Wilson R. P., Steinfurth A., Ropert-Coudert Y., Kato A., Kurita M., 2002. Lip- reading in remote subjects: an attempt to quantify and separate ingestion, breathing and vocalisation in free-living animals using penguins as a model. Marine Biology, 140, pp. 17-27. Wilson R. P., White C. R., Quintana F., Halsey L. G., Liebsch N., Martin G. R., Butler P. J., 2006. Moving towards acceleration for estimates of activity- specific metabolic rate in free-living animals: the case of the cormorant. Journal of Animal Ecology, 75, pp. 1081-1090. Wilson R. P., Wilson M.-P. T., Link R., Mempel H., Adams N. J., 1991. Determinations of movements of african penguins Spheniscus demersus using a compass system: dead reckoning may be an alternative to telemetry. Journal of Experimental Biology, 157, pp. 557-564. Yoda K., Sato K., Niizuma Y., Kurita M., Bost C.-A., Le Maho Y., Naito Y., 1999. Precise monitoring of porpoising behaviour of Adelie penguins determined using acceleration data loggers. Journal of Experimental Biology, 202, pp. 3121-3126.
Chapter 2
Animal-borne sensors to study the demography and behaviour of small species
Olivier Guillaume, Aurélie Coulon, Jean-François Le Galliard, and Jean Clobert
1. Introduction
One of the main characteristics of ecological systems is their hierarchi- cal organisation – communities are collections of species interacting with each other, species are groups of populations distributed spatially and connected through dispersal, and populations are made up of individu- als. Despite the widespread opinion that individual variation is the raw material of ecological and evolutionary dynamics, ecological approaches at the level of communities or ecosystems have tended to ignore the large variation among individuals seen in their morphology, behaviour, or life histories (Bolnick et al., 2003). One of the reasons for this is that there are serious methodological constraints in our ability to identify and track individuals of most animal species within complex communities. Indeed, most communities are made up of relatively small species, which are extremely challenging to mark and equip with sensors. For example, a large part of the world’s mammals weigh less than 100g (see figure 1, Gardezi and da Silva, 1999) and the median body size of birds is around 30-40g (Blackburn and Gaston, 1994). Yet, small animal species contrib- uted a lot to our understanding of ecological and evolutionary processes within natural populations, including population demography, dispersal ecology and evolutionary ecology (e.g. Clobert et al., 2001). They also play an important part in many terrestrial and aquatic ecosystems on Earth, where they include a large number of herbivores, small predators, as well as parasitoids, pollinators and plant mutualists. 44 Ecophysiology and animal behaviour
Figure 1: Relationship between species diversity (number of species) and species body mass (log-transformed, kg) in a large data set of the world’s mammal species after Gardezi and da Silva (1999). The distribution is significantly skewed to the right and indicates maximal species diversity for body mass around 25-63g. The dashed arrow represents the range of body mass (more than 100g) where animals can currently be fitted with some of the smallest bio-tracking and bio-logging devices (assuming the device must weigh less than 3-5% of the body mass, see table 1). This review focuses on animals weighing less than 100g, which represents most vertebrate species on earth and almost all invertebrate species. © The University of Chicago Press, 1999.
Several challenges for the development and proper implementation of animal-borne sensors on small terrestrial and aquatic animal species are identified. Appropriate detection and marking techniques are needed for the population ecology of many animal species, especially when rare and elusive ones are involved. In the field of movement ecology, gathering data on individual behaviour and movements of individuals in space and time, by using appropriate tracking technologies (also sometimes called bio-telemetry and referred to here as bio-tracking), is also fundamental. Finally, ecophysiological studies require gathering information on indi- vidual physiological state as well as on environmental conditions using micrometeorological and physiological sensors. This chapter aims at i) describing methods and sensors that have been used to collect such behavioural and demographic data on individuals of small animal species and ii) identifying the main current limitations and challenges of existing technologies and the most urgent areas of development in this domain. Part I – Chapter 2 45
Note that the methods reviewed here are also relevant for the juvenile forms of many large species. We define and review sensors in the broad sense so as to include indi- vidual marking techniques, bio-tracking techniques, and micrometeoro- logical and physiological sensors sensu stricto. Animal-borne sensors may be defined as any device installed on an animal that measures a physi- cal or chemical quantity and converts it into a signal which can be read by an observer or by an instrument, for example a micrometeorological quantity (e.g. temperature) or a physiological quantity (e.g. glucose level). Some tag systems that give information on the identity and devices that provide information on location of study animals, for example through the use of radio emitters or harmonic radar tags carried by the individuals, can be also considered as a type of biological sensors. Note that the cou- pled system of sensors and data loggers installed on the animal is called a bio-logger, and differs from simple bio-telemetry tools in the sense that data are stored locally in the memory of the devices and not transferred in real time via radio waves or other transmitting means. The development and use of bio-loggers with large animals is reviewed in chapter 1 of this book by Ropert-Coudert et al., and we will briefly review its applications with small animals here.
2. Methodological issues
Studying small animals in their natural environment or in experimental infrastructures without interfering with their normal behaviour presents major challenges to ecologists. This is especially true when studies require equipping model species with animal-borne sensors to measure and record environmental or physiological parameters. The main methodological issue with small animals lies in their weight and volume, which poses an upper limit on the size of sensors and constrains any marking and attach- ment technique. Indeed, tags and sensors must obviously neither interfere with the natural behaviour, nor influence the survival of the animals that carry them. Guidelines generally recommend that sensors must weight less than 5% of the animal body mass for vertebrates, and less than 10% for invertebrates (Cochran, 1980; Cant et al., 2005). However, those figures must be adjusted according to the study species and populations. For example, in species for which running, flying or swimming is essential, the maximum tolerated weight of the device should even be lighter (e.g., Bedrosian and Craighead, 2007). The evaluation of the impacts of sen- sors, of their components, like the antenna, and of the fixation procedure has been performed in various contexts, such as meta-analyses of survival, growth or fecundity, which showed some significant detrimental effects 46 Ecophysiology and animal behaviour
(for example Bridger and Booth, 2003; Weatherhead and Blouin-Demers, 2004; Whidden et al., 2007). Hence, careful tests of animal-borne sen- sors and associated protocols should be conducted prior to deployment in field or experimental conditions. In addition, small animals or juvenile forms are often characterised by fast growth and high mortality, and some of these species may also be difficult to capture even when their populations are abundant. These demographic facts put some challenges on the ability to recapture animals, relocate their tags and design appropriate attachment techniques. For example, juvenile forms of many lizards must be equipped with sensors that must not interfere with their fast growth and strong susceptibility to predators, but at the same time should be inexpensive because of the high chance to loose sensors in the course of a standard demographic study (Le Galliard et al., 2011). In the following section, we review available animal-borne sensors used to study the demography and behaviour of animals, and dis- cuss their applications and limits in the context of their implementation on small species.
3. Specifications of animal-borne sensors for small species
We list in table 1 some animal-borne sensors that can be commonly used to study demography and behaviour including bio-tracking, where the focus is on animal movements and data are often obtained remotely using telemetry, and bio-logging, where combined information on ani- mal behaviour, physiology and habitat are typically recorded and data are stored locally on an autonomous device installed on free-ranging animals. Bio-tracking technologies compatible with demographic and behavioural studies of small animals include small tags used to mark and identify individuals from a short distance like magnetic wire tags and passive Radio Frequency Identification tags (RFID, Canner and Spence, 2011; Courtney et al., 2000; Bergman et al., 1992), also called passive integrated transponders (PIT). Other tracking devices such as harmonic radars, VHF transmitters or satellite based transmitters allow the localisa- tion of small animals from a longer distance and therefore were used more often to study movement behaviour. Moreover, a range of more advanced bio-loggers can be used on small animals when remote transmission is not feasible and data should be stored locally. We review and discuss the use of these animal-borne sensors and techniques for small species. Other methods like video or camera traps have been used in some instances to obtain information on the abundance and ecology of small animals but they are less flexible and informative than bio-tracking and bio-logging technologies (see IV, 2, for a case study). Part I – Chapter 2 47
3. 1. Marking and tracking small animals with passive RFID tags It may be difficult to identify small animals with the help of techniques that avoid undue pain and stress, have no effect on fitness traits, and pro- duce marks that are not easily lost. Non-invasive techniques, such as paint marks or bead-tags, do not ensure the identification of a large number of individuals and are often only temporary, except for rings. Thus, most demographic studies of small species involve more invasive techniques such as branding, toe-clipping or scale-clipping. To avoid the potential pain and stress caused by these techniques, passive integrated transpond- ers (PIT) tags, also called passive RFID (radio frequency identification) tags, have been recommended because these ones provide permanent and reliable individual marks. This technology uses communication via radio waves to exchange data (identification number) between a reader and a passive electronic tag attached to the animal (Gibbons and Andrews, 2004). Examples of their use include tracking studies where “antennas” positioned in the natural habitat are connected to the reader. However, because the animal can only be detected at a short distance (see table 1), antennas must be located at places of maximal use by animals (e.g. dis- persal corridors, nests or burrows, runways, etc.). We used this technol- ogy to study the spatial ecology of small mammals in northern Europe, even during the winter snow period (Hoset et al., 2008; Le Galliard et al., 2007). The custom-made system was developed by Harald Steen and Lars Korslund from the University of Oslo (Korslund and Steen, 2006). It consists of a tube-shaped single coil antenna (20 × 4cm) placed on the ground along runways to maximize recording rates, and attached to Trovan® LID665 oEM PIT tag decoders (LID665, EID Aalten bV, Aalten, Netherlands) that record PIT-tag number, date and hour each time a tagged vole passed through the antenna (see figure 2). There are how- ever potential difficulties with PIT tags: they cannot be injected on the juvenile forms of most species, tags may get lost through the injection site, and injection as well as retention of these tags can cause small species pain and harm (e.g. Le Galliard et al., 2011).
3. 2. Harmonic radars for tracking very small animals Harmonic radar is becoming common to track small animals. Several studies recorded trajectories of hundreds of metres of very small animals, especially flying insects like beetles, honeybees or butterflies. Those studies focused on migration, dispersal, foraging, or flight behaviour and assessed environmental factors influencing movements (e.g. Chapman et al., 2011 and references therein). The harmonic radar system consists of an emitter that generates a micro-wave signal of a determined frequency. The tag – composed of a diode connected to a wire antenna – receives 48 Ecophysiology and animal behaviour
Category Sensor type Information Data acquisition Detection range
Identity and Scanning with a specific < 1 m with on- Passive RFID tag location radiofrequency source ground antenna Visual reading after Identity Requires recapture extraction Passive wire tag Identity and Scanning with a specific 3 cm with hand- location magnetic source held antenna Up to hundreds Scanning with a specific Harmonic radar Location of meters with on- Bio- radiofrequency source ground radars tracking Radio Identity and VHF receiver connected Up to 50 m with on- transmitter location to an antenna ground antenna
Satellite relay or VHF Global with Identity and GPS tracking receiver connected to an satellites, hundreds location antenna meters with VHF Identity and Satellite PTT Satellite relay (Argos) Global location Direct communication GLS logger Location Requires recapture port to data logger Direct communication GPS data logger Location Requires recaptures port to data logger Light, temperature; humidity, tilt; Bio- pressure and logging depth; magnetic field strength; Direct communication Data storage tag Requires recapture pitch and roll; port to data logger conductivity, salinity, dissolved oxygen, pH, imaging, EEG, ECG, EMG, etc Data storage tags, and GPS Global with and GLS loggers Satellite relay or VHF Combined Bio-logging + bio- satellite, hundreds combined with receiver connected to an tracking meters with VHF devices VHF and/or antenna relay Argos for remote transmission
Table 1. Specifications of some available sensors used in demographic and behavioural studies of small animals. Bio-tracking refers to the process of gathering remotely identity and/or location data on the study animal using passive or active devices, even if data may be stored locally prior to retrieval. Bio-logging differs from telemetry because data (location, animal physiology, or environment) are stored locally in the memory of the devices and must be downloaded via a communication port. Combined devices allow to store Part I – Chapter 2 49
Approximate lifespan Accuracy Smallest device of smallest device
< 1m 1 × 6 mm, 7.15 mg Unlimited
- 0.25 × 0.5 mm, small weight Unlimited
Few meters 12 mm long, 3 mg Unlimited
Few meters 10 × 5 mm, antenna 70 mm, 0.2 g Ten days
From hours in real-time to months Few meters 15g depending on the data acquisition and retrieval procedures
Hundreds of meters 5 g Several months
Hundred kilometres 0.5 g Up to several years
From hours to months depending Meters 22 × 14 mm, antenna 7 mm, 2 g on the numbers of locations
iButton (temperature): 17 × 6 mm, 1.49 g > 1 year
Geolocating archival tags (geolocation, internal and external temperature, - 6 months pressure, light, sea water switch): 8 × 20 × 6.7 mm, 1.9 g Archival Tag (temperature and pressure): 1 year 8 × 32 mm, 3.4 g EEG + GPS: 66 × 36 × 18 mm, 35g < 47 hours Argos + temperature sensor: 7.5 × 24 mm, Up to months depending on data antenna 200 mm, 5 g acquisition and retrieval procedures Meters with GPS, Argos + GPS + Temperature + activity: ten meters with Up to 3 years VHF, hundred 64 × 23 × 16.5 mm, antenna 178 mm, 22g meters with Argos Video AVED + VHF: 14 g Ten minutes
together both location, physiological and environmental data from various sensors and to transmit the stored data via a VHF radio or a satellite data relay network. RFID : RadioFrequency IDentification ; VHF : Very High Frequency ; GPS : global positioning system ; PTT : Platform Terminal Transmitter ; GLS : Global Location Sensing, AVEDs : animal-borne video and environmental data collection systems. 50 Ecophysiology and animal behaviour
Figure 2: Use of passive radiofrequency identification (RFID) tags for demographic studies of small mammals. A. Subcutaneous injection of a RFID tag on the back of a juvenile root vole (Microtus oeconomus). B, C. Custom made reader connected to a battery and an antenna, tube-shaped single coil antenna placed on the ground. © J.-F. Le Galliard. the signal and reemits at a doubled frequency that can be picked up by the receiver antenna. Two different transmitter-receiver systems are used to detect the tags (see figure 3). The first is a hand-held unit, originally designed to locate avalanche victims who wear tags on their clothes (e.g. Recco). This system is efficient to locate more or less 10cm tags from 50m above ground, less than 10 m on the ground and about 10cm below the ground surface (Mascanzoni and Wallin, 1986; o’Neal et al., 2005). The second, a ground-based scanning station, uses conventional radar plan position indicator technology (PPI) that gives the coordinates (range and azimuth) of the diodes (see Riley and Smith, 2002 for a detailed descrip- tion). This system can be used to track smaller tags (more than 1cm) on Part I – Chapter 2 51 horizontal landscapes (Cant et al., 2005; ovaskainen et al., 2008) or for vertical looking up to 900m (Riley et al., 2007). The advantages of the harmonic radar system are multiple. Since tags are passive and do not require batteries, they have a potentially unlimited lifespan, their weight can be very low (down to a few milligrams), and they are rather cheap (less than 1€). However, this system also presents drawbacks. All diodes reemit at the same frequency. Therefore, for studies that need to individually identify the target, a complementary method is required (e.g. a visual mark to identify the individual once found). In addition, the radar signals can be absorbed or disturbed by landscape components that may work as barriers or reemit a background noise. The best performances have been obtained with PPI in agricultural landscapes (Cant et al., 2005; Ovaskainen et al., 2008) but this was achieved with the use of a heavy and complex system made out of a trailer, which is often uneasy to use in the field. Finally, the external antenna of the device can hinder the movements of individuals, especially for ground-dwelling spe- cies, and it also complicates the fixation process (Langkilde and Alford, 2002; o’Neal et al., 2005; Pellet et al., 2006).
Figure 3: Use of harmonic radar to track flying insects. Left to right, a tag attached on a honeybee (from Riley et al., 2007), the ground-based scanning station used to track flying insects (from ovaskainen et al., 2008). © J.R. Riley/outlooks on Pest Management, © O. Ovaskainen.
To summarise, harmonic radars represent one of the most promising opportunities to study movements on small spatial scales for small spe- cies, but only a limited number of research groups were able to use this 52 Ecophysiology and animal behaviour technology and few succeeded on other fauna than flying insects (Lovei et al., 1997; Langklide and Alford, 2002; O’Neal et al., 2005; Pellet et al., 2006). Technical advances are still necessary to improve the perform- ances of these radars and to extend the field of investigations to new spe- cies and various ecological problems. Harmonic radars provide to date the best method to track very small animals (Chapman et al., 2011), but their use may require complex equipments that are not necessarily easy to implement under field conditions. Another major drawback of this sys- tem is that tracking simultaneously several individuals is difficult because signals from different individuals can not be differentiated.
3. 3. VHF radio tracking of small species Very high frequency (VHF, 30-300MHz) radio tracking systems consist of three parts: i) an emitter attached to an animal; ii) an antenna that picks up the signal sent by the emitter; iii) a receiver, to which the antenna is plugged, that decodes the signal to the operator. The antenna can be located on a car or hand held. The closer the orientation of the antenna is to the direction of the signal, the stronger the signal is received. The strength of the signal is used to calculate the location of the monitored animal using triangulation. VHF radio tracking (also called radio teleme- try) has been used to monitor large animals since the early 1960s (e.g. large carnivores or large mammals, Lemunyan et al., 1959), because the size of the first emitters precluded its use on smaller species. Miniaturisation efforts have then allowed the application of the system on smaller species, e.g. bats and birds. The smallest emitters currently available weigh circa 0.2g, allowing thus the monitoring of species as small as arthropods, small reptiles and amphibians (e.g. Naef-Daenzer et al., 2005, Rock and Cree, 2008). For example, recent advances in radio telemetry allowed tracking movements of a Neotropical orchid bee, which is a small insect pollinator (Wikelski et al., 2010, see figure 4). This technology provided new and valuable knowledge to understand the ecology and evolution of pollina- tion of orchids. However, the miniaturisation also comes at a cost since the duration of the miniaturised emitter is generally short (i.e. a few days to a few weeks for emitters < 1g), and the range of detectability of the emitter (i.e. the maximum distance at which the signal can be detected by the antenna- receptor system) is reduced to a few hundred meters for emitters < 1g compared to several kilometres for the biggest emitters assuming ground- to-ground conditions where the signal from an emitter on the ground picked up by a hand-held antenna. In the study by Wikelski et al. (2010) cited above, the lifespan of the emitters was around 10 days, which pre- cluded, for example, to draw firm conclusions on the size of the home Part I – Chapter 2 53 ranges of the bees. Yet, the monitoring of the signal of fast moving ani- mal can be improved by detecting the signal from the air with mobile antenna installed on e.g. an helicopter or with a fixed network of antenna installed on towers (Kays et al., 2011, Wikelski et al., 2010). For example, an automated radio telemetry system was built from receivers mounted on 40m towers topped with arrays of directional antennas and was later used to track the activity and location of radio-collared animals in a tropi- cal rainforest (Kays et al., 2011; IV, 2) This automated platform can be installed on a permanent study plot for long-term monitoring of medium and large sized animals, but the system encounters difficulties when it comes to detect activity and movements of the smallest species (around 4-100g., Kays et al., 2011). Thus, there is definitely room for technological improvements of radio tracking systems for most juveniles of the small species still cannot be fitted with emitters, and because the lifespan of the smallest emitters precludes any long-term monitoring of individuals and reduces long range detection.
Figure 4: VHF tracking of small species. The transmitter (300mg) is glued to the bee thorax. © C. Ziegler from Wikelski et al. (2010).
3. 4. Satellite-based tracking for small species Satellite-based telemetry like Argos system utilises a platform transmit- ter terminal (PTT) attached to the animal that transmits an ultra high frequency (UHF, 300-3000MHz) signal by pulses according to pre-pro- grammed time laps. This UHF signal can be detected by more than one of the satellites from the Argos network when these satellites pass over the tags. During this measurement window lasting a few minutes, the satel- lites can calculate the animal’s location based on the Doppler effect (i.e., the shift in pulse radio frequency due to the movement of the satellite 54 Ecophysiology and animal behaviour relative to the tag) and the satellite then relays this information to receiv- ing-interpreting sites located on the ground like the Argos system data processing centres. Satellites network from the Argos system allow theo- retically locating a PTT anywhere on the earth with an accuracy of about 150 meters (Wikelski et al., 2007; Bridge et al., 2011). For determination of more accurate locations down to a few meters, a GPS (global positioning system) can be installed on the animal. The GPS tag sends an UHF signal to a specific satellites network different from the Argos satellite constellation. These satellites return then the signal to the tag which calculates its own position by trilateration. The number of location records per unit time is pre-programmed by users and data are then sent to a satellite relay and available using an internet interface for example (e.g. the Argos network). This GPS technology is more precise than direct telemetry by the Argos system (less than 10m accuracy) and now Argos devices often integrate a GPS tag that sends locations within the Argos transmission. However, the GPS system coupled to the Argos relay is also more energy-consuming, especially for real-time location. Thus, GPS devices require powerful batteries and have a relatively shorter lifespan than PTT devices of the same weight. Currently, the smallest available Argos transmitter weights approximately 5g, while the smallest device for Argos satellite-relay GPS tracking weights 22g with a maximum lifespan of several months depending on the frequency of data sampling and retrieval (Guilford et al., 2011, see table 1). Therefore, the use of sat- ellite-based tracking technologies is still restricted to animals weighting more than 100g especially when it comes to research projects that require long-term tracking data (Wilkeski et al., 2007). To fulfill the lack of tech- nology applicable to smaller animals, the Icarus initiative, abbreviation of the International cooperation for animal research using space headed by Martin Wikelski at the Max Planck Institute in Germany, aims at estab- lishing a remote sensing platform for tracking over large spatial scales with transmitters as small as 1g (www.icarusinitiative.org). The project will require the deployment of low orbit altitude satellites to track the weak UHF signals of low weight transmitters located on the ground. A test of the method using an antenna attached to the International space station will start by 2014. An alternative to the very energy-consuming GPS tracking in real time is GPS data logging, where location data are stored locally on the tag attached to the animal. Several manufacturers develop GPS data loggers for wild- life tracking with a total weight starting at values as small as 2 grams and with an accuracy within 2.5m for on-ground measurements (table 1). The main limitation of these small GPS loggers is that the retrieval of the data requires recapturing the animals to download from the data logger. Thus, some of these devices also allow downloading the location data using Part I – Chapter 2 55 remote communication via satellites or via VHF, which brings the total weight of the equipment up to about respectively 15g and 30g. In addi- tion, solar panels can be used to power the GPS directly in direct sunlight and to charge the battery of the logger. This technology helps reduce the weight of the battery while at the same time increasing the overall lifetime possibly up to several years and increasing the capacity to collect more location data. Nevertheless, the solar panels also cause some overload and the smallest available devices of this type weigh approximately 5g, which is still too heavy to fit on numerous small animal species. Irrespective of the method used (satellite Argos PTT or GPS tracking), another important limiting factor of positioning systems using satellites is that marine species cannot be tracked during the diving phase because the UHF signal used for satellite communication propagates badly in the sea depths.
3. 5. Bio-logging for small species More complex environmental and physiological parameters can be collected with the help of various types of data loggers (see table 1). Autonomous bio-loggers like data storage tags and iButton® devices can be set up on some small animals because of their relatively low weight (around 2g), but the data from these loggers cannot be retrieved from a distance. Animals must therefore be recaptured to upload the data, which may be time-consuming and difficult to practice under field conditions. For example, the iButton thermal loggers were used to gather data on body temperatures of small reptiles in the laboratory because these log- gers are small and cheap (Lovegrove, 2009; Robert and Thompson, 2003). Similar thermal data can be obtained by using temperature-sensitive pas- sive integrated transponders. In addition to this, data storage tags offer the possibility to measure a great variety of parameters including light intensity, tilt, pressure, depth, magnetic field strength, pitch and roll, con- ductivity, dissolved oxygen, pH, electroencephalograms, electrocardio- grams, or electromyograms (Walsh and Morgan, 2004; Van der Kooij et al., 2007; Jonsson et al., 2010) with minimum weights starting at approxi- mately several grams. For example, a 3.4 g archival tag allows to measure both temperature and pressure during a minimum period of one year (table 1). However, more integrated and complex systems become inevita- bly heavier. For example, a neurologger used in combinaison with a GPS has been developed by Vyssotski et al. (2008) to analyse the neuronal activity of navigating homing pigeons and weights 35 grams. Data loggers combined with GPS, VHF or satellite transmitters are essen- tial to our understanding of the behaviour of large animals (see I, 1) since these technologies offer the possibility to correlate location or movement with various bio-physical and physiological parameters (Ropert-Coudert 56 Ecophysiology and animal behaviour and Wilson, 2005). Yet, these loggers can be difficult to mount on small animals due to their large volume and their weight typically larger than 10-20g, which is the minimum weight for a current GPS data logger (see section 3.4 above). When the sole focus is on location, geolocators (or GLS logging) can provide an alternative lightweight technology for track- ing individual movements on large spatial scales. The method consists to measure ambient light level during the day from which sunset and sun- rise times are estimated from thresholds in light curves. The latitude and longitude are then derived from characteristics of daytime light cycles. Archival GLS loggers are now as small as 0.5g and can be fitted on small animal species, but the main limitations against the deployment of this technique are that i) animals must be recaptured to retrieve the data and ii) the accuracy of the measurements is rarely better than 100 km (Phillips et al., 2004). Nevertheless, this technology is adequate for tracking sea birds, migratory passerines or pelagic species (e.g. marine mammals, pen- guins) in order to determine long-distance movements, breeding season foraging ranges or broad-scale habitat preference from several months to multiple years. In addition to this range of bio-loggers, animal-borne video and environ- mental data collection systems (AVEDs) have been used for a long time to study large marine or terrestrial species (Moll et al., 2007), and have been recently upgraded to be suitable for a wider range of smaller species (Rutz and Bluff, 2008). For example, a device weighting about 14g and adapted to crows seems promising to collect animals’ eye view of resource use and social interactions along a known movement trajectory. Unfortunately, the system is still too heavy to fit on many small species and its current recording capacity does not exceed 1 hour. Advances are obviously neces- sary to improve the performances of this technology and to extend the field of investigations to smaller model species.
4. Designing the future generation of sensors for small animals
Our comparative analysis of available sensors highlights that current sys- tems present major limits. The first one is that even basic sensors constitute a real burden for the small animals that have to carry them. Therefore, the main technical challenge is the miniaturisation of the components. An extra challenge is that this miniaturisation must be done without too much loss of capacity in the number of measured parameters, autonomy, or detectability. Another necessary improvement is to develop compatible fixation procedures. For large animals, various techniques (including glues, harnesses, subcutaneous or intraperitoneal implants, or ingestible tags) have Part I – Chapter 2 57 been experimented to attach the animal-borne device with a minimum hin- drance. However, ecologists who work on small species must often develop their own attachment techniques and materials since few standard are avail- able or adapted to their models, and this process generally adds annoyances to a burden yet significant. The miniaturisation of sensors and associated equipments could limit this problem; however, technological advancements for harmless and weakly invasive fixation techniques are necessary. In addi- tion, to address further questions in the field of animal ecology, we may wish to obtain novel kinds of information. This could be done by combin- ing the “classical” data obtained with standard animal-borne sensors with new data obtained from other environmental sensors. In particular, a wide range of techniques are now available to assess the thermal environments (thermal imagery techniques, Lavers et al., 2005; Hristov et al., 2008), the weight of individuals (for automatic measurement of growth, body mass regulation and feeding activity, Rands et al., 2006), or the acoustic environ- ment (for characterisation of behaviour and social interactions, see chapter by Huetz and Aubin). With such an integration in mind, the next gen- eration of sensors for small animals will therefore allow investigating more accurately the internal and external factors influencing the responses of organisms to environmental variations.
Authors’ references
Olivier Guillaume, Jean Clobert: Station d’Écologie Expérimentale du CNRS à Moulis, CNRS USR 2936, Moulis, Saint-Girons, France
Aurélie Coulon: Muséum National d’Histoire Naturelle, Département Écologie et Gestion de la Biodiversité, Unité Conservation des Espèces, Restauration et Suivi des Populations (CERSP), MNHN-CNRS UMR 7204, Brunoy, France
Jean-François Le Galliard: Université Pierre et Marie Curie, Laboratoire Écologie et Évolution, CNRS UMR 7625 Paris, France École Normale Supérieure, Centre de recherche en écologie expérimen- tale et prédictive (Cereep) – Ecotron Ile de France, CNRS UMS 3194, St-Pierre-Lès-Nemours, France
Corresponding author: Olivier Guillaume, olivier.guillaume@EcoEx- moulis.cnrs.fr 58 Ecophysiology and animal behaviour
Acknowledgement
Jean-François Le Galliard acknowledges the support of the ANR grant Extinction (07-JCJC-0120) and a Marie Curie Fellowship.
References
Bedrosian B., Craighead D., 2007. Evaluation of techniques for attaching transmitters to common raven nestlings. Northwestern Naturalist, 88, pp. 1-6. Bergman P. K., Haw F., Blankenship H. L., Buckley R. M., 1992. Perspectives on design, use and misuse of fish tags. Fisheries, 17, pp. 20-25. Blackburn T. M., Gaston K. J., 1994. The distribution of body sizes of the world’s bird species. Oikos, 70, pp. 127-130. Bolnick D. I., Svanback R., Fordyce J. A., Yang L. H., Davis J. M., Hulsey C. D., Forister M. L., 2003. The ecology of individuals: incidence and implications of individual specialization. American naturalist, 161, pp. 1-28. Bridge E. S., Thorup K., bowlin M. S., Chilson P. b., Diehl R. H., Fléron R. W., Hartl P., Kays R., Kelly J. F., Robinson W. D., Wilkeski M., 2011. Technology on the move: recent and forthcoming innovations for tracking migratory birds. BioScience, 61, pp. 689-698. Bridger C. J., Booth R. K., 2003. The effects of biotelemetry transmitter presence and attachment procedures on fish physiology and behaviour. Review in Fisheries sciences, 11, pp. 13-34. Canner J. E., Spence M. 2011. A new technique using metal tags to track small seeds over short distances. Ecological Research, 26, pp. 233-236. Cant E. T., Smith A. D., Reynolds D. R., Osborne J. L. 2005. Tracking butterfly flights with harmonic radar across the landscape. Proceedings of the Royal society B, 272, 785-790. Chapman J. W., Drake V. A. Reynolds D. R., 2011. Recent insights from radar studies of insect flight. Annual Review of Entomology, 56, pp. 337-356. Clobert J., Danchin E., Dhondt A. A., Nichols J. D., 2001. Dispersal. Oxford University Press, Oxford. Cochran W. W., 1980. Wildlife telemetry, in: Schemnitz S.D. (Ed.), Wildlife management techniques manual, 4th ed., Wildlife Society, Washington DC, pp. 507-520. Courtney D. L., Mortensen D. G., Orsi J. A., Munk K. M., 2000. Origin of juvenile Pacific salmon recovered from coastal southeastern Alaska identified by otolith thermal marks and coded wire tags. Fisheries Research, 46, pp. 267-278. Gardezi T., Da Silva J., 1999. Diversity in relation to body size in mammals: A comparative study. American Naturalist, 153, pp. 110-123. Part I – Chapter 2 59
Gibbons J.W., Andrews K. M., 2004. PIT tagging: simple technology at its best. Bioscience, 54, pp. 447-454. Guilford T., Äkesson S., Gagliardo A, Holland R. A., Mouritsen H., Muheim R., Wiltschko R., Wiltschko W., Bingman V. P., 2011. Migratory navigation in birds: new opportunities in an era of fast-developing tracking technology. Journal of Experimental Biology, 214, pp. 3705-3712. Hoset K. S., Le Galliard J.-F., Gundersen G., Steen H., 2008. Home range size and overlap in female root voles: effects of season and density. Behavioural Ecology, 19, pp. 139-145. Hristov N. I., Betke M., Kuntz T. H., 2008. Applications of thermal infrared imaging for research in aeroecology. Integrative and Comparative Biology, 48, pp. 50-59. Jonsson G. K., Thorsteinsson V., Tomasson G. G., 2010. Identification of vertical and horizontal movement patterns in cod behavior in Spink A.J., Grieco F., Krips O.E., Loijens L.W.S., Noldus L.P.J.J. , Zimmerman P.H. (Eds), Proceedings of Measuring Behavior 2010, 7th International Conference on Methods and Techniques in Behavioral Research, Eindhoven, pp. 303-305. Kays R., Tilak S., Crofoot M., Fountain T., Obando D., Ortega A., Kuemmeth F., Mandel J., Swenson G., Lambert T., Hirsch B., Wikelski M., 2011. Tracking animal location and activity with an automated radio telemetry system in a tropical rainforest. The Computer Journal, 54, pp. 1-17 Korslund L., Steen H., 2006. Small rodent winter survival: snow conditions limit access to food resources. Journal of Animal ecology, 75, pp. 156-166. Langkilde T., Alford R., 2002. The tail wags the frog: harmonic radar transponders affect movement behaviour in Litoria lesueuri. Journal of Herpetology, 36, pp. 711-715. Lavers C., Franks K., Floyd M., Plowman A., 2005. Application of remote thermal imaging and night vision technology to improve endangered wildlife resource management with minimal animal distress and hazard to humans. Journal of Physics: Conference Series, 15, pp. 207-212. Le Galliard J.-F., Gundersen, G., Steen H., 2007. Mother-offspring interactions do not affect natal dispersal in a small rodent. Behavioural Ecology, 18, pp. 665-673. Le Galliard J.-F., Paquet M., Pantelic Z., Perret S., 2011. Effects of miniature transponders on physiological stress, locomotor activity, growth and survival in small lizards. Amphibia-Reptilia, 32, pp. 177-183. Lemunyan C. D., White W. M., Nyberg E., Christian J. J., 1959. Design of a miniature radio transmitter for use in animal studies. Journal of Wildlife Management, 23, pp. 107-110. Lovegrove b. G., 2009. Modification and miniaturization of thermochron Ibuttons for surgical implantation into small animals. Journal of Comparative Physiology B, 179, pp. 451-458. Lovëi G. L., Stringer I. A. N., Devine C. D., Cartellieri M., 1997. Harmonic radar – A method using inexpensive tags to study invertebrate movement on land. New Zealand Journal of Ecology, 21, pp. 187-193. 60 Ecophysiology and animal behaviour
Mascanzoni D., Wallin H., 1986. The harmonic radar: a new method of tracing insects in the field. Ecological Entomology, 11, pp. 387-390. Moll R. J., Millspaugh J. J., Beringer J., Sartwell J., He Z., 2007. A new ‘view’ of ecology and conservation through animal-borne video systems. Trends in Ecology and Evolution, 22, pp. 660-668. Naef-Daenzer B., Früh D., Stalder M., Welti P., Weise E., 2005. Miniaturization (0.2g) and evaluation of attachment techniques of telemetry transmitters. Journal of Experimental biology, 208, pp4063-4068. O’Neal M. E., Landis D. A., Rothwell E., Kempel L., Reihard D., 2005. Tracking insects with harmonic radar: a case study. American Entomologist, 50, pp. 212-218. Ovaskainen O., Smith A. D., Osborne J. L., Reynolds, D. R., Carreck, N. L., Martin A. P., Niitepold K., Hanski I., 2008. Tracking butterfly movements with harmonic radr reveals an effect of population age on movement distance. Proceedings of the National academy of sciences of the USA, 49, pp. 19090-19095. Pellet J., Reichsteiner L., Skrivervik A., Zurcher J.-F., Perrin N., 2006. Use of the harmonic radar direction finder to study the terrestrial habitats of the european tree frog (Hyla arborea). Amphibia-Reptilia, 27, pp. 138-142. Phillips R. A., Silk J. R. D., Croxall, J. P., Afanasyev, V., Briggs, D. R., 2004. Accuracy of geolocation estimates for flying seabirds. Marine Ecology Progress Series, 266, pp. 265-272. Rands S. A., Houston A. I., Cuthill I. C., 2006. Measurement of mass change in breeding birds: a bibliography and discussion of measurement techniques. Ecology, 23, pp. 1-5. Riley J. R., Chapman J. W., Reynolds D. R., Smith A. D., 2007. Recent applications of radar to entomology. Outlooks on Pest Management, 18, pp. 62-68. Riley J. R., Smith A. D., 2002. Design considerations for an harmonic radar to investigate the flight of insects at low altitude. Computers and Electronics in Agriculture, 35, pp. 151-169. Robert K.A., Thompson M.B., 2003. Reconstructing thermochron ibuttons to reduce size and weight as a new technique in the study of small animal thermal biology. Herpetological Review, 34, pp. 130-132. Rock J., Cree A., 2008. Extreme variation in body temperature in a nocturnal thigmothermic lizard. Herpetological Journal, 18, pp. 69-76. Ropert-Coudert y., Wilson R. P., 2005. Trends and perspectives in animal- attached remote sensing. Frontiers in Ecology and the Environment, 3, pp. 437-444. Rutz C. G., bluff L. A., 2008. Animal-borne imaging takes wing, or the dawn of ‘Wildlife video-tracking’. Trends in Ecology and Evolution, 23, pp. 292-294. Van der Kooij J., Righton D., Strand E., Michalsen K., Thorsteinsson V., Svedäng H., Neat F.C., Neuenfeldt S., 2007. Life under pressure: insights Part I – Chapter 2 61
from electronic data-storage tags into cod swimbladder function. ICES Journal of Marine Science, 64, pp. 1293-1301. Vyssotski A. L., Serkov A. N., Itskov P. M., Dell’Omo G., Latanov A. V., Wolfer D. P.,Lipp. H. P., 2006. Miniature neurologgers for flying Pigeons: multichannel EEG and Action and field potentials in combination with GPS Recording. Journal of Neurophysiology, 95, pp. 1263-1273. Walsh S. J., Morgan M. J., 2004. Observations of natural behaviour of yellowtail flounder derived from data storage tags. ICES Journal of Marine science, 61, pp. 1151-1156. Weatherhead P. J., Blouin-Demers G., 2004. Long-term effect of radiotelemetry on black ratsnakes. Wildlife Society Bulletinn, 32, pp. 900-906. Whidden S. E., Williams C. T., breton A. R., buck C. L., 2007. Effects of transmitters on the reproductive success of Tufted Puffins. Journal of Field Ornithology, 78, pp. 206-212. Wikelski, M., Kays, R.W., Kasdin, N.J., Thorup, K., Smith, J.A., Swenson, G.W., 2007. Going wild: what a global small-animal tracking system could do for experimental biologists. Journal of Experimental Biology 210 (2), 181-186. Wikelski M., Moxley J., Eaton-Mordas A., López-Uribe M. M., Holland R., Moskowitz D., Roubik D. W., Kays R., 2010. Large-range movements of neotropical orchid bees observed via radio telemetry. PLoS One, 5, pp. 6.
Chapter 3
Passive hydro-acoustics for cetacean census and localisation
Flore Samaran, Nadège Gandilhon, Rocio Prieto Gonzalez, Federica Pace, Amy Kennedy, and Olivier Adam
1. Introduction
Scientific observations of cetacean species (whales, dolphins, and relatives) are nowadays intensively collected by biologists for several purposes: identi- fying the species to monitor their population, locating individuals to collect data on the abundance and seasonal distribution of each species, obtain- ing detailed information about individuals (e.g. individual behaviour, food and health) and describing interactions among individuals (social group relationships, cross-species interaction). These observations are critical to measure the impacts of anthropogenic activities (e.g. underwater noise, ship strikes, fishery tools and bycatch) on populations and communities of cetaceans. Recently, Simmonds and Isaac (2007) and Simmonds and Eliott (2009) suggested to use whale observations to include relevant indi- cators in prediction models that describe the global warming of the planet. Odontocetes are top predators and mysticetes eat krill. Global warming will impact their alimentary resources since we know, for example, that krill mass is decreasing in the Austral Ocean. Moreover, some species use hot-spots for foraging and breeding. Observations of changes in the use of these hot-spots by cetaceans will give further insights into changes of their environment. 64 Ecophysiology and animal behaviour
Figure 1: Common observation methods for cetaceans. These methods include visual approaches by human observers, animal-borne sensors (I, 1), genetic methods and acoustic tools. The methods allow to obtain demographic, behavioural and genetic information. © Gandilhon/Breach, © Kennedy/NMML.
There are several methods available to observe and collect scientific data on cetaceans (figure 1). A classical and often-used method is based on visual observations, when cetaceans are identified by sight from a distance by a trained human observer. This technique is simple and trained observers can easily identify species, count the number of individuals or even identify indi- viduals within a group using photo-identification techniques. Behavioural observations can also be conducted with this technique. However, this approach has also some drawbacks, such as the dependence on weather and availability of observation sites (shore, boat, etc), the impossibility to col- lect data at night, and the potential bias from the observers. Another and more advanced solution to observe and collect scientific data on cetaceans is to use automatic animal-borne sensors. This approach is newer and faces numerous technical challenges, including, for example, the need for sensors able to resist pressure differentials due to the water depth, and autonomous enough in terms of power supply and data storage (see chapter I, 1). We describe the principles and challenges of advanced detection methods based on passive acoustics monitoring (PAM). PAM consists in passively listening to the sounds emitted by animals. It is an attractive alternative to classical visual observations because cetaceans are vocally active and their sound can travel far underwater (Samaran et al., 2010b). In addition, species Part I – Chapter 3 65 of cetaceans can be detected remotely, according to the main features of their sounds (acoustic intensity, bandwidth), the acoustic propagation and the ambient noise level. Note that the extraction of individual acoustic sig- nature is still a scientific challenge. The first step with acoustic methods consists in choosing the right electronic instrumentation to make record- ings (hydrophone, amplifier, filters, data acquisition board, data transmis- sion, storage unit). The following steps are dedicated to signal processing and pattern recognition to detect cetacean sounds in the recordings, to identify, date, and localise species observations, and estimate the number of individuals. These acoustic events can then be used by biologists to map the presence of even distribution (locations). The diversity of sounds produced by cetaceans is overviewed in the first part of this chapter. Then, a detailed account of the technological solutions to record and to analyse these sounds for the purpose of ecological studies of cetaceans is presented. Lastly, the application of these techniques is illustrated with case studies taken from our current projects.
2. Acoustic signatures of cetaceans’ species
Cetaceans are marine mammals that include 83 species from two major taxo- nomic units named Mysticetes (cetaceans without teeth) and Odontocetes (cetaceans with teeth). Cetaceans are present in all oceans, and their con- servation statuses are different according to the species (http://www.iucn- redlist.org/). The sound repertoire of different cetacean species may vary from highly stereotyped repetitive sounds like the ones emitted by fin whales (Balaenoptera physalus) and the monotonous clicks of sperm whales (Physeter macrocephalus) to more complex communication calls, like those of bottlenose dolphins (Tursiop truncatus), killer whales (Orcinus orca) and hump- back whales (Megaptera novaeangliae). A summary of the frequency range and source levels of sounds emitted by common cetacean species is given in table 1. These sounds are hypothesised to play a crucial role for communica- tion among cetaceans, including group cohesion, searching for food, strategy for eating, mother-calf contact, individual recognition (acoustic signature), hunting, socialising, looking for a partner, delineating a territory, establishing a hierarchy, detecting predators and dangers, and orientation (Frankel, 1998). The intrinsic sound characteristics are non-linearly distorted from the acoustic propagation. Acoustic sound attenuation depends on distance between the cetacean source and the hydrophone and is also proportional to the frequency squared. Leroy’s equation gives the attenuation coefficient α (dB/km) depending on the frequency: 2 = 3 2 + fr f , ()f 6.10 f 0.16 2 2 fr + f 66 Ecophysiology and animal behaviour where f (kHz) is the frequency of the acoustic wave and fr is the relaxa- tion frequency of the boric acid B(OH)3 present in the sea water. The fr value depends on the location and the empirical value is 1.6 kHz in the Mediterranean Sea for example. In addition, acoustic waves can be reflected at the sea surface and bot- tom, where the level of absorption depends on the nature of the seabed. Furthermore, the propagation speed of the sound in seawater is not con- stant: it depends importantly on salinity, pressure and temperature. All these parameters can be included in a mathematical model to describe acoustic propagation, based on the Helmholtz equation, where P is the pressure, and c 0 is the propagation speed of the sound in the seawater: 2 2 1 P = . P 2 2 0 c0 t For harmonic solution where P = pe i t and ω is the angular frequency (ω=2πF with F the frequency), the Helmholtz equation writes like: 2 2 p + k 2 p = 0, with the wavenumber k = = . c To solve this equation, different mathematic models and associated soft- wares could be applied including ray tracing (e.g. Bellhop software), nor- mal modes models (e.g. Kraken software), the parabolic equation (e.g. RAM software) or the wavenumber integration (e.g. oases software). All of these can be downloaded on the Ocean acoustic library website (http:// oalib.hlsresearch.com).
Table 1: Features of sounds emitted by some common cetaceans’ species (adapted from Simmonds et al., 2004). To compare, a large tanker generates low frequency noise (<500Hz, 186db re 1µPa at 1m).
Broadband Frequency range Source level (kHz) (dB re 1µPa at 1m) Sperm whale clicks 163-223 [0.100 – 30] Spinner dolhin bursts 108-115 Bottlenose dolphins whistles 125-173 [0.8 – 24] Bottlenose dolphins clicks 212-228 [110 – 130] Risso’s dolphin 120 65 Fin whale moans 155-186 [0.03 – 0.75] Blue whale moans 155-188 Humpback whale song 144-174 [0.03 – 8] Humpback whale fluke 183-192 [0.03 – 8] and flipper slap Pilot whale 180 [0.5 – 20] Snapping shrimp 183-189 Part I – Chapter 3 67
The analysis of these sounds is difficult because of the presence of noise and non-linear distortion. Usually, different approaches are used to ana- lyse vocalisations or short transient sounds (e.g. odontocete clicks, right whale gunshots). The first step consists in using the bandpass filter corres- ponding to the species studied, and possibly applying an enhancement of the signal if the amplifier gain is too low for the recordings. To detect the emitted sounds, one method is to set a threshold on the temporal energy signal. The use of the Teager-Kaiser operator represents an alternative that takes into account the fact that acoustic signals vary within the same spe- cies or individual, where signal variation is given by equation: ( s ) = s2 ()n sn()1 sn()+1 , where s(i) is the ith sample of the recorded signal (voltage corresponding to the acoustic pressure on the hydrophone, V/dB unit). The main advan- tages of this non-linear operator (in addition to its easy implementation) is that only three successive samples are needed to calculate Ψ. This opera- tor, taking into account the signal variations, is used to track the instan- taneous energy of the signal (for example for amplitude modulation and frequency modulation). Parametric models as autoregressive (AR) or/and mean average (MA) models are also a solution, especially for the detection of vocalisations or non-stationarity in the acoustic recordings. To analyse cetacean sounds, specialists in marine biology make extensive use of the Fourier transform or the spectrogram for the time evolution of frequency variations. More up-to-date methods exist such as the wavelet transform or the Hilbert Huang transform (Adam, 2008).
3. Acoustic observatories
Passive acoustics is based on the use of at least one hydrophone to record underwater sounds during a given time. These recordings typically include a large variety of sounds such as natural sounds (waves, rain, wind…), bio- logical sounds (fish, shrimp, coral reef…) or sounds from human activities (sonar, airgun, marine traffic…). The choice of the recording equipment depends on several factors including the acoustic intensity and frequency bandwidth of sounds emitted by the cetacean species under investiga- tion, the bathymetry of the area where it is deployed, and the ambient noise level, including those resulting from human activities. The choice of hydrophone also depends on the objectives of the study including the willingness to detect one or more specific cetacean species. If the target of the study is to detect cetaceans’ vocalisations over very long distances, the device will be chosen to maximise the amplitude of the signal up to 68 Ecophysiology and animal behaviour the saturating level that may result from marine noise and the minimum amplitude of vocalisations that are to be detected. The solution is to record continuously and keep the whole signal in memory to provide an a posteriori analysis. This recordings can be continuous time or during a specific period every day to optimise the memory size and the power consuming. The other solution is to process cetacean sounds immediately in situ, such that only time corresponding to cetaceans’ presence is stored in memory. This approach can save substantial energy and memory space.
3.1 Instantaneous acoustic observatories Here, the objective is to make recordings from a ship. Therefore, the acoustic system should be light and easily manoeuvrable. Typically, the equipments required are one or more hydrophones, their amplifier, the digital recorder or the data acquisition board and a computer. It is possible to buy this material for 6,000 to 7,000€ for a single hydrophone (sensi- tivity -170dB re 1V/uPa, [10-90kHz] omnidirectionnal). Digital record- ers can now record at 192kHz on 24 bits and it is possible to use data acquisition cards with higher sampling rates and the possibility of differ- ent synchronised channels. Instantaneous acoustic observatories are usu- ally used for opportunistic detections of cetaceans in a specific area. This approach may be supplemented by visual observations. The objective is to give information about the cetaceans’ distribution and localise potential hot-spots in a local region. We used this approach in Guadeloupe (French West Indies) to confirm the presence of elusive species like beaked whales, and in Madagascar to record different humpback whale singers.
3.2 Semi-permanent and permanent acoustic stations Semi-permanent or permanent acoustic stations are used for the purpose of monitoring a specific area (hot-spot of whale activity, a channel, a strait, or a harbour…). This system can be deployed with a buoy at the sea surface or can be anchored on the seabed. A good example of the first case is a new prototype that we deployed in Guadeloupe (French West Indies) around the end of 2010 (figure 2). This system was built by CeSigma (www.cesigma.com) and PLK Marine (www.plkmarine.com), and was implemented in a Fish Aggregating Device (FAD) in agreement with the fisheries committee of Guadeloupe. Technical features were a sampling frequency of 200kHz, samples coded on 24 bits, 4 channels, and 700Gb of storage, a Wifi transmission capacity, and continuous recordings (Gandilhon et al., 2010). This so-called “sonobuoy” is used to observe different species including sperm whales, humpback whales, bot- tlenose dolphins, spotted dolphins, rough-toothed dolphins. Part I – Chapter 3 69
Figure 2: Design of the acoustic system developed by CeSigma and PLK Marine for the scientific program Gualiba I coordinated by the team “Dynamique des écosystèmes Caraïbes” de l’Université des Antilles et de la Guyane and the Centre de neurosciences Paris Sud of University Paris Sud orsay. The sonobuoy consists of an autonomous recording and transmission device connected to a hydrophone recording ocean sounds continuously.
Another device is the autonomous underwater acoustic recorder for lis- tening, manufactured and distributed by Multi-Electronique (www.multi- electronique.com). The entire system is based on a hydrophone, a data acquisition card, an external hard drive and power batteries. An acoustic latch releases the material from its anchor, back to the surface by a buoy. This type of material was used to monitor populations of great whales and some odontocetes for a year in the Mozambique Channel and on the south of Tromelin Island in the Indian Ocean. The buoy was deployed during the mission Eparses 2009 funded by the Marine protected areas, Terres australes et antarctiques françaises (Taaf), Centre d’études biologi- cal Chizé (CEbC-CNRS) and the Laboratoire domaines océaniques de l’université de Bretagne occidentale (LDO-UBO). The analysis was done a posteriori to automatically detect the cetacean sounds and extract the presence rate during the different months of the year. Other instruments of the same type have been employed since 2007 in collaboration with 70 Ecophysiology and animal behaviour the Pacific marine environment laboratory (NoAA), CNRS-CEbC, and the LDO-UBO to monitor large whales as part of the Subantarctic Indian ocean Program (see figure 3).
Figure 3: Design of the autonomous hydrophone of the Pacific marine environment laboratory (NoAA) and used for the scientific mission DEFLo-HyDRo. The sample frequency of the hydrophone is 250Hz and the data is coded 16 bits with a storage capacity of 80Go. The autonomy can be chosen from 18 to 24 months depending on if the recordings are continuous or during a short specific period of the day (10 min recordings every 6 hours for example). © NoAA/PML Vents program/Acoustics group (http://www.pmel.noaa.gov/vents/multimedia.html).
Regarding permanent observatories, marine ecologists using acoustic methods have exploited permanent infrastructures from other disciplines, like those of geophysics or physical oceanography. For example, in Europe, the ESONET project (European Seafloor Observatories NETwork, www. esonet-emso.org) brings together specialists in geophysics, chemistry, biochemistry, oceanography, marine biology and fisheries. Different underwater observatories are distributed all around the European coasts. In 2007, in collaboration with Prof. H. Glotin (LSIS, www.lsis.org), we proposed to add the marine mammal observation task to the neutrino detector installed on the ANTARES observatory (http://antares.in2p3. fr) deployed in the Mediterranean Sea (Hyères, France). The advantage of this system is that it provides permanent acoustic recordings throughout the whole year. The objective is to give an indication of the presence of sperm whales and fin whales even during winter, when weather conditions are not optimal for visual observations. Passive acoustics, in this case, is a unique solution to provide this kind of information of cetacean presence. There are other permanent observatories using acoustic detectors, such as the Mars ocean observatory testbed in Monterrey Bay (www.mbari.org/ mars) and the Neptune project in the Northeast Pacific ocean (www. neptune.washington.edu). The modules of these infrastructures are placed on the seabed, and power supply and data can be transmitted by a cable connected to facilities on shore. These projects have a section dedicated to marine biology and especially to cetacean observations. Part I – Chapter 3 71
4. Case studies
We worked on several cetacean species and the diversity of sounds that we dealt with explains the different methods we investigated.
4.1 Analysis of Sperm whale clicks Sperm whales emit two major types of clicks: regular clicks characterised by high intensities and inter-click intervals greater than 0.5ms, and creaks that are successions of clicks of variable amplitude and spacing of less than 0.5sec. The bio-mechanic of click generation is pretty well described in the literature by the model first introduced by Mohl et al. (2003) and subsequently modified by Laplanche et al. (2006). However, automatic detection of clicks can be challenging because the inter-click interval fluc- tuates over time, and the signal-to-noise ratio varies considerably accord- ing to the ambient noise and the position of the whale relative to the recording device (figure 4). Therefore, one needs to trade detection capac- ity off the risks of false alarms when setting and adjusting the detector.
Figure 4: Sperm whale clicks with underwater noise (upper panel) detected by using the Teager-Kaiser operator described previously (lower panel). Sperm whale clicks were recorded off the coast of Toulon (France) in August 2004. one sperm whale was presented the data shows regular clicks. 72 Ecophysiology and animal behaviour
Figure 5: Data retrieved from the acoustic sensing of Sperm whale clicks. A. Estimation of the Sperm whale length by measurement of the delay between the first pulse and the second pulse. The y-axis measures the relative amplitude of the pulse (normalized correlation between the original pulse and its reflection) and the z-axis corresponds to the number of successives clicks (here, 50 clicks). Sperm whale clicks are multipulsed signals. The first pulse comes from the “monkey lips” close to the blowhole and the other pulses come from reflections from the distal sac Part I – Chapter 3 73 and the frontal sac respectively located at the beginning and the end of the head. We can measure the time between 2 successive pulses: this delay is the time it takes for the acoustic wave to cross the head. From the known celerity of the acoustic wave in the head, this allows to estimate the head length and therefore the body length (Lopakta et al., 2006). B. Classification of Sperm whale clicks. Clicks were classified with the use of three parameters calculated after the Schur coefficients (for details, see Lopatka et al., 2006). Black circles: sperm whale clicks. Red circles: clicks from striped dolphins. C. This figure describes the different steps during the Sperm whale dives. Step 1, Sperm whale emits regular clicks at time intervals greater than 0.5sec to detect the presence of preys. Step 2, Sperm whale emits buzz at time intervals less than 0.5sec to get a better resolution of the “acoustic image” of this volume. Step 3, Sperm whale stops emitting clicks at the end of this part of the dives and will go back to step A for another prey research.
Sperm whale clicks have high frequency components (greater than 5kHz), so it is possible to use a high pass filter cutoff frequency exceeding 1kHz or more to overcome the ambient noise mostly due to the presence of maritime traffic. We tested several approaches to detect sperm whale clicks: Teager-Kaiser operator (see figure 4), autoregressive and moving average models, Schur algorithm, spectrogram and wavelet decomposi- tion (Adam et al., 2005; Lopatka et al., 2005). The goals were firstly to minimise the false alarm rate, and secondly to obtain accurate estimates of the time of occurrence of clicks in order to localise the individuals by triangulation. We found that sperm whale clicks were characterised the best by reflection coefficients of the Schur model (Lopatka et al., 2006). By using the Schur model, it was then possible to characterise the mor- phology of individuals (the size of the sperm whale can be deduced from the inter-pulse), distinguish clicks of sperm whales from other transient sounds, and locate the individual with precision enough to rebuild its dive profiles (figure 5).
4.2 Detection and localisation of blue whales We have also used passive acoustic monitoring to study Antarctic blue whale populations in the Southern Ocean. Most of the year, Antarctic blue whale emits a low frequency call (from 28Hz to 20Hz, see figure 6A) that lasts for 15-20 sec with high intensity every minute. Algorithms for automatic whale call detection, extraction and discrimination were developed and used on a one-year continuous acoustic dataset (2003- 2004) recorded in the station located in sub Antarctic area near Crozet Islands under the framework of the International monitoring system of the comprehensive-test-ban treaty organisation (IMS-CTBTO). All data are available under contract with the Direction des applications militaires du commissariat à l’énergie atomique (CEA-DAM). The aim was to assess 74 Ecophysiology and animal behaviour
Figure 6: Acoustic study of Antarctic blue whale. A. Two successive Antarctic blue whale calls. Calls are stereotypical sounds, with 2 main frequencies at 28Hz and 20Hz. Antarctic blue whales emit these calls between 2 breathing phase. From this specific time-frequency pattern, these calls can be automatically detected. b. Number of Antarctic blue whale calls detected from the recordings of one hydrophone deployed in the North of Crozet Island from May 2003 to April 2004. the seasonal occurrence of blue whale in a specific area. The detection procedure was based on a matched filter model (Samaran et al., 2008). More than 170,000 blue whale calls were detected all year-round indicat- ing their continuous presence in the region (figure 6B). Results revealed the seasonal occurrence and migration patterns of blue whales, providing information about ecology and habitats in this former commercial whal- ing area (Samaran et al., 2010c). A mathematical model RAM (Range-dependent Acoustic Model) was used to predict how sound levels changed with distance between vocalising whales and IMS receivers. This approach allowed estimating the size of the monitored area, which was estimated to be a radius of 200km. The distri- Part I – Chapter 3 75 bution of the estimated distances confirmed the presence of whales close to the Crozet Islands, showing the importance of this sub-Antarctic area for these endangered species especially during the austral summer feeding season (Samaran et al., 2010b). In addition, the triangular configuration of the calibrated hydrophones of the station allowed localising and tracking calling whales from observed differences in arrival times of the same sig- nals at the three hydrophones. We could therefore estimate the movement and detection range between the recording system and the animals, which are critical data to understand the habitat of calling whales without human disturbance. The sound levels of received calls may also be used to estimate the level of sound emitted by the vocalising whales (Samaran et al., 2010a). The last objective of our analysis was to estimate the total number of vocalising whales. This task was not trivial and could be conducted along two methodological approaches. First, we could search for individual acoustic signature. Unfortunately, this is difficult for cetaceans in their natural environment, especially when they are far from the hydrophone, because of the non-linear distortion of sound due to the acoustic propaga- tion (Musikas et al., 2009). Second, we could use the distance sampling method. This method is well-known for a wide range of applications with visual observations of wildlife, but it is particularly difficult to apply when little or no information is available about the call rate (Marques et al., 2009). Therefore, we suggested a third method based on a joint estimation of the number of calls emitted by each individual and the number of indi- viduals in a specific area (Valsero et al., 2010). The two estimates are given by their probability functions, assumed to follow a Poisson distribution:
k s ()s PB()()s = k = e k! k μt ()μt PC()()t = k = e k! where B(s) is the number of whales in the area s (around the hydrophone) and C(t) the number of detected calls during time t. The estimated number of individuals in the study area at a time t can then be obtained by the method of moments using the maximum likelihood estimation (Valsero et al., 2010): 2 N μˆ = N N 2 ˆ = N 2 N s ()N 76 Ecophysiology and animal behaviour where N is the number of detected calls in the area s during t, N is the 2 mean and N is the variance. The method gives estimates of the number of individuals present in the Crozet Archipelago during the year between 0 and 4 individuals (95% confidence interval) based on the distribution of the time between successive calls and between 2 and 12 individuals based on the number of calls. This type of results is only possible with passive acoustics because visual observations cannot be conducted throughout the whole year in this area, highlighting the importance of having a per- manent recording system installed there.
Table 2: Estimation of the number [min, max] of whale individuals around the Crozet Archipelagos by 2 distributions based on the distribution 1) of the time between successive calls and 2) of the number of calls in a specific area (Valsero et al., 2010).
Confidence interval (1) Confidence interval (2) Month 90% 95% 90% 95% May [1, 8] [1, 8] [3, 12] [2, 13] June [1, 8] [1, 9] [3, 12] [2, 13] July [1, 7] [0, 7] [2, 10] [2, 11] August [1, 7] [0, 7] [3, 11] [2, 12] September [0, 6] [0, 6] [3, 11] [2, 12] October [1, 6] [0, 7] [3, 10] [2, 11] November [1, 7] [0, 8] [3, 11] [2, 12] December [0, 6] [0, 7] [3, 11] [2, 12] January [0, 5] [0, 6] [3, 11] [2, 12] February [1, 7] [1, 8] [2, 10] [2, 11] March [0, 3] [0, 3] [2, 10] [2, 11] Avril [0, 6] [0, 7] [2, 10] [2, 11] Mean [0, 3] [0, 4] [3, 11] [2, 12]
4.3 Analysis of Humpback whale songs During the breeding season, Humpback whale males emit songs struc- tured in a hierarchical manner where the basic building blocks are called sound units according to Payne and McVay (1971). Until now, whale experts classified these sounds manually by listening and labeling their recordings on spectrograms. Their objective was to extract the leitmotiv of the song for a specific population in a specific area for the breeding season (Noad et al., 2000; Cerchio et al., 2001). Of course, it is clear that Part I – Chapter 3 77 the choice of the well-adapted hydrophones for this application is crucial to improve the correct classification. To help the biologists, our objective was to develop an automatic analysis of Humpback whales songs (figure 7). This goal is quite difficult because, even when the hydrophone was just positioned in front of one singer, the recordings simultaneously included a lot of vocalisations from the other singers considered as background noise. So, to go further in the automatic classification task, we extracted different features of each sound unit, as duration, bandwith, harmonics, and frequency slopes. In addition, we proposed an original approach to increase the performance of our clas- sifier: we defined the concept of subunits, which means that the sound units defined by Payne and McVay (1971) can be decomposed into one or more subunits. By this definition, we are willing to use the tonal informa- tion and the sound prosody (variation of the features in the frequency domain) to increase the correct classification. Our underlying idea is to show that the diverse number of sound unit types performed by the sing- ers can be explained by a limited short number of subunits.
Figure 7: Segmentation of Humpback whale song recording into units (green area) and ambient underwater noise (red area).
The second step of our work was to extract information for each detected sound units. We first focused on the variation of the derivative of the five first main energetic frequencies. We then also defined different types of sound unit shape: sound unit with constant harmonic signals or chirps hav- ing increased or decreased, convex, concave or linear frequency variations. These features were extracted with different methods largely used in the Speech processing method (Kay, 1988) and based on the presence of at least one frequency in the sound units (Pace et al., 2009a). Classes were obtained by applying the unsupervised k-means algorithm, a statistical classification method, and the Davies-Bouldin criterion was used to evaluate the similar- ity within and between classes. This criterion made it possible to determine 78 Ecophysiology and animal behaviour the optimal number of classes for all vocalisations present in our dataset. Based on this method, 18 different groups of subunits were detected in a sample of 424 vocalisations (Picot et al., 2008; Pace et al., 2009b). We finally worked on hidden Markov models to take into account the possible variant duration of the subunits and to characterise the link between successive subunits (i.e. the syntax, see figure 8). The complete method for automatic classification of these sound units is a valuable tool for biologists willing to investigate the song evolution and the interactions between or within popu- lations. In the future, we expect that this method will also make it possible to assign an acoustic signature to a specific Humpback whale individual.
Figure 8: Segmentation and classification of Humpback whale sound units. The sound units are not emitted in a random order by the singers. Thus we used the hidden markov model (HMM) to take advantage of the specific order of these sound units to detect and classify them. MFCC: Mel-frequency cepstrum coefficients, Δs: a measure the temporal rates of change of the MFCCs.
5. Conclusion and future work
Using passive acoustic monitoring to assess cetacean populations has sev- eral benefits in comparison with conventional survey methods such as Part I – Chapter 3 79 visual sightings. The animals can be studied continuously without any negative impact. This method is also less dependent on weather condi- tions than visual methods, and does not rely on animals surfacing in order to be detected. It can be applied globally, including remote areas where visual sightings are usually either too sparse, difficult to gather, or costly. Other advantages of passive acoustic monitoring are that it helps to identify areas of cetacean concentration, seasonal occurrence and dis- tribution patterns; it can facilitate the long-term monitoring of cetacean abundance through variations in call rates over the years, and inform on where to establish marine protected areas. Passive acoustics is therefore an interesting complementary method for the cetacean observations.
Table 3: Advantages and drawbacks of two permanent acoustic systems
Advantages Drawbacks System with Electrical power by solar panels Weather conditions: buoy at the and/or wind turbines movement of waves, wind, sea surface Setting parameters (amplifier bad weather gain, sampling frequency) Risk of damage and theft Data transmission via HF, Wifi Real-time application System Discreet Electrical power deployed Less susceptible to surface Access to instrument on the sea activities (parameters settings) bottom Data transmission
Several passive acoustic approaches are possible, from instantaneous observations with a light deployable hydrophone to continuous observa- tions with sonobuoys deployed either at the sea surface or on the seabed. The solutions offered by permanent observatories have advantages and drawbacks listed in table 3, and future work could be dedicated to build a system that retains the advantages of the two main techniques and elimi- nates the drawbacks. For this purpose, the power supply, data storage, and data transmission limits must be circumvented, especially for real- time applications. The best feasible solution is probably to set up fully cabled systems (data and power) at the sea bottom, even if this solution is quite expensive. Another important task is to develop automatic real-time analysis for the detection and, if possible, the localisation of cetaceans. These analyses would be best conducted in situ and the results could be directly sent to the biologists and the managers of marine protected areas and/or coastal areas. This could provide authorities with real-time moni- toring tools to diminish the risk of collision between ships and cetaceans for example. 80 Ecophysiology and animal behaviour
Authors’ references
Flore Samaran: Université de la Rochelle, Observatoire Pelagis, Centre de Recherche des Mammifères Marins, ULR-CNRS UMS 3419, La Rochelle, France
Nadège Gandilhon: Université des Antilles et de la Guyane, Laboratoire de Biologie Marine, Equipe dynamique des écosystèmes Caraïbes, Pointe-à-Pitre, Guadeloupe, France
Rocio Prieto Gonzalez: Universidad de Valladolid, Departamento de Estadística e Investigación Operativa, Valladolid, Spain
Rocio Prieto Gonzalez, Amy Kennedy, Olivier Adam: Université Paris Sud orsay, Centre de Neurosciences Paris Sud, UPS- CNRS UMR 8195, Orsay, France
Federica Pace: University of Southampton, Institute of Sound and Vibration Research, Southampton England
Amy Kennedy: Alaska Fisheries Science Center, National Marine Mammal Laboratory, Seattle, USA
Olivier Adam: Université Pierre et Marie Curie, Institut Jean d’Alembert, Lutheries acoustique musicale, UPMC-CNRS UMR 7190, Paris, France
Corresponding author: Olivier Adam, [email protected]
References
Adam, O. Lopatka M., Laplanche C., Motsch J.-F., 2005. Sperm whale signal analysis: comparison using the autoregressive model and the wavelet transform. World Academy of Science, Engineering and Technology, 4, pp. 188-195. Adam O., 2008. Segmentation of killer whales vocalisations using the Hilbert Huang Transform. EURASIP Journal on Advances in Signal Processing, 2008, pp1-11. Part I – Chapter 3 81
Cerchio S., Jacobsen J. K., Norris T. F., 2001. Temporal and geographical variation in songs of humpback whales, Megaptera novaeangliae: synchronous change in Hawaiian and Mexican breeding assemblages. Animal Behaviour, 62, pp. 313-329. Frankel A. S., 1998. Sound production in: Perrin W. F., Wurisg B., Thevissen J. M. G. (Eds), Encyclopedia of Marine Mammals. Academic Press, San Diego, USA, pp. 1126-1137. Gandilhon N., Gervain P., Nolibe G., Louis M., Adam O., 2010. Creation of an autonomous system on moored Fish aggregating device (FAD) for a permanent acoustic monitoring of marine mammals and other perspectives for marine environment attention, Guadeloupe, F.W.I. 63th Annual Gulf and Caribbean Fisheries Institute (GCFI), Puerto Rico. Kay S. M., 1988. Modern Spectral Estimation: Theory and application. Prentice Hall, Upper Saddle River, USA. Laplanche C., Adam O., Lopatka M., Motsch, J.-F., 2006. Measuring the off-axis angle and the rotational movements of phonating sperm whales using a single hydrophone. Journal of the Acoustical Society of America, 119, pp. 4074-4082. Lopatka M., Adam o., Laplanche C., Zarzycki J., Motsch, J.-F., 2005. An attractive alternative for sperm whale click detection using the wavelet transform in comparison to the Fourier spectrogram. Aquatic Mammals, 31, pp. 463-467. Lopatka M., Adam O., Laplanche C., Zarzycki J., Motsch, J. F., 2006. Effective analysis of non-stationary short-time signals based on the adaptive Schur filter. Transactions on Systems, Signals and Devices, 1, pp. 295-319. Marques T. A., Thomas L., Ward J., Dimarzio N., Tyack P. L., 2009. Estimating cetacean population density using fixed passive acoustic sensors: an example with Blainville’s beaked whales. Journal of the Acoustical Society of America, 125, pp. 1982-1994. Mohl b., Wahlberg M., Madsen P., Heersfordt A., Lunds, A., 2003. The monopulsed nature of sperm whale clicks. Journal of the Acoustical Society of America, 114, pp. 1143-1154. Musikas, T., Samaran, F., Aupetit, M., Adam, O., 2009. Density estimation of Antarctic blue whales using automatic calls detection. First International workshop on density estimation of marine mammals using passive acoustics, Italy. Noad M. J., Cato D. H., Bryden M. M., Jenner M. N., Jenner K. C. S., 2000. Cultural revolution in whale songs. Nature, 408, pp. 537. Nosengo N., Riccobene G., Pavan, G., 2009. The neutrino and the whale. Nature, 462, pp. 560-561. Pace F., White P., Adam O., 2009a. Comparison of feature sets for humpback whale song classification. Proceedings of the fifth IOA International Conference on Bio-Acoustics, 31, pp. 136-144. 82 Ecophysiology and animal behaviour
Pace F., White P., Adam o., 2009b. Characterisation of sound subunits for humpback whale song analysis. 4th International Workshop on Detection and Localization of Marine Mammals using Passive Acoustics, Italy. Pace F., benard F., Glotin H., Adam o., White P., 2010. Subunit definition and analysis for humpback whale call classification. Applied Acoustics, 11, pp. 1107-1114. Payne R. S., McVay S., 1971. Songs of Humpback whales. Science, 173, pp. 585-597. Picot G., Adam O., Bergounioux M., Glotin H., Mayer F.-X., 2008. Automatic prosodic clustering of humpback whales song. New Trends for Environmental Monitoring using Passive Systems, pp. 1-6. Samaran F., Adam O., Motsch J.-F., Guinet C., 2008. Definition of the Antarctic and pygmy blue whale call templates. Application to fast automatic detection. Canadian Acoustics, 36, pp. 93-103. Samaran F., Adam O., Motsch J.-F., Cansi Y., Guinet C., 2010a. Source level estimation of two blue whale subspecies in southwestern Indian Ocean. Journal of the Acoustical Society of America, 127, pp. 3800-3808. Samaran F., Adam o., Guinet C., 2010b. Detection range modeling of blue whale calls in southwestern Indian ocean. Applied Acoustics, 71, pp. 1099-1106. Samaran F., Adam O., Guinet C., 2010c. Discovery of a mid-latitude sympatric area for two Southern Hemisphere blue whale subspecies. Endangered Species Research, 12, pp. 157-165. Simmonds M. P., Dolman S., Weilgart L. (Eds) 2004. Oceans Of Noise. The Whale and dolphin conservation society, Chippenham, UK. Simmonds M. P., Isaac S., 2007. The impact of climate change on marine mammals: early signs of significant problems. Oryx, 41, pp. 1-8. Simmonds M. P., Eliott W. J., 2009. Climate change and cetaceans: concerns and recent development. Journal of the Marine Biological Association of the United Kingdom, 89, pp. 203-210. Valsero, M. C., Prieto Gonzalez, R., Samaran, F., Adam, O., 2010. A spatio- temporal Poisson model to estimate the density of the Antartic blue whales in the Austral ocean. International Workshop on Applied Probability, Espagne. Chapter 4
Bioacoustics approaches to locate and identify animals in terrestrial environments
Chloé Huetz, Thierry Aubin
1. Needs for non-invasive methods to identify and locate animals
Population assessment and a proper understanding of behavioural strate- gies are central and urgent tasks in conservation biology. Nevertheless, up to now, field-based biological researches are held back by the difficulty, cost and intrusiveness of marking and tagging animals, and the relative ineffectiveness of manual data collection and analysis thereafter. Indeed, the monitoring of wild animals almost systematically presupposes their catching first. This invasive stage is not necessary when animals are acous- tically monitored (Gilbert et al., 1994; Hartwig, 2005). Almost all vocal species possess unique acoustic patterns that differ significantly from one to another individual, while following a common structure typical of the species. By using acoustic analysis methods, it is then possible to iden- tify individuals or species emitting vocalisations (insects, frogs, birds and mammals). Sound sources have also the property to be localisable. Until recently, localisation of wild animals by acoustic methods was not widely used. This was mainly due to technical limitations, as the monitoring of simultaneous acoustic sources is problematic in the field. Indeed, among several requirements, simultaneous field recordings devices have to share features such as being wireless, waterproof, easily transportable, and with large memory capacities. Now that technologies exist to overcome these limitations, it has become possible to localise and track the movements of animals that generate sounds. Such systems have been first called acoustic 84 Ecophysiology and animal behaviour location systems (ALS) by McGregor et al. (1997), but are now currently named automatic acoustic survey systems (AASS). An accurate AASS must fulfil two conditions: it must allow to localise the sound source with precision and identify the emitter. Until now, AASS have been used mostly to detect marine mammals (e.g. Stafford et al., 1998; Mellinger and Clark, 2003; Clark and Clapham, 2004; see chapter I, 3). In contrast, AASS dedicated to the monitoring of terrestrial species are rather uncommon (even though Mennill et al. 2006 used it for a case study with birds). For cetaceans, sensors are hydrophones and AASS spa- tial precision for animal localisation is in the kilometre range. Animals can also be tagged with localisation systems such as Argos for more accu- rate localisation, but even these systems cannot provide an accuracy below the metre range. For terrestrial species, more accurate locations are often required, for example, to determine the relative positions of neighbouring birds. In addition, tagging small terrestrial animals is often impossible due to several technical reasons (small size, difficulty of catching animals, see chapter I, 2). Moreover, marine and terrestrial environments differ from an acoustic point of view, the latter being often less homogeneous, with more obstacles. Here we review and discuss the accuracy of AASS for monitoring the position of animals vocalising in different terrestrial environments. We first introduce the principles and purposes of acoustic location systems. Then, we propose a non-exhaustive review of the different methods that can be implemented in an AASS in order to automatically locate ani- mals from the sounds they produce. We also present several existing methods used to extract from a vocalisation the individual and the spe- cies signatures. Finally, we illustrate the use of these technologies with some recent field applications concerning the location of birds in forest habitats.
2. Localisation system: time delay of sound arrival estimation and triangulation
2.1. Principle of Automatic Acoustic Survey Systems Acoustic location systems use simultaneous recordings from an array of acoustic sensors (microphones or hydrophones) scattered over a particu- lar area. From these recordings, two measurements can be extracted to compute the sound-source location: the level (or amplitude) differences between recordings, and the time delays between the sound arrival times at spatially separated microphones (see for review McGregor et al., 1997; Mennill et al., 2006). The latter, usually designated as the time-of-arrival Part I – Chapter 4 85 difference (TDoA) is estimated either i) by pair-wise cross-correlations of the sound waveforms recorded from the time-synchronised micro- phones or ii) by beamforming algorithms, which can be defined as the sum of all signals or their energy in the time domain (Valin et al., 2004) or in the frequency domain (Chen et al., 2006) properly time-delayed. The cross-correlation between two recordings shows one peak, corres- ponding to the TDoA of the same sound source at the two microphone positions.
Figure 1: Representation of our automatic acoustic survey system (AASS) localisation procedure. The first step consists in setting up the microphones and the wireless recording system (1). All pairwise distances between microphones are measured with a lasermeter (1.1), and all channels are recorded synchronously on a laptop computer (1.2). The offline analysis (2) follows three steps: from the 6-channels recordings, the animals’ vocalisations are extracted and band-pass filtered (2.1) to extract putative noise sources, then pairwise cross-correlations of the waveforms are performed for each vocalisation (2.2) in order to compute the time-of-arrival differences, which are then used to locate each vocalisation (2.3).
Sound source location can then be determined using the so-called trian- gulation principle. Indeed, the TDoAs of a sound between each micro- phone pair constrain the emitter’s location to a hyperboloid, and its 86 Ecophysiology and animal behaviour precise location can be resolved by intersecting hyperboloids from many pairs of receivers. Another possibility is to divide the space around the microphones into cells by a grid and compute all possible TDoAs for each cell. The location of a sound source is then computed by minimis- ing the difference between the measured TDoAs with the theoretical ones. When using the beamforming algorithms, TDoA are estimated by searching the delays between recordings that maximise the output signal of the beamformer. Lastly, some methods use simultaneously the TDoA, the level (amplitude) difference, and the reflection of the sound on the surface and ground to locate with a greater precision the emitter (Cato, 1998).
Figure 2: Example of the output of our localisation system. In this case, a loudspeaker was placed outside the microphone configuration (red star) and was emitting seven vocalisations. Each vocalisation was then localised (blue circle). As shown on the zoom, the localisation error is small, on the centimetre range. our AASS consisted of an array of six wireless omnidirectional micro- phones recording simultaneously up to six-channel sound files on a lap- top computer via an emitter-receiver station (figure 1). The microphones were set up throughout a given area in the field. Exploiting the speed of sound propagation through air, the system triangulates the position of sound sources on the basis of TDoAs at the microphones. These TDoAs Part I – Chapter 4 87 are estimated using a cross-correlation of the band-pass filtered wave- forms. Instead of using the standard triangulation equations, we applied a simple exhaustive search in space like Hammer and Barrett (2001). Test experiments were conducted in the field in Brazilian rainforest and in European temperate forests in order to probe the accuracy of the system. A loudspeaker was placed at a given location around the microphones, and its distance to the microphones was measured with a lasermeter. Several sounds were played and recorded on the AASS. Figure 2 shows an example of the localisation system’s result for the playback of seven bird songs (computed source in blue circles). The “real position” of the emitter (the loudspeaker) is indicated with a red star and it can be seen that all songs are localised very close to the “real” position measured with the lasermeter.
2.2. Limits of Automatic Acoustic Survey Systems Acoustic localisation systems are prone to the same constraints and usu- ally present the same limits as any kind of acoustic survey system. Indeed, sound localisation of animals is difficult when emitters are distributed widely in the open and when reverberations degrade temporal patterns in vocalisations (Spiesberger, 1999). Selective filtering due to obstacles in the environment can also modify the signal during propagation (Spiesberger, 1999; 2005). In noisy environments, multiple sounds coming from dif- ferent species or individuals, and their background, can overlap with the signal of interest in the temporal and frequency domains, producing a jamming effect and leading to false localisations. An insufficient signal- to-noise ratio in turn impairs the localisation process (Quazi and Lerro, 1985) and methods to enhance the signal-to-noise ratio (e.g. a band-pass prefiltering) must be applied when possible. In general, any factor dete- riorating the cross-correlations between recordings strongly affects the TDOA estimation and therefore can diminish the accuracy of the locali- sation process (Spiesberger, 1999; 2005). In addition, several constraints are specific to the use of an AASS for the purpose of localisation. The first necessary prerequisite of any localisation system is to know accurately the relative positions of the microphones (Quazi and Lerro, 1985). Indeed, in order to make use of the TDOAs and convert them into distances relative to the microphones, the whole microphone configuration has to be known, and subsequently, several pairwise distances between microphones have to be measured. The accu- racy and the space coverage of the localisation process strongly depend on the inter-microphone distances. In fact, the optimisation of the spatial configuration of microphones faces trade-offs between a large spacing, an accurate measurement of their relative positions, and a good signal-to- 88 Ecophysiology and animal behaviour noise ratio of the recorded vocalisation on the different microphones. If most of the studies assess that, in theory, three microphones are necessary to localise in 2D, and four in 3D, it has been shown that one additional microphone is needed to obtain accurate estimates of sound location (4 in 2D and 5 in 3D, Spiesberger, 2001). Moreover, some redundancy can help the localisation process (Chen et al., 2003). Depending on the needed accuracy of the output localisa- tion and on the studied environment, several methods can be used to measure the relative microphone positions. The position of the micro- phones can be estimated by the coordinates of a global-positioning sys- tem – GPS (Mennill et al., 2006). However, GPS positioning can hardly be used when animals need to be localised with great accuracy (less than 1m), or in obstructed environments such as dense forests. In these cases, a lasermeter can be used under the condition that no obstacle stops the laser beam and thus prevents the distance measures between pairs. This constrains the inter-microphone distances in thick vegeta- tion or in irregular topography. Another constraint on the microphone configuration is the perfect time-synchronization that must be achieved between devices. A wireless or a wired synchronization system is there- fore needed.
3. Identification systems: review of acoustic methods to extract species and individual signatures from animals’ vocalisations
3.1. Principles of identification systems The fact that animals can recognise one another by voice alone has been demonstrated repeatedly, especially in birds (for a review see Falls, 1982; Dhondt and Lambrechts, 1992) and mammals (for example Balcombe and McCracken, 1992, for bats; Caldwell et al., 1990, for dolphins; Sèbe et al., 2010, for lambs; Tooze and Harrington, 1990, for wolves). The infor- mation content of a signal is represented by structural sound features such as its spectrum or the temporal evolution of its amplitude and frequency modulations (see figure 3). A signal would be ideal for individual or spe- cies recognition if it is highly stereotyped within each individual or spe- cies, and if it significantly differs between individuals or species. Different methods are available for determining which parameters encode acoustic identity and allow recognition between species or between individuals. Part I – Chapter 4 89
Figure 3: Examples of display calls (above: spectrograms; below: oscillograms) emitted by three individuals of King penguins Aptenodytes patagonicus (from Aubin and Jouventin, 2002). 90 Ecophysiology and animal behaviour
Among the possible means to identify animal vocalisations, the most classical method is the visual inspection and labelling of oscillographs or spectrographs. This process is time consuming and dependent upon the judgement of one observer (Kogan and Margoliash, 1998). Beside this “manual” approach, some more or less automatic classification meth- ods can be used. A first method is based on variance calculations realised on the temporal, amplitude and frequency parameters of a given signal. For each call parameter, the between-individual and within-individual coefficients of variation (CVbi and CVwi) can be calculated as follows: 1 CV = 100 (SD/Xmean )(1 + [ 4 n ]), where SD is the standard deviation, Xmean the mean of the sample and n the sample size (Sokal and Rohlf, 1995). To assess the potential for the coding of individual identity (potential of individual coding: PIC) for CVbi each parameter, the ratio E(CVwi), where E(CVwi) is the mean value of the CVwi of all individuals, is calculated (Scherrer, 1984; Robisson et al., 1993). For a given parameter, a PIC value greater than 1 means that this parameter may correspond to an individual parameter as its intra-individual variability is smaller than its inter-individual variability. In the same way, the potential of specific coding (PSC) can be evalu- ated with the same formula by using the between-species and within- species coefficients of variation. Beside this univariate approach using parameters with high PIC or PSC values, multivariate analyses (discrimi- nant function analysis, DFA, principal component analysis, PCA, and artificial neural network analysis, ANN) can be performed (figure 4). All these methods provide classification procedures that assign each recorded call to its appropriate emitter (correct assignment) or reject the assignment (see Terry and McGregor, 2002). Other existing vocalisation classification techniques are based on tradi- tional automatic speech recognition methods (Rabiner and Juang, 1993). one conventional method, the dynamic time warping (DTW; Anderson et al., 1996), is well suited for the detection of pure tones such as those in bird, bat and cetacean songs. This method compares the spectrograms (frequency versus time representation, see figure 3) of input sounds with those of a training data set of predefined templates (representative of sounds to detect and chosen by the investigator) by successive cross-cor- relations (Clark et al., 1987). These templates are the targeted database for the matching process. In contrast to the deterministic template matching of DTW, another method, the hidden markov model (HMM, Rabiner, 1989) uses a statistical representation. Briefly, an HMM is typically a col- lection of finite sets of states. Each state represents spectral properties in the form of Gaussian mixtures of spectral features, while temporal properties are represented by state transition probabilities. Each state has Part I – Chapter 4 91 a probability distribution over the possible output cases. Therefore, the sequence of cases generated by a HMM provides information about the sequence of states and are thus especially adapted for temporal pattern recognition of sounds such as sequences of successive notes in songs. This list of classifiers presented here is not exhaustive and it is possible to find in the literature a lot of other acoustic identification methods such as for example the spectral peak tracks method (SPTM) recently suggested by Chen and Maher (2006).
Figure 4: A principal component analysis taking into account 2 factors (F1 + F2 = 54% of the total variance) and based upon 18 acoustic parameters measured in the song of a tropical bird, the White-browed Warbler Basileuterus leucoblepharus. on this basis, the PCA separates 71% of the 21 individuals analysed. Each polygon corresponds to one individual (from Aubin et al., 2004).
3.2. Limits of identification systems All the methods described above are well suited to classify sounds, but they also have some limitations. The most important requirement for the reliability of the identification of vocal signature is that emitters produce individualised and stable vocalisations. Another necessity is to have a prior knowledge of the structure of the vocalisations emitted by the individuals 92 Ecophysiology and animal behaviour who will be then automatically identified. For example, as underlined by Terry et al. (2005), discriminant function analysis may assign all vocali- sations to particular individuals if all individuals are known; thus this method cannot accommodate vocalisations from new individuals. All these methods, qualified as “supervised” methods, use in fact a training data set or templates that must be “learned” by the classifier. Instead, the HMM model is based on a statistical calculation; it can therefore accu- mulate more information and possibly generalise better than techniques based on fixed templates. The ANN classifiers often require a very high computational complexity. The DTW method does not use amplitude normalisation, so the results may be sensitive to amplitude differences between signals. The DTW and HMM methods do not perform well in noisy environments or for sounds with short duration and variable ampli- tude. In a word, all of the proposed methods accommodate well tonal or harmonic sounds, but are inappropriate when vocalisations containing aperiodic or noise-like components are involved.
4. Field applications
Studying animals in their natural habitat with a minimised influence on their behaviour is a key issue in ecology and ethology. Locating animals by the sounds they produce has the first advantage to be passive, meaning that the effect of the AASS on animal behaviour is insignificant. The second benefit is that terrestrial AASS can be used in habitats where visual location is difficult, such as dense vegetation (tropical forests, bushes, reed-beds). These systems may be also useful for studying secretive animals, difficult to observe because of large home ranges or nocturnal activity (see chapter IV, 2). They also ensure an accurate investigation of biological questions such as territorial defence, nest site, and mate fidelity. For example, they may be used to monitor multiple individuals simultaneously and, therefore, study behaviours such as territorial interactions between neighbours and duetting (Bower and Clark, 2005; Burt and Vehrencamp, 2005; Mennill and Vehrencamp, 2008). Thus, these systems seem particularly suitable for studying communication networks. For example, we are currently studying in the Amazonian forest the acoustic network of a typical bird of this habi- tat, the screaming piha Lipaugus vociferans. This species shows a remarkable form of lek display: females are attracted by singing assemblies of males and come for mating. Up to 25 males distributed on an area of about 600m of diameter are usually observed at a single lek. Each male has its own song posts and it counter-sings the other males. The use of an AASS will enable us to decipher this complex vocal organisation and particularly, to examine the singing interactions and movements of all the birds from a lek. Part I – Chapter 4 93
In a more general way, AASSs appear extremely useful to provide, with very little human interference, quantitative and qualitative indices of animal diversity in poorly accessible environments. Starting from this method, it is possible to develop concrete applications with an acoustic platform designed for the estimation of biodiversity of the fauna in different catego- ries of environments. An automatic acoustic survey may help in the evalua- tion of the biological quality of a given habitat and appears as a useful tool for measuring the abundance of species or the impact of human activity on biodiversity (Sueur et al, 2008; see also II, 1). This method should sub- stitute to the traditional human visual or hear sampling realised on point counts (i.e. Uezu et al., 2005) to estimate the abundance of vocal species in remote or obstructed environments, such as rainforests. Thus, we believe that acoustic location and identification technology may provide a valu- able and versatile tool for ecologists and ethologists.
5. Conclusion: future orientations, developments and needs for new sensors?
AASSs offer one of the best solutions for a non-invasive sensor that will enable biologists both to locate and identify individuals of a large number of “singing” species within a population in cheap, fast and automatic man- ner and in a wide range of environments. The emergence of this method is sped up by recent technological advances. Thus, commercially available autonomous digital recorders are now able to collect thousands of hours of audio data. The automatic identification of vocalisations is not always per- fectly accurate, but the development of new algorithms methods in auto- matic human speech recognition has recently improved the process. Source localisation algorithms for monitoring vocal wildlife populations are now efficient, the limitations being mainly due either to imprecise microphone coordinates or the presence of a particular constellation of competing sound sources in the field. In the first case, the development of more accurate GPS would be an alternative to laser measurements to geolocate more precisely microphones. With such systems, it would also be possible to increase the distance between the microphone and thus monitor a wider area. In the second case, the noise generated by parasitic sounds overlapping the target sound, should be more or less removed by using specific artificial neural network analysis. Despite significant progress in source localisation theory and sensor network systems, progress toward developing prototype AASSs has been greatly slowed down by the absence of integrated platforms suit- able for monitoring wild animals. An emphasis must be now placed on fully integrated systems (hardware and software) that are robust enough to be deployed in all kind of environments, and user-friendly enough to be used by biologists with little or no technical expertise. 94 Ecophysiology and animal behaviour
Authors’ reference:
Chloé Huetz, Thierry Aubin: Université Paris-Sud, Centre de Neurosciences Paris-Sud, UPS-CNRS UMR 8195, Orsay, France
Corresponding author: Thierry Aubin, [email protected]
Aknowledgement
We thanks F. Sèbe, H. Courvoisier and M.L. da Silva for technical sup- port and help in the field. The work in the Amazonian forest (Brazil) was financially supported by a FP7-people-IEF grant.
References
Anderson. S E., Dave A. S., Margoliash D., 1996. Template-based automatic recognition of birdsong syllables from continuous recordings. Journal of the Acoustical Society of America, 100, pp. 1209-1219. Aubin T., Jouventin P., 2002. How to identify vocally a kin in a crowd? The penguin model. Advances in the Study of Behaviour, 31, pp. 243-277. Aubin T., Mathevon N., Da Silva M.L., Vielliard J.M.E., Sèbe F., 2004. How a simple and stereotyped acoustic signal transmits individual information: the song of the White-browed Warbler Basileuterus leucoblepharus. Anais da Academia brasileira de Ciencias, 76, pp. 335-344. Balcombe J. P., McCracken G. F., 1992. Vocal recognition in Mexican free-tailed bats: do pups recognize mothers? Animal Behaviour, 43, pp. 79-87. Bower J.L., Clark C.W., 2005. A field test of the accuracy of a passive acoustic location system. Bioacoustics, 15, pp. 1-14. Burt J.M., Vehrencamp S.L., 2005. Dawn chorus as an interactive communication network, in: McGregor P. K. (Ed.), Animal Communication Networks. Cambridge University Press, New York, pp. 320-343. Caldwel, M.C., Caldwell D.K., Tyack P. L., 1990. Review of the signature- whistle hypothesis for the Atlantic bottlenose dolphin, in: Leatherwood S., Reeves R.R. (Eds.), The bottlenose dolphin. Academic Press, New York, pp. 199-234. Cato D.H., 1998. Simple methods of estimating source levels and locations of marine animal sounds. Journal of the Acoustical Society of America, 104, pp. 1667-1678. Chen C.E., Ali A.M., Wang H., 2006. Design and testing of robust acoustic arrays for localization and enhancement of several bird sources, in: IPSN Part I – Chapter 4 95
Proceedings of the 5th international conference on Information processing in sensor networks. Association for Computing Machinery, New york, pp. 268-275. Chen Z., Maher R.C., 2006. Semi-automatic classification of bird vocalizations using spectral peak tracks. Journal of the Acoustical Society of America, 120, pp. 2974-2984. Clark C.W., Clapham P.J., 2004. Acoustic monitoring on a humpback whale (Megaptera novaeangliae) feeding ground shows continual singing into late spring. Proceedings of the Royal society of London B, 271, pp. 1051-1057. Clark C.W., Marler P., Beeman K., 1987. Quantitative analysis of animal phonology; an application to swamp sparrow song. Ethology, 76, pp. 101-115. Dhondt A.A., Lambrechts M.M., 1992. Individual voice recognition in birds. Trends in Ecology and Evolution, 7, pp. 178–179. Falls J.B., 1982. Individual recognition by sound in birds, in: Kroodsma D.E., Miller E.H. (Eds), Acoustic Communication in Birds, vol. 2. Academic Press, New York, pp. 237-278. Gilbert G., McGregor P.K., Tyler G., 1994. Vocal individuality as a census tool: practical considerations illustrated by a study of two rare species. Journal of Field Ornithology, 65, pp. 335-348. Hammer O., Barrett N., 2001. Techniques for studying the spatio-temporal distribution of animal vocalizations in tropical wet forests. Bioacoustics, 12, pp. 21-35. Hartwig S., 2005. Individual acoustic identification as a non-invasive conservation tool: an approach to the conservation of the African wild dog Lycaon pictus (Temminck, 1820). Bioacoustics, 15, pp. 35-50. Kogan J.A., Margoliash D., 1998. Automated recognition of bird song elements from continuous recordings using dynamic time warping and hidden Markov models: a comparative study. Journal of the Acoustical Society of America, 103, pp. 2185-2196. McGregor P.K., Dabelsteen T., Clark C.W., Bower J.L., Tavares J.P., Holland J., 1997. Accuracy of a passive acoustic location system: empirical studies in terrestrial habitats. Ethology Ecology Evolution, 9, pp. 269-286. Mellinger D. K., Clark C. W., 2003. Blue whale (Balaenoptera musculus) sounds from the North Atlantic. Journal of the Acoustical Society of America, 114, pp. 1108-1119. Mennill D. J., Burt J. M., Fristrup K. M., Vehrencamp S. L., 2006. Accuracy of an acoustic location system for monitoring the position of duetting songbirds in tropical forest. Journal of the Acoustical Society of America, 119, pp. 2832-2839. Mennill D. J., Vehrencamp S. L. 2008. Context-dependent functions of avian duets revealed through microphone-array recordings and multispeaker playback. Current Biology, 18, pp. 1314-1319. Quazi A. H., Lerro D.T., 1985. Passive localization using time delay estimates with sensor positional errors. Journal of the Acoustical Society of America, 78, pp. 1664-1670. 96 Ecophysiology and animal behaviour
Rabiner. L R., 1989. A tutorial on hidden markov models and selected applications in speech recognition, in: Proceedings of the IEEE, 77, pp. 257-286. Rabiner L. R., Juang B. H., 1993. Fundamentals of Speech Recognition, Prentice-Hall, Englewood Cliffs, New Jersey. Robisson P., Aubin T., Brémond J.C., 1993. Individuality in the voice of emperor penguin Aptenodytes forsteri: adaptation to a noisy environment. Ethology, 94, pp. 279-290. Scherrer B., 1984. Biostatistique. Gaëtan Morin, Chicoutimy, Canada. Sèbe F., Duboscq J., Ligout S., Aubin T., Poindron P., 2010. Early vocal recognition of mother by lambs: contribution of low and high-frequency vocalizations. Animal Behaviour, 79, pp. 1055-1066. Sokal R. R., Rohlf F. J., 1995. Biometry, third ed. W.H. Freeman and Company, New York, USA. Spiesberger J. L., 1999. Locating animals from their sounds and tomography of the atmosphere: experimental demonstration. Journal of the Acoustial society of America, 106, pp. 837-846. Spiesberger J. L., 2001. Hyperbolic location errors due to insufficient numbers of receivers. Journal of the Acoustical Society of America, 109, pp. 3076-3079. Spiesberger J. L., 2005. Probability distributions for locations of calling animals, receivers, sound speeds, winds, and data from travel time differences. Journal of the Acoustical Society of America, 118, pp. 1790-1800. Stafford K. M., Fox C. G., Clark D. S., 1998. Long-range acoustic detection and localization of blue whale calls in the northeastern Pacific Ocean. Journal of the Acoustical Society of America, 104, pp. 3616-3625. Sueur J., Pavoine S., Hamerlynck O., Duvail S., 2008. Rapid acoustic survey for biodiversity appraisal. PLoS One, 3, pp. 1-9. Terry A.M.R., McGregor P.K., 2002. Census and monitoring based on individually identifiable vocalizations: the role of neural networks. Animal Conservation, 5, pp. 103-111. Terry A.M.R., Peak, T., McGregor P.K., 2005. The role of vocal individuality in conservation. Frontiers in Zoology, 2, 10. Tooze Z. J., Harrington F. H., Fentress, J. C., 1990. Individually distinct vocalisations in timber wolves, Canis lupus. Animal behaviour, 40, pp. 723-730. Uezu A., Metzger J. P., Vielliard J. M. E., 2005. Effects of structural and functional connectivity and patch size on the abundance of seven Atlantic Forest bird species. Biological Conservation, 123, pp. 507-519. Valin J. M., Michaud F., Hadjou B., Rouat J., 2004. Localization of simultaneous moving sound sources for mobile robot using a frequency- domain steered beamformer approach, in: Proceedings of IEEE International conference on Robotics and Automation, 1, pp. 1033-1038. II B iODIVERSITY
Chapter 1
Global estimation of animal diversity using automatic acoustic sensors
Jérôme Sueur, Amandine Gasc, Philippe Grandcolas, Sandrine Pavoine
1. Introduction
The estimation of biodiversity can be considered as one of the main chal- lenges in modern biology. When dealing with ecology, evolutionary biol- ogy and conservation biology, there is an inescapable need to describe the composition and dynamics of biological diversity (Magurran, 2004). In ecology, the concept of biological diversity is mainly species-oriented, even if other evolutionary units or traits can also be used. In this context, biodiversity potentially refers to all species encountered in a given area at a specified time, including every potential species from underground bacteria to giant trees. Therefore, biodiversity assessment can turn out to be a time-consuming and complex task, as it relies on species inventory that may involve very different taxonomic groups. Exhaustive approaches such as the all taxa biodiversity inventory (ATBI) programs aim at inventorying the whole biodiversity mainly in tropical habitats (Gewin, 2002), but these programs are highly sensitive to the logistic and time-constraints of most inventory studies. An alternative to these approaches is to focus on one or a few taxa and consider them as biodiversity indicators (Pearson, 1994), but the choice of representative taxa is not trivial (Lawton et al., 1998). In addition, it is well known that patterns of species diversity for different taxa are sensitive to the observa- tion scale. More precisely, there is a general congruence for species diver- sity between different taxa at a large area scale (more than 1km2) but not at a fine scale (less than 1km2, Weaver, 1995). This renders difficult the definition of an indicator taxon or even of several indicator taxa suppos- edly representative of the diversity in other forms of organisms (Ricketts 100 Biodiversity et al., 1999). Irrespective of the taxonomic breadth of any biodiversity assessment, the estimation of species biodiversity relies on inventories and species examination by one or several taxonomic experts that can be sup- ported with genetic barcoding techniques (see chapter II, 3). Sampling in the field and identification in museum collections can require a con- siderable effort when the objective is to sample a large region for a long time period. To improve the rate of specimen collection, non-specialist taxonomic workers – or para-taxonomists – can separate morpho-species instead of identifying valid species. This solution is advocated by the rapid biodiversity assessment (RBA) programs that have been especially devel- oped for the rapid exploration of biodiversity in tropical habitats (Oliver and Beattie, 1993; 1996; Oliver et al., 2000). Biodiversity assessment is often restricted to species richness, i.e. to the counting of the total number of species. However, a collection of spe- cies cannot be described solely by the number of items it includes. The abundance of each species has to be assessed to provide an estimation of species evenness. Evolutionary, ecological and life history characters of the species also describe facets of biodiversity (Brooks and McLennan, 1991; Vane-Wright et al., 1991; Grandcolas, 1998; Pavoine et al., 2009; Petchey et al., 2009). Species turnover along time and/or spatial scales is also required to take into account biodiversity dynamics. All of these requirements led to a plethora of biodiversity indices that have been devel- oped for decades (Magurran, 2004; Buckland et al., 2005; Pavoine and Bonsall, 2010). In practice, a measure of biodiversity can be achieved with direct or indi- rect sampling. In the latter case, the use of a sensor should be ideally con- sidered by employing a simple tool that returns an index of biodiversity. A human observer or a network of human observers might be considered as a “biodiversity sensor”, but may be biased by the experience of the observ- ers and cannot be “deployed” in rough terrains for long periods of time. Another possibility is to work with local image, video capture instru- ments or with global satellite imagery. Satellite-based earth observations, or remote sensing, can produce environmental parameters from biophysi- cal characteristics that can be indirectly used to assess species ranges and species richness patterns (Kerr et al., 2003, Turner et al., 2003; Wang et al., 2010; see also IV, 2). These methods are undoubtedly very attractive, but they rely on extremely expensive equipment and are difficult to adapt to small spatial scales. Like other methods, remote sensing often requires a time-consuming validation step. For instance, vegetation mapping with satellite images is based on a colorimetric calibration of pixels with a large set of direct vegetation samples. The use of a given sensor should be made according to a sampling strat- egy designed and evaluated carefully with respect to the type of data to Part II – Chapter 1 101 be collected. It is particularly important to identify the precision of the measurement system (data quality) and the level of accuracy that has to be reached along both time and space scales (data quantity). In most sam- pling strategies, there is a basic trade-off between precision and accuracy. In this context, we are currently developing an acoustic sensor that would produce a biodiversity index by analysing the sound produced by local animal communities. This approach could provide a portable, cheap, rea- sonably accurate and non-invasive animal diversity sensor that could be used at different space and time scales.
2. Sensing diversity through bioacoustics
Some animal species, including taxa often used in biodiversity studies, produce active sounds during their social interactions or in other contexts. For example, some fish and reptiles, most amphibians, birds, mammals, insects and other arthropods use sound for communication, navigation or predation acts. These acoustic signals generally produce a species-specific signature and several techniques in bioacoustics were developed to exploit these signals as an indication of species occurrence and as a tool for bio- diversity studies (Obrist et al., 2010). The most elementary application is sensing by observers. This is usually achieved when following animal pop- ulations through aural listening and identification (e.g., Cano-Santana et al., 2008, for crickets). When based on a massive network of listeners, such a census method can generate large datasets of strong interest to ecologists (e.g. Devictor et al., 2008, for birds). Nonetheless, volunteer- based call surveys tend to be replaced by the automated digital recording system (ADRS), which is an electronic equipment that allows automatic data collection and generates a large amount of high-quality information about species biodiversity (e.g. Acevedo and Villanueva-Rivera, 2006, for amphibians). The problems alluded above for species identification with museum specimens is also true for species identification with sound. Acoustic identification is based on the experience of the observers, which can be biased due to sensory or training differences. As any other identification, it also relies on a taxonomic database providing information on the cor- respondence between every species and its acoustic signature. Automatic identification of the different songs embedded in the recording is rather complex and still suffers errors (e.g. Skowronski and Harris, 2006, for bats). These approaches are also difficult to deploy in complex acoustic environments like tropical forest soundscapes, where tens of signals mix up and many species still remain unknown (Riede, 1993). Reliable results can be obtained only when focusing on a single species with a rather 102 Biodiversity simple and loud call as demonstrated with the neotropical bird Lipaugus vociferans (see I, 4) and the Blue Whale Balaenoptera musculus in a marine context (see I, 3). Keeping in mind these constraints, we applied the concept of RBA to sounds produced by animals and even pushed the concept one step fur- ther. We recently suggested tackling the problem of diversity assessment at the community level by using bioacoustic methods (Sueur et al., 2008a). In the case of bioacoustics, the unit to work with is the acoustic com- munity, which is defined as the sum of all sounds produced by animals at given location and time. The signals produced by different species can overlap, interfere and consequently reduce signal transmission between the emitter and the receiver of a focal species. Sound produced by other species is indeed considered as noise for the focal species and acts as a severe constraint on the evolution of conspecific signals (Brumm and Slabbekoorn, 2005). Consequently, species sharing the same acoustic space are supposed to show an over-dispersion of the frequency and time- amplitude parameters of their songs reducing the risk of interference. This has been reported in several acoustic communities (e.g., Lüddecke et al., 2000, for amphibians; Sueur, 2002, for cicadas; Luther, 2009, for birds). A measure of sound complexity could then work as a proxy of commu- nity richness and composition. The acoustic indices we are developing are mainly based on this concept of acoustic partitioning. We hereafter review the recording equipment and analysis we used to try and build an animal diversity acoustic sensor.
3. Listening and measuring acoustic diversity
A biodiversity sensor provides a measure of a single or a set of variables characterising biodiversity. Even if a sensor is composed of several probes and data analysers, it is often viewed as a all-in-one equipment that senses and analyses the environment concomitantly. Our method currently relies on two different equipments that are not used at the same time. However, we here consider that these sub-units constitute together a single sensor (figure 1). The first sub-unit is a digital sound-recorder that can be settled outdoor. The second sub-unit is a computer installed with soft- ware specifically developed to analyse sound diversity. Further statistical analyses on the acoustic indices, i.e. the biodiversity variables measured by the sensor, are not considered as part of the sensor but as part of data analysis processes. We hereafter detail the sampling protocols based on a single recorder or an array of recorders, the properties of the autono- mous recorder currently in use and, eventually, the algorithm developed to compute the diversity indices from sound files. Part II – Chapter 1 103
Figure 1: Diagram showing the successive steps of the global estimation of animal diversity. Here, the biodiversity sensor is considered as the combination of dif- ferent processes: recording, audio file conservation, signal analysis, and indices computation. These processes – in situ recording step with autonomous equipment and ex situ calculation of indices from stored acoustic data – are currently separated from each other. However, a portable all-in-one system might be developed in the future.
3.1. From a single manual recording spot to a network of autonomous recorders The sampling protocol is mainly constrained by the recording equipment available. Our method was first tested with a comparison between two closely spaced dry lowland coastal forests in Tanzania. The recording of the animal communities inhabiting these forests was achieved with a digital recorder (Edirol© R09) equipped with an omnidirectional micro- phone (Sennheiser© K6/ME62). Recordings were done by a single person at three times of the day and successively in the two forests. This proce- dure limited the sampling to a few days and to only two sampling sites. Such digital recorders also provide internal microphones that can be used to reduce costs. In this case, several items can be purchased to cover a wider area and a longer period of time. However, the recorders still have to be triggered and stopped manually, a condition that makes field work rather challenging. Unattended recorders were not available since the North American com- pany Wildlife Acoustics© provided an autonomous digital field recorder (see details about this recording package in section 3.2). An autonomous system was absolutely necessary to design sampling protocols with syn- chronised units such as regular, cluster, multi-level, or stratified protocols. We first used three of these recorders to assess animal diversity within tem- perate woodlands by simultaneously recording a mature forest, a young forest and an edge forest (figure 2A, Depraetere et al., 2012). We then increased the number of recorders to estimate biodiversity endemism of three New-Caledonian sites. We planned a stratified sampling with four recorders set in each site. This ensured a repetition per site and allowed comparisons within and between study sites (figure 2B). Later, we tracked acoustic diversity of a typical tropical forest by deploying a network of 12 104 Biodiversity recorders regularly spaced on a 100 ×100m grid in French Guiana (see IV, 2). Each recorder was equipped with a microphone settled 1.5m high and a second microphone placed 20m high in the canopy (figure 2C). This 3D regular sampling covered 12ha of forest for more than 40 days. Eventually, we tried to transfer our method to freshwater habitats like forest ponds. This was achieved by adapting the autonomous recorder with a Reson© hydrophone and an Avisoft© pre-amplifier (figure 2D). This high-quality equipment is expensive (about 2700€ per unit) and sampling was there- fore limited to three recording units. We therefore designed a rotating sampling by regularly moving the hydrophone position along transects.
Figure 2: The autonomous Wildlife Acoustics© recorder installed for outdoor studies. A. First version of the recorder (SM1) settled in a temperate woodland to estimate local bird acoustic community (Rambouillet, France). B. Second version of the recorder (SM2) with a single microphone in action (Mandjelia, New- Caledonia). C. The same recorder with two microphones, one 2 m high and the other one ready to be set 20 m up in the canopy (Nouragues experimental station, French Guiana). D. Recorder connected to an hydrophone to record underwater sound of a pond (Rambouillet, France).
3.2. The Song Meter: an autonomous acoustic sensor Wildlife Acoustics© developed two generations of autonomous digital recorders, namely the Song Meter SM1 and SM2 (figure 3). These stereo recorders, which weigh 1.6kg each and measure 20.3 × 20.3 × 6.4cm, Part II – Chapter 1 105 possess a stereo recording system with omnidirectional microphones that have a flat frequency response between 0.02 and 20kHz. These microphones can be directly connected to the main box, where data are stored, or can be settled up to 50m away from it. Given that terrestrial animals produce sound with an intensity of ca. 80dB at 1m re. 2×10-5 μPa (Sueur, unpublished data) and given that the microphones have a sensitivity of -36 ± 4dB, we can estimate that in a closed habitat, such as a forest, the microphone detects sounds up to around 100m from the source. A SM2 platform would then cover an area of approximatively 3,1ha. The recording sampling rate can be set from 4 to 48kHz with the stan- dard SM2 and up to 384kHz with the ultrasonic SM2 option. The SM2 recorders are currently working with a lossless compression format (.wac) that can be written on four secure digital (SD) cards. The four SD slots provide 128Go storage space. Choosing an adequate sampling rate is not an easy task as it results from a trade-off between cost, data storage and the sound frequency used by animals. Increasing the sampling rate to high frequency requires a specific and expensive motherboard and, above all, generates very large sound files that are difficult to handle and to anal- yse. However, this is the only solution to record the acoustic activity of some insects and bats that emit ultrasound signals for communication or navigation. Up to now, we sampled the animal acoustic communities at a 44.1kHz sampling rate. A network of recorders generates thousands of files that need to be stored and analysed (see section 4.2). Using a higher sampling rate will certainly preclude the estimation of acoustic diversity by generating too high an amount of data. Electrical power is provided by four alkaline or LR20 batteries ensuring a maximum of 240 hours of recordings. Energy can also come from an external 12V battery potentially connected to a solar panel. Eventually, the SM2 platform provides also an internal temperature sensor and a connection for an external sensor. The additional data are written on the SD cards together with sound files. The main advantage of the Song Meter is that it can be easily programmed to record on simple time-of-day schedules or to implement complex monitoring protocols, even scheduling recordings relative to local sunrise, sunset and twilight. For instance, a schedule can be programmed to record regularly all day and night long, but also to record more intensively around sunrise and sunset, when dawn and dusk choruses of birds, insects and amphibians occur. 106 Biodiversity
Figure 3: The second version of the recorder (SM2) opened to show the main characteristics. A cable can be used to set the microphones away from the main box. Detailed characteristics can be obtained at http://www.wildlifeacoustics.com. © Wildlife Acoustics.
3.3. Computing the acoustic indices Biodiversity is traditionally decomposed into two levels, the average diversity within communities, or α diversity, and the diversity between communities, or β diversity. We therefore developed two acoustic indices aiming at estimating these two components of biodiversity (Sueur et al., 2008a). Both indices can be computed with the package seewave (Sueur et al., 2008b) of the free R environment (R Development Core Team, 2012). The first index, named H, is a Shannon-like index. The index H gives a measure of the entropy of the acoustic community by considering both temporal and frequency entropy. H is computed according to: H = Ht × Hf with 0 ≤ H ≤ 1, and Ht = - ∑ (A(t) × log(A(t)) / log (n)), and Hf = - ∑ (S( f ) × log(S( f )) / log (N)), where n = length of the signal in number of digitized points, A(t) = prob- ability mass function of the amplitude envelope, S( f ) = probability mass function of the mean spectrum calculated using a short term Fourier transform (STFT) along the signal with a non-overlapping Hamming win- dow of N = 512 points (figure 4). Part II – Chapter 1 107
Figure 4: The main two transforms used on raw recordings. A. Waveform or oscillogram of a sound recording. b. Amplitude envelope, A(t), obtained through the Hilbert transform. C. Mean spectrum, S( f ), obtained through a Fourier transform. Note the different x axes and that all y axes are in relative amplitude along a linear scale.
The H index increases logarithmically from 0 to 1 with species richness and evenness when considering species-specific calls (Sueur et al., 2008a). The index will be particularly high for a signal that has a flat amplitude envelope and a flat frequency spectrum. When only considering the spec- tral component of the index, a flat or multi-peak spectrum will give a higher Hf index than a single peak spectrum (figure 5 A, b). The H index was applied in Tanzania, and correctly revealed a higher acoustic diversity in the preserved part of the forest than in the disturbed part (Sueur et al., 2008a). However, Hf is not reliable when dealing with recordings made in the temperate woodland, where the acoustic activity is low and polluted 108 Biodiversity with environmental noise. In this particular case, we developed another index, named Acoustic Richness AR which was computed according to: AR = rank(Ht) × rank(M) × n-2, with 0 ≤ AR ≤ 1, where rank is the value position along the ordered samples, M is the median of the amplitude envelope and n the number of recordings (Depraetere et al., 2012). The second index, named D, is a simple acoustic dissimilarity measure. D is similarly composed of two sub-indices based on a difference between amplitude envelopes and frequency spectra respectively (figure 5 C). D is calculated like following: D = Dt × Df with 0 ≤ D ≤ 1, and
Dt = 0.5 × ∑ |A1 (t) – A2 (t)|, and
Df = 0.5 × ∑ |S1 ( f ) – S2 ( f )|, where A1(t), A2(t) are probability mass functions of the amplitude enve- lope for the two recordings under comparison, and S1( f ), S2( f ) are prob- ability mass functions of the mean spectrum for the two recordings to be compared. The D index increases linearly with the number of unshared species between the two recordings, or communities (Sueur et al., 2008a). Both indices may suffer a bias as some species naturally produce signals with high temporal and/or spectral entropy. This is particularly the case of cicadas whose noise-like sound can be mistakenly interpreted as a high local diversity. Such bias can be buffered with a large sampling including a high number of time and space repetitions. The indices can also produce false values when background noise overlaps with the sound produced by the animal community (see section 4.1). Both indices are currently tested in different temperate and tropical habitats in this respect. Other acoustic indices have been developed elsewhere to monitor habitat state or community activity. Qi et al. (2008) divided the soundscape of an ecosystem following three frequency bands: the anthrophony, between 0.2 and 1.5kHz, the biophony, which starts at 2kHz with a peak at 8kHz, and the geophony, which can cover the entire spectrum with dominant low frequency. By computing a ratio between biological and anthropo- genic signals, they coined an ecological estimator of ecosystem health. This original procedure does not give an estimation of local diversity but assesses the level of biological sound activity relative to anthropogenic activities. Pierreti et al. (2010) and Farina et al. (2011) designed an acoustic complexity index (ACI). This index computes time and frequency vari- ability of a sound extrapolated from a spectrogram. The ACI appears to be correlated with the number of vocalisations produced by a bird com- munity. However, this index assesses neither species diversity nor commu- nity turnover. The ACI index proved to be poorly sensitive to invariant noise, such as continuous noise from cars or aircrafts, but can be impacted Part II – Chapter 1 109 by unpredictable noise such as wind, running water or irregular human activity. All these acoustic indices, including H, D, and others in current development probably do not quantify the same facet of animal acoustic diversity.
Figure 5: Illustration of a spectral analysis on recordings made in two sites in New- Caledonia (France). A. A recording showing a broadband frequency spectrum with a high Hf index and a high number of peaks. b. A recording with a single dominant frequency peak generating a lower Hf index and less frequency peaks. C.
The difference between the two spectra used to compute the Df index. 110 Biodiversity
4. Sensitivity to noise level, sensor size and autonomy
4.1. Everything but noise Background noise is probably the primary issue in bioacoustics. Noise can significantly impairs acoustic observations and experiments by masking or distorting both time-amplitude and frequency parameters (Hartmann, 1998; Vaseghi, 2000). There are three main sources of noise to consider when recording outdoor: i) anthropogenic noise due to machinery, car, boat, plane, train traffic, or any other human activity, ii) biotic noise due to the activity of surrounding species, and iii) environmental noise due to rain, wind, river stream, waterfall, or sea wave (Brumm and Slabbekoorn, 2005; Laiolo, 2010). The estimation of animal diversity through acoustics is based on the recording of a whole community and as such does not face the classical problems encountered when trying to record a single species in the background noise generated by surrounding active species. However, anthropogenic and environmental noise can have negative effects on the results. In a few instances, anthropogenic noise can be removed by apply- ing classical frequency filters (Stoddard, 1998). Recordings made close to an airport or a road with a regular traffic can be cleaned with a high-pass filter that will remove the low frequency band generated by plane or car engines. Such filters might exclude low pitch animal calling songs, but this can be accounted for when computing diversity indices. The main difficulty arises when recordings are polluted with unpredictable and/or broadband noise that can be interpreted erroneously as animal sounds. Removing such chaotic sound is a challenge to be solved in bioacoustics as well as in other acoustic disciplines (Rumsey and McCormick, 1992; Hartmann, 1998; Stoddard, 1998; Vaseghi, 2000). Usual frequency filters cannot be used as noise may overlap the frequency band used by the ani- mal community. Other noise reduction algorithms use noise spectrum as a reference to be convoluted with the original signal. This solution might appear elegant but still suffers important limitations. First, the noise has to be constant in its frequency content, a condition rarely met in a natural acoustic environment. Second, it is necessary to identify accurately a time window where only noise occurs. This latter condition is very difficult to meet when faced with hundreds or thousands of recordings. Fortunately, some upstream solutions can be considered to reduce the anthropogenic and environmental noise (Obrist et al., 2010). When using an outdoor acoustic sensor, the most important parameter to consider is the direction and the protection of the microphone. The microphone can be oriented in a horizontal or vertical position as soon as its direc- tivity pattern is omnidirectional. A vertical upward position should be avoided when possible, as rain drops might directly strike the microphone Part II – Chapter 1 111 membrane. A vertical upside-down orientation might be the best solu- tion in avoiding rain and lateral wind effects. More generally, adapting the orientation of the microphone to the local main sources of noises is usually advocated. For instance, the noise of running water or passing-by cars can be reduced by orienting the microphone perpendicularly to the source, and windscreens should be used to attenuate wind noise. Another upstream solution is to exclude data potentially corrupted with environ- mental noise. This can be achieved in three ways. The first option consists in cutting off the recording session when weather conditions are too bad. It is not yet available but could certainly be implemented quickly, given the availability of climate sensors in sound meter devices. The second option is to apply a signal-to-noise algorithm that indicates the occurrence of an important background noise. A threshold could be used as a refer- ence to keep or to remove the files from the dataset. This solution is under development in our group. The third and last option, which is currently in use, is to gather climatic parameters from a local station and identify the time periods when the weather was too bad to allow a correct estimation of the acoustic diversity. This identification can be achieved automati- cally with a threshold applied on the climatic parameters or by running a redundancy analysis (RDA, Rao, 1964) to the acoustic indices with the climatic parameters as factors (Depraetere et al., 2012).
4.2. Optimal size of recorders As described above, the SM2 recorder weighs around 1.6kg and can be fitted with two microphones (figure 3). Hence, handling several of these units in a hard-to-reach environment requires a significant effort. A reduced size and weight would make field work easier and could also allow settling more units in the habitat. However, this has to be traded off against the size of the data that needs to be stored and analysed. A typical .wav file, which is the most popular uncompressed audio format, has a size of around 690kb/s (= 84ko/s) when sampled at a 44.01kHz rate. This means that one minute of recording is roughly equivalent to 5Mo for a single channel (mono) or around 10Mo for two channels (stereo). Sampling quickly generates x×102 hours of recording in x×103 files for a total x×102 osG data. A detailed above, the recorders have storage capacity of 128Go, which is enough for most applications sampled at 44.01kHz, but might appear limited for an over-month or over-year survey or for a long ultrasound monitoring. The next step of data transfer onto a hard disk for storage and conservation can take a significant time as writing speed is usually slow (around 6Mb/s = 0.7Mo/s). Eventually, the long- term storage of teraoctets of data can encounter some limits with a stan- dard hard disk or server capacity. 112 Biodiversity
Regarding the calculation of indices, the larger the file, the slower the analysis process. Even if automated with R scripts, the analysis of thou- sands sound files is time-consuming. This is due to three main factors: i) the number of files to be analysed, ii) the size of each file, and iii) the time taken by R to work with large files. There is no easy way yet to coun- teract these three caveats. The number of files will increase as samples will be larger. The size of each file cannot be reduced. Compressed formats in particular, such as .mp3, cannot be used for obvious reasons of signal quality. The platform R is very convenient as it is free and open-source. It makes it a perfect tool for sharing our research and transferring our tech- niques to other laboratories. However, it may be relevant to look for other software solutions (see section 5.2).
4.3. Energy The SM2 recorder was developed to consume as less energy as possible, but current batteries ensure 240 hours of recording and therefore put a strong limit on the duration of sampling. A solution is to connect the recorder with a 12V battery fuelled with solar energy. However, if such autonomous energy system properly works in sunny areas, it is not adapted to cloudy or shaded areas like the understory of a tropical forest where a very low percentage of solar radiation reaches the ground.
5. What’s next?
5.1. Sampling Our method needs to be tested, validated and eventually applied in sev- eral acoustic conditions from different habitats. So far, we have tested it with both simulated and field acoustic communities (Sueur et al., 2008a; Depraetere et al., 2012). Tests on field communities concerned African tropical coast forest and temperate forest habitats. The latter test implied a modification of the indices to take into account the background noise and the low activity of the acoustic community. We are currently sam- pling several other places including mountain tropical forests in New Caledonia, neotropical evergreen forest in French Guiana, and evergreen monsoon forest in India. We are also transferring the technique to fresh- water habitats by using hydrophones immerged in ponds. One of the aims of our method is to provide a long-term and large-scale sampling. We are currently sampling species diversity with a network of 10-16 sensors working about 40 days long over approximately 16 ha of tropical forest in French Guiana and India. This time period is too short Part II – Chapter 1 113 to track seasonal variations of species diversity. We would like to extend it to at least one year or even longer periods. Moreover, we plan to increase the number of sensors to monitor a larger area. Increasing the sampling time and network size will generate serious storage issues. A cut-off system that stops recordings when the meteorological conditions are not good enough could constitute a nice and cheap solution to overcome this dif- ficulty. Another way could consist in sending directly the data from the recorder to a server through a satellite connection, as wireless connec- tion to a base radio may be too slow for heavy sound files (see IV, 2 sec- tion 2.3). However such technological improvement mainly depends on the industry and may take some time to emerge.
5.2. Improving the indices As explained earlier, background noise is a central issue, and our indices, especially the index H, are particularly sensitive to noise. It is therefore necessary to develop new indices that are noise-resistant. Current research is ongoing in our laboratory to develop a new measurement of the rich- ness based on the frequency peaks of the Fourier spectrum (figure 5). The spectrum can be smoothed or residual peaks due to noise can be filtered out so as to improve the measure in case of rain or wind noise. Amplitude or frequency threshold will be also applied on the envelope and the frequency spectrum respectively, to try to increase the signal-to- noise ratio. Whatever the index in use is, we also need to exactly identify which biodiversity information is collected by using the acoustic com- munity as a proxy of animal diversity. Does the H index only embed a richness-evenness value or does it include phylogenetic and/or functional diversity information? Eventually, as outlined above, the signal analysis can be slow due to R process. Software directly written in C language will be developed on the next years to significantly speed up the analysis process.
5.3 Sharing the method with other scientists and citizens There is an important ethical requirement for making available the bio- acoustic sensor and primary biodiversity data for later uses in terms of knowledge, engineering or conservation (e.g. Graham et al., 2004; Suarez and Tsutsui, 2004). The recording equipment we used so far can be pur- chased to the company Wildlife Acoustics©. The H and D index can be computed with the free R package seewave. The sensor and integrated bio- acoustic system is therefore available to anyone. However, R does not have a user-friendly interface and we plan therefore to share the method soon through an interactive website. Any user will be able to upload recording 114 Biodiversity files for analysis. The acoustic indices will be returned to the user together with an optional graphical representation of the sound analysed (e.g., waveform, envelope, spectrogram, spectrum). On a long-term scale, the recorder and the signal analysis would not be separated but associated in a small and light all-in-one system. This system could be a ‘smartphone’ including a free application that computes the indices. Smartphones were proved to work as nice sensors for mapping the noise level of European cit- ies (Maisonneuve et al., 2010). A similar citizen-science experience could be undertaken to assess animal acoustic diversity inside or around cities.
Authors’ references
Jérôme Sueur, Amandine Gasc, Philippe Grandcolas: Muséum national d’Histoire naturelle, Département Systématique et Évolution, UMR 7205 Paris, France
Sandrine Pavoine: Muséum national d’Histoire naturelle, Département Écologie et Gestion de la Biodiversité, UMR 7204 Paris, France
Corresponding author: Jérôme Sueur, [email protected]
Aknowledgement
This research has been supported by the INEE (CNRS) with a PEPS pro- gram and a PhD grant awarded to AG. Sampling in French Guiana was achieved thanks to a CNRS Nouragues 2010 grant. Sampling in New Caledonia was realised thanks to the ANR BIONEOCAL grant to PG. Main part of research was financed with the FRB BIOSOUND grant (Fondation pour la Recherche sur le Biodiversité). Marion Depraetere, Vincent Devictor, Stéphanie Duvail, Olivier Hamerlynck, Frédéric Jiguet, Isabelle Leviol, Pierre-Yves Martel, and Raphaël Pélissier participated at different steps to the development of this new sensing method or pro- vided field data with which to compare our acoustic indices.
References
Acevedo M.A., Villanueva-Rivera L.J., 2006. Using automated digital recording systems as effective tools for the monitoring of birds and amphibians. Wildlife Society Bulletin, 34, pp. 211-214. Part II – Chapter 1 115
B . rooks D R., McLennan D. A., 1991. Phylogeny, ecology, and behavior: a research program in comparative biology. University of Chicago Press, Chicago, USA. Brumm H., Slabbekoorn H., 2005. Acoustic communication in noise. Advances in the Study of Behavior, 35, pp. 151-209. Buckland S. T., Magurran A. E., Green R. E., Fewster R. M., 2005. Monitoring change in biodiversity through composite indices. Philosophical Transactions of the Royal Society of London B, 360, pp. 243-254. Cano-Santana Z., Romero-Mata, A., Perez-Escobedo H. M., 2008. Modified Petraborg index applied to the sampling of male crickets by aural detection. Journal of Orthoptera Research, 17, pp. 111-112. Depraetere M., Pavoine S., Jiguet F., Gasc A., Duvail S., Sueur, J., 2012. Monitoring animal diversity using acoustic indices: implementation in a temperate woodland. Ecological Indicators, 13, pp. 46-54. Devictor V., Julliard R., Couvet D., Jiguet F., 2008. Birds are tracking climate warning, but not fast enough. Proceedings of the Royal Society of London B, 275, pp. 2743-2748. Farina A., Pieretti N., Piccioli L., 2011. The soundscape methodology for long- term bird monitoring: a Mediterranean Europe case-study. Ecological Informatics, 6, pp. 354-363. Gewin V., 2002. All living things, online. Nature, 418, pp. 362-363. Graham C. H., Ferrier S., Huettman F., Moritz C., Peterson A. T., 2004. New developments in museum-based informatics and applications in biodiversity analysis. Trends in Ecology and Evolution, 19, pp. 497-503. Grandcolas P., 1998. Phylogenetic analysis and the study of community structure. Oikos, 82, pp. 397-400. Hartmann W.M., 1998. Signals, sound, and sensation. Springer, Berlin, Germany. Kerr J.T., Ostrovsky M., 2003. From space to species: ecological applications for remote sensing. Trends in Ecology and Evolution, 18, pp. 299-305. Laiolo P., 2010. The emerging significance of bioacoustics in animal species conservation. Biological Conservation, 7, pp. 1635-1645. Lawton J. H., Bignell D. E., Bolton B., Bloemers G. F., Eggleton P., Hammond P. M., Hodda M., Holt R. D., Larsenk T. B., Mawdsley N. A., Stork N. E., Srivastava D. S., Watt A. D., 1998. biodiversity inventories, indicator taxa and effects of habitat modification in tropical forest. Nature, 391, pp. 72-76. Lüddecke H., Amézquita A., Bernal X., Guzmàn F., 2000. Partitioning of vocal activity in a neotropical highland-frog community. Studies on Neotropical Fauna and Environment, 35, pp. 185-194. Luther D., 2009. The influence of the acoustic community on songs of birds in a neotropical rain forest. Behavioral Ecology, 20, pp. 864-871. 116 Biodiversity
Magurran A. E., 2004. Measuring biological diversity. Blackwell Science Ltd, Malden, USA. Maisonneuve N., Stevens M., ochab b., 2010. Participatory noise pollution monitoring using mobile phones. Information Polity, 15, pp. 51-71. Obrist M. K., Pavan G., Sueur J., Riede K., Llusia D., Márquez R., 2010. Bioacoustics approaches in biodiversity inventories. Abc Taxa, 8, pp. 68-99. Oliver I., beattie A.J., 1993. A possible method for the rapid assessment of biodiversity. Conservation Biology, 7, pp. 562-568. Oliver I., beattie A.J., 1996. Invertebrate morphospecies as surrogates for species: a case study. Conservation Biology, 10, pp. 99-109. Oliver I., Pik A., Britton D., Dangerfield J. M., Colwel, R. K., Beattie A. J., 2000. Virtual biodiversity assessment systems. BioScience, 50, pp. 441-450. Pavoine S., Love M. S., bonsall M. b., 2009. Hierarchical partitioning of evolutionary and ecological patterns in the organization of phylogenetically- structured species assemblages: application to rockfish (genus: Sebastes) in the Southern California Bight. Ecology letters, 12, pp. 898-908. Pavoine S., Bonsall M. B., 2010. Measuring biodiversity to explain community assembly: a unified approach. Biological Reviews, 86, pp. 792-812. Pearson D. L., 1994. Selecting indicator taxa for the quantitative assessment of biodiversity. Philosophical Transactions of the Royal Society of London B, 345, pp. 75-79. Petchey o. L., o’Gorman E., Flynn D. F. b., 2009. A functional guide to functional diversity measures, in: Naeem S., Bunker D., Hector A., Loreau M., Perrings, C. (Eds) Biodiversity, Ecosystem functioning, and Human wellbeing: An Ecological and economic perspective. oxford University Press, Oxford, pp. 49-60. Pieretti N., Farina A., Morri F.D., 2010. A new methodology to infer the singing activity of an avian community: the Acoustic Complexity Index (ACI). Ecological Indicators, 11, pp. 868-873. Qi J., Gage S. H., Wooyeong J., Napoletano B., Biswas, S., 2008. Soundscape characteristics of an environnement. A new ecological indicator of ecosystem health, in: Wei J. (Ed) Wetland and water resource modeling and assessment, a watershed perspective. CRC Press, Boca Raton, pp. 201-211. R Development Core Team, 2012. The R Project for Statistical Computing. Accessed on February 14, 2012, http://www.r-project.org/. Rao C. R., 1964. The use and interpretation of principal component analysis in applied research. Sankhya A, 26, pp. 329-359. Ricketts T. H., Dinerstein E., Olson D. M., Loucks C., 1999. Who’s where in North America? BioScience, 49, pp. 369-381. Riede K., 1993. Monitoring biodiversity: analysis of Amazonian rainforest sounds. Ambio, 22, pp. 546-548. Rumsey F., McCormick T., 1992. Sound and recording: an introduction. Focal Press, Boston, USA. Part II – Chapter 1 117
Skowronski M. D., Harris J. G., 2006. Acoustic detection and classification of Microchiroptera using machine learning: lessons learned from automatic speech recognition. Journal of the Acoustical Society of America, 119, pp. 1817-1833. Stoddard P. K., 1998. Application of filters in bioacoustics, in: Hopp. S. L., Owren M. J., Evans C. S. (Eds) Animal acoustic communication. Springer, Berlin, pp. 105-127. Suarez A. V., Tsutsui N. D., 2004. The value of Museum collections for research and society. Bioscience, 54, pp. 66-74. Sueur J., 2002. Cicada acoustic communication: potential sound partitioning in a multispecies community from Mexico (Hemiptera: Cicadomorpha: Cicadidae). Biological Journal of the Linnean Society, 75, pp. 379-394. Sueur J., Pavoine S., Hamerlynck O., Duvail S., 2008a. Rapid acoustic survey for biodiversity appraisal. PloS One, 3, e4065. Sueur J., Aubin T., Simonis C., 2008b. Seewave: a free modular tool for sound analysis and synthesis. Bioacoustics, 18, pp. 213-226. Turner W., Spector S., Gardiner N., Fladeland M., Sterling E., Steininger M., 2003. Remote sensing for biodiversity science and conservation. Trends in Ecology and Evolution, 18, pp. 306-314. Vane-Wright R. I., Humphries C. J., Williams P. H., 1991. What to protect? – Systematics and the agony of choice. Biological Conservation, 55, pp. 235-254. Vaseghi S.V., 2000. Advanced digital signal processing and noise reduction, second ed. Wiley, Chichester, UK. Wang K., Franklin S.E., Guo X., Cattet M. 2010. Remote sensing of ecology, biodiversity and conservation: a review from the perspective of remote sensing specialists. Sensors, 10, pp. 9647-9667. Weaver J.C., 1995. Indicator species and scale of observation. Conservation Biology, 9, pp. 939-942.
Chapter 2
Assessing the spatial and temporal distributions of zooplankton and marine particles using the Underwater Vision Profiler
Lars Stemmann, Marc Picheral, Lionel Guidi, Fabien Lombard, Franck Prejger, Hervé Claustre, Gabriel Gorsky
1. Introduction
The last two decades, international multidisciplinary programs such as the Census Of Marine Life (COML), Joint GlObal Flux Studies (JGOFS), Global Ocean Ecosystem Dynamics (GLOBEC), Integrated Marine Biogeochemistry and Ecosystem Research (IMBER) conducted numer- ous cruises and sampled large areas of the oceans, often focusing on the first hundred meters of the water column. In parallel, advances in remote sensing technologies from satellites allowed synoptic descriptions of some physical and optical properties of the ocean surface used to assess epipelagic particle biomasses and primary production at a global scale (see III, 3). By contrast, pelagic ecosystems of mesopelagic water layers – also known as mid-water (100-1000m) – and deeper water layers remain widely unknown. Observing these pelagic ecosystems requires the use of large and often costly instruments launched from research vessels such as pumps, multinets, remotely operated vehicles (ROV), or submersibles. Furthermore, fragile zooplankton (ctenophores, medusae, siphonophores, appendicularians) or fragile aggregates are destroyed during collection with plankton nets, in situ water pumps, and/or sediment traps, which pre- vents the analysis of their spatial distribution. This challenge can partly be overcome by using non intrusive underwater optical and imaging tech- nologies that appear to be promising tools for the study and quantifica- 120 Biodiversity tion of zooplankton community structures, diversity, as well as marine particles size spectra. The description of the meso- and bathypelagic fauna began to emerge with the use of ship-tethered cameras hooked on ROV (Lindsay et al., 2004; Lindsay and Hunt, 2005; Robison, 2004; Robison et al., 2005a; Steinberg et al., 1997). However, the deployment of these cameras is time- consuming and financially expensive, which prevents their wide use. Smaller instruments hooked on conventional gears – such as a rosette – or on autonomous platform – such as gliders and profilers (see IV, 1), may be more cost efficient and would provide valuable dataset on the spatial and temporal distributions of organisms and non living particles. Relatively few available instruments allow simultaneous in situ measure- ments of oceanic particles and zooplankton. Particles can be detected and measured by the laser in situ scattering and transmissometry (Agrawal and Pottsmith, 2000) based on scattering intensity. However, this instrument does not provide information on the shape of the particles and limits its use for zooplankton identification. The laser optical plankton coun- ter records a shape approximation of particles crossing an array of light beams and can hardly set one particle apart another among various classes of particles and organisms (Herman et al., 2004). More recently, several instruments that employ image analysis to cha racterise and enumerate oceanic zooplankton have been developed and tested in the field (Benfield et al., 2007), including i) the video plankton recorder (Davis et al., 2005), ii) the shadowed imaged particle profiling and evaluation recorder (Sipper, Samson et al., 2001), iii) the in situ ich thyoplankton imaging system (Isiis, Cowen and Guigand, 2008), and iv) the zooplankton visualisation and imaging system (Zoovis, Benfield et al., 2007). Most of these instruments detect relatively large organisms (more than 100µm); however, there is an increasing interest in quantifying nano- and microplankton particles (Olson and Sosik, 2007; Sosik and Olson, 2007). Several systems using holographic imaging have been developed for this purpose (Alexander et al., 2000; Hobson et al., 1997; Katz et al., 1999; Pfitsch et al., 2007). Whether designed for small or large plankton, all these instruments collect images of a defined volume of water that can be processed to obtain unique information about the distribution, abun- dance, and behaviour of plankton on scales that cannot be investigated by conventional sampling systems such as nets and pumps. Most of the time, these instruments were used to document the in situ behaviour, taxonomic diversity, spatial distribution, and relative abundance of planktons. They were also used independently to study the dynamic of non-living particles in the water column. Ideally, both plankton and non-living particles should be studied simulta- neously because of their interactions in the pelagic realm. These interac- Part II – Chapter 2 121 tions include for example zooplankton feeding on detritus produced at the surface leading to particle aggregation, fragmentation, and reminerali- sation in the water column. These interactions affect the transfer of large amounts of carbon from the surface to the deep sea – a process known as the “biological pump” – and contribute significantly to climate variability (Sarmiento and Le Quere, 1996; Volk and Hoffert, 1985). Therefore, in order to better understand the biological pump, it is crucial to evaluate simultaneously the distribution of the particulate matter and the zoo- plankton in the water column. The underwater vision profilers (UVPs) were designed and constructed in our laboratory at Villefranche-sur-Mer in order to achieve this goal (figure 1). Yet, particle and plankton-imaging systems present new challenges to the studies of aquatic biota. In this paper, we describe the fifth generation of the UVP (UVP5) design and cal- ibrations. Moreover, we expose experimental results from different cruises showing the possibility of studying the biodiversity of zooplankton and the size spectra of particles.
Figure 1: Pictures of the underwater vision profiler UVP4 (A) and UVP5 as stand alone (b) and picture of UVP 5 in a 24 bottles Rosette CTD system (conductivity, temperature and depth, C). UVP4 is a large stand-alone package of nearly 1 m3 (300 kg) and incorporates a CTD, fluorometer and nephelometer sensors (Gorsky et al., 1992; Gorsky et al., 2000). The latest version called UVP5 (Picheral et al., 2010) is a smaller instrument (30kg) that can equip a standard rosette frame, interfaced with the CTD, and used down to 6000m deep instead of 1000m deep for UVP4. 122 Biodiversity
2. Description of the underwater vision profiler (UVP)
2.1. Main characteristics The underwater vision profilers (UVPs) were designed and constructed at the laboratory of Villefranche-sur-Mer to quantify simultaneously large particles (more 100 μm) and zooplankton in a known volume of water (Picheral et al., 2010). The UVP versions 2 to 4 had been operating since 1991 and they provided a database of more than 1300 inter-calibrated profiles of particle size distribution covering the global ocean. However these instruments required dedicated winch time on research ships, their maximum operating depth was 1000m, and the image acquisition at the ocean surface was limited because of daytime light saturation. In addi- tion, their complexity required an onboard trained technician, which lim- ited spreading their use over the oceanographic community. Nowadays, the UVP5 overcomes these limitations and can be set up for short or long-term deployments either as an autonomous system or as a comple- ment to CTD (conductivity, temperature and depth) system. The UVP5 dimensions allow its incorporation into autonomous underwater vehicles (AUV), remotely operated vehicles (ROV), or drifting or geostationary mooring. In the near future, the ongoing miniaturisation of the sensors
Table 1: Underwater Vision Profiler 5 details
Camera housing pressure rated 6000 m Housing 2 independent glass cylinders for the lighting Camera 8 with internal memory storage Data storage Optional external drive 1.3 Megapixelup to 11fps processed images Camera and image 9 mm fixed focal lens analysis Pass band Filter centered on 625 nm Lighting Flash duration down to 100 μs Persistor CF2 piloting processor Piloting board Analog to digital conversion for external sensors Digital to analog output to CTD Power management Connection (camera Serial interface 100Mb network housing) Pressure digital sensor with 0.01% accuracy Embedded Sensors Pitch sensor Internal temperature sensor Rechargeable lithium-ion 6.3 A/29 V battery pack Power Continuous monitored during data acquisition Part II – Chapter 2 123 will lead to the development of autonomous camera systems that could be mounted on drifters and gliders working in network allowing real time “visual” monitoring of the biogeochemistry and the biology of the ocean (see IV, 1). The UVP5 instrumental package contains an intelligent camera and a lighting system encompassed into independent housings (figure 1). In addition, pressure and angle sensors are included to the system in order to monitor the UVP5 deployments and data acquisition. The hardware is also composed of an acquisition and piloting board, internet switch, hard drive, and dedicated electronic power boards whose details and charac- teristics are presented in table 1. Images can be recorded in fields of view ranging from 8 × 6 to 22 × 18cm at a distance of 40cm from the camera in red light environment in order to reduce zooplankton phototactic behav- iour and to prevent contamination by the sunlight at the surface.
2.2 Calibration The manufacturing process of the UVP5 produces light-emitting diodes (LED) lighting systems and glass housings with unique optical character- istics. Therefore, each instrument requires individual calibration. In order to be able to estimate accurate concentrations and sizes of in situ marine particles, calibrations of the water volume and the size of particle within an image have to be done prior to the first deployment. A short descrip- tion of the method is presented below but details can be found in Picheral et al. (2010). The calibration of the volume of the image has to be done independently for each of the two lights. A white sheet of paper, immerged in a tank with seawater, is placed at different distances from the LEDs. Pictures of the light field projected on the white paper are recorded and gathered in order to reconstruct the volume in 3D (Picheral et al., 2010, figure 2C). The size calibration protocol defines the equation and enables the conversion from a particle defined by a number of pixels to size (area) in metric unit. Due to light-scattering in the water, this relationship is not linear for small targets. It follows the rule Sm=A×SpB, where Sp is the surface of the particle in pixels and Sm is the surface in squared-millimetres. The calibration and determination of A and B involves diverse objects sorted into three major qualitative optical groups (dark, transparent, and heterogeneous) in order to represent the diversity of natural particles present in the environment. 124 Biodiversity
2.3. Zooplankton identification Since 2001, the UVP4 and UVP5 have provided images of macrozoo- plankton over the globe. All profiles have been analysed following the same protocol and using custom software routines to extract large objects (i.e. 500µm in maximum length). This size threshold was selected because most of the organisms cannot be identified below that size due to cur- rent insufficient resolution of the images. The sorting of the objects is computer-assisted as for the laboratory Zooscan system (Gorsky, 2010) and the computer prediction is visually validated by specialists to identify taxa. The size of the organisms is reported as well as its area or major and minor axes of the best fitted ellipse. This measure is best suited for dark and opaque organisms such as chaeognathes, radiolarians, fish, and large crustaceans, but cannot be used for gelatinous organisms.
3. Study of particle dynamics and zooplankton community structures at different spatial scales
3.1. Marine particles The UVPs were deployed more than 3000 times covering almost all oceans on Earth (figure 2). The first versions of the UVPs (2 and 3) were not able to efficiently distinguish the non-living particles from the zooplank- tonic organisms. Therefore, earlier studies focused on the size spectra of all particles, assuming that most of them were nonliving particles. This hypothesis was then confirmed by the use of UVPs 4 and 5 showing that zooplanktons account for only 0.1 to 10% of the total number of particles in the water column (see next section). The most important biogeochemical information provided by the UVPs consists on the size spectra of large particles (more than 100µm). These particles, in the form of aggregates of individual particles of different origins, are the main vector of the vertical flux of carbon to the deep sea. In order to correctly estimate this flux, the concentration of particle per size bin (number per centimetre) must be converted to biovolume (cm3.cm-3) and to biomass (mg DryWeight.cm-3) assuming relationships between size and mass (Stemmann et al., 2008a). Then, the known rela- tionship between size and settling speed can be used to estimate vertical flux (Guidi et al., 2008; Stemmann et al., 2004b). The coupling between small and meso-scale (scales from 5km to 100km) physical and biological processes in highly dynamic environments such as frontal zones, filaments, and equatorial systems was shown to influence the spatial patterns of carbon export. Vertical profiles of particle flux can Part II – Chapter 2 125 be analysed in a spatial context in order to provide estimates of carbon sequestration by the oceans at different scales. Previous deployments of the UVPs at high spatial resolution revealed that particle spatial patterns can be observed at scales as small as 10 to 100 km (Gorsky et al., 2002a; Gorsky et al., 2002b; Guidi et al., 2007; Stemmann et al., 2008c). Particle size spectra were also used in time series to constrain mathematical mod- els of particle flux to the interior of the ocean (Stemmann et al., 2004a; 2004b). These analyses led to formulate the hypothesis that zooplankton organisms can detect large settling particles and can fragment them in numerous smaller parts that have slower settling speed. This process may generally affect carbon sequestration in the deep ocean.
Figure 2: Global map showing the location of sites that were studied using the different versions of the UVP (dark blue = UVP2, green = UVP3, light blue = UVP4, red = UVP5).
3.2. Comparison between zooplankton and non living particle size spectra The improvements of the optics and illumination of UVP4 and UVP5 enabled simultaneous estimations of the vertical distributions of both particles and zooplankton size spectra (figure 3). 126 Biodiversity
Figure 3: A. Vertical abundance (relative units) of two size classes of large particulate matter (LPM lines) and vertical day (upper right) and night (upper left) distributions of copepods during the California current ecosystem long-term ecological research (CCELTER) cruise off the Californian coast in autumn 2008. b. Typical UVP5 images of individuals from different macrozooplankton groups including copepoda (1), radiolarian (2), chaetognate (3), medusae (4), appendicularia (5), and euphausid (6). Part II – Chapter 2 127
Acoustic, optical, and imaging systems all face the same challenge when trying to distinguish between plankton and other particles in the water column. Plankton larger than 500µm includes crustacean (e.g. copepods and euphausiids), gelatinous taxa (e.g. medusae, tunicates), and eggs and fish larvae. other particles of the same size range include aggregates, abandoned houses of larvaceans, mucous webs of pteropods and all asso- ciated material, including living (protozoa and bacteria) and dead materi- als. Many of these “other particles” are fragile and are not retained and/ or preserved by filters or nets meshes (Gonzalez-Quiros and Checkley, 2006). Therefore, the contributions of organisms to the total number or the biomass of particles is not well known. Misrecognition between organ- isms and particles can have deep implication for the estimation of avail- able biomass for higher trophic levels and for the estimation of vertical carbon fluxes. The laser optical plankton counter (LOPC) potentially dis- tinguishes automatically zooplankton from particles based on the opacity and size of the recorded objects (Checkley et al., 2008; Gonzalez-Quiros and Checkley, 2006; Jackson and Checkley, 2011). However, results pro- vided by this instrument consist in a proxy for zooplankton since the recognition cannot be validated nor the taxa recognised. The UVP’s dis- tinction is based on the automatic sorting of particles larger than 500µm followed by manual image analysis and visual verification of the plank- ton identifications by experts (Stemmann et al., 2008b; Stemmann et al., 2008d). During the Boum cruise on the Mediterranean Sea (summer 2008), the UVP was deployed on a longitudinal transect from the East to West basin for short-term stations and 3 sites were selected for their oligotrophic cha– racteristic (figure 4). The comparison between particles and zooplankton size spectra for the same size range (500µm-few mm) shows that the dominant zooplankton in abundance were radiolaria. More interesting, the results show almost for the first time that living organisms were only 1-15% of total particles detected by the UVP in the more than 500µm size range. These ratios are slightly lower than those reported earlier for the OPC (25%) and LOPC (20+/-14%) in the Californian Current sys- tem (Gonzalez-Quiros and Checkley, 2006; Jackson and Checkley, 2011). More data of such type should be acquired in different oceans to test whether the strong dominance of non-living particles is a common feature of pelagic ecosystem. 128 Biodiversity
Figure 4: Particles and zooplankton normalised number spectra obtained by the underwater vision profiler at 3 locations in the Eastern (left), Central (middle) and Western (right) Mediterranean Sea during the BOUM cruise in July 2008 (adapted from Stemmann and Boss, 2012). Particles were counted automatically from 60µm in equivalent spherical diameter (ESD) and thus include non living particles and zooplankton organisms. The different taxa were counted manually on the images only for size larger than 500µm from which they can be identified.
3.3. Appendicularians and the biological pump Appendicularians are zooplanktonic pelagic tunicates. They produce a mucous external filtration device called “the house” which allow them to filter small particles (0.2-50µm, see Lombard et al., 2011) from the seawater. Up to 26 houses can be produced within a day by a single indi- vidual (Sato et al., 2003), and once clogged, are discarded contributions to marine snow (Alldredge, 2005; Alldredge and Silver, 1988). Thus, the biogeochemical action of appendiculiarians includes mostly “repackag- ing” by filtering small particles and producing large ones. This effect on the biogeochemistry of particles and therefore on carbon fluxes was shown to be potentially important (Berline et al., 2011; Robison et al., 2005b). However, these organisms have been largely understudied until now mainly because of instrument limitations. Imaging systems such as the UVP overcome these limitations and provide simultaneous observations of their distribution and relation to particle stocks and fluxes. Appendicularians repackaging action were estimated from observations in the northeastern Atlantic ocean by the UVP4. Combined data of appendicularians and associated fluxes from UVP observations and from sediment traps suggested that the estimated pro- Part II – Chapter 2 129 duction of particulate matter by sub-surface appendicularians exceeded the observed total sinking flux at 200m (Lombard et al., 2010). This study supports the hypothesis that appendicularians play an important role in the production of particle fluxes (Alldredge, 2005). In addition, labora- tory observation on discarded houses showed that empty appendicular- ian houses undergo a rapid deflation and compression process, decreasing their size and increasing their sinking speed (Lombard and Kiørboe, 2010). This process, combined with the previous estimation of discarded houses production, leads to the conclusion that up to 20-40% of the 300-500µm particles observed by the UVP in the upper 100m of the water column may be of appendicularian origin. In addition to producing discarded houses in the epipelagic layers, appendicularians are also supposed to be efficient at repackaging small particles by grazing into larger aggregates (more than 1mm) in the deep ocean (Alldredge, 2005). Using the UVP4 observations, the relation- ship between the changes in the vertical distributions of particles and zooplankton, including appendicularians, was investigated during the Mareco cruise in the North Atlantic (Stemmann et al., 2008b). The gelatinous fauna were consistently the most numerous between 400- 900m and in particular the appendicularians, that occurred mostly below 300m (figure 5). Particles vertical profiles showed that the equiv- alent spherical volume of particles (100µm Figure 5: Vertical distribution of appendicularians (upper panel, bars are mean abundance and the stems are the standard deviation) and particles (lower panel, 100µm 3.4. Macrozooplankton spatial distribution in the mesopelagic layer The mesopelagic layer of oceans is located between the photic zone (the illuminated surface zone, where light penetrates the water down to a depth of 100m) and a depth of 1000m. It is bathed in half-light, which is why it is often referred to as the “twilight zone”. The mesopelagic zone represents one of the largest habitat on Earth, yet it is still widely unknown, especially when it comes to its biological composition. Since 2001, we have studied the in situ vertical (0-1000m) distribution of macro- zooplankton during 12 cruises in 6 oceans (Mediterranean Sea, North Atlantic shelves, Mid-Atlantic ridge, tropical Pacific ocean, eastern Indian ocean, and sub-Antarctic ocean). Nine regions were identified based on the hydrological properties of the water column. They corre- spond to nine of the biogeochemical provinces defined by Longhurst (1995). Part II – Chapter 2 131 We tested if the zoogeography of macrozooplankton in the mesopelagic layer corresponds to these biogeochemical provinces (Stemmann et al., 2008d). The zooplankton community was sorted in 21 morphotypes and more than 5000 organisms were identified in the 100-1000 m depth layer. The numerically dominant groups were crustaceans (24%) followed by the medusae (18%), appendicularians (14%) chaetognathes (11%), fish (7%) and single-cell sarcodines of the group Star (6%, see figure 6). The other taxo- nomic groups were less than 5% of the total count each. However, pooling all single-cell sarcodines moved this group to second rank (23%) in term of frequency of occurrence. From a trophic perspective, the assemblages Figure 6: Frequency of occurrence for the 20 taxonomic groups in the 9 regions. Note that the numerically dominant group of Crustacean has been removed from the list to increase the details in the other groups. Appendicularians (App.), Thaliacae (Thal.), Fish, Haliscera spp. medusa (Hal.), S. bittentaculata (Sol.), Aglantha spp. (Agl.), Aeginura grimaldii (Grim.), “other medusae” (Med.), chaetognath (Chaet.), lobate ctenophore (Lob.), cydippid ctenophore (Cyd.), siphonophore (Sipho.), single-cell sarcodine grouped by four (Radio CS.), colonial radiolarians (Radio C.), colonial radiolarians with double line (Radio CD.), Phaedorian (Phaeo.), single-cell sarcodine with spines (Spine.), double-cell sarcodine with spines (Spine 2.), spheres (Sphere.), and sarcodine with hairs (Star.). The regions are defined as: Northeast Atlantic shelves (NECS), Atlantic Arctic (ARCT), North Atlantic drift (NADR), Atlantic Subarctic (ARC), Subantarctic, ocean (SANT), North Atlantic Subtropical ocean, (NAST), South Pacific Subtropical Gyre (SPSG), Western Australia (AUSW), Mediterranean Sea (MEDI). The order of the region is set so the proportion of carnivorous organisms (in grey from Chaet. to Sipho.) decreases from left to right (modified from Stemmann et al., 2008). 132 Biodiversity of zooplankton could be lumped into three categories: gelatinous carni- vores (cydippid stenophores, lobate ctenophores, medusae, siphonophores, chaetognathes), filter feeder detritivores (appendicularians and salps) and omnivores (sarcodines, crustaceans and fish). Interestingly, the proportion of carnivores decreased from 95% to 15%, from the high latitude regions (Northest Atlantic shelves, Atlantic Arctic, North Atlantic drift, Atlantic Subarctic, Subantarctic ocean) to the low latitude regions (Mediterranean sea, western Australia, South Pacific subtropical Gyre). The similarity in the community assemblages of zooplankton in the layer between 100 and 1000m was significantly higher within regions than between regions, for most cases. The regions with comparable compositions were located in the North Atlantic with adjacent water masses, suggesting that the assemblages were either mixed by advective transport or that environmental conditions were similar in mesopelagic layers. The data suggest that the spatial struc- turing of mesopelagic macrozooplankton occurs at large scales (e.g. basin scales) but not necessarily at smaller scales (e.g. oceanic front). 4. Conclusion Results obtained using the UVP but also several other in situ imaging instru- ments have shown that bio-imagery techniques can provide useful data on plankton and particles spatial and temporal distribution in the upper kilo- metre of the ocean. In the next decade, rapid technological evolution toward miniaturisation in the optical sensors is expected, and will make possible the use of these sensors on autonomous platforms. Their extensive use may set a revolution in ocean plankton sciences equivalent to the revolution in medical practices for the last 15 years. Broader spatial and longer temporal coverage of plankton size spectra will soon be possible for global moni- toring programs (see chapter IV, 1). Mathematical models for individual physiological and population change rates, biomasses flow between trophic levels, and functions of organisms or particle size, were also developed in the last decade. The new sets of data obtained by the wide use of imaging instruments are well adapted to calibrate and validate these models. Authors’ references Lars Stemmann, Marc Picheral, Franck Prejger, Hervé Claustre, Gabriel Gorsky: Université Pierre et Marie Curie, Laboratoire d’Océanographie de Villefranche, UPMC-CNRS UMR 7093, Villefranche-sur-Mer, France. Part II – Chapter 2 133 Lionel Guidi: University of Hawaii, Department of Oceanography, C-MORE, Center for Microbial Oceanography: Research and Education, Honolulu, USA Fabien Lombard: Université de la Méditerranée, Laboratoire d’Océanographie Physique et Biogéochimique, UMR 6535, Campus de Luminy, Marseille, France Corresponding author: Lars Stemmann, [email protected] Aknowledgement The authors would like to thank the many colleagues who helped us to develop our knowledge on plankton and particles dynamics as detected using their optical properties. Lars Stemmann was supported by funding from the 7th European Framework Program (JERICO) and by the PICS program of the CNRS. Fabien Lombard was supported by the French pro- gram ANR-10-PDOC-005-01 ‘Ecogely’. Lionel Guidi was supported by Center for Microbial Oceanography, Research and Education (C‑MORE; NSF grant EF-349 0424599), the HOT program (NSF grant OCE09‑26766) and the Gordon and Betty Moore Foundation (P.I. Pr David M. Karl). References Agrawal y. C., Pottsmith H. C., 2000. Instruments for particle size and settling velocity observations in sediment transport. Marine Geology, 168, pp. 89-114. Alexander S., Anderson S., Hendry D. C., Hobson P. R., Lampitt R. S., Lucas- Leclin B., Nareid H., Nebrensky J. J., Player M. A., Saw K., Tipping K., Watson J., Craig G., 2000. HoloCam: a subsea holographic camera for recording marine organisms and particles. SPIE Proceedings series: Optical Diagnostics for Industrial Applications, 4076, pp. 111-119. Alldredge A. L., 2005. The contribution of discarded appendicularian houses to the flux of particulate organic carbon from oceanic surface waters, in: Gorsky, G. Youngbluth M. J. (Eds.), Response of Marine Ecosystems to Global Change: Ecological Impact of Appendicularians. Gordon and Breach scientific publishers, Paris, pp. 309-326. Alldredge A. L., Silver M. W., 1988. Characteristics, dynamics and significance of marine snow. Progress in Oceanography, 20, pp. 41-82. Benfield M. C., Grosjean P., Culverhouse P. F., Irigoien X., Sieracki M. E., Lopez- Urrutia A., Dam H. G., Hu Q., Davis C. S., Hansen A., Pilskaln C. H., 134 Biodiversity Riseman. E M., Schultz H., Utgoff P. E., Gorsky G., 2007. Rapid: Research on automated plankton identification. Oceanography, 20, pp. 172-187. Berline L., Stemmann L., Vichi M., Lombard F., Gorsky G., 2011. Impact of appendicularians on detritus and export fluxes: a model approach at DyFAMed site. Journal of Plankton Research, 33, pp. 855-872. Checkley D. M., Davis R. E., Herman A. W., Jackson G. A., 2008. Assessing plankton and other particles in situ with the SOLOPC. Limnology and Oceanography, 53, pp. 2123-2136. Cowen R. K., Guigand C. M., 2008. In situ ichthyoplankton imaging system (Isiis): system design and preliminary results. Limnology and Oceanography: Methods, 6, pp. 126-132. Davis C. S., Thwaites F. T., Gallager S. M., Hu, Q., 2005. A three-axis fast-tow digital Video Plankton Recorder for rapid surveys of plankton taxa and hydrography. Limnology and Oceanography: Methods, 3, pp. 59-74. Gonzalez-Quiros R., Checkley D. M., 2006. occurrence of fragile particles inferred from optical plankton counters used in situ and to analyze net samples collected simultaneously. Journal of Geophysical Research – Oceans, 111, pp. 5-6. Gorsky G., Aldorf C., Kage M., Picheral M., Garcia Y., Favole J., 1992. Vertical distribution of suspended aggregates determined by a new underwater video profiler, in: Nival P., Boucher J., Bhaud M. (Eds.), 3ème Colloque du Programme National sur le Déterminisme du Recrutement, Nantes (France), 1-3 Oct 1991. Annales de l’Institut océanographique, 68, pp. 275-280. Gorsky G., Ohman M. D., Picheral M., Gasparini S., Stemmann L., Romagnan J.-B., Cawood A., Pesant S., Garcia-Comas C., Prejger F., 2010. Digital zooplankton image analysis using the ZooScan integrated system. Journal of Plankton Research, pp. 285-303. Gorsky G., Picheral M., Stemmann L., 2000. Use of the underwater video profiler for the study of aggregate dynamics in the North Mediterranean. Estuarine Coastal and Shelf Science, 50, pp. 121-128. Gorsky G., Prieur L., Taupier-Letage I., Stemmann L., Picheral M., 2002b. Large particulate matter in the Western Mediterranean – LPM distribution related to mesoscale hydrodynamics. Journal of Marine Systems, 33, pp. 289-311. Guidi L., Jackson G. A., Stemmann L., Miquel J., Picheral M., Gorsky G., 2008. Relationship between particle size distribution and flux in the mesopelagic zone. Deep Sea Research Part I: oceanographic Research Papers, 55, pp. 1364-1374. Guid L., Stemmann L., Legendre L., Picheral M., Prieur L., Gorsky G., 2007. Vertical distribution of aggregates (> 110 µm) and mesoscale activity in the northeastern Atlantic: effects on the deep vertical export of surface carbon. Limnology and Oceanography, 52, pp. 7-18. Herman A. W., beanlands b., Phillips E. F., 2004. The next generation of Optical Plankton Counter: the Laser-OPC. Journal of Plankton Research, 26, pp. 1135-1145. Part II – Chapter 2 135 Hobson. P R., Krantz E. P., Lampitt R. S., Rogerson A., Watson J., 1997. A preliminary study of the distribution of plankton using hologrammetry. Optics and Laser Technology, 29, pp. 25-33. Jackson G. A., Checkley D. M., 2011. Particle size distributions in the upper 100m water column and their implications for animal feeding in the plankton. Deep-Sea Research Part I: Oceanographic Research Papers, 58, pp. 283-297. Katz J., Donaghay P., Zhang J., King S., Russell K., 1999. Submersible holocamera for detection of particle characteristics and motions in the ocean. Deep-Sea Research Part I: oceanographic Research Papers, 46, pp. 1455-1481. Lindsay D. J., Furushima Y., Miyake H., Kitamura M., Hunt J. C., 2004. The scyphomedusan fauna of the Japan Trench: preliminary results from a remotely-operated vehicle. Hydrobiologia, 530-531, pp. 537-547. Lindsay D. J., Hunt J. C., 2005. Biodiversity in midwater cnidarians and ctenophores: submersible-based results from deep-water bays in the Japan Sea and north-western Pacific. Journal of the Marine Biological Association of the United Kingdom, 85, pp. 503-517. Lombard F., Kiørboe T., 2010. Marine snow originating from appendicularian houses: Age-dependent settling characteristics. Deep-Sea Research Part I: Oceanographic Research Papers, 57, pp. 1304-1313. Lombard F., Legendre L., Picheral M., Sciandra A., Gorsky G., 2010. Prediction of ecological niches and carbon export by appendicularians using a new multispecies ecophysiological model. Marine Ecology Progress Series, 398, pp. 109-125. Lombard F., Selander E., Kiørboe T., 2011. Active prey rejection in the filter- feeding appendicularian Oikopleura dioica. Limnology and Oceanography, 56, pp. 1504-1512. Longhurst A., 1995. Seasonal cycles of pelagic production and consumption. Progress in Oceanography, 36, pp. 77-167. Olson R. J., Sosik H. M., 2007. A new submersible imaging-in-flow instrument to analyze nano-and microplankton: Imaging FlowCytobot. Limnology and Oceanography: Methods, 5, pp. 195-203. Pfitsch D. W., Malkiel E., Takagi M., Ronzhes Y., King S., Sheng J., Katz J., 2007. Analysis of in situ microscopic organism behavior in data acquired using a free-drifting submersible holographic imaging system. Oceans, vols 1-5, pp. 583-590. Picheral M., Guidi L., Stemmann L., Karl D. M., Iddaoud G., Gorsky G., 2010. The Underwater Vision Profiler 5: An advanced instrument for high spatial resolution studies of particle size spectra and zooplankton. Limnology and Oceanography: Methods, 8, pp. 462-473. Robison B. H., 2004. Deep pelagic biology. Journal of Experimental Marine Biology and Ecology, 300, pp. 253-272. Robison B. H., Raskoff K. A., Sherlock R. E., 2005a. Ecological substrate in midwater: Doliolula equus, a new mesopelagic tunicate. Journal of the Marine Biological Association of the United Kingdom, 85, pp. 655-663. 136 Biodiversity Robison B. H., Reisenbichler K. R., Sherlock R. E., 2005b. Giant larvacean houses: Rapid carbon transport to the deep sea floor. Science, 308, pp. 1609-1611. Samson S., Hopkins T., Remsen A., Langebrake L., Sutton T., Patten J., 2001. A system for high-resolution zooplankton imaging. IEEE Journal of Oceanic Engineering, 26, pp. 671-676. Sarmiento J. L., Le Quere C., 1996. Oceanic carbon dioxide uptake in a model of century-scale global warming. Science, 274, pp. 1346-1350. Sato R., Tanaka Y., Ishimaru T., 2003. Species-specific house productivity of appendicularians. Marine Ecology Progress Series, 259, pp. 163-172. Sosik H. M., olson R. J., 2007. Automated taxonomic classification of phytoplankton sampled with imaging-in-flow cytometry. Limnology and Oceanography: Methods, 5, pp. 204-216. Steinberg D. K., Silver M. W., Pilskaln C. H., 1997. Role of mesopelagic zooplankton in the community metabolism of giant larvacean house detritus in Monterey bay, California, USA. Marine Ecology Progress Series, 147, pp. 167-179. Stemmann L., Boss E., 2012. Plankton and particle size and packaging: From determining optical properties to driving the biological pump. Annual Review of Marine Science, 4, in press. Stemmann L., Eloire D., Sciandra A., Jackson G.A., Guidi L., Picheral M., Gorsky G., 2008a. Volume distribution for particles between 3.5 to 2000µm in the upper 200m region of the South Pacific Gyre. Biogeosciences, 5, pp. 299-310. Stemmann L., Hosia A., Youngbluth M.J., Soiland H., Picheral M., Gorsky G., 2008b. Vertical distribution (0-1000 m) of macrozooplankton, estimated using the Underwater Video Profiler, in different hydrographic regimes along the northern portion of the Mid-Atlantic ridge. Deep Sea Research Part II: Topical Studies in Oceanography, 55, pp. 94-105. Stemmann L., Jackson G. A., Gorsky G., 2004a. A vertical model of particle size distributions and fluxes in the midwater column that includes biological and physical processes – Part II: application to a three year survey in the NW Mediterranean Sea. Deep Sea Research Part I: oceanographic Research Papers, 51, pp. 885-908. Stemmann L., Jackson G. A., Ianson D., 2004b. A vertical model of particle size distributions and fluxes in the midwater column that includes biological and physical processes – Part I: model formulation. Deep-Sea Research Part I: Oceanographic Research Papers, 51, pp. 865-884. Stemmann L., Prieur L., Legendre L., Taupier-Letage I., Picheral M., Guidi L., Gorsky G., 2008c. Effects of frontal processes on marine aggregate dynamics and fluxes: An interannual study in a permanent geostrophic front (NW Mediterranean). Journal of Marine Systems, 70, pp. 1-20. Stemmann L., Robert K., Hosia A., Picheral M., Paterson H., Youngbluth M. J., Ibanez F., Guidi L., Gorsky G., Lombard F., 2008d. Global zoogeography Part II – Chapter 2 137 of fragile macrozooplankton in the upper 100-1000 m inferred from the underwater video profiler. ICES Journal of Marine Science, 65, pp. 433-442. Volk T., Hoffert M. I., 1985. Ocean carbon pumps: analysis of relative strengths and efficiencies in ocean-driven atmospheric CO2 changes, in: Sundquist E. T., Broecker W. S. (Eds.), The Carbon Cycle and Atmospheric CO2: Natural Variations Archean to Present. Geophysical monography series, 32, pp. 99-110. Chapter 3 Assessment of three genetic methods for a faster and reliable monitoring of harmful algal blooms Jahir Orozco-Holguin, Kerstin Töbe, Linda K. Medlin 1. General introduction Harmful marine algal taxa are globally distributed from tropical to polar latitudes, and occupy ecological niches ranging from brackish water, such as the Baltic Sea, to oceanic environments. It is generally acknowledged that the occurrence of harmful algal blooms is increasing and that pollu- tion in coastal waters has contributed to this increase. Harmful algae can produce secondary compounds that are toxic to fish and shellfish (e.g. oysters, mussels) who feed on toxic algae. These compounds can cause amnesic, paralytic or diarrhetic traumas when fish and shellfish are eaten by humans (Hallegraeff, 1993). To mitigate health problems for human populations and negative effects to fisheries, aquaculture and tourism, accurate and cost-effective systems for identification and detection of toxic algae are urgently required. The monitoring of harmful algal blooms (HABs) is an European Union requirement and usually relies on visual confirmation of water discol- oration, fish kills, and laborious cell counts. These techniques are very time-consuming, require specialised or trained personnel, expensive equipment, and are ineffective when many samples have to be routinely analysed. Currently, up to five working days may be needed between specimen collection and the delivery of a diagnostic report of the species present, leaving little time for mitigation responses, which usually involve moving the caged pinfish or the mussel rafts to a new location. Other 140 Biodiversity mitigation efforts, such as precipitation of the bloom with clay particles, are not practised in Europe. Molecular methods are potentially faster and more accurate than tradi- tional light microscopy methods, and have been used for identification of phytoplankton (Ayers et al., 2005; Diercks et al., 2008a; 2009; Gescher et al., 2008; Greenfield et al., 2006; O’Halloran et al., 2006). Recently, the analysis of small-subunit (SSU) ribosomal RNA (rRNA) genes has been established as an efficient way to characterise complex microbial samples (Amann et al., 1990). The method has proven to be of special value for the analysis of picophytoplankton samples, which are difficult to monitor by conventional methods because of their small size (Moon-van der Staay et al., 2001). This method also circumvents the selective step of laboratory cultivation (Giovannoni et al., 1990). Direct cloning and sequencing of the small subunit (SSU) ribosomal DNA (rDNA) from natural samples provide a broader view of the structure and composition of communi- ties (López-Garcia et al., 2001). Because the rRNA database is increasing continually, it is possible to design probes from higher taxonomic groups down to the species level (Guillou et al., 1999; Groben et al., 2004). The species-specific probes can be applied for the analysis of phytoplankton communities with detection by flow cytometry, epifluorescence micros- copy (Miller and Scholin, 1998; Lim et al., 1999), or other methods that take advantage of the hybridisation principle (Metfies and Medlin, 2004). These well-established approaches have the major disadvantage that they can only be used to identify one or a few organisms at a time, which makes it very time-consuming to get a broad view of a microbial sample. Thus, other methods have been developed in which multiple species can be identified at the same time (Metfies and Medlin, 2004). The identification of a region of the rRNA molecule that is species or group specific is essentially a genetic barcode. However most people work- ing in the barcode field do not take into consideration how they could possible apply their barcode in a real life situation. For monitoring pur- poses, rRNA barcodes are put to use, either as a probe for hybridisation to a target or as a primer to amplify the target in a PCR reaction (see below). In designing a barcode, the position of the mismatch must be taken into account if using the barcode as a probe (mismatch in the mid- dle) or as a primer (mismatch at the 3’ end) is chosen. The use of rRNA probes in fluorescence in situ hybridisation (FISH) reactions has often been used for identification of harmful microalgal species in field samples (e.g., Groben and Medlin, 2005), although it is not regularly included in monitoring programmes. FISH enables the direct visualisation of target cells by epi-fluorescence microscopy or by automated cytometric tech- niques. The whole cell stays intact and co-occurring phytoplankton spe- cies can be discriminated when counterstained with an overall DNA stain. Part II – Chapter 3 141 However, weak probe penetration, loss of cells during preparatory steps and autofluorescence of the target cells can mask the fluorescence signal. Thus, an unambiguous species differentiation may be difficult to achieve. Moreover, an extensive analysis of environmental samples with FISH is very time consuming, thus being inadequate to achieve the high sample throughput needed in routine monitoring programs (Touzet et al., 2009). The limited number of fluorochromes available also restricts the use of multiple probes at one time. Several cell free formats (DNA only) are cur- rently available and have overcome the problems associated with FISH and the whole cell format. These include real-time quantitative polymer- ase chain reaction (qPCR), biosensors and microarrays. In the following sections, we will address issues related to the assessment of these labora- tory methods for the monitoring of toxic algae and their possible use in automated devices and in situ monitoring programs of HAB. The moni- toring of aquatic pathogens is essential to guarantee the safety and health of aquatic resources and the application of cell free methods to routine monitoring programs offers the best solution to assess good environmen- tal status of all waters in a rapidly changing environment. 2. Quantitative polymerase chain reaction-based method 2.1. General principles The polymerase chain reaction (PCR) is one of the most powerful tech- nologies in molecular biology. With PCR specific sequences, the number of DNA target molecules is amplified exponentially with each ampli- fication cycle. The direct sequence amplification in PCR approaches enables the detection of low abundance targets and the detection of “hidden” DNA like target species consumed by a predator. However, an adverse aspect of PCR is the impossibility to visualise the target species to ensure unequivocally their presence in the sample and to assess any cross-reaction to any non-target organism (Kudela et al., 2010). With traditional qualitative “endpoint” PCR, no information about the quan- tity of starting material in the sample is available. Whereas in qPCR approaches, data are collected over the entire PCR cycle by using flu- orescent markers that are incorporated into the PCR product during amplification. The quantity of the amplified product is proportional to the fluorescence generated during each cycle, which is monitored with an integrated detection system during the linear exponential phase of the PCR (Saunders, 2004). The accumulation of the PCR amplicon is measured as a change in fluorescence and is directly proportional to the amount of starting material (see figure 1). 142 Biodiversity Figure 1: Amplification plot of 28S rDNA from Alexandrium tamarense in two different environmental samples from the Scottish East Coast (TaqMan approach). The excited fluorescence is plotted against the cycle number. The delta Rn is the magnitude of the signal generated by the given PCR conditions relative to a standard. The qPCR can single out base pair differences: thus closely related spe- cies or populations can be distinguished. For environmental samples, an external standard for quantifying the amplified DNA is used. This could be a dilution of plasmids or DNA derived from laboratory cultures with a known concentration of the target template. To infer concentrations of the target species in an unknown sample, a standard curve must be made for each target species because of differences in DNA content per cell (Handy et al., 2008). When using a plasmid standard for quantifying cell numbers, it is essential to know the copy number of the ribosomal gene of the target species. In addition, one should take into account that the copy number of the rDNA genes may vary among different strains of an organism and species (Erdner et al., 2010). The most common used qPCR method is the SybR Green approach. In this assay, the fluorescent dye SYBR Green binds to the minor groove of double stranded DNA (dsDNA), which results in an increase of the fluo- rescent emission proportional to an increase in the dsDNA PCR ampli- con formation after each cycle. Proper primer design is critical to avoid primer-dimers, which would be counted as amplified DNA because of the unspecific binding of SybR Green to all dsDNAs. Therefore, a melting curve analysis is performed that identifies primer-dimers by their lower Part II – Chapter 3 143 melting temperature compared to that of the target amplicon (Nolan, 2004). In other more sensitive and specific qPCR approaches, specific or non-specific primers together with a specific fluorigenic oligonucle- otide probe are applied (e.g., TaqMan approach, molecular beacon and hybridisation probe assay). These assays apply fluorescence resonance energy transfer (Fret), which is the transfer of energy from an excited fluorophore, the donor, to another fluorophore, the acceptor, to generate enhanced fluorescence upon binding of the specific probe to its target (Cardullo et al., 1988). The use of specific primers and oligonucleotide probes, labelled with unique fluorescent dyes with different excitation wavelength, enables a rapid and quantitative enumeration of several organisms within one sample (multiplex PCR). The number of detectable target genes in one sample is limited by the number of available fluorescence reporter dyes for the separate probes. Consequently the detection is limited to six spe- cies in one sample. However, multiplex qPCR experiments have to be carefully optimised, and often require an elaborate adaptation, notably with increasing target species in one assay (Kudela et al., 2010). Potential drawbacks and limitations of qPCR could be, for instance, different DNA extraction yields depending on the extraction method used and the pres- ence of humic substances that could influence the PCR reaction. These problems could be resolved or at least minimised by applying a high- quality DNA isolation method. Quantitative PCR can be easily performed immediately after in situ sam- pling onboard ship or on shore, but preserved samples can also be used. However, the preservation method can influence the results or even inhibit the reaction. The sensitivity of qPCR is considerably lower with formalin and glutaraldehyde preservation than with no preservation. Preservation using ethanol and freezing is preferred because it is still pos- sible to detect and quantify target cells from fixed field samples after three years (Hosoi-Tanabe and Sako, 2005). Another commonly used fixative for phytoplankton samples is Lugol’s iodine, which has been reported to lower the sensitivity of some qPCR experiments (Bowers et al., 2000) but has also been successfully applied in others (Kavanagh et al., 2010). In HAB studies, multiplex qPCR experiments are applied less frequently, because of the extensive required optimisations to apply different primers and/or probes together in one environmental sample. Handy et al. (2006) successfully tested multiprobing using a single primer set with species spe- cific probes in one assay versus multiplexing using specific primers and specific probes. They found that multiplexing was more efficient, albeit both methods were successful in detecting multiple raphidophyte species. More recently a new technology for qPCR has emerged. It is termed drop- let qPCR and involves the Illumina® or 454 sequencing method. Tewhey 144 Biodiversity et al. (2009) were able to perform 1.5 million PCR with primers targeting 435 exons of 47 genes to screen genetic variation in large human popula- tions. In this method, the genomic DNA template mixture contains all of PCR components except for the primers. The template is prepared by fragmenting genomic DNA using DNaseI to produce 2-4kb fragments. The template mixture is made into droplets and paired with primer pair droplets and both droplets enter the microfluidic chip at a rate of about 3,000 droplets per second. As the primer pair droplets are smaller than the template droplets, they move faster through the channels until they contact the preceding template droplet. Field-induced coalescence of these droplet pairs results in the two droplets merging to produce a single PCR droplet, which is collected and processed as an emulsion PCR reaction (Tewhey et al., 2009). 2.2. Case studies of harmful algae Quantitative PCR has been used as a sensitive and accurate alternative to microscopic cell counts for estimating changes in cell densities of harm- ful algal species in natural phytoplankton samples. In particular, qPCR enables the differentiation between morphologically similar species, such as the dinoflagellate Cryptoperidiniopsis brodyi (Steidinger et al., 2006), which co-occurs with Pfiesteria species and is indistinguishable by light microscopy, but is easily identified by using qPCR (Park et al., 2007). Another benefit over microscopic counts is the sensitive enumeration and identification of fragile species, which might not be easily preserved, such as raphidophytes (Handy et al., 2008) or species that can only be reliably identified with electron microscopy, such as Prymnesium cells. Multiplexing has also been successfully applied to detect the harmful spe- cies Prymnesium parvum (Manning and La Claire, 2010). Moreover, qPCR can be much faster and more reliable than traditional counting methods and cryptic species can be more easily identified (Manning and La Claire, 2010). Therefore, qPCR has become a standard method in detecting harmful algae (Fitzpatrick et al., 2010). The target genes for the primers and probes in HAB qPCR applications are the internal transcribed spacer I-5.8S rRNA gene, or the 18S/28S rRNA gene, depending on the DNA sequence divergence between closely related spe- cies. A variety of primers targeting these ribosomal genes are hitherto avail- able for the detection and identification of various HAB species in qPCR applications (table 1). Part II – Chapter 3 145 Table 1: Summary of qPCR studies for the detection of HAB species and the toxins they produce. QPCR On Midtal Taxon Toxins Study Approach Phylochip Galluzzi et al. Genus Alexandrium SYBR Green Yes 2004 Toxic North American clade of Dyhrman et the A. catenella/ Saxitoxin SYBR Green Yes al. 2006 and fundyense /tamarense 2010 species complex Unknown Zhang and Pfiesteria shumwayae SYBR Green No fish killer Lin 2005 Zamor et al. Prymnesium parvum prymensins SYBR Green Yes 2011, Galluzzi et al. 2008 Genus Pseudo- Domoic Fitzpatrick et SYBR Green Yes nitzschia acid al. 2010 Cysts of toxic North American clade of Erdner et al. the A. catenella/ Saxitoxins SYBR Green Yes 2010 fundyense /tamarense species complex Unknown Taqman Bowers et al. Pfiesteria species No fish killer probes 2000 Toxic North American and Temperate Asian Hosoi-Tanabe Taqman clade of the A. Saxitoxin Yes and Sako probes catenella/fundyense 2005 /tamarense species complex A. minutum, A. ostenfeldii, A. tamutum, Mediterranean, North American and Saxitoxins Taqman Töbe et al. Western European and Yes probes unpublished ribotypes of the A. sprilloids catenella/ fundyense /tamarense species complex from European waters Cysts of the Temperate Asian ribotype of A. Taqman Kamikawa et Saxitoxins Yes catenella/ fundyense/ probes al. 2007 tamarense species complex Table 1 – to be continued 146 Biodiversity Table 1 – continued QPCR On Midtal Taxon Toxins Study Approach Phylochip Bowers et al. 2006, Coyne et al. Harmful Unknown Taqman Yes 2005, Handy raphidophytes Fish killer probes et al. 2006, Kamikawa et al. 2006 Okadaic Hybridization Kavanagh et Dinophysis species Yes acid probes al. 2010 Hybridization Touzet et al. A. minutum Saxitoxins Yes probes 2009 Lingulodinium Hybridization Moorthi et al. Not toxic No polyedrum probes 2006 2.3. Future prospects Multiplex qPCR assays should be improved for routine testing in HAB studies, with several probes recognising different HAB species in one single environmental sample, in order to accelerate the identification of several species and lower substantially analysis costs and time per sam- ple. Further development of primer and probes for HAB species and the application of the variety of available probes will alleviate the deployment of the method, circumventing long primer and probe testing procedures. Costs of real-time PCR instruments are decreasing, so that in future qPCR instruments will be standard tools in HAB studies. New high throughput technologies, such as the open array technology using the qPCR method, are available. Here, the benefits from microarrays and the data quality of PCR are combined. The open array is a new nanoliter fluidics platform for low volume solution phase reactions, which enables the analysis of thou- sands of samples in parallel. The application of such a high throughput method will also considerably alleviate the analysis of large amounts of environmental samples in HAB studies in future. In addition, qPCR has now been adapted for use in a buoy (Preston et al., 2011) and it is only a matter of time until qPCR primers for toxic algae are added because the environmental sample processor buoy is already capable of detecting toxic algae using a sandwich hybridisation method with chemiluminescence detection (see section 3.3 below). Part II – Chapter 3 147 3. Electrochemical biosensor-based methods 3.1. General principles DNA (RNA)-based biosensors have been used in numerous fields rang- ing from medical diagnosis to forensic and environmental research (Heidenreich et al., 2010; Liu et al., 2010). An electrochemical biosensor is a self-contained integrated device capable of providing specific (semi)- quantitative analytical information using a biological recognition element (biochemical receptor), which is retained in direct spatial contact with an electrochemical transduction element (Thévenot et al., 1999). This trans- ducer transforms the recognition event into a measurable signal by means of a potentiostat (see complete set-up in figure 2). Whereas a chemilu- minescent biosensor requires a spectrophometer or a camera to record a change of colour, electrochemical genosensors use molecular probes to detect target nucleic acids in a sample by recording changes in an electro- chemical signal. DNA (RNA) strands are used as the recognition element that can discriminate any target within a mixed assemblage by specific hybridisation of the capture and signal probes to the complementary tar- get strand. The method provides a high selectivity, sensitivity and accu- racy, typically from the millimolar to the femtomolar range with more or less 5% accuracy. Figure 2: Principles of a lab-benched electrochemical genosensor. A. Experimental set-up including the electrochemical sensor interface, the chip connector and the chip. B. Design of the three electrode cell printed on a ceramic substrate (© Dropsens, http://www.dropsens.com/). 148 Biodiversity Biosensors are also powerful tools for species detection. Although some chemiluminescent biosensors have been developed (Scholin et al., 1996), those based on the direct electrochemical detection of nucleic acid tar- get molecules have successfully been applied by linking DNA or RNA hybridisation events onto an oligonucleotide-modified electrode surface (Drummond et al., 2003). The simplicity, low power requirements, speed and accuracy of electrochemical biosensors have made them attractive candidates to overcome traditional limitations in HAB studies (Diercks et al., 2008b; 2011; Metfies et al., 2005). Moreover, the ability of electro- chemical or chemiluminescent sensors to identify directly nucleic acids in complex samples is a valuable advantage over other approaches, such as qPCR that requires target purification and amplification (Liao et al., 2007). The study of toxic algal blooms with genosensors is greatly facilitated by the use of DNA (rRNA) probes. The detection strategy is usually based on a sandwich hybridisation assay (SHA) in which a target DNA or RNA is bound by both a capture and a signal probe (Rautio et al., 2003; Zammatteo et al., 1995). Only one of the two probes needs to be specific for the target species. A capture probe is immobilised on a semiconduct- ing transducer platform, e.g. carbon or gold. If the target sequence binds to the capture probe in the first hybridisation event, its detection takes place via a second hybridisation event with a signal probe linked to a recorder molecule, such as fluorochromes or digoxigenin. An antibody linked to the recorder molecule is coupled to a horseradish-peroxidase (HRP) enzyme for electrochemical signal amplification. HRP converts electrochemically inactive substrates to an electroactive product that can be detected amperometrically, where the measured current is pro- portional to the analyte concentration in a sample (Metfies et al., 2005). Figure 3 shows the typical amperometric signal expected for a DNA- based biosensor using gold as transducer platform (signal c) as compared to the negative control and background (signals b and a, respectively). For monitoring of water samples, a calibration curve has to be deter- mined for each probe set to assess the current density (nA.mm-2) for 1 ng RNA. For each target species, the RNA concentration per cell has to be investigated. Subsequently the cell concentration of the target species in a water sample can be calculated from the electrochemi- cal signals. By using a different substrate the anti-digoxigenin anti- body conjugated to HRP reacts to produce a green coloured product, the intensity of the reaction can be measured in a spectrophoto- meter or captured by a camera, thus forming a chemiluminescent biosensor. Part II – Chapter 3 149 Figure 3: Analytical signal recorded by a DNA-based biosensor at a fix potential of -0.15 V. The signal includes (a) background noise, (b) negative control current, and (c) positive control current with a DNA-based biosensor using gold as transducer surface. The signal of the positive control relative to the negative control is proportional to DNA concentration in the sample. Detection assays of oligonucleotide probes involving the amplification of hybridisation signals through enzyme tracer molecules have the advantage of being ultrasensitive (Ronkainen-Matsuno et al., 2002). The assay format maximizes discrimination of the target sequences, and RNA purification is not required. Reactions are rapid, easy to execute and amenable to auto- mation. Quantification of the target species can be performed by using smaller, portable and inexpensive instrumentation and several probe sets have already been developed for this purpose. Such probes can simultane- ously be measured by using the multichip connector shown in Figure 2. However, few reliable genosensors have been applied so far because one of their main drawbacks is their lack of robustness. Conditions, such as pH, temperature and ionic strength, and short-term stability, must be considered. 3.2 Case studies of harmful algae The sandwich hybridisation assay with chemiluminescent detection was first introduced by Scholin et al. (1996) for the detection of Pseudo- nitzschia species in California waters. A benchtop device is available and other prototypes operated from a buoy are currently being tested. Probes were initially designed with sequence data from local populations, which proved to be non-specific when applied to other areas. This difficulty underlies the need to use a sequence database based on global isolates 150 Biodiversity when probes are designed to make them universal. Currently, DNA probes for Pseudo-nitzschia spp., Alexandrium spp., Heterosigma akashiwo, Chattonella spp., Fibrocapsa japonica, a variety of Karenia spp., Karlodinium veneficum and Gymnodinium aureolum are available using the chemilumi- nescent detection system in a semiautomatic robotic system (Scholin et al., 2003). Other species for whom sandwich hybridisation probes have been developed (e.g. Coccholodinium polykrikoides, Mikulski et al., 2008), have yet to be applied in a semiautomatic system. In New Zealand, this method has gained national accreditation and is used to monitor shellfish harvests (Ayers et al., 2005). A chemiliuminescent SHA detection in a microliter plate format was also adapted as a rapid means to test probe specificity and some probe sets for 10 toxic algae were validated (Diercks et al., 2008c). Fibre optic genosensors have been applied to harmful algal cell enumera- tion of Alexandrium fundyense, Pseudo-nitzschia australis and Alexandrium ostenfeldii (Anderson et al., 2006). Biosensors for the detection and identi- fication of the toxic dinoflagellate Alexandrium ostenfeldii and A. minutum were developed by Metfies et al. (2005). Development and adaptation of a multiprobe biosensor for the simultaneous detection of 16 target spe- cies of toxic algae has been conducted but is not commercially available (Diercks et al., 2008a; Diercks et al., 2011). More recently, elucidation of the different steps of the biosensor fabrication process from the elec- trochemical point of view, proof of concept with different algal species, and evaluation of the influence of the transducer platform geometry and material has been published (Orozco et al., 2011a). Probe orientation and effect of the digoxigenin-enzymatic label in a sandwich hybridization for- mat to develop toxic algae biosensors have also been evaluated (Orozco et al., 2011b). However, a system enabling the identification of a broad spectrum of toxic algal species and the in situ quantification of very low cell concentrations of cells is still unavailable and very much needed. 3.3 Future prospects In the past decade, the application of biosensor technology has gained significant impact in microbial ecology. In the European Union (EU) FP6-project Alagadec, a portable semi-automated electrochemical bio- sensor-system was developed in order to facilitate the detection of toxic algae in the field. This device enables the electrochemical detection of microalgae from water samples in less than two hours, without the need of expensive equipment. In the future, autonomous biosensors will be combined with in situ measurement systems for monitoring of the marine environment. The Scholin chemiluminescent SHA has been adapted for real time measurements in a buoy, the environmental sample processor Part II – Chapter 3 151 (ESP, Greenfield et al., 2006). Toxin analysis by antibody/antigen detec- tion methods (Elisa) and most recently qPCR (Preston et al., 2011) have also been added to this platform, which will serve the need for high reso- lution monitoring of marine phytoplankton in order to evaluate conse- quences of environmental change in the oceans. Whereas the chemiluminescent robotic system is already commercial- ised, the hand held electrochemical device is still a prototype and is not available for purchase. Nevertheless, probes exist only for a limited number of phytoplankton and must be validated for each region where they are applied, calibration curves must be generated for each probe set, and high sample volume (ca. 5L) are required if the cell densities are relatively low (under the limit of detection of the methods). Validation of probe signals against total rRNA and over the growth cycle of the algae under different environmental conditions has to be carried out to infer cell numbers before the method can be applied to wild samples. In addition, manual RNA isolation should be done by a trained molecular scientist, because a large amount of good quality target rRNA is required for these assays. The manual isolation of RNA is currently the limiting factor of all systems. It has been found out that different users can isolate different qualities of rRNA from the same sample. An automated RNA isolation, developed during the Algadec-project and the lysis methods available in Scholin’s environmental sample processor should overcome these difficulties. 4. Microarrays-based method 4.1. General principles At the core of the DNA microarray technology is a DNA microchip that contains an array of oligonucleotides, PCR-products, or cDNAs spot- ted onto a small surface, e.g. a glass slide. Recently developed DNA- microarray-technology allows the simultaneous analysis of up to 250,000 probes at a time (Lockhart et al., 1996). DNA-microarray technology has enormous potential to be used as a method to analyse samples from com- plex environments, because it provides a rapid tool without a cultivation step. Target nucleic acids are labelled with a fluorescent dye prior to their hybridisation to the probes on the DNA chip. The fluorescence pattern on the DNA-chip after the hybridisation of the target DNA is then ana- lysed with a fluorescent laser-scanner (DeRisi et al., 1997, see figure 4). When these probes detect species, the microarray has been termed a phy- lochip and these are essentially barcodes for species and their automated application. 152 Biodiversity Figure 4: Spotting scheme for the first generation Midtal microarray. The microarray consists in two supergrids made out of four grids installed on a microscopic slide. Each position in the grid represents a spot of ca. 50µm in diameter where a given probe is immobilised. Each probe (colour coded) is spotted four times. This generation of the microarray has 960 spots, covering 112 probes for toxic algal species and higher taxon levels, and various positive and negative control probes. The basis of the immuno-microarray is an immunoassay that has been min- iaturised. In immunoassays, a competitive format is applied where the anti- gens are miniaturised in diminutive spots on a small surface. Fluorescently labelled antibodies compete with the analytes in the sample to conjugate with the miniaturised antigen. Unbound antibodies from the sample are then free to bind to the microarray, and the lower the signal, the higher the concentration of analytes (toxins) present in the field sample. Fluorescence emission is measured with techniques such as confocal or CCD microarray scanners. For optimal use in a monitoring program, it is absolutely neces- sary to validate the signal intensity against known cell counts under various environmental conditions, because absolute cell numbers are the basis for HAb studies. The principal advantage of the method is that thousands of probes can be miniaturised on a single chip. The primary disadvantage is Part II – Chapter 3 153 the high cost, which is a ca. 25€ per sample (Gescher et al., 2010) and the need to make a calibration curve for each probe on the chip. The first DNA-microchip to study microbial diversity was to analyse sam- ples from nitrifying bacteria, which are difficult to study by cultivation (Guschin et al., 1997). A hierarchical set of oligonucleotide probes targeting the 16S ribosomal RNA was created to analyse the bacterial samples on the DNA-chip. Hierarchical ribosomal RNA probes are now available for many species of algae (Groben and Medlin, 2005), and some of them are available in a microarray format (Metfies and Medlin, 2004, Gescher et al., 2008a; 2008b, Midtal: www.midtal.com). These techniques have been tested in the field and have shown congruence with results obtained by flow cytometry and FISH hybridisation (Metfies et al., 2010; Gescher et al., 2008b). 4.2. Case studies of harmful algae There are no published case studies directly applying microarrays to toxic algae. However, microarrays have been the subject of several EU projects. In FP5 Picodiv and Micropad projects, microarrays were developed for algae and protozoa, and results from chip hybridisation were favourably compared to other measurements of diversity, i.e., direct cell counts and clone libraries (Medlin et al., 2006). In these two projects, the microarrays were in early stages of development and proof of principle was the major outcome because it was discovered that probes made for fluorescence in situ hybridization (FISH) could not be directly transferred to a microarray chip format (Metfies and Medlin, 2008). With a few exceptions, nearly every probe had to be modified for a successful use in the microarray chip for- mat. Problems with transferring FISH probes to a microarray chip format led workers in the EU project Midichip to modify their probes and micro- arrays for cyanobacteria. The updated method involved additional steps as compared to the one step hybridisation found on most microarrays. The FP6 project FISH AND CHIPS made use of prototype findings to develop a microarray chip for phytoplankton at the class level. Field data were ana- lysed over three years with rRNA as the preferred target molecule (Gescher et al., 2008b, Metfies et al., 2010). A microarray for toxic species in the dinoflagellate genus Alexandrium was also developed but not field tested (Gescher et al., 2008a). In the EU project Aquachip, pathogenic bacteria were the target of interest. This project developed a chip for five bacteria but they were not widely tested with environmental samples. In addition, the detection system developed for this chip was based on a microtiter plate system with detection under a fluorescent microscope. This is not a stan- dard protocol that can be used in a commercial microarray chip reader, and therefore was never commercialised. In EU FP7 project Midtal (www.mid- tal.com), a species microarray with 163 hierarchical probes for species of 154 Biodiversity toxic algae is in early stages of development (figure 4). Field testing has just begun with high correlation between microarray signal intensity and field counts. The toxin microarray in Midtal, which is based on surface plasmon resonance, detects changes in mass when the antibody binds to the toxin. This microarray can simultaneously detect 4 different toxins (saxitoxins, neosaxitoxin, okadaic acid, and domoic acid) in a competitive assay format. 4.3. Future prospects A common problem in all of the phylochip assays is the wide variation in signal strength of the various probes. In Midtal, the signal of the probes on the microarrays was enhanced by increasing probe length to 25 nucleo- tides instead of 18 and also by adding a longer spacer region to lift the probes above the surface of the microarray so that there is more space for the hybridisation to occur. A fragmentation protocol to break the RNA into small pieces has also been optimised to prevent strong secondary structure formation for signal enhancement (figure 5). Calibration curves will be produced for each probe on the microarray. This is the most time- consuming step needed to make the microarray quantitative because cul- ture experiments have to be established to measure the amount of RNA per cell under different abiotic conditions and to equate RNA content to cell numbers accurately. However, once these calibrations are done, the microarray becomes a very valuable and fast tool to measure community responses over broad ranges in space and time. Figure 5: Hybridisation of fragmented RNA in increasing fragmentation temperature. The probes tested in this study are: bathy01 (a), Pra507 (b), Chlo02 (c), Crypt01 (d), CryptoA (e), Crypt01-25A (f), Crypt03-26 (g), ATWE03 (h), DinoE-12 (i), LPoLyJ (j), Prym01-A (k), Pela02 (l), PNFRAGA (m), Psnmulti A-17 (n), Psn seriA+11 (o), and NS04 (p). Probes with lower signals are enhanced by fragmentation by increasing temperatures from 40°C to 70°C. Data shown are the signal to noise ratio of the hybridisation signal according to the probes tested at different temperatures. Part II – Chapter 3 155 Authors’ references Linda K. Medlin: Université Pierre et Marie Curie, Laboratoire d’Océanographie Micro bienne, UPMC-CNRS UMR 7621, Banyuls-sur-mer, France Jahir Orozco-Holguin: University of California, San Diego (UCSD), Department of Nano Engineering, La Jolla, USA Kerstin Toebe: Alfred Wegeneer Institute for Polar and Marine Research, Bremerhaven, Germany Corresponding author: Linda K. Medlin, [email protected] Aknowledgement J. O. was supported by a Postdoctoral Fellowship from The Institut National des Sciences de l’Univers (INSU), France. This work was partially sup- ported by EU FP7 MIDTAL. References Anderson D. M., Kulis D., Erdner D., Ahn S., Walt D., 2006. Fibre optic microarrays for the detection and enumeration of harmful algal bloom species. African Journal of Marine science, 28, pp. 231-235. Amann R. I., Binder B. J., Olson R. J., Chrisholm S. W., Devereux R., Stahl D. A., 1990. Combination of 16S rRNA-targeted oligonucleotide probes with flow cytometry for analyzing mixed microbial populations. Applied and Environmental Microbiology, 56, pp. 1919-1925. Ayers K., Rhodes L. L., Tyrrell J., Gladstone M., Scholin C. A., 2005. International accreditation of sandwich hybridization assay format DNA probes for micro-algae. New Zealand Journal of Marine and Freshwater research, 39, pp. 1225-1231. Bowers H. A., Tengs T., Glasgow H. B., Burkholder J. M., Rublee P. A., Oldach D. W., 2000. Development of real-time PCR assays for rapid detection of Pfiesteria piscicida and related dinoflagellates. Applied and Environmental microbiology, 66, pp. 4641-4648. Bowers H. A., Tomas C., Kempton J. W., Lewitus A. J., Tengs T., oldach D. W., 2006. Raphidophyceae (Chadefaud ex Silva) systematics and rapid 156 Biodiversity identification: Sequence analyses and real-time PCR assays. Journal of Phycology, 42, pp. 1333-1348. Bowers H. A., 2008. Development and application of real-time PCR assays for the study of Harmful Algal Bloom (HAB) species, PhD thesis. University of Kalmar, Sweden, Dissertation Series 55. Cardullo R. A., Agrawal S., Flores C., Zamecnik P. C., Wolf D. E., 1988. Detection of nucleic acid hybridization by nonradiative fluorescence resonance energy transfer. Proceedings of the National Academy of Sciences of the USA, 85, pp. 8790-8794. Coyne, K.J., Handy, S.M., Demir, E., Whereat, E.B., Hutchins, D.A., Portune, K.J., Doblin, M.A., Cary, S.C., 2005. Improved quantitative real-time PCR assays for enumeration of harmful algal species in field samples using an exogenous DNA reference standard. Limnolology and oceanography: Methods, 5, 389-391. DeRisi J. L., Iyer V. R., Brown P. O., 1997. Exploring the metabolic and genetic control of gene expression on a genomic scale. Science, 278, pp. 680-686. Diercks S., Gescher C., Metfies K., Medlin L. K., 2009. Evaluation of locked nucleic acids for signal enhancement of oligonucleotide probes for microalgae immobilised on solid surfaces. Journal of Applied Phycology, 21, pp. 657-668. Diercks S., Medlin L. K., Metfies K., 2008a. Colorimetric detection of the toxic dinoflagellate Alexandrium minutum using sandwich hybridization in a microtiter plate assay. Harmful Algae, 7, pp. 137-145. Diercks S., Metfies K., Medlin L. K., 2008b. Development and adaptation of a multiprobe biosensor for the use in a semi-automated device for the detection of toxic algae. Biosensors and Bioelectronics, 23, pp. 1527-1533. Diercks S., Metfies K., Medlin L. K., 2008c. Molecular probe sets for the detection of toxic algae for use in sandwich hybridization formats. Journal of Plankton Research, 30, pp. 439-448. Diercks-Horn S., Metfies K., Jäckel S., Medlin L. K., 2011. The Algadec device: a semi-automated rRNA biosensor for the detection of toxic algae. Harmful Algae, 10, pp. 395-401. Drummond T. G., Hill M. G., barton J. K., 2003. Electrochemical DNA sensors. Nature Biotechnology, 21, pp. 1192-1199. Dyhrman S. T., Erdner D., La Du J., Galac M., Anderson D.M., 2006. Molecular quantification of toxic Alexandrium fundyense in the Gulf of Maine using real-time PCR. Harmful Algae, 5, pp. 242-250. Dyhrman S. T., Haley S. T., Borchert J. A., Lona B., Kollars N., Erdner D. L., 2010. Parallel analyses of Alexandrium catenella cell concentrations and shellfish toxicity in the Puget Sound. Applied and Environmental Microbiology, 76, pp. 4647-4654. Erdner D. L., Percy L., Keafer B., Lewis J., Anderson D. M., 2010. A quantitative real-time PCR assay for The identifcation and enumeration of Alexandrium cysts in the marine sediments. Deep Sea Research Part II: Topical Studies in Oceanography, 57, pp. 279-287. Part II – Chapter 3 157 Fitzpatrick E., Caron D. A., Schnetzer A., 2010. Development and environmental application of a genus-specific quantitative PCR approach for Pseudo- nitzschia species. Marine Biology, 157, pp. 1161-1169. Galluzzi L., Penna A., bertozzini E., Vila M., Garcés E., Magnani M., 2004. Development of a real-time PCR assay for rapid detection and quantification of Alexandrium minutum (a dinoflagellate). Applied and Environmental Microbiology, 70, pp. 1199-1206 Galluzzi L., bertozzini E., Penna A., Perini F., Pigalarga A., Graneli E., Magnani M., 2008. Detection and quantification of Prymnesium parvum (Haptophyceae) by real-time PCR. Letters in Applied Microbiology, 46, pp. 261-266 Gescher C., Metfies K., Medlin L. K., 2008a. The Alex Chip – Development of a DNA chip for identification and monitoring of Alexandrium. Harmful Algae, 7, pp. 485-494. Gescher C., Metfies K., Frickenhaus S., Knelfelkamp B., Wiltshire K. M., Medlin L. K., 2008b. Feasibility of assessing community composition of prasinophytes at the Helgoland Reede sampling site with a DNA- Microarray. Applied and Environmental Microbiology, 74, pp. 5305-5316. Gescher C., Metfies K., Medlin L. K., 2010. Hybridization and microarray fluorescent detection of phytoplankton, in: Karlson B., Cusack C., Bresnan E. (Eds): Microscopic and molecular methods for quantitative phytoplankton analysis, Intergovernmental Oceanographic Commission of UNESCO (IOC), Manuals and guides, 50, pp. 77-86. Giovannoni S. J., Britschgi T.B., Moyer C.L., Field K. G., 1990. Genetic diversity in Sargasso Sea bacterioplankton. Nature, 345, pp. 60-63. Greenfield D. I., Marin R., Jensen S., Massion E., Roman B., Feldman J., Scholin C. A., 2006. Application of environmental sample processor (ESP) methodology for quantifying Pseudo-nitzschia australis using ribosomal RNA-targeted probes in sandwich and fluorescent in situ hybridization formats. Limnology and Oceanography: Methods, 4, pp. 426-435. Groben R., John U., Eller G., Lange M., Medlin L. K., 2004. Using fluorescently labelled rRNA probes for hierarchical estimation of phytoplankton diversity, a mini review. Nova Hedwigia, 79, pp. 313-320. Groben R., Medlin L. K., 2005. In situ hybridisation of phytoplankton using fluorescently-labelled rRNA Probes. Methods in Enzymology, 395, pp. 299-310. Guillou L., Chrétiennot-Dinet M.-J., Medlin L. K., Claustre H., Loiseaux-de Goer S., Vaulot D., 1999. Bolidomonas: a new genus with two species belonging to new algal class, the Bolidophyceae (Heterokonta). Journal of Phycology, 35, pp. 368-381. Guschin D. Y., Mobarry B. K., Proudnikov D., Stahl D. A., Rittmann B. E., Mirzabekov A. D., 1997. Oligonucleotide microchips as genosensors for determinative and environmental studies in microbiology. Applied and Environmental Microbiology, 63, pp. 2397-2402. 158 Biodiversity Hallegraeff G. M., 1993. A review of harmful algal blooms and their apparent global increase. Phycologia, 32, pp. 79-99. Handy S. M., Hutchins D. A., Cary S. C., Coyne K. J., 2006. Simultaneous enumeration of multiple raphidophyte species by quantitative real- time PCR: capabilities and limitations. Limnology and oceanography: Methods, 4, pp. 193-204. Handy S. M., Demir E., Hutchins D. A., Portune K. J., Whereat E. B., Hare C. E., Rose J. M., Warner M., Farestad M., Cary S. C., Coyne K. J., 2008. Using quantitative real-time PCR to study competition and community dynamics among Delaware Inland Bays harmful algae in field and laboratory studies. Harmful Algae, 7, pp. 599-613. Heidenreich b., Pohlmann C., Sprinzl M., Gareis M., 2010. Detection of Escherichia coli in meat with an electrochemical biochip. Journal of Food Protection, 73, pp. 2025-2033. Hosoi-Tanabe S., Sako Y., 2005. Species-specific detection and quantification of toxic marine dinoflagellates Alexandrium tamarense and A. catenella by real-time PCR assay. Marine Biotechnology, 7, pp. 506-514 Kamikawa, R., Asai, J., Miyahara, T., Murata, K., Q yama, K., yoshimatsu, S., Yoshida, T. Sako, Y., 2006. Application of a real-time pcr assay to a comprehensive method of monitoring of harmful algae. Micro. Environ. 21, 163-173. Kamikawa R., Nagai S., Hosoi-Tanabe S., Itakura S., Yamaguchi M., Uchida Y., Baba T., Sako Y., 2007. Application of real-time PCR assay for detection and quantification of Alexandrium catenella cysts from marine sediments. Harmful Algae, 6, pp. 413-420. Kavanagh S., Brennan C., O’Connor L., Moran S., Salas R., Lyons J., Silke J., Maher M., 2010. Real-time PCR detection of Dinophysis species in Irish coastal waters. Marine Biotechnology, 12, pp. 534-542. Kudela R. M., Howard M. D. A., Jenkins b. D., Miller P. E., Smith G. E., 2010. Using the molecular toolbox to compare harmful algal blooms in upwelling systems. Progress in Oceanography, 85, pp. 108-121. Liao J. C., Mastali M., Li Y., Gau V., Suchard M. A., Babbitt J., Gornbein J., Landaw E. M., McCabe E. R. B., Churchill B. M., Haake D. A., 2007. Development of an advanced electrochemical DNA biosensor for bacterial pathogen detection. Journal of Molecular Diagnostics, 9, pp. 158-168. Lim E. L., Dennett M.R., Caron D.A., 1999. The ecology of Paraphysomonas imperforata based on studies employing oligonucleotide probe identification on coastal water samples and enrichment cultures. Limnology and Oceanography, 44, pp. 37-51. Liu M. C., Zhao G. H., Tang Y. T., Yu Z. M., Lei Y. Z., Li M. F., Zhang Y., Li D. M., 2010. A simple, stable and picomole level lead sensor fabricated on DNA-based carbon hybridized TiO2 nanotube arrays. Environmental Science and Technology, 44, pp. 4241-4246. Part II – Chapter 3 159 López-Garcia P., Rodriguez-Valera F., Pedros-Alio C., Moreira D., 2001. Unexpected diversity of small eukaryotes in deep-sea Antarctic plankton. Nature, 409, pp. 603-607. Manning S. R., La Claire J. W., 2010. Multiplex PCR methods for the species- specific detection and quantification of Prymnesium parvum (Haptophyta). Journal of Applied Phycology, 22, pp. 587-597. Medlin L. K., Metfies K., Mehl H., Wiltshire K., Valentin K., 2006. Picoplankton diversity at the Helgoland time series site as assessed by three molecular methods. Microbial Ecology, 167, pp. 1432-1451. Metfies K., Huljic S., Lange M., Medlin L. K., 2005. Electrochemical detection of the toxic dinoflagellate Alexandrium ostenfeldii with a DNA-biosensor. Biosensors and Bioelectronics, 20, pp. 1349-1357. Metfies K., Gescher C., Frickenhaus S., Niestroy R., Wichels A., Gerdts G., Knefelkamp B., Wiltshire K., Medlin L. K., 2010. Contribution of the class Cryptophyceae to phytoplankton structure in the German Bight. Journal of Phycology, 46, pp. 1152-1160. Metfies K., Medlin L. K., 2004. DNA-microchips for phytoplankton: The fluorescent wave of the future. Nova Hedwigia, 79, pp. 321-327. Metfies K., Medlin L. K., 2008. Feasibility of transferring FISH-probes to an 18S rRNA gene-phylochip and a mapping of signal intensities. Applied and Environmental Microbiology, 74, pp. 2814-2821. Metfies K., Medlin L. K., 2007. Refining cryptophyte identification with DNA- microarrays. Journal of Plankton Research, 29, pp. 1071-1076. Mikulski C. M., Park Y. T., Jones K. L., Lee C. K., Limb W. A., Lee Y., Scholin C. A., Doucette G. J., 2008. Development and field application of rRNA- targeted probes for the detection of Cochlodinium polykrikoides Margalef in Korean coastal waters using whole cell and sandwich hybridization formats. Harmful Algae, 7, pp. 347-359. Miller P. E., Scholin C. A., 1998. Identification and enumeration of cultured and wild Pseudo-nitzschia (Bacillariophyceae) using species specific LSU rRNA –targeted fluorescent probes and filter-based whole cell hybridization. Journal of Phycology, 34, pp. 371-382. Moon-van der Staay S. Y., De Wachter R., Vaulot D., 2001. Oceanic 18S rDNA sequences from picoplankton reveal unsuspected eukaryotic diversity. Nature, 409, pp. 607-610. Moorthi S. D., Countway P. D., Stauffer B. A., Caron D. A., 2006. Use of quantitative real-time PCR to investigate the dynamics of the red tide dinoflagellate Lingulodinium polyedrum. Microbial Ecology, 52, pp. 136-150 Nolan T., 2004. Getting started-the basics of setting up a qPCR assay, in: Bustin S. A. (Ed.), A-Z of quantitative PCR. International University Line, La Jolla, pp. 527-544. O’Halloran C., Silver M. W., Holman T. R., Scholin C. A., 2006. Heterosigma akashiwo in central California waters. Harmful Algae, 5, pp. 124-132. 160 Biodiversity O rozco J., Medlin L. K., 2011a. Electrochemical performance of a DNA-based sensor device for detecting toxic algae. Sensors and Actuators B: Chemical, 153, pp. 71-77. Orozco J., baudart J., Medlin L. K., 2011b. Evaluation of probe orientation and effect of the digoxigenin-enzymatic label in a sandwich hybridization format to develop toxic algae biosensors. Harmful Algae, 10, pp. 489-494. Park T. G., de Salas M. F., Bolch C. J. S., Hallegraeff G. M., 2007. Development of a real-time PCR probe for quantification of the heterotrophic dinoflagellate Cryptoperidiniopsis brodyi (Dinophyceae) in environmental samples. Applied and Environmental Microbiology, 73, pp. 2552-2560. Penna A., Magnani M., 1999. Identification of Alexandrium (Dinophyceae) species using PCR and rDNA-targeted probes. Journal of Phycology, 35, pp. 615-621. Preston C. M., Harris A., Ryan J. P., Roman B., Marin III R., Jensen S., Everlove C., birch J., Dzenitis J. M., Pargett D., Adachi M., Turk K., Zehr J. P., Scholin C. A., 2011. Underwater application of quantitative PCR on an ocean mooring. PloS One, 6, e22522. Rautio J., Barken K. B., Lahdenpera J., Bretenstein A., Molin S., Neubauer P., 2003. Sandwich hybridization assay for quantitative detection of yeast RNAs in crude cell lysates. Microbial Cell Factories, 2, pp. 4. Ronkainen-Matsuno N. J., Thomas J. H., Halsall H. B., Heinemann W. R., 2002. Electrochemical immunoassay moving into the fast lane. Trends in Analytical Chemistry, 21, pp. 213-225. Saunders N. A., 2004. Introduction to Real-Time PCR, in: Edwards K. Logan J., Saunders N. A., (Eds.), Real-Time PCR. An Essential Guide. Horizon Bioscience, Norfolk, pp. 1-11. Scholin C. A., buck K. R., britschig T., Cangelosi G., Chavez F. P., 1996. Identification of Pseudo-nitzschia australis (Bacillariophyceae) using rRNA-targeted probes in whole cell and sandwich hybridization formats. Phycologia, 35, pp. 190-197. Scholin C. A., Marin III R., Miller P. E., Doucette G. J., Powell C. L., Haydock P., Howard J., Ray J., 1999. DNA probes and a receptor-binding assay for detection of Pseudo-nitzschia (Bacillariophyceae) species and domoic acid activity in cultured and natural samples. Journal of Phycology, 35, pp. 1356-1367. Scholin C. A., Vrieling E., Peperzak L., Rhodes L., Rublee P., 2003. Detection of HAB species using lectin, antibody and DNA probes, in: Hallegraeff G. M., Anderson D. M., Cembella A. D. (Eds), Manual on harmful marine microalgae 11. Intergovernmental oceanographic commission of Unesco, Paris, pp. 131-164. Steidinger K. A., Landsberg J. H., Mason P. L., Vogelbein W. K., Tester P. A., Litaker R. W., 2006. Cryptoperidiniopsis brodyi gen. et sp. nov. (Dinophyceae), a small lightly armoured dinoflagellate in the Pfiesteriaceae. Journal of Phycology, 42, pp. 951-961. Part II – Chapter 3 161 Thévenot D., Toth K., Durst R. A., Wilson G., 1999. Electrochemical biosensors: recommended definitions and classification. Pure and Applied Chemistry, 71, pp. 2333-2348. Touzet N., Keady E., Raine R., Maher M., 2009. Evaluation of taxa-specific real time PCR, whole-cell FISH and morphotaxonomy analyses for the detection and quantification of the toxic microalgae Alexandrium minutum (Dinophyceae), Global Clade ribotype. FEMS Microbiology Ecology, 67, pp. 329-341. Tewhey R., Warner J. b., Nakano M., Libby b., Medkova M., David P. H, Kotsopoulos S. K., Samuels M. L., Hutchison J. B., Larson J. W., Topol E. J., Weiner M. P., Harismendy O., Olson J., Link D. R, Frazer K. A., 2009. Microdroplet-based PCR enrichment for large-scale targeted sequencing. Nature Biotechnology, 27, pp. 1025-1031. Tyrrell J. V., Connell L. B., Scholin C. A., 2002. Monitoring for Heterosigma akashiwo using a sandwich hybridization assay. Harmful Algae, 1, pp. 205-214. Wang J., 2006. Electrochemical biosensors: Towards point-of-care cancer diagnostics. Biosensors and Bioelectronics, 21, pp. 1887-1892. Zammatteo N., Moris P., Alexandre I., Vaira D., Piette J., Remacle J., 1995. DNA probe hybridization in microwells using a new bioluminescent system for the detection of PCR-amplified HIV-1 proviral DNA. Journal of Virological Methods, 55, pp. 185-197. Zamor R. M., Glenn K. L., Hambright K. D., 2011. Incorporating molecular tools into routine HAB monitoring programs: using qPCR to track invasive Prymnesium. Harmful Algae, in press. Zhang H., Lin S., 2005. Development of a cob-18S rDNA real-time PCR assay for quantifying Pfiesteria shumwayae in the natural environment. Applied and Environmental Microbiology, 71, pp. 7053-7063. Chapter 4 Automatic particle analysis as sensors for life history studies in experimental microcosms François Mallard, Vincent Le Bourlot, Thomas Tully 1. Introduction Biodiversity is expected to be heavily damaged by the alarming effects of climate change in the next decades and centuries (Leadley et al., 2010). Assessing how populations will respond to environmental change is cru- cial if one wants to predict the consequences of global change on biodi- versity. The density, phenotypic structure and genetic composition of a population are shaped by extrinsic variations of the environment, intrin- sic regulatory mechanisms such as density dependence mechanisms, and complex interactions between both extrinsic and intrinsic factors. These processes determine whether deleterious environmental changes can lead either to a local population extinction (Drake and Griffen, 2010; Griffen and Drake, 2008; Sinervo et al., 2010) or its rescue through plastic or genetic adaptation (Bell and Gonzalez, 2009; 2011; Chevin and Lande, 2010). Thus, it is especially important to monitor population changes and get accurate measurements of the factors that regulate the size and struc- ture of a population so as to understand how they will react to environ- mental changes. Studying how life history traits (individual fitness or demographic com- ponents) respond to environmental changes is widely done in evolution- ary ecology. In this context, the most common traits under study in population dynamics are age and size at maturity, reproductive output (fecundity, egg size, intervals between reproductive events), longevity and mortality rates (Braendle et al., 2011). These traits are determined 164 Biodiversity by a combination of genetic and environmental effects. When the value of a trait changes according to an environmental factor, it is considered as phenotypically plastic (Scheiner, 1993). The shape of its relationship with the environmental factor (also called a reaction norm) is essential to assess the population responses to environmental change (Flatt, 2005). By linking life-history variation with the genetic makeup of an organism, the interplay between population dynamics and evolutionary dynamics can also be addressed (Saccheri and Hanski, 2006). As a matter of fact, most populations are composed of a mixture of different categories of individuals and, even in an extreme case of a clonal population where all individuals share the same genotype, there is variation in age, size or body condition for instance. Life-history variation has to be taken into account and measured to bet- ter understand how a population behaves (De Roos, 2008; Tuljapurkar et al., 2009). Thus, one of the dreams of a population ecologist would be to follow in parallel the dynamics and structure of a population, and the life-histories of every individual the population is made up of. This would enable the understanding of how population growth, popu- lation dynamics, individual phenotype and life-history traits influence one another (Pelletier et al., 2007; Coulson et al., 2006; Ozgul et al., 2009; Pettorelli et al., 2011). Unfortunately, following simultaneously the dynamic of a whole population and the growth and reproductive trajectory of its individuals remains a Holy Grail quest especially for animal populations in the wild because of the mobility and elusive- ness of the tracked individuals. In this chapter, we discuss the use of a new generation of sensors, based on automated image analysis from microcosm experiments, to address limitations of ecological methods in previous studies. 2. State of the art and objectives Several studies in the wild have quantified to which extent the population growth and dynamics are controlled by the underlying life-history strate- gies of individuals within the population (Coulson et al., 2006; Pettorelli et al., 2011). These studies are based on longitudinal follow-up both at the population and individual levels, the individuals being marked and recaptured. A longitudinal follow-up of individuals is crucial if one wants to address fundamental questions about the patterns of life-history traits throughout life. However, a longitudinal follow-up of wild animals is in general very time-consuming and can only be applied to a part of the studied population. Part II – Chapter 4 165 Given the difficulty of gathering relevant demographic data on wild animals both at the population and individual levels, researchers have looked for complementary and more convenient experimental model approaches, in particular microcosms experiments (Benton et al., 2007). Because of their relatively short generation time and ease of rearing, sev- eral small arthropods have been used as model organisms in microcosm experiments in order to study in parallel population dynamics and life- history traits including Daphnia (Drake and Griffen, 2009; Hebert, 1978), Drosophila (Mueller et al., 2005), mites (Benton and Beckerman, 2005), and collembola (Ellers et al., 2011; Tully and Ferrière, 2008). Data collec- tion, however, is often made visually in most microcosm experiments. This may be accurate enough when populations are sufficiently small and close to extinction (Drake and Lodge, 2004; Pike et al., 2004) but it becomes very time-consuming or impossible to do for larger popula- tions. For instance, experimental mite populations are studied by daily counts of individuals using a binocular microscope and hand-held counter. When the density is too high to be measured visually, it is estimated by extrapolating measurements made on a sub-sample of the population (Bowler and Benton, 2011; Plaistow and Benton, 2009). This procedure not only takes time but it is also prone to errors, including differences among observers, and it only gives coarse measurement of the life-history. Alternatively, measurements can be made on digital images of indi- viduals or populations. Digital images are ideal sources of information for phenotypic and demographic studies in microcosm experiments, because images can be collected very rapidly, they are cheap and can be stored and re-observed if necessary, and the procedure of taking images is generally harmless. For example, Stemman et al. (II, 2) describes how image analysis can be used to identify and separate particles and plank- ton species by size in pelagic, marine environments. In images from microcosm experiments, the extraction of life-history data can be made by hand on a computer using appropriate image analysis software such as ImageJ (Abramoff et al., 2004) to estimate for instance egg and body sizes (Tully and Ferrière, 2008). However, this non-automated procedure is time-consuming and quickly becomes impractical when one wants to closely follow hundreds of individuals each laying hundreds of eggs, or dozens or more populations composed of hundreds of individuals each. Automatic counting and measuring is then needed. Some com- mercial softwares provide such services but they are often unaffordable and closed source. Previous studies designed and proposed image analysis methods to auto- matically track or count small organisms such as small arthropods in the laboratory (Krogh et al., 1998, Auclerc et al., 2010; Lukas et al., 2009). 166 Biodiversity The method of Krogh et al. (1998) is adapted for white collembolan species and requires immobilising the individuals by CO2 anaesthesia and transferring them on an even, black surface. The method therefore requires in practice long and delicate manipulations. Other image ana lysis methods usually require a very contrasted and even background, which is rarely the case in microcosms, or a polarised filter to prevent light reflection on the background. Hooper et al. (2006) developed an image analysis setup for measuring Daphnia population size, but this method does not allow recursive automatic counting because non- daphnid objects (noise and impurities on the glass) had to be manually deleted before automatic enumeration of the Daphnia on their pictures. Using these different methods as a routine for counting and measuring individuals in an experimental population must therefore be banished. Some authors have used morphological image processing tools to reduce the noise in the background and help to identify the individuals (here collembolan) on the images (Marçal and Caridade, 2006). However, this method is prone to errors: it does not permit the elimination of particles that look like collembolans and dead collembolans will be counted and measured. We present hereafter a method that we developed to automate the meas- urements of i) some fundamental life-history traits (size, growth, fecun- dity) of a collembolan that is used as a model organism and ii) the density and fine scale size structure of collembolan populations reared in microcosms. Our method can be applied in general to count and measure some animals (or some particles) that are moving on a motion- less background. It requires simple material (a digital camera, a stand and a good lighting device) and the freely available, open-source image processing software ImageJ (Abràmoff et al., 2004). We first give some details about our model organism before presenting the principle of our method and the device settings that we use. We then explain more precisely how the images are processed to extract relevant information from the background. Lastly, examples of analyses with large data sets collected a on large number of individuals are presented, such as the follow-up of growth trajectory of isolated individuals, the growth of a cohort, or the fluctuations of both population size and structure. Part II – Chapter 4 167 3. Description of the methods 3.1. The biological model In our study, we used the springtail Folsomia candida (Collembola, Isotomidae) as a model organism. This species is convenient to breed in the laboratory (Fountain and Hopkin, 2005) and is used as a model organ- ism in ecotoxicology, ecology and evolution (Fountain and Hopkin, 2001; Tully and Lambert, 2011). The collembolans are bred in small boxes of about 5cm in diameter whose bottoms have been filled with a 2 to 3cm thick layer of plaster of Paris. Plaster of Paris is perfect for growing col- lembolans since it keeps a high level of moisture in the boxes. It also has the advantage of providing a flat two-dimensional environment, which is ideal to observe and count all the creepy-crawlies wandering in the box. To enhance the contrast between the collembolans that are white and their background, the plaster was darkened with some Indian ink. However, a completely dark and homogeneous background is not neces- sary for efficient image processing (see below). More details about our rearing conditions of the collembolan can be found in Tully and Ferrière (2008) and Tully and Lambert (2011). 3.2. Camera settings and lighting unit We used a digital camera (Nikon D300) equipped with a 60mm macro lens and fixed on a camera stand. The camera is connected to a computer and is driven through the software Camera Control Pro from Nikon that enables the adjustment and control of the camera settings (figure 1). We took some 8 bit grey pictures of 4,288 × 2,848 pixels, saved as slightly compressed .JPEG files. We used several LED bulbs such as powerful Pikaline bulbs (16W, 650 lumens) that generate relatively constant, homo- geneous and strong lighting. The lighting unit has to provide a light as homogeneous as possible but our image analysis can compensate for some heterogeneity in the lighting as discussed below. However, the stability of lighting conditions between pictures in a stack is more important. The powerful lighting unit enables to shoot with short aperture time (1/100) and small aperture (F36), which ensures sharp pictures with large depth of field. Artificial lighting using fluorescent light bulbs is not recommended because the light intensity fluctuates at 50Hz frequency, which generates substantial lighting heterogeneity between pictures when using an aper- ture time shorter than 1/50s. We also avoid using incandescent light bulbs since they produce a lot of heat that can harm or disturb our organisms. 168 Biodiversity Figure 1: The camera stand with the camera and the LED lighting unit on, ready to take pictures of the rearing box in the centre. 3.3 Principle of analysis Our method consists in taking a set of several (usually three to five) images of our rearing boxes under the same conditions. Between each image of a box, we blow lightly in it to ensure that each living individual has moved between the first and the last shot. These images are then compared using the ImageJ software (http://rsbweb.nih.gov/ij/) to generate a new image composed of all elements that remained motionless within the set (fig- ure 2). This generated image, called the still background, is then subtracted to each picture. This produces a stack of images that only contains the mobile elements, here the collembolans that have moved between pictures. These elements are then counted and measured after scaling the images. Part II – Chapter 4 169 Figure 2: The principle of image analysis using the ImageJ multitracker procedure. A. one of the images of the original set. Left, the rearing box contains a population of collembolans and can be scaled relative to a graph paper (on the top) or a black square automatically recognised by the plugin. Right, a detailed view of one area of the rearing box. B. The motionless image obtained by comparing four different pictures from the same set to keep the elements that did not move. This motionless image is then used to detect the outline of the box – the area of interest to detect particles (C). D. A subtraction between the original image and the motionless picture followed by a thresholding procedure yields a new image where the particles can be counted and measured. 170 Biodiversity The structure and size of the population is stored in a text file. The princi- ple of this analysis is inspired by the particle analysis procedure developed in the ImageJ multitracker plugin (Kuhn, 2001). An almost perfectly even and contrasted background is not needed to calculate the still background, which allows the measurements to be taken directly in the rearing boxes and minimise disturbances (see figures 2, 3 and 4). Only the moving par- ticles are measured so that dead animals are discarded from the counting. 3.4. Image analysis Once the still background is removed from the set of pictures, the ImageJ software can be used to count and measure the particles. A thresholding procedure is then needed to transform our 8-bits image that contains 256 levels of grey into a black and white 2-bits image. One has to choose a threshold value that is the grey level above which the pixels will become white and under-which they will become dark. It is then possible to count and measure the white particle on the black background (figure 3). To facilitate this thresholding procedure when a single picture is analysed, users have to i) control the overall luminosity on each picture and to ii) maximise the contrast between the particles of interest and its back- ground. The first constraint ensures selecting particles with the same pre- cision within a picture (homogeneous lighting). The second one allows getting a straight discrimination between particles and the background such that precise and repeatable measurements can be obtained with dif- ferent thresholds in a broader range of values. Our measurement method guarantees great precision while allowing a reliable automatic thresholding: removing the motionless background corrects relative lack of enlightenment homogeneity and increases the contrast between moving particles and the background that becomes homogeneously black (figure 3). The user no longer needs to choose by hand an appropriate threshold and the software can be programmed to automatically run the analyses. However, our method is sensitive to local variations of lighting between pictures within the stack. In some cases, the moving particles can create shadows that darken their surrounding substrate, which may reduce the efficiency of the removal of the motion- less pixels. This may generate some noise in the background. Providing omnidirectional lighting reduces the formation of shadows and easily prevents these annoying effects. Part II – Chapter 4 171 Figure 3: Comparison of two particle analysis methods. A. The original picture to be analysed: the collembolans lay on an inhomogeneous substrate. B, C and D. The same picture from which a thresholding procedure has been applied. In B, the motionless background has been removed before the process and the particles are reliably detected. Without removal of the motionless background, no threshold level gives satisfying results: in C, the threshold level is too low and some unwanted background elements are kept (red arrow); whereas in D, the level is both too high to select all the individuals (red arrow) and too low to remove all the non-living elements. 3.5. Time requirement and versatility This automatic particle measuring and counting method enabled us to analyse a large number of samples in a relatively short amount of time. For instance, it takes about a little less than two hours to shoot a hun- dred populations (5 pictures/populations) and to sort these pictures on the computer. It then takes about one hour for the plugin to analyse the whole 500 pictures and count and measure all the individuals in these populations (20 to 30sec per set of 5 pictures on a 2.5GHz computer). Although the procedure was developed and adapted to our collembolan system, it is versatile enough to be tuned and adapted to many different systems as long as there is a still background on which some particles randomly move. To scale our measurements into millimetres rather than 172 Biodiversity pixels we designed an automatic scaling script based on the recognition of a black square of a known area (see figure 2). Including the camera (Nikon D300s, about 1,200€), the lens (Nikon 60mm f/2.8G ED AF-S Micro NIKKOR, about 550€), the stand (starting at about 90€ for the Kaiser 5361) and the software Camera Control Pro (about 130€, running on both Mac and Windows computer), our measurement system costs about 2,000€. 4. Case studies We use our automated particle analysis as a routine procedure in the laboratory. We developed several specific java-coded plugins that we can directly run within ImageJ. The software allows us to choose between a large number of measurements performed on each particle including posi- tion on the picture, area, and centre of mass, but also bounding rectangles or fitted ellipse parameters and many more. We present below different types of analyses based on this method to suggest ideas for broader appli- cations. We recommend using the program R to manipulate and analyse such data (R Development Core Team, 2011). 4.1. Movement analysis Our automated particle analysis was used to track an individual. We col- lected a set of pictures (1 every 6sec during about 3/4h) with a fixed web- cam controlled by a webcam capture application (Dorgem, Fesevur). The different steps of the analyses are described on figure 4. Some difficulties may arise from temporal variation in the still background, which is likely to occur from variation in light intensity or from some changes in the background due for instance to the drying of the plaster in long term follow-ups. One way to circumvent these difficulties is to calculate the motionless background picture and analyse images on shorter periods. But this is very time-consuming and is prone to error when an individual does not move during this period. The image analysis gives the differ- ent particle positions during the follow-up. The few tiny particles that may also be detected can be easily discriminated from our collembolan by setting for instance a size threshold in ImageJ before the analysis. In figure 4, one can see that the animals first explore relatively exhaus- tively the rearing box before finding refuge in the upper left side of the box. The method was used to track a single animal, but tracking several animals at the same time is also possible using the MultiTracker plugin from ImageJ. Part II – Chapter 4 173 Figure 4: Image analysis method customised for tracking individuals. A. The first picture of the set shows the starting position of the collembola. B. This picture shows the still background. C. This picture is obtained from first picture substracted from the background. D. This picture is the sum of the 300 pictures in the stack from which the background has been removed. The particle analysis is robust to the substrate or lighting heterogeneities. E. The successive positions of the individual are measured by particle analyses. 174 Biodiversity 4.2. Life-history analysis One of the major advantages of our automated image analysis method is that it provides a straightforward estimate of the number and the size of the particles without altering their shape by smoothing or other image treat- ments. We used it as a routine procedure to measure and counts cohorts of collembolan, but also to get fecundity measurements by isolating the eggs and then counting the active juveniles once the eggs have hatched. The method helps characterise life-history traits (Tully and Ferrière, 2008): it allows measuring growth trajectories and reproductive events for instance. On figure 5, we illustrate some data collected from a cohort of more than 3,500 individuals that has been kept and followed under controlled condi- tions of temperature, food and density. The adults were transferred regu- larly to fresh rearing boxes and the eggs kept separately until hatching to be counted. Body length (figure 5B) and fecundity (figure 5C) were measured using our automatic particle analysis. Although the counting measure- ments is pretty accurate (figure 2 and 3), the body length measurements can be biased when some individuals are curved or are measured in weird positions. To avoid that, we scored each particle using ratios of its width, length, perimeter and area (figure 5A). We used this score to select the col- lembolan on which a reliable measurement can be made. This allows us to reach a precision of about 0.1 to 0.15mm on the mean cohort body length (figure 5B). These two types of measurements (number and size of parti- cles) can also be coupled together to follow populations made of a mixture of cohorts. The data can be conveniently arranged to draw histograms of the relative abundance by size class (figure 5D), which allows calculating the size distribution of the population. 4.3. Population dynamics Another application of our method is the survey of a population dynamics by following the population’s size and structure. This was done on a weekly basis for about a year on our collembolan populations. A standard popula- tion survey focuses on counting the number of individuals at each time step to produce a time series of the population size or density. Although the information is incomplete to embrace all the richness of a population dynamic, it is an essential first step. This kind of information is easily obtained from the results of the automatic particle measuring process by calculating the average number of particles per date. Figure 6A shows such a time series. Large fluctuations of the population size can be observed. It rises from 366 individuals to more than 2000 with no specific temporal pattern. Rapid increase in the number of individuals can be explained by the simultaneous hatching of numerous clutches. But it is not possible with this method to tell if the observed decreases are caused by death of adults or of juveniles, even though these two mortality processes have very different meanings regarding the future dynamic of the population. Part II – Chapter 4 175