Unmanned aerial systems in conservation biology

Unmanned Aerial Systems in Conservation Biology Ph.D. Margarita Mulero Pázmány 2015 Utilización de sistemas aéreos no tripulados en biología de la conservación Tesis doctoral 1 Unmanned Aerial Systems in Conservation Biology Utilización de sistemas aéreos no tripulados en biología de la conservación Ph.D. / Tesis doctoral Margarita Mulero Pázmány University of Seville. 2015 2 3 Supervisor Dr. Juan José Negro Balmaseda Department of Evolutionary Ecology Doñana Biological Station (DBS-CSIC) Tutor Dr. Carlos Antonio Granado Lorencio Department of Vegetal Biology and Ecology University of Seville 4 5 Dedicated to my family 6 7 Contents • Abstract 9 • Introduction 11 • Aims 20 • Structure 21 • Chapter 1: Environmental impact assessment of infrastructures. A low cost way for assessing bird risk hazards in power lines: Fixedwing small unmanned aircraft systems. 23 • Chapter 2: Management of endangered species. Remotely Piloted Aircraft Systems as a Rhinoceros Anti-Poaching Tool in Africa. 37 • Chapter 3: Conservation in a human dominated landscape. The eye in the sky: combined used of unmanned aerial systems and GPS data-loggers for ecological research and conservation of small birds. 61 • Chapter 4: Conservation in a protected area. Could Unmanned Aerial Systems replace biologging as a tool for animal spatial distribution studies? 79 • General discussion 97 • Conclusions 102 • Literature cited 103 • Acknowledgements 117 8 Abstract Unmanned Aerial Systems (UAS) have been used for decades in the military field, mainly for dangerous or tedious missions where it is preferable to send a vehicle equipped with sensors taking data automatically than to use conventional manned aircrafts. In recent years technology has advanced, the market has grown exponentially, prices have descended and the use of the systems is simpler, which has led to the incorporation of the UAS to the civilian world. UAS have proven useful in ecology related tasks, such as animals monitoring and habitats characterization and their potential for ecology has been pointed out, but to date, there are just a few studies addressing their use in conservation biology. This Ph.D. attempts to fill the gap of knowledge in the use of UAS in conservation biology. It describes for the first time the use of these systems in an immediately applicable way in impact assessment of infrastructures for wildlife and protection of endangered species. Furthermore, it presents UAS as a tool for obtaining high-resolution spatiotemporal information which helps to understand habitat use in rapidly changing human dominated areas and demonstrates that these systems can provide information as valid as the obtained by conventional techniques on the spatial distribution of species in protected areas. The experiments performed in the frame of this Ph.D. indicate that UAS can provide useful information for conservation biology, such as high spatio-temporal resolution aerial images obtained by embarked cameras that allow to monitor the environment at the researcher’s desired frequency and revisiting sites to perform systematic studies. The results also revealed that UAS use in conservation biology presents some constraints, mainly related with the scope of the missions, the limiting costs of the systems, operating constrains associated to weather conditions, legal limitations and the need of specialized personnel for operating the systems, as well as some limitations of data analysis related with image analysis. Overall, given the novelty of the subject and the importance it is expected to have in the near future, I consider that providing information on the capabilities and limitations of these systems, based on solid experiments in conservation biology, is not 15 • Commercial aerial surveillance. • Media industry: sports, filming companies. • Oil gas and mineral exploitation and production. • Disaster relief and medical assistance. • Archaeology research. • Homeland security: coastal patrol, domestic police missions, border surveillance, public protests monitoring, drug plantations detection. • Environmental monitoring: wildlife census, animal tracking, and invasive plants assessment. • Land management: forest fire damage assessment, forest fire mapping, forest fire communications, retardant application,. • Agriculture: crops productivity assessment, crops spraying, vineyards monitoring (Berni et al. 2009). UAS integration in environmental research Given the advances in UAS technology and its growing diffusion in the media, it is not surprising that scientists started to explore the use of UAS for environmental monitoring. The rapprochement between UAS and environmental research has taken place along the last 15 years in two contexts differentiated by project budget and UAS access scenarios, which led to the parallel existence of two lines of work at different scales and thematic. 1) Large scale projects: mainly conducted by NASA and NOAA (with some exceptions for large research groups) using large and medium UAS with capacity to carry payload formed by advanced sensors, high range (>25 km), and autonomy (>4 hours). UAS including payload prices are generally over 100.000 ! and require high operational costs i.e.: Global Hawk, Manta, Scan Eagle, Altair, Aerosonde, Ikhana, SIERRA, R100 Marine and Aerocam. Research topics (conducted and planned) are mostly related with earth science: climate change, atmospheric research (meteorology and chemistry), soil studies, large scale fire studies, vegetation structure, composition and canopy chemistry, glacier and ice sheet dynamics, surface deformation, imaging spectroscopy, topographic mapping, gravitational acceleration measurements, Antarctic and Artic exploration surveys, magnetic fields measurements, river discharge, soil 16 moisture and freeze, landfall and physical oceanography (Williamson 2011; NOAA 2014). 2) Local scale projects: conducted by university or research centers’ departments using small UAS equipped with basic payloads (generally RGB still photo or video cameras and thermal ensors and less frequently: meteorological sensors, broadband or narrowband pyranometer-type radiometric sensors, or lightweight miniaturized hyperspectral radiometers) (Anderson and Gaston 2013), short operational ranges (<25km) and autonomy (<4 hours). UAS including payload prices are generally below 100.000 ! and operations are mostly performed by the members of the research group. Small UAS are frequently fully or partially self-made or acquired in the amateur market, although there are also some professional/ commercial systems, i.e.: UAS developed by Conservation Drones, DIY, Gatewing x100, Quest UAV, SUMO, MMAV, BAT-3, CSIRO, Droidworx multicopter, Iris+, Aero, Draganflyer X4, Nova2, Fulmar, Cryowing, Mikrokopter and MLB Foldbat (more platforms with actual and possible uses in environmental research are reviewed in Anderson & Gaston, 2013). Research topics are often a further extension of the subject the group was already studying by other means. The main works that have been performed, or are currently being conducted in this field (excluding the ones presented in this Ph.D.) are: 1) Wildlife surveys: these studies are mainly focused on the evaluation of the systems for the different species detection and feasibility for a more generalized use. Birds: water birds surveys in Florida (Jones 2003; Frederick et al. 2009; Watts et al. 2010), black headed gulls in North East of Spain (Sardà-Palomera et al. 2012), geese in Canada (Chabot and Bird 2012), sandhill cranes in Colorado (Farrell 2013), steller's sea eagle nests in Russia (Potapov and Utekhina I.G., McGrady M.J. 2013), gull colonies in Germany (Grenzdörffer 2013) and ospreys in Montana (Averett 2014). Terrestrial mammals: roe deer in Germany (Israel 2011), rhinoceros for antipoaching in South Africa (Dewar 2013; WcUAVC 2013) and in Zimbabwe (Olivares- Mendez and Bissyand 2013), elephants in Democratic Republic of Congo (Linchant et al. 2013), Mozambique (Mander 2013), Burkina Faso: (Vermeulen et al. 2013), Kenya (Schiffman 2014) (and with rhinoceros and orangutans) in Indonesia (Gemert et al. 2014). Marine mammals and fish: manatees (and alligators) in Florida (Jones 2003; Jones et al. 2006), marine mammals in Washington (Koski et al. 2009), humpback 17 whales and dugongs in Western Australia (Pyper 2008; Hodgson et al. 2013) and Chinook salmon in Oregon, Washington and Utah (AeryonLabs 2014). 2) Habitat characterization: mainly focused on vegetation and landscape characterization, although some have a more ecological orientation, including animal species habitat selection studies. Ecological restoration in Latin America, (Watts et al. 2008), Mediterranean riparian forest in Southern France (Dunford et al. 2009), coastal research (Pereira et al. 2009), rangeland monitoring in New Mexico (Laliberte et al. 2011), biodiversity in forests in Germany (Getzin et al. 2012), tropical forest in Indonesia (Koh and Wich 2012), habitat mapping using Terrestrial LiDAR and UAV imagery in Russia (Olsoy et al. 2013), wetlands in Canada (Chabot and Bird 2013; Chabot et al. 2014), river mapping (location not specified) (Room and Ahmad 2014), community-based forest monitoring in Malaysia, Nepal and Indonesia (Paneque-Gálvez et al. 2014), marshlands monitoring (ABC 2014) and environmental variables related to epidemiology in Malaysia and Philippines (Fornace et al. 2014). 3) Methodological studies: focused on advances in techniques for data processing or the design of systems specifically for environmental purposes. Development of a UAV for wildlife surveillance (Lee 2004), algorithm for automatic bird detection (Abd-elrahman et al. 2005), geo-referencing techniques (Wilkinson 2007), estimating distribution of hidden objects (Martin et al. 2012), estimating the surface area of sampling strips (Lisein et al. 2013), bio-logged wildlife tracking (Soriano et al. 2009; Körner et al. 2010; Leonardo et al. 2013), remote water sampling (Schwarzbach et al. 2014). Finally, there are two interesting articles that review the use, capabilities and limitations of UAS in Remote Sensing (Watts et al. 2012) and spatial ecology (Anderson and Gaston 2013). Why UAS in conservation biology? Conservation biology is a mission-oriented science that focuses on how to protect and restore biodiversity, dealing with issues where quick action is critical and its goal is to provide principles and tools for preserving biological diversity. To effectively inform policy and management authorities, conservation research must address the most pressing problems and the most threatened systems and organisms (Soulé 2007; SCB 2014). 18 Despite the explosion of projects with UAS for environmental applications in the recent years, there is still a lack of studies that fit the philosophy of conservation biology, i.e. providing solutions that are immediately applicable to solve concrete and urgent environmental problems. To address these problems, the Doñana Biological Station-CSIC in collaboration with several institutions, among which the School of Engineering of the University of Seville is noteworthy, has participated since 2005 in three multidisciplinary projects focusing on the development of systems and techniques for the application of UAS to conservation biology, and in which this Ph.D. is framed. SADCON (Andalusian Government, Project for Excellence, 2005 / TEP-375). Distributed autonomous systems for environmental conservation. AEROMAB (Andalusian Government, Project for Excellence, 2007, P07-RNM- 03246). Aerospace technologies for biodiversity conservation. PLANET (7th Framework Program, cooperation FP7-257649) Platform for the deployment and operation of heterogeneous networked cooperating objects. 19 Aims The overall objective of this Ph.D. is to evaluate the use of UAS in conservation biology, identifying their capacities and limitations. The specific aims are: ! How can UAS contribute to environmental impact assessment of infrastructures? • How can UAS contribute to management of endangered species? • Conservation in a human dominated landscape: Can UAS constitute a useful tool for obtaining high-resolution spatiotemporal information on animals habitat use? • Conservation in a protected area: Are UAS capable of providing information as valid as the obtained by conventional techniques on the spatial distribution of species in protected areas? 20 Structure In addition to the general introduction, this Ph.D. consists of four chapters that explore the specific objectives stated above examining four representative use cases. To accomplish the objectives, we (a multidisciplinary team from Doñana Biological Station-CSIC) conducted several field campaigns using small UAS along the last five years. The first three chapters correspond to published papers and the last one to a submitted manuscript. CHAPTER 1: Environmental impact assessment of infrastructures We describe the use of low cost small unmanned aircraft systems (sUAS) equipped with on-board cameras for power line surveillance to subsequently assess their impact on birds in a low cost way. We characterized four power lines, geo-referenced every pylon in selected portions, and assessed their hazard for birds. We compare the effectiveness of two variants of the sUAS method for data acquisition and two methods of plane control. CHAPTER 2: Management of endangered species We describe the use of a small low cost RPAS equipped with three different types of cameras to test their ability to support rhinoceros anti-poaching tasks in the KwaZulu-Natal province of South Africa. We performed several flights in order to test the technical capabilities of the system to detect rhinoceros, to reveal simulated poachers and to do fence surveillance. We evaluated the influence of flight altitude, time and habitat type in the effectiveness of the system. Considering the most common modus operandi of poachers, we alsoanalyzed the aspects that affect remotely piloted aircraft’s integration in anti-poaching operations. CHAPTER 3: Conservation in a human dominated landscape We describe the combined use of GPS data loggers and environmental information recorded by UASs to study habitat selection of a small bird species, the lesser kestrel Falco naumanni, living in a highly dynamic landscape. After downloading the spatio-temporal information from the kestrels, we programmed the UASs to fly and document with pictures the paths of those same birds shortly after their flight, extracting environmental information at quasi-real time that we used to study the availability of different habitat types along the bird flightpath. CHAPTER 4: Conservation in a protected area 21 We assess the usefulness of UAS overflights to: i) get data to model the distribution of free-ranging cattle by comparing with results obtained from GPS-GSM collared cattle, and ii) predict species densities by comparing with actual density in Doñana Biological Reserve (South of Spain). 22 “Nobody realizes that there are some people who spend excessive energy just to look normal”. Albert Camus Chapter 1 Environmental impact assessment of infrastructures [This Chapter is published as: Mulero-Pázmány, M., Negro, J. J., & Ferrer, M. (2014). A low cost way for assessing bird risk hazards in power lines: Fixed-wing small unmanned aircraft systems. Journal of Unmanned Vehicle Systems, 2(1), 5–15. doi:0.1139/juvs-2013-0012] ABSTRACT Accidents on power lines are one of the most important causes of man-induced mortality for raptors and soaring birds. The factors that condition the hazard have been extensively studied, and currently there are a variety of technical solutions available to mitigate the risk. Most of the resources in conservation projects to reduce avian mortality now are invested in fieldwork to monitor the lines, which diverts the resources available to install actual corrective measures to mitigate bird hazard. Little progress has been achieved in the methodology to characterize line risk, which is an expensive, tedious and time-consuming task. In this work we describe the use of low cost small Unmanned Aircraft Systems (sUAS) equipped with onboard cameras for power line surveillance. As a case study, we characterized four power lines, geo-referenced every pylon in selected portions and assessed their hazard for birds. We compare the effectiveness of two variants of the sUAS method for data acquisition and two ways of plane control. This work provides evidence of the usefulness of sUAS as a fast, inexpensive and practical tool in conservation biology, adding to their already known applications in wildlife monitoring the environmental impact assessment of infrastructures. INTRODUCTION Bird mortality on power lines is an important conservation issue recognized decades ago (Olendorff, R.R., Miller, A.D. and Lehman 1981; Crivelli, A.J., Jerrentrup, 23 H., Mitchev 1988; Ferrer et al. 1991). Raptor and large bird species are especially prone to electrocution, mostly on distribution lines (Negro, J. J., Ferrer 1995), and collision with cables is more frequent on transmission lines, affecting gregarious species or birds that fly at times with reduced visibility (Negro 1987; Ferrer, M., Negro 1992; Ferrer, M., Janss 1999). The distribution of bird accidents on power lines has a significant tendency to accumulate on certain pylons or spans (cable length between two pylons) (Ferrer, M, Hiraldo 1991; CLAVE S.L. 1992). Thus, effectively correcting a small fraction of all pylons and/or spans of a given line it is possible to reduce total mortality drastically (Ferrer, M, Hiraldo 1991; López-López et al. 2011). Bird nesting on pylons is another situation that may increase electrocution risk and also produces damage to the infrastructures; both result in economic losses and reduce service quality for utility companies (Eléctrica 2005; Ferrer 2012). Currently, the bulk of the effort in terms of time and costs to mitigate the bird hazard of power lines is invested in the fieldwork for the characterization phase of the study. Line monitoring is normally done by car or on foot (Katrasnik et al. 2008), identifying pylon design, recording pylon location with a GPS, identifying bird mortalities and surveying habitat types, all factors that would contribute to the assignment of risk values (Ferrer, M, Hiraldo 1991). There are other possibilities for power line study, such as using conventional aircraft with automatic video surveillance systems (Whitworth et al. 2001; Ma and Chen 2004), satellite images, rotary-wing unmanned aircraft systems (UAS) (Campoy, P., García, P.J., Barrientos, A., Cerro, J.D., Aguirre, I., Roa, A., García, R., Muñoz 2001; Peungsungwal, S., Pungsiri, B., Chammongthai, K., Okuda 2001; Ma and Chen 2004; Jones et al. 2005; Katrasnik et al. 2008; Li and Ruan 2010) and more sophisticated solutions, including climbing-flying robots (Katra"nik et al. 2008), but they are too expensive to be applied routinely in conservation biology studies or they have not been implemented realistically in the field yet. Fixed-wing small Unmanned Aircraft Systems (sUAS) are undergoing remarkable development, which has led to a decrease in prices and a greater variety of equipment available. Their use has increased considerably for different purposes in military and civil applications. SUAS have been recently incorporated in wildlife conservation, mainly focusing on aerial wildlife surveys and habitat studies (Jones et al. 24 2006; Pereira et al. 2009; Watts et al. 2010; Chabot and Bird 2012; Rodríguez et al. 2012a; Sardà-Palomera et al. 2012; Getzin et al. 2012). Here we describe the use of fixed-wing sUAS technology as a tool to characterize power lines to subsequently assess their impact on birds in a low cost way. We also compare the usefulness of two different types of cameras to identify and georeference power pylons and test as well two alternative variants of plane control. MATERIAL AND METHODS Study area Fieldwork was conducted in two locations in southwestern Spain: an agricultural area in Dos Hermanas, Seville (5º56´16.1816´´W, 37º15´22.462´´N) and a preserved area within Doñana National Park, Huelva (6º31´58.8522´´W, 37º6´53.2887´´N). Surveys took place in March, April and December 2012. SUAS technical specifications We used the radio controlled Easy fly St-330 (St-models, China) propelled by a brushless electrical motor. Wingspan is 1.960 m and it has a Maximum Take-Off Weight (MTOW) of 2 kg with a 250 g payload (Figure 1). Its maximum range is 10 km, endurance 50 minutes and it can take off and land manually in small patches of flat and open terrain. Operations can be carried out in two different ways and it is possible to switch from one to the other during the flight. Automatic mode: the plane is controlled and guided by the autopilot system. No intervention from the pilot is required during the flight (only taking-off and landing are performed manually). The autopilot provides flight stabilization and the capability to program waypoints, and if the control signal is lost, the autopilot activates the “return home” mode. First Person View system (FPV): the pilot controls the plane in real time using virtual reality glasses and sees telemetry data superimposed on the video. The FPV system includes a long-range radio control receiver. In both control modes, On-Screen Display (OSD) function provides real time flight information (course, altitude, speed, waypoints and artificial horizon) superimposed on the video signal from a camera located on the plane´s nose, which can 31 Australia, declares to spend $80 million a year on inspection (Li and Ruan 2010). In the Andalusia region (Spain), approximately 20% of the total budget spent in retrofitting dangerous distribution power poles to protect the endangered Spanish imperial eagle (Aquila adalberti) was the cost for identification of power pole design, which was around 500,000 ! (López-López et al. 2011). It is important to point out that this kind of surveillance of the poles it is necessary not only during pole characterization prior to select which ones must be modified, but also a periodic survey of the anti-electrocution devices is needed. Limited life span of insulation protective devices requires periodic inspections to assure effective protection. Similarly, large bird nests on power poles require periodic surveys in order to prevent outages. Consequently, reduction in the total cost and time using sUAS would be greater. As reference, for the sUAS inspection of the 12 km of lines surveyed for this study, 4 flights were needed. On each one of them, the two operators invested a total of 2 hours for the sUAS preparation, flight and data processing. Our study is the first one demonstrating that low cost fixed-wing sUAS are a useful tool for power lines monitoring and offer advantages in cost and time investment versus other methods. Our system, valued at 7,800 !, has been able to geo-reference and characterize power lines providing the information needed to assess bird electrocution and collision hazard. Thus, their use can help to minimize the resources invested in the fieldwork phase of the work, to allocate most of the funds into actual corrective measures. ACKNOWLEDGEMENTS We thank Esteban Guerrero and Miguel Ángel Aguilar, the technician and pilot, respectively, responsible for the UAS. This study was conducted within the Projects: Aeromab, (Andalusia Government, Project for Excellence, 2007, P07-RNM-03246) and Planet (European Commission 7th FP Grant Agreement 257649). We also thank Manuela de Lucas, Marcello D’Amico and Manuela González for their reviews, and for providing valuable comments on this manuscript. 32 FIGURES Fig. 1. Description of our small unmanned aerial system: (a) aerial platform, (b) antennas, (c) ground control station, (d) wing-mounted forward-pointing camera, and (e) wing-mounted nadir-pointing camera. 33 Fig. 2. Surveyed power lines with pylons geo-referenced by three different methods: (a) Dos Hermanas area, (b) Doñana area. Circle, from GPS at the base of the pylon; square, from sUAS using nadir-pointing camera; triangle, from sUAS using forward-pointing camera. 34 Fig. 3. Example of pylon designs recorded from the UAS (pylon designs classified following Clave 1992). 35 Fig. 4. Geo-referencing precision of the two types of cameras: (a) nadir-pointing; and (b) front-pointing. Fig. 5. White stork nests on the pylons: (a) and (b) recorded by forward-pointing camera; (c) and (d) recorded by nadir-pointing camera. 36 “Africa, amongst the continents, will teach it to you that God and the Devil are one”. Karen Blixen Chapter 2 Management of endangered species [This Chapter is published as: Mulero-Pázmány, M, Stolper R, van Essen LD, Negro JJ, Sassen T. 2014. Remotely Piloted Aircraft Systems as a Rhinoceros Anti-Poaching Tool in Africa. PloS ONE 9(1): e83873] Abstract: Over the last years there has been a massive increase in rhinoceros poaching incidents, with more than two individuals killed per day in South Africa in the first months of 2013. Immediate actions are needed to preserve current populations and the agents involved in their protection are demanding new technologies to increase their efficiency in the field. We assessed the use of remotely piloted aircraft systems (RPAS) to monitor for poaching activities. We performed 20 flights with 3 types of cameras: visual photo, HD video and thermal video, to test the ability of the systems to detect (a) rhinoceros, (b) people acting as poachers and (c) to do fence surveillance. The study area consisted of several large game farms in KwaZulu-Natal province, South Africa. The targets were better detected at the lowest altitudes, but to operate the plane safely and in a discreet way, altitudes between 100 and 180 m were the most convenient. Open areas facilitated target detection, while forest habitats complicated it. Detectability using visual cameras was higher at morning and midday, but the thermal camera provided the best images in the morning and at night. Considering not only the technical capabilities of the systems but also the poachers´ modus operandi and the current control methods, we propose RPAS usage as a tool for surveillance of sensitive areas, for supporting field anti-poaching operations, as a deterrent tool for poachers and as a complementary method for rhinoceros ecology research. Here, we demonstrate that low cost RPAS can be useful for rhinoceros stakeholders for field control procedures. There are, however, important practical limitations that should be considered for their successful and realistic integration in the anti-poaching battle. INTRODUCTION 37 The two species of African rhinoceros, the black rhinoceros (Diceros bicornis) and the white rhinoceros (Ceratotherium simum) were driven to near extinction in the 1990’s (Emslie and Brooks 1999). Numbers of both species are raising in Africa since 2007 (Knight 2011), but from 2010 the continued escalation in population growth has slowed down (Emslie et al. 2013), and the two species are still vulnerable, with white rhinoceros classified as Near Threatened and black rhinoceros listed as Critically Endangered according to IUCN criteria (Emslie 2012). South Africa holds more rhinoceros than any other country in the world, with 83% of Africa’s individuals, and also experiences the highest absolute levels of poaching, which is the main threat for their conservation (Emslie et al. 2013). Over the last years, and despite the anti-poaching efforts, there has been a massive increase in the number of rhinoceros poaching incidents. In 2010 there was an average of 0.9 rhinoceros killed per day; in 2011 it increased to 1.2; this number escalated to 1.8 in 2012, (resulting in 668 deaths along the year) and it has reached a staggering historical record of 2.2 per day in the two first months of 2013 (up to February 20th) (Emslie et al. 2013). The rhinoceros poaching is a complex problem with multiple causes and potential solutions (Eustace 2012). Their horn is considered to be a traditional medicine for a variety of ailments in Asia (Lever 2004), with the highest demands from China, Hong Kong, South Korea and Southeast Asian countries, and it is used for ceremonial purposes in Yemen (Loon and Polakow 1997; Milledge 2007). Due to the high demand and the illegal nature of the trade, the prices fetched by the horn in the black market are high. This constitutes a temptation to rural people with scarce resources, as the market value of one horn-set may be equal to the salary of several years for the poacher (Eustace 2012). There are various long and medium-term strategies in progress to reduce the illegal trade of rhinoceros horn, and they remain in constant discussion: horn control, legislation, cooperation with the horn purchasing countries, environmental education and rural development projects in rhinoceros areas, most of them conducted by public institutions or NGOs (Milledge 2007; Knight 2011). These general strategies are also supported by immediate anti-poaching actions in the field, directed by the management authorities or the landowners, and carried on by either park rangers or security companies. 38 In South Africa, around a quarter of the total population of rhinoceros live on private land (Knight 2011). The owners of these reserves and game farms are increasingly hiring specialized companies that focus on the protection of wildlife and the apprehension of poachers. The service of protecting valuable wildlife has led to an emergence of this type of business in recent years. They employ techniques based on operational methods of the police and armed forces. The basis of this strategy is to deploy ground based patrol units that spend multiple days tracking animals and poachers, and monitoring the fence lines for breaks. While the cost of employing these companies is high (around 10,800 ! per year to maintain 1 guard patrolling 700-800 ha), they are the most popular alternative to reduce the number of poaching incidents in private land. Both private companies and public agents working in rhinoceros antipoaching are demanding new technologies to increase their efficiency to detect and intercept poachers before a rhinoceros is killed. The need to be more effective in addressing the poaching problem was expressed by the IUCN/SSC African Rhinoceros Specialist Group (Knight 2011). Discussions with security companies and conservation agencies have indicated that aerial monitoring may be of assistance in covering more ground, and remotely piloted aircraft systems (RPAS hereinafter) have been suggested to do this work (Eustace 2012). Some security firms already patrol the vast farms by flying twice a day with a micro light aircraft and directing the “boots on the ground” to the whereabouts of the rhinoceros. Remotely Piloted Aircraft Systems (RPAS), sometimes also referred as Unmanned Aerial Vehicles (UAVs), Unmanned Aerial Systems (UASs) or drones (the ones for military purposes), are aircrafts (fixed or rotary wings) that are equipped with cameras and/or other sensors and can be sent (using manual, semi-automatic or automatic control) to a destination to gather information. These aircrafts act like an “eye in the sky” (Rodríguez et al. 2012a) with the operator at the ground control station receiving data or sending orders to the aerial platform. RPAS have been used for locating “enemies” in military applications for the last 20 years (Zenko 2013), and more recently they have started to play a role in many civilian tasks, including wildlife monitoring (Jones et al. 2006; Watts et al. 2010; Koh and Wich 2012; Rodríguez et al. 2012a; Sardà-Palomera et al. 2012; Vermeulen et al. 2013). In this paper, we describe the use of a small low cost RPAS equipped with three different types of cameras to test their ability to support rhinoceros anti-poaching tasks 39 in cooperation with a specialized security company working in the KwaZulu-Natal province of South Africa. We performed several flights in order to test the technical capabilities of the system to detect rhinoceros, to reveal simulated poachers and to do fence surveillance. We evaluated the effectiveness of the system at different altitudes and times of the day and night, and over the two main habitat types in the area: open grassland and forest. Considering the most common modus operandi of poachers, we analyzed the aspects that affect remotely piloted aircraft’s integration in anti-poaching operations. MATERIAL AND METHODS Ethics statement At present, no regulations are in place for the use of RPAS in South Africa. Draft regulations pertaining to the use of UAVs have been published by the South African Civil Aviation Authority (SACAA) but these have not been ratified to date. The Recreational Aviation Authority of South Africa (RAASA) indicated that the flights could be performed as long as they were conducted over wildlife areas with low manned aircraft activity and not close to registered active airfields. The study therefore complies with the current South African legislation involving aviation safety. The RPAS operators had the required international radio operator licenses to operate in the frequencies used for this work. To get an insight into the poaching problem, we met four people involved in rhinoceros protection at different levels. These interviews did not contain personal or ethically sensitive information, therefore ethics approval was deemed unnecessary by both the Ethics Committee of Animal Welfare of Doñana Biological Station (CEBAEBD) and the Animal Ethics Committee (AEC - Faculty of Natural and Agricultural Sciences), a sub-committee of the Committee for Research Ethics and Integrity of the University of Pretoria. All four interviewed people provided their verbal informed consent to take part in the study once informed about the nature and objectives of the investigation. The participants gave their implied consent through cooperation and it was therefore deemed unnecessary to obtain written consent. All aspects of these personal communications were written down as part of the data collection process of the entire project. Ethics committee approval was deemed unnecessary to approve this consent procedure. We thank farm owners and the security company for providing 40 valuable information used in this study, the lodging and the logistics for the field campaign. Study area The study area comprised 13 farms whose areas ranged between 1,500 and 25,000 ha, covering a total of 100,000 ha located in KwaZulu-Natal province, South Africa. The habitat on the farms is a combination of forest patches and grassland, and is utilized mainly for ecotourism and hunting. The rhinoceros population (both black and white) in the area is approximately 500 individuals. The field campaign was performed during August 2012. Rhinoceros safety requirements definition To define poachers’ way of operation and actual anti-poaching surveillance methods, we separately met four people at the onset of the fieldwork: the security company manager, the rangers’ coordinator and two rangers of the farms of the study area, all of them responsible for different aspects of rhinoceros safety. Remotely Piloted Aircraft System description -Airframe The fixed-wing RPAS is a commercially available radio control plane airframe Easy Fly St-330 (St-models, China) modified by our team. It has a wingspan of 1,960 mm and a maximum take-off weight of 2 kg with a 350 g payload (Fig. 1). It has a maximum range of 10 km; an endurance of 50 minutes and it is launched by hand and landed manually in small patches of open terrain. It is propelled by a brushless electrical motor using a lithium polymer battery. The plane is capable to operate in three different modes, and it is possible to switch from one to the next during the flight: automatic (using the abilities of the autopilot), FPV (“first person view mode”) and manually (radio control conventional mode, also called “third person mode”). It is equipped with an onboard FPV video camera, a GPS (10 Hz, Mediatek, model FGPMMOPA6B), a data-logger with a barometric altitude sensor Eagletree GPS logger V.4 (Eagletree systems, WA, USA) and an autopilot (Ikarus, Electronica RC, Spain) which provides flight stabilization and On Screen Display (OSD). The OSD provides GPS information about the position, speed, height and course of the aircraft. The data combined with the FPV video signal from the camera are sent to the ground station. For nocturnal flights we equipped the 47 poachers were identified in a wide range of altitudes from 29 to 158 m in grassland and forest habitat, although it was more difficult to distinguish some individuals in the forest, especially certain rangers in camouflage clothing because they offered less contrast with the surroundings. Fence surveillance results were acceptable at morning and midday hours, with the pictures presenting enough quality to zoom in and find people along it. At the lowest altitude (40 m) it was also possible to detect footprints in the sand, but the quality was not sufficient to check the condition of the fence wires along the entire fence route. (Fig. 2) The quality of the images was best at midday (80% of the pictures had high quality in this time period) with vertical sunlight, and the results were worse when the shades of the trees produced dark areas, which happened in the morning (66% high quality) and in the evening, when this effect is accentuated because the air is less clean causing a blurry effect (100% medium quality pictures). Video data The HD video camera provided good resolution below 40 m AGL, but due to the wide angle of the lens (fov 127º), flights above 50 m altitude AGL had not enough quality to identify people or to survey the fences. These results led us to cancel the planned flights for rhinoceros detection, as we considered the altitude had to be so low to identify objects that it could be dangerous for operating the airplane and might also disturb the rhinoceros. (Fig. 3 and Video S1 in supplementary material) The thermal camera provided the finest images in the early morning, when the ground was coldest and there is more contrast between it and any animal or person. We confirmed the presence of targets at altitudes as high as 155 m, but in general, it was difficult to identify them at the species level, as they appear in the video as diffuse (although very contrasting) white spots. Only 5% of the images taken with this camera presented high quality, 24% medium and 71% low. At the earliest hours of the night, the results obtained did not allow us to confirm that any of the spots we detected when overflying a rhinoceros was actually a rhinoceros, and low altitude was needed to identify the people using details such as body shapes. After hours of working with thermal video and “training the eyes” we noticed a considerable improvement on detection and shapes identification. Resolution offered by the thermal camera was enough to follow fence posts and to detect individuals, but fence wires were not distinguishable at all. (Fig. 4 and Video 2 in supplementary material) 48 DISCUSSION Rhinoceros poaching is a pressing issue that needs immediate solutions in the field. Rhinoceros stakeholders are demanding new technologies (Knight 2011); social media have already suggested the use of drones (Wild 2013) and WWF announced in 2012 that it sponsors an on-going remotely aerial survey system and anti-poaching program in cooperation with Google to protect tigers, rhinoceros and elephants (WWF 2013). RPAS have already proved their efficacy for military and civil applications in general, and wildlife monitoring in particular. Now the question is how to integrate RPAS in rhinoceros anti-poaching tasks. To answer this question there are two main aspects to consider: capabilities (technical and practical) and current limitations. Technical considerations The still photo camera provided the best results in terms of image quality (94% of the pictures taken by this camera allowed us to confirm the targets) and precision in the location. That is why this is the most attractive and currently the method of choice in conservation biology studies (Chabot and Bird 2012; Getzin et al. 2012). However, it is a relatively slow procedure, as images must be downloaded after RPAS lands and then reviewed and post processed. Even if pictures were transmitted in real time to the ground station (which is technically possible) accelerating the process, it would still take time to review them. Therefore, the use of a still photo camera would not be suitable to support real time anti-poaching tasks like poachers location during a pursuit. A positive aspect is that still photos would be the best method to provide image proofs against poachers because it offers the best resolution. Video offers real time data, so it seems a better option than still images for poaching control. It is recommended to use a video camera with a narrower view field and zoom capabilities to identify the targets at safe altitudes (over 100 m AGL) in real time with enough magnification. Although video offers less precision on target location, according to the interviews with the people involved in rhinoceros safety, accuracy is not so important for anti-poaching purposes, or at least it is less important than immediacy. As far as we know, this study offers the first nocturnal tests for wildlife monitoring using thermal cameras onboard a fixed-wing small RPAS, which is the only option for RPAS nighttime surveillance. The camera we used provided acceptable results when flying low, but the quality does not guarantee to identify some targets and 49 it is possible to miss some, even one as conspicuous as a rhinoceros, when thermal contrast is low or flying at high altitudes. 29% of the thermal images allowed us to confirm the targets, and the rest presented low quality, precluding identification. It is important to consider that the last are still useful, as in a real anti-poaching situations, the dubious objects could be further inspected either overflying lower the RPAS or by other means (as ground patrols). Additionally, the quality and resolution of the thermal sensor can be improved and therefore the detection. As expected, habitat type had an influence on target detection, which is more noticeable when using visual cameras, either video or still photo. Although rhinoceros are large enough to be detected from high altitudes with still photo cameras, people, especially if wearing camouflage clothes or hidden under a thick tree may not be detectable if flying at high altitudes. Time of day had an influence on target detection. Our results indicated that best time for the use of visual cameras was from early morning to midday, and decreased along the evening. Thermal camera provided better results when temperature contrast is higher (Israel 2011), mainly at early morning and night. The detectability limitation linked to the hourly cycle, which is related to light conditions and air-ground thermal contrast, is important, as this means that the usefulness of RPAS as monitoring tools does not remain constant throughout the day. This effect would be accentuated when the temperatures are higher and humidity increases, as we would expect in the area where we performed the tests during summer, or in places with high humidity levels (tropical or coastal areas). There is a compromise in deciding flight altitude for anti-poaching. Lowest altitudes provide the best results in terms of image or video resolution, but the surveyed area is smaller. Flying low implies more risk for the plane in case of failure and easier detection of the plane from the ground (therefore disturbing the rhinoceros or being more easily detected by poachers). Our results suggest that an altitude range between 100 and 180 m AGL is suitable for detecting rhinoceros or people, and to do fence surveillance with acceptable quality levels, it is a safe altitude for the plane and it is not very noticeable from the ground. Practical considerations 50 Considering poachers modus operandi and current security procedures, there are some limitations for the integration of RPAS in routine anti-poaching work in a realistic and efficient manner. Legal aspects South Africa, as with many other countries in the world, does not yet have a legal framework for operating unmanned aerial systems. The absence of regulation for flying beyond line of sight constrains the range of work of the aircrafts, strongly limiting the actual technological capacities of the systems to just short range operations of RPAS operated by manual radio control (SAMAA 2001), as the ones we presented in this paper. Some authors already addressed this issue arguing that operations that do not pose a safety threat to humans in the air or on the ground should be permitted (Ingham et al. 2006). They suggested Light UAVs for poaching site surveillance and proposed ideas including UAV corridors, avoiding inhabited areas and frequently used airspace, all in order to fly these aircrafts safely. We support these proposals, as rhinoceros distribution coincides with very low populated areas where the risk of hitting a person or crashing with another aircraft or infrastructure is low, especially flying at altitudes below 300 m AGL. The South African Civil Aviation Authority (SACAA) has published draft UAS regulations (SACAA 2008; Mamba 2009) that include exceptional permits for public interest uses of UASs (as anti-poaching could be classified). However, to date there has been no official notice that the SACAA has approved any protocol for UASs flights. Scale of work and range Scale of work is a limiting factor in using RPAS for anti-poaching tasks. The territories rhinoceros inhabit are large and population density is low (1 rhinoceros/200 ha on average in our study area). We demonstrated that it is possible to have an “eye in the sky”, but this eye cannot look everywhere all the time, so that logistics have to be evaluated. How many eyes are necessary and how often do they have to look? The management and application of a RPAS or multiple RPAS is a key question that rhinoceros safety stakeholders need to consider and define before planning RPAS use. Small low cost RPAS typically fly for 30-40 min and their range is limited up to 10-15 km. Roughly considering that a RPAS flying at 150 m AGL could cover 711 ha, to survey the 100,000 ha of our study area would take around 140 hours (5.8 days). And that excludes the time to move the Ground Control Station from one point to another, taking off and landing, changing and charging the batteries, data processing, and 51 assuming 24 hours personnel availability. Obviously, that time would be reduced if having more RPAS available, but that would entail higher associated costs. There is a compromise between the area to control and the frequency of this control. A reasonable solution would be to focus RPAS for monitoring hot spots: either rhinoceros preferred locations or most sensitive poaching areas, which are generally known by security companies or park rangers, or areas where access by anti-poaching patrols and/or vehicles is complicated by other factors such as difficult terrains etc. Weather conditions Small RPAS are safe to fly up to 15-20 km/h wind speed. They are not suitable to operate in rainy conditions because the electronics can be damaged and the data obtained by the cameras in low light levels would not be useful. Temperature and terrain altitude affect air density, which influences the power needed to fly the plane, aircraft battery consumption and consequently endurance and range. These variables also influence the power required for takeoff, which is higher the colder it is, or in higher terrains. This can also translate into more failed takeoffs. In experiments performed for other purposes, we found that our system lost 10 minutes of endurance (around 30%), when comparing sea level in summer in Spain to winter at 2,000 m in South Africa. RPAS possible negative effects Rhinoceros did not show any alarm or discomfort reactions during our flights. However, there is no proof that RPAS could not disturb them or other animals if their use is continuous, so further investigation of this aspect is needed. Some farms that have rhinoceros also offer ecotourism activities that bring important income. Therefore, visitor acceptance to the presence of RPAS in those areas would be important. Choosing the right RPAS The range of RPAS available is extensive and growing by the day. From micro systems that fit in the palm of a hand up to 2 tons airplanes, there is a huge variety in market offer. Considering the scale of work, the funding limitations and the sensor requirements, “close range” (Blyenburgh 1999) RPAS seem to be the best choice for anti-poaching purposes. RPAS’ users always want to improve system performances to maximize endurance, range and sensors capabilities (data quality), and to minimize another set of characteristics associated with the RPAS: price (of the system and spares), logistics (size, transportation, taking off and landing requirements), and experience level needed 52 for its operation. Unfortunately, any improvement in the system performances entails an undesirable effect in one or more of the second set of characteristics that would make RPAS less affordable or practical. Thus, the most suitable choice is a balanced compromise the user has to accept considering all the pros and cons for his specific purposes. Costs and benefits The recommended close range RPAS are typically lighter than 5 kg, have 30-45 minutes endurance and offer an operational range between 5-20 km. The price, capacities and reliability vary according to the manufacturer. In general, there is an investment in a whole system, composed by the ground control station, antennas, and two or three planes that need to be repaired or substituted when they reach a certain number of flights. As a reference, the system we used has performed more than 500 flights with an approximate total investment of 14,000 ! including the sensors payload (see Table 1). There are more affordable options available in the market, but from our experience, reliability of some very cheap components like servos, batteries or even tripods is not guaranteed and their failure may cause serious problems affecting expensive components, so it is worth to get at least medium quality spares. The benefit of integrating RPAS in anti-poaching work is difficult to evaluate in economic terms, as its calculation would involve to put a price on the life of a rhinoceros and to evaluate how many could be saved by using RPAS. It has been pointed out (Ferreira et al. 2012) that white rhinoceros carry two types of values: a commercial value (live rhinoceros trade and rhinoceros hunting) and a conservationist or aesthetic value. The first one could be calculated (white rhinoceros average price in 2012 was 17,330 !, record price in 2012 was 53,784 !; black rhinoceros record price in 2012 was 44,969 !) but the second one is hardly translated into numbers. Currently there is not real work using RPAS to be able to estimate the number of rhinoceros that could be saved by RPAS use or to calculate other types of surveillance costs that might be reduced by using this technology. As a reference, the investment needed for a small low cost RPAS (including spare platforms, spares, tools, etc.) that could last for about two years being used weekly (around 30,000 !), plus around 6,000 ! to train operators, could be assumed by a medium size security company or institutions that control areas between 50,000-100,000 ha (Security company manager, pers. comm.). The business of anti-poaching is growing, especially in private land, with the result that RPAS will be 53 not only appreciated for their real usefulness, but also as a competitive asset for those companies that include them in their surveillance programs. RPAS integration in anti-poaching tasks Considering both the technical and practical aspects we propose three alternatives for RPAS integration into anti-poaching work: 1-As a secret tool for surveillance. Security companies and public entities could use RPAS as a “hidden” tool to monitor systematically poaching hot spots or sensitive areas in order to get data, detect intruders, check rhinoceros presence and safety, as well as provide evidence that could be used on court against poachers. In this case, RPAS must be as discrete as possible. This would entail minimize the noise and camouflage the plane itself and to prevent locals to know about its use. 2-As a supporting tool during poaching incidents. The role of RPAS could be to support ground patrols during the pursuit of poachers, providing real time information about suspect numbers, locations and movements. Images taken may be used as evidence in court if needed. RPAS require less logistics than conventional aircraft, but they still do require some. For this type of very immediate use, technical efforts should be concentrated on developing mobile units integrated in small trailers or 4x4 vehicles that could permit a fast deployment. 3-As a deterrent tool. Security company managers suggested that by making widely known that the area is under constant vigilance by RPAS, it would discourage locals to poach. That would include performing demonstrations to the local communities and appearing in media with awareness campaigns, which could make them afraid and aware that they can be detected even without notice. In this case, it would be convenient to focus the effort with RPAS on farm perimeters surveillance and to get proof of irregular use of the area, giving media coverage to them. The three alternatives may be combined in different times or areas to optimize the use of the system. For example - keep RPAS use secret until they contribute to catch a poacher and then publicize it widely in the local area. There is also a fourth use for RPAS, not related to poaching but also involving rhinoceros conservation. RPAS can provide quasi-real time information of habitat changes affecting species movement behavior (Rodríguez et al. 2012a). Thus, combining high-resolution images of the areas with individuals’ locations, RPAS can contribute to answer ecological questions that have been identified as key conservation factors, such as population density, nutrition and diet (Knight 2011). 54 We also foresee a promising field of work using other sensors (like static surveillance cameras and movement detectors) that could work together with RPAS forming an heterogeneous cooperating objects network for sensitive areas surveillance. CONCLUSSIONS-MANAGEMENT IMPLICATIONS Our study is the first approach using remotely piloted aircraft systems for antipoaching tasks and it can be expanded to other areas or species that suffer from the same problem. Some other African and Asiatic countries have rhinoceros poaching problems too, (Milledge 2007; Martin and Martin 2010) and large mammals such as elephants also suffer from illegal hunting (Dublin 2011). We have demonstrated that current low cost RPAS present enough technical capabilities to provide useful data, but there are also important practical and technical limitations that must be considered, evaluated and solved by users and authorities before these systems can be deployed in a realistic way (see Table 3 for a summary of the best and worst scenarios). The role RPAS can play in anti-poaching should not be overestimated and investment in this technology should be proportional to the results obtained because the resources for rhinoceros conservation are limited. ACKNOWLEDGEMENTS We thank farm owners and the security company (who preferred to remain anonymous due to commercial reasons and their rhinoceros’ safety) for providing valuable information used in this study, the lodging and the logistics for the field campaign. We also thank the logistical support provided by the Centre for Wildlife Management, University of Pretoria and CSIR during the campaign. We are grateful to Esteban Guerrero and Miguel Ángel Aguilar, the technicians of the Aeromab and Planet projects, who piloted the aircraft and worked in data processing. Additionally, we wish to thank Airam Rodríguez, Marcello D’Amico and Manuela González for their reviews and for providing valuable comments on this manuscript. 55 TABLES Table 1. Cost of the RPAS equipment (Material bought in Spain in June 2012) Component Price (!) Airframe with the electronic system 1,000 Ground control station (antennas included) 6,000 Stills Photo Camera 450 HD Video camera 300 Thermal camera 6,000 Total 13,750 56 Table 2. Flights results. Camera Time period Time start Time end Target Habitat Result Altitude (m) (Min-Max) Still photo Morning 09:03 09:26 People Grassland, Mixed Confirmed 32-149 Fences Mixed Confirmed 40-175 09:05 09:38 Rhinoceros Forest Confirmed 57 09:42 10:02 People Mixed, Forest, grassland Confirmed 29-82 Fences Mixed Confirmed 42-72 09:52 10:12 Rhinoceros Forest Confirmed 31-137 Midday 10:16 10:39 Fence Mixed Confirmed 50-175 People Grassland Confirmed 123-158 11:22 11:43 Rhinoceros Grassland, Forest Confirmed 38-239 13:14 13:56 People Forest Not confirmed Evening 17:19 17:38 Rhinoceros Forest Confirmed 82 People Grassland, Forest Not confirmed Fences Mixed Not confirmed Thermal video Morning 07:51 08:11 Fence Mixed, Grassland Confirmed 27-155 People Mixed, Grassland Confirmed 31-100 08:21 08:55 Fence Mixed Confirmed 37-98 People Mixed Not confirmed 08:27 08:56 Fence Mixed Not confirmed 09:25 10:03 Fence Mixed Confirmed 48-54 People Mixed Not confirmed Midday 10:27 10:46 Rhinoceros Forest Not confirmed 10:40 11:07 Rhinoceros Forest Not confirmed 12:32 13:04 Rhinoceros Forest, Grassland Not confirmed Night 18:19 19:02 People Grassland, Forest Confirmed 12-125 Fences Mixed Not confirmed 18:41 19:00 Rhinoceros Forest Not confirmed 19:17 19:40 Fence Mixed Not confirmed People Grassland Confirmed 36 19:27 19:45 Rhinoceros Grassland Not confirmed Visual video Midday 11:08 11:27 Fences Mixed, Forest, Grassland Confirmed 10-17 People Mixed, Forest, Grassland Confirmed 10-35 We provide the minimal and maximum altitude at which a target was confirmed in each flight. When only one value is presented it means that the target was located just once. 63 century. However, the population has been considered stable for the last two decades, and consequently, it has been recently downlisted from ‘Vulnerable’ to ‘Least Concern’ according to IUCN criteria (BirdLife International 2011). Presumably, the main cause of the decline of the lesser kestrel in western Europe was habitat loss and degradation as a result of agriculture intensification (BirdLife International 2011). During the chick rearing period, lesser kestrels select field margins and cereal field as foraging areas (Tella et al. 1998; Franco et al. 2004). In addition, kestrels associate with grain harvesters to catch the arthropods flushed by these machines. One of the most important structural changes associated with agriculture intensification is field enlargement, and consequently, the reduction of field margins (Rodríguez and Wiegand 2009). Likewise, the use of machines to harvest cereal fields has reduced the time of harvesting at a locality to just some weeks or days. So, both factors are concurrently limiting kestrel foraging opportunities. Study area Due to the lesser kestrel decline and also for research purposes, several breeding programs have been put in place in Spain in recent years (Pomarol 1993; Negro et al. 2007; Alcaide et al. 2010). One of these reintroductions was carried out in the roof of our own institute (Doñana Biological Station, Seville, Spain), where we conducted this study. In 2008, a hacking program was started releasing to the wild a total of 149 nestlings (51, 58 and 40 in 2008, 2009 and 2010, respectively) originating from a captive breeding program (DEMA, Almendralejo, Spain, www.demaprimilla.org). In addition, injured adult birds (1–4 individuals) were maintained during four breeding seasons (2008– 2011) at an external cage (6x2x2 m) to facilitate conspecific attraction at the colony. Breeding pairs established themselves at the colony after the second year (one, three, six and three breeding pairs in 2009, 2010, 2011, 2012, respectively). The colony is formed by two elongated constructions on the roof of a five-floor building. Forty wooden nest boxes with sliding doors to capture the birds at the nests from inside the building are open to the north wall (see Figure S1). Although the colony is located within the urban area of Seville, it is in the northernmost edge of the city facing agricultural fields and the communication ring of the city (highways, railroads, and a high density of powerline corridors). Agricultural fields extend toward the northwest, the nearest ones being no more than 500 m away from the colony. 64 Unmanned Aerial Systems (UASs) The aerial platform was built into a ST-model Easy Fly plane (St-models, China) with a wingspan of 1.96 m and a weight of about 2,000 g (Figure S2). It is propelled using a brushless electrical engine (lithium polymer battery). The UAS was controlled from a ground station using a long-range radio control system. It carried an onboard video camera, a GPS (10 Hz, Mediatek, model FGPMMOPA6B), a data logger with a barometric altitude sensor Eagletree GPS logger V.4 (Eagletree systems, WA, USA), an Ikarus autopilot (Electronica RC, Spain), which provided flight stabilization and On Screen Display (OSD), and a Panasonic Lumix LX-3 digital photo camera 11MP (Osaka, Japan). The camera was integrated in the plane wing aimed to the ground, and was activated using a mechanical servo, set in speed priority mode and in its widest zoom position. The Ikarus OSD provided GPS information about the position, speed, height and course of the aircraft. These data were combined with the video signal from the camera and sent to the ground station in 2,4 GHz. The autopilot provides stabilization of the aircraft, waypoint following capability (including altitude) and an ‘‘emergency return home’’ function. The take-off and landing of the plane is by manual control. The ground station is composed by a monitor, a DVD recorder, the video receiver and the control signal transmitter with their associated antennas. It also includes a Laptop PC to program the autopilot, to store the pictures and data logs, and to decode in-flight telemetry allowing to track the position of the UAS in real time on a Microsoft map (Redmond, WA, USA). Experimental procedures During the 2011 nestling period (June–July), we fitted 5 g GPS data loggers to both members of two breeding pairs of kestrels using Teflon ribbon backpack harnesses (Micro size, TrackPack, Marshall Radio Telemetry, North Salt Lake, Utah, USA). Two GiPSy2 GPS data loggers (2361566 mm, 1.8 g plus 3.2 g battery, Technosmart, Italy) were programmed in continuous mode (1 fix/ sec) for a fours hour period. To avoid monitoring abnormal behavior due to capture stress and harness fitting, birds were first captured and fitted with a harness and a 5 g dummy GPS data logger. One week later birds were recaptured and the dummy substituted by a real GPS data logger programmed to start recording data the next day after recapture. To download the data from the data loggers, birds were recaptured at their nest boxes when they were delivering food to their nestlings, after batteries were exhausted one day latter. 65 After the download of the bird tracks, six flights were made by the UAS. Three of them with the aim of repeating the flights made by the lesser kestrels from their nests to their foraging areas, and three additional flights following random transects over the agricultural fields. Random flights connected locations randomly selected in a straight line. Pictures of the area overflown were taken using the onboard photo camera that was shooting continuously while the aircraft was following the routes. Data analysis Given that the accuracy on altitude measurements of the GPS used for navigation is relatively low, to georeference the pictures taken by the camera onboard we used information provided by an Eagletree GPS logger V.4 (Eagletree systems, WA, USA) that includes a barometric altitude sensor. The pictures were georeferenced using a customized extension of ENVI software that used Eagletree data to generate GeoTIFF files. Images taken from the UAS let us clearly identify six types of field crops (or land uses): harvested cereal, fully grown cereal (unharvested), olive trees, sunflowers, fallow land and ‘others’ (e.g., farm houses, barns, roads, streams). Using ArcGIS v.10 (ESRI, Redlands, CA, USA), we measured the percentage of total distance overflown by the UAS over each field type, as well as the number of field margins crossed by the UAS. To evaluate the capacity of UAS to follow kestrels’ routes, we used the tool ‘NEAR’ implemented in ArcGIS to calculate the distance between each kestrel fix to the nearest UAS fix. For this analysis, we deleted the part of kestrel tracks related to active hunting activities and distinct from displacement flights between the colony and the actual foraging grounds (easily recognizable by changes in elevation, direction and speed between consecutive fixes at the distal part of tracks; see Figure 1). RESULTS We obtained 4,460 high resolution images along six different flights (three following the kestrels plus three random transects), but there was a high degree of overlap, and we finally selected 466 of them to build the photo-mosaics. The kestrel actual flights recorded by the bird data loggers were always included in the imagery taken by the UAS (Figure 1). UASs followed the kestrel tracks with high precision, with the majority of recorded distances between kestrel and UAS fixes lower than 50 m. The 75th and 90th percentiles were 85.9 and 128.9 m, respectively (Figure 2). Spatial 66 resolution of imagery depends on the altitude at which images are taken (Figure S3). Our UAS flew at a mean altitude of 184 m, and thus, the mean spatial resolution of imagery was 7.7 cm. The area overflown by kestrels is intensively cultivated, being divided into small plots of sunflower, cereal (mainly wheat), olive groves, and other minor cultivations. Proportions of overflown field types did not show significant differences between flights (i.e. go, return and random transect flights; Table 1), so that kestrels flew them in proportion to their availability. Additionally, go and return flights did not differ from the random flights performed by the UASs in relation to the proportion of habitat types. This suggests that the kestrels did not follow specific prospecting strategies when getting to the foraging areas or leaving them. However, local environmental conditions affecting kestrel flight decisions at a microscale, such as wind gusts, could not be recorded in our aerial photographs. DISCUSSION The lesser kestrel is one of the smallest raptors in Eurasia and its size, and particularly body mass, poses a serious limit to the weight of biotelemetry devices or loggers that can be attached (about 5– 6 g maximum, depending on the individual) to record spatial position or behavioral activity. During the course of our investigations on the lesser kestrel, that began in 1988 (Negro 1997), we have always pursued to get an accurate knowledge of their daily movements at their breeding grounds. Applying radio transmitters and direct behavioral observations of unmarked individuals we have been able to determine foraging habitat preferences (Donázar et al. 1993; Ursua et al. 2005; Ribeiro 2007) but soon realized that we lost track of the birds more often than we located them, biasing our studies to locations near the breeding colony. Later on, geolocators have permitted us to determine that kestrels from southern Spain wintered in the Sahel area of western Africa (Rodríguez et al. 2009). While this was a breakthrough with conservation implications, due to the low spatial precision of the technology, it was useless to monitor movements at the breeding grounds. It was not until recently that programmable GPS data loggers small enough to be fitted in a lesser kestrel became available. This technology has revealed that individual kestrel sometimes forage 15–20 km away in straight line from the breeding colony (data not shown). A question emerged as what type of habitats the kestrels were selecting out of the available ones. Lesser 67 kestrels are colonial birds that exploit sudden outburst of invertebrate prey (Cramp and Simmons 1980). They defend no foraging grounds and flocks of several birds may be sighted hovering and diving at times on ground-based or low flying potential prey (Cramp and Simmons 1980). Although information on crop types may be obtained from satellite images, kestrels are known to respond to rapid structural changes of vegetation in their environment (Ribeiro 2007). A flock of kestrels may hunt on a particular harvested field for one or two days and never be back. Keeping this in mind, we used the UAS, as it could be deployed immediately after we downloaded GPS data from individual kestrels. The results presented here are meant as a demonstration of the capabilities of the UAS to obtain a mosaic of images corresponding to the actual full foraging trips of free-ranging small birds. The UAS flight paths reproduced the kestrel flights reliably, as indicated by the fact that their trajectories tended to be less than 100 m apart (see Figure 2). The precision fit of the UAS autopilot depends on the number of waypoints included in the settings (note that our Ikarus autopilot admits 32 waypoints), as well as the meteorological conditions, so we foresee precision will be improved using better autopilots. In addition, images taken by the camera installed in the UAS flying at average altitude of 184 m above sea level covered an area on the ground that always contained the bird track projection (Figure S4). Post-processing of the pictures resulted in a mosaic of georeferenced images allowing an evaluation of habitat types as well as plot sizes and other landscape features, such as grassy field margins, roads, power lines, or even the presence of harvesters in the fields (data not shown; but see Figure 1 for examples of field margins and roads). In fact, UAS images taken from a mean altitude of 184 m showed a higher resolution (7.7 cm) than freely available satellite images (e.g. those coming from MODIS, 250 m, or Landsat TM or ETM+, 30 m), under request commercial satellite images (e.g. DigitalGlobe, Colorado, USA, 30–65 cm) or orthophotographies (e.g. Junta de Andalucía, Spain, 1–1.5 m). To obtain habitat information, there are other alternative (or complementary) options (see Table S2). The most basic would be to get to the study area and survey it by foot or using a ground vehicle. This is time consuming, it has logistical complications and some landscape variables (at large scales) may not be easily quantified. Stationary cameras or sensors scattered in the landscape can provide interesting information about environmental changes, but they involve a huge economic investment and previous knowledge of animal movements, long post-processing of the 68 data, and it is always risky for the equipment, especially in open areas where they can be damaged or stolen. Satellite images are very useful for spatial studies, but their spatial and temporal resolution may not suit research objectives. In our study case, freely available satellite images do not reach the necessary spatial and temporal resolution to distinguish changes in the highly dynamic habitat (e.g. harvested vs. nonharvested fields). For example: NASA’s Earth Observing System Data and Information System (EOSDIS) can provide only 250-m resolution images from MODIS sensor twice a day for Spain; but they are affected by clouds and have a spatial resolution too low for our aims. Commercial satellite images with the appropriate spatial resolution could be available, but at a high cost and there is greater delay in data acquisition compared with UAS. Aerial photographs can be ordered from specialized firms, but a mosaic of georeferenced images of the landscape would be quite expensive, and it would be logistically problematic to obtain the pictures when needed, i.e. at the desired temporal resolution. In the case of small birds, the recreation of flight paths of birds has been achieved using radio-tracking devices and miniaturized video cameras (Rutz et al. 2007; Millspaugh et al. 2008; Rutz and Bluff 2008; Bluff and Rutz 2008; Sakamoto et al. 2009; Grémillet et al. 2010). However, if home range is large enough to lose the radio signal or there is no previous information on where the birds are moving, this methodology may bias the results (see (Millspaugh et al. 2008)). In larger birds, cameras have been attached on them (e.g. seabirds (Sakamoto et al. 2009; Grémillet et al. 2010)), but in a non-systematic way and with no possibility to get zenithal images of enough high quality that could be processed in a statistical manner. In our case, there is admittedly a delay of several hours between the flight of the bird and that of the UAS, but this is of little relevance for answering most of our ecological questions. In our study, GPS data for bird positions was obtained at a frequency of one fixper-second. In the trade-off among fix frequency vs. length of the registration period, we favored the former for improved spatio-temporal accuracy. Our decision rested on two facts: one, this configuration let us to distinguish among soaring, gliding and hunting flights (i.e. hovering and strikes) according to elevation, direction and speed of fixes; and two, the kestrels we were tracking, even if free-ranging, were easily captured in the colony situated on the roof of our headquarters. This condition, the easy of retrieving the GPS data logger to download data, is not met in a majority of investigations on wild birds (Millspaugh et al. 2008). Therefore, future technological 69 advances to finely track a wider range of small sized species should include remote wireless downloading of the GPS information by GSM, Bluetooth or radio. For the moment, this technology has only been incorporated to relatively large devices that can only be mounted on correspondingly large bird species (see www.celltracktech.com, www. technosmart.eu, (van Diermen et al. 2009)). In addition, UASs can be configured to carry on board additional sensors, such as barometers, thermometers or video cameras. These capabilities of the UAS as a nonintrusive tool for ecological research can also be envisaged as extremely useful in studies of flight dynamics (e.g. recording atmospheric parameters such as temperature, wind direction and strength, or barometric pressure (Shepard et al. 2011)), predator-prey interactions (e.g. recording UV light from prey urine tracks which may attract to predators (Viitala et al. 1995)), social dynamics (e.g. monitoring birds of different species during migration (Chabot and Bird 2012)) or behavioral decisions related to the conservation of species (e.g. recording what shearwater fledglings would see when they are fatally attracted to artificial lights during their first flights from nest-burrows to sea (Rodríguez and Rodríguez 2009; Rodríguez et al. 2012b)). As a future refinement, UASs may also be used to locate and track at a safe distance animals equipped themselves with radio transmitters or other locating devices. All the heavy equipment, such as video or still cameras, would go in the UAS and the animal would just carry a light weight location device. Our UAS flew programmed routes, providing georeferenced images of the area overflown by kestrels. The combination of the GPS position provided by the data loggers and the images provided by the UAS recreate the trajectory of a bird carrying a camera. It improves, however, the performance of the other techniques available to date to study the environment as conventional fieldwork, satellite imagery, aerial pictures or stationary cameras. ACKNOWLEDGEMENTS We thank Manuel Baena for fitting a video camera to monitor the nests, Carlos Marfil and Miguel Fernández for their help during the captures of lesser kestrel in the colony, and David Aragonés and Isabel Afán for their advice with GIS work at LASTEBD. Esteban Guerrero and Miguel Ángel Aguilar, of the Aeromab project team, manned the UAS and prepared the image mosaic. Rafa Silva and Pepe Ojeda built the simulation flight of the kestrel and the UAS. Todd Katzner, Pascual López-López, Katherine Renton and an anonymous reviewer made useful comments on early drafts of 70 this manuscript. We thank Junta de Andalucía and UE for granting the research projects (i.e., AEROMAB, HORUS and PLANET) TABLES Table S1 Budgetary cost of the equipment used in this study. 71 Table S2 Pros and cons of commonly used techniques for recording environmental information. This table is based on our study case, i.e. an actual case to study the habitat selection of Lesser Kestrel using the kestrel flight tracks. Note that advantages/disadvantages may change according to the aims of the studies. 72 FIGURES Fig. 1. Track of a lesser kestrel foraging flight over the images obtained by an unmanned aerial system. A White and black tracks correspond to unmanned aerial system and lesser kestrel flights, respectively. The circle indicates the hunting area. The rectangle indicates the enlarged area in B. B High resolution images showing sunflowers, olive trees, road and harvested cereal fields. 79 a non-invasive methodology able to provide accurate spatial data useful for ecological research, wildlife management and rangeland planning. INTRODUCTION Assessing the distribution of animal species among available environments and the reasons behind those patterns are recurrent ecological questions that may also affect human activities and conservation efforts (Morrison et al. 2006). Resource utilization, wildlife management, conservation planning, ecological restoration and prediction of possible future impacts of land use or climate changes are all applied areas that benefit from spatial distribution models of individuals, populations, species and communities (Collinge 2010; Qamar et al. 2011). Numerous methodologies are available to collect spatial data for animals. Direct methods include observation, capture, biotelemetry, radar, laser and cameras, whereas indirect methods are dependent on some evidence of animal activity in an area or specific site (e.g. bed sites, feces, nests or tracks) (Mcdonald et al. 2012). Bio-logging consists in the remote data collection from free-ranging animals using attached electronic devices (Cooke et al. 2004). This is an increasingly popular option among ecologists because it provides valuable information on the animals’ movements and habitat use. This method has experienced a remarkable development thanks to the continuous technological advances, especially those regarding tags miniaturization in recent years. Nevertheless, bio-logging techniques present some constraints, including logistical challenges, possible undesirable effects on the animals during the capture, handling and along the period on which the individuals are tagged (see Murray & Fuller 2000 for a review) and the limitation in the number of animals that can be studied, constrained by the number of tags deployed, which are often expensive (Rutz and Hays 2009). Reliable estimates for species abundance at large spatial scales are highly demanded in order to establish bases on which management schemes can be sustained. It is well known that wildlife population abundance is not easily estimated and lots of methods are described for this purpose in the scientific literature (e.g. Morellet et al. 2010). For a given species, the effort required to apply each method is highly variable and it determines their applicability to be used, mainly at large spatial scales. Obviously, the efforts required to determine the abundance of a species at large spatiotemporal scales exclusively from fieldwork are unworkable for most of the studies. 80 Thus, surveying a number of representative populations, on which the relationships between species abundance and the environmental conditions can be determined, is a way to forecast the abundance in unsampled territories, by generalizing the adjusted species abundance–environmental gradients relationships (e.g. Etherington et al. 2009; Acevedo et al. 2014). At this regard, to record reliable information of species abundance is one of the challenges for wildlife management. Unmanned Aerial Systems (UAS hereinafter) have proven useful to address various ecological challenges involving animal monitoring (Jones 2003; Watts et al. 2010; Sardà-Palomera et al. 2012; Vermeulen et al. 2013) and habitat characterization (Koh and Wich 2012; Getzin et al. 2012). The potential value of UAS for spatial ecology is enormous (Anderson and Gaston 2013) but to date, there are just a few studies that have explored their possibilities (Rodríguez et al. 2012a; Chabot et al. 2014). The aims of this work are to test the suitability of aerial images obtained from UAS flights for i) modeling spatial distribution patterns of animals as compared against a widely used method (bio-logging using GPS-GSM collars), and ii) predicting species abundance by comparing with real abundance in the study area. We use as model species free-range cattle Bos taurus inhabiting Doñana Nature Reserve (Southwest of Spain) under a traditional husbandry system. Cattle are large mammals that offer logistical advantages for bio-logging deployment, are easily detectable in UAS images and actual abundance data are available. In addition, the knowledge of the spatial distribution of these large herbivores is critical for ecosystem management (Lazo 1995; Bailey et al. 1996). Researchers and park managers are specially interested because cattle presence and their foraging impact in the protected area is a controversial issue (Espacio Natural Doñana 2000). Health issues are also at stake, as cattle share habitat, resources and diseases such as tuberculosis with wild ungulates (see Gortázar et al. 2008). MATERIAL AND METHODS Study site and species Doñana Nature Reserve (DNR hereinafter, 37°0' N, 6°30' W) is located in the right bank of the Guadalquivir river estuary in the Atlantic Coast (Andalusia, Southwest of Spain). DNR covers 1,008 km2 and hosts a unique biodiversity and ecosystems including marshlands, lagoons, scrub woodland, forests and sand dunes that led to its 81 declaration as a World Heritage Site and Biosphere Reserve (UNESCO 2014). The area has a Mediterranean climate classified as dry sub-humid with marked seasons. We performed the field work during the dry season, when study area includes the following main habitats: (LT1) dense scrub dominated by Erica scoparia and Pistacia lentiscus, (LT2) low-clear shrub land, mainly of Halimium halimifolium, Ulex minor and Ulex australis (LT3) herbaceous grassland, (LT4) Eucaliptus sp. and Pinus sp. woodlands, (LT5) bare lands, sandy dunes and beaches, (LT6) water bodies and vegetation associated with watercourses covered mainly by Juncus sp. patches (Fig 1). A northsouth oriented longitudinal humid ecotone can be identified between the scrublands and the edge of the dry marshlands, dominated by Scirpus maritimus and Galio palustris with Juncus maritimus associations. The study area in DNR is divided in four Management Areas (MA hereinafter) from South to North named respectively: Marismillas (MA1), Puntal (MA2), Biological Reserve (MA3), and Sotos (MA4). Our model species is free-ranging cattle Bos taurus that occupy different MA and are captured just once per year for sanitary handling. Since 2000, cattle are managed according to the Cattle Use Plan (Espacio Natural Doñana 2000) which determines the number of individuals allowed on each MA (MA 1=318, MA 2=152, MA 3=168, MA 4=350). Doñana cattle is an autochthonous breed, named Mostrenca, although some cross-breeds exist in some herds. Unmanned Aerial Systems (UAS) methodology We completed a total of 192 km of UAS diurnal aerial tracks of two types (eastwest and north-south oriented transects) on each cattle management area with six replicates (Fig 1). UAS surveys took place during August and September 2011, the end of the dry season and a time when food resources become more limiting for herbivores in DNR in terms of water and forage availability (see Bugalho & Milne 2003) between 3 p.m. and 8 p.m. (local time). The tracks were performed at an average speed of 40 km/h at 100 m altitude above ground level. The covered strips were approximately 4 km long and 100 m wide (Fig 1). The flights were performed with a small UAS (1.96 m wingspan; see Fig 2) assembled at Doñana Biological Station using a foam fuselage of an Easy Fly plane (Stmodels, China) propelled by an electrical engine. It is equipped with an Ikarus autopilot (Electronica RC, Spain), which provides waypoint following capability and an Eagletree GPS logger V.4 (Eagletree systems, WA, USA) with a barometric altitude 82 sensor. The digital photo camera Panasonic Lumix LX-3 11MP (Osaka, Japan) is integrated in the plane wing nadir pointing and the shoot is activated by a mechanical servo. The images were taken in speed priority mode and in its widest zoom position with continuous shooting. Total price of the system was around 5,700 ! as of June 2012. We geo-referenced the images using the information provided by the UAS with a customized extension of ENVI software using Eagletree data to produce GeoTIFF files. Accuracy of our UAS locations is estimated in the range of 10-50 m (Mulero- Pázmány et al. 2014, authors' unpublished data) before post-processing, and was improved up to 1-3 m after GIS corrections (superimposing the image on orthophotos and manually correcting it by using reference points). We traced the animals in the images and processed them over a 1 ha approximated patch size (grid) as proposed in detailed studies on ungulate behavior (Gibson and Guinness 1980). GPS-GSM methodology Twelve Mostrenca breed cattle were equipped with GPS-GSM collars along July 2011 in the Biological Reserve (MA3) (Fig 2), during routine veterinary inspections with the animals restrained in a cattle chute. The collars included a satellite position capture system (GPS) and a Global System for Mobile communications (GSM) (Microsensory System, Spain) (Cano et al. 2007). The price per collar is 2,750! plus sms service, covered by the manufacturers in our case. The collars were programmed to take a GPS location every hour, sending encoded packets with 20 positions to the central station when mobile phone coverage allowed. Data collected included date, time, geographic coordinates and Location Acquisition Time (LAT hereinafter, precision measure to obtain a fix; range from 0 to 160 sec). We screened our data using LAT # 154 sec to detect anomalous fixes (manufacturer's technical data; Microsensory System, Spain). We obtained a fix-rate of 93.95%, which is acceptable considering that fix rate success of < 90% can cause habitat-induced bias in resource selection studies (Frair et al. 2004). Positional error associated with GPS locations was 26.64 m on average, SD= 23.5 m, according to stationary tests carried out in the center of our study area. Data analysis Landscape covariates. 83 Environmental variables were estimated from thematic cartography 1:10,000 scale (Consejería de Medio Ambiente y Ordenación del Territorio. 2013) using Quantum GIS version 1.8.0 Lisboa (QGIS Development Team 2012) and were determined following the information of the factors potentially regulating ungulates spatial abundance in the study area (Braza and Alvarez 1987). For each 1 ha grid of the study area (total = 29,532 grids, including the 10.1% corresponding to UAS track grids; n=2,983) and for each 26 m radius buffer (according to GPS positional error) around each GPS cattle (used) and random (available) locations (Jerde and Visscher 2005), we calculated: distance to nearest artificial water hole (DW); distance to nearest marshshrub ecotone (DE); exact grid area (GA) to control the variation in UAS image areas in the case of UAS track grids, and proportion of the different land cover types (LT1- LT6). Distances, areas and land cover type proportions were treated as continuous variables (Table S1) and cattle management area (MA) as a categorical variable. Distance variables were obtained as the shortest distance from each grid and buffer centroid to the nearest environmental feature. To correct visibility reduction produced by vegetation cover for cattle detection in UAS images, we calculated detection coefficients for LT1 and LT4 land cover types. We estimated the detection proportion of 100 random circle points (1 m2 size) created in QGIS from ten different habitat images (1 ha) of each cattle management area and land cover type (80 images analyzed). Detection coefficients used in statistical analysis were 0.544 for LT1, and 0.360 for LT4 respectively. Colinearity between explanatory variables was tested with Spearman’s pairwise correlation coefficients r > |0.5| (Hosmer and Lemeshow 2000). Cattle distribution modeling We tested the factors affecting the spatial distribution of cattle by (i) using UAS images as a first approach; and (ii) using GPS-GSM collar locations as a second approach, by means of Generalized Linear Models (GLM). For the UAS model, we only included the east-west UAS track data, because north-south UAS tracks showed low habitat feature variation (these data were later used for model validation). The response variable was the number of detected animals per UAS grid and was modeled with a negative binomial distribution and logarithmic link function. The final UAS model was obtained using a backward stepwise procedure based on Akaike’s information criterion (AIC) (Akaike 1974). 84 For the GPS model, we used Resource Selection Function (RSF) logistic regression (Manly 2002) where used locations (only considering the ones obtained during the same period-hours of UAS flights) were coded as 1, and random locations (available, ten per used GPS location), inside the individual Fixed Kernel (95% utilization distribution) home ranges, as 0. The response variable is presence/absence of cattle in the grid, and the model included the variables selected for UAS approach except the MA categorical factor (since the collared animals were restricted in MA3). Validation and comparison between the two methods UAS model validation was performed by mean of Pearson’ correlations with independent (20%) data of the east-west tracks and all information in north-south UAS track dataset. GPS model validation was performed by assessing the predictive capacity of each model with the area under a relative operating characteristic (ROC) curve (AUC), to rate the probability that the models correctly discriminated between used and random locations. The AUC ranges from 0.5 for models with no discrimination ability to 1 for models with perfect discrimination (Pearce and Ferrier 2000). Spatial predictions of both final models were transferred to MA3 area where visual and quantitative comparisons were conducted to verify correspondence between predictions of UAS and GPS approaches by Spearman’s pairwise correlation. All statistics were performed in R version 3.0.1 (R Foundation for Statistical Computing, Vienna 2013). We also compared the densities (number of animals / surface) predicted by the UAS model with the current density in the different MAs (data provided by Doñana Biological Reserve and Doñana National Park authorities) evaluating cattle aggregation in the grids by variance to mean ratio (Elliot 1977). RESULTS A total of 358 individual cattle were identified and located on the UAS track images along DNR (Fig 2). We did not observe any disturbance reactions to the UAS during the overflights from the cattle nor from other ungulates present in the area. Overall, the GPS collars fixed 1,752 locations of the 12 marked animals during the same period of UAS flights. Table S1 illustrates the descriptive statistics for the analyzed continuous landscape covariates in the UAS track grids, GPS (used and available) location buffers and total MA3 and DNR grids. 85 Results of the variables included in the spatial distribution models selected by the stepwise procedure ($AIC), estimated coefficients, standard errors and significance are summarized in Table 1 for each approach. The best UAS fitting model (AIC = 397, $AIC from saturated model = -32) found that the environmental covariates influencing cattle distribution are mainly related to landcover types, with a positive effect of grasslands on the ungulates distribution and a negative effect of the distance to the ecotone and to shrubs. UAS best fitting model also revealed a significant effect of the management area on cattle abundance. GPS method identified all the included variables as significant and showed the same effect of them over cattle presence. Validation of the model predictive performance on independent UAS track datasets showed that the selected best spatial distribution model performed adequately with significant Pearson’s rank correlations (east-west data: r = 0.30, p < 0.001, n = 258; and north-south data: r = 0.32, p < 0.001, n = 852). The assessment performed for the GPS location model showed a high predictive capacity (AUC = 0.945). These validation results permitted the transference of the models to the MA3 by using total 1 ha grids (Fig 2). The map representing predicted spatial distribution of cattle shows common distribution patterns throughout MA3 between UAS and GPS approaches. High relations were found between the predicted values of UAS and GPS methods in the MA3 by Spearman’s rank correlation: r = 0.716, p < 0.001, n=6,501. The mean of predicted densities calculated by the UAS approach for each MA were higher than the densities provided by DNR authorities, showing differences between the four MA of DNR, with more overestimated values in the MA with higher aggregation coefficients (Table 2). DISCUSSION In an effort to assess the ability of UAS to contribute to spatial ecology studies, we compared the predicted spatial patterns of free-ranging cattle in Doñana Biological Reserve obtained by using animal locations from UAS overflights images against locations from bio-logged cattle (GPS-GSM collars). Both models, using the same environmental covariates, performed well and provided similar spatial distributions of cattle at a very fine scale (1 ha grids). Models results 86 The environmental variables selected by the UAS model to explain the abundance of cattle are those expected to be more important from an ecological perspective. The positive influence of herbaceous grasslands on ungulates distribution reflected by our models has been previously identified by other authors (Bailey et al. 1996) indicating the need to forage on green pastures during the dry season. The ecotone between the shrub lands and the marshlands is the richest area of DNR, keeping more humidity than other areas and offering not only grasslands but also shade and refuge which are valuable for ungulates in the dry season (see Braza & Alvarez 1987). Models also showed a negative effect of dense and low-clear shrub on cattle presence, that tend to avoid those land types in favor of the open grassland areas (Casasús et al. 2012). However, this work is limited to data obtained at a specific time of the day, as our main goal is to compare two methods in the same conditions, and therefore general habitat use by cattle should be addressed in a more complete study performed all day round. Although the UAS method worked successfully for predicting cattle spatial pattern, it overestimated cattle density in all the management areas (Table 2). This discrepancy may be explained because the flight locations were biased towards the areas where cattle is more concentrated, a problem which could be solved by performing stratified surveys in the different habitats. Also, the overestimation is not homogeneous along DNR, but higher in those areas with a more aggregated distribution. This fact has been proven relevant for animal surveys in general and manned aerial censuses, more related with UAS, in particular (Tellería 1986; Fleming and Tracey 2008). There are various protocols to assess this effect (Redfern et al. 2002; Tracey et al. 2008) and techniques to correct it (Bayliss and Yeomans 1989; Fleming and Tracey 2008) that should be considered if the researcher main objective was estimating abundance, for instance increasing sampling effort as cattle spatial aggregation does. Methods comparison Although bio-logging and UAS approaches proved to be useful in our study, there are several factors that condition their general applicability in spatial ecology. On the basis that the most desirable in spatial ecology studies is to maximize sampling size, data accuracy, diversity and frequency for both the animals and the habitat, while 87 minimizing impact, cost, logistic and data processing effort, we provide below an analysis of the pros and cons of each method. Sampling size Sampling size for bio-logging is limited by financial constrains and/or trapping success (Cooke et al. 2004; Rutz and Hays 2009). This involves a risk of data bias caused by the selection of animals to be fitted with tags, including that produced by the non-random selection in relation to age, sex and geographic location, which increases if the trapping method is not selective. Deployed tags can fail because they may stop sending data or become lost further reducing the sample size, a fact that can lead to biased inferences by focusing on the space use of a few individuals while ignoring the position of non-tagged animals (con- or hetero-specifics). Sampling size for UAS monitoring depends on the area the system is able to cover during the flights (which depends on its range and autonomy) and their detection capacity. Fleming & Tracey (2008) analyzed efficacy of manned aerial surveys, also applicable to UAS, identifying the size, shape, color, shadow and contrast against background of the animals, as well as their response to the aircraft as relevant factors for detection. UAS flight altitude must be a compromise between obtaining adequate resolution to distinguish the species under investigation and the size of the area to cover. Cattle and other smaller ungulate species were easily spotted in our images obtained with an embarked 11 MP commercial camera at 100 m altitude above ground level, in contrast with other studies (Vermeulen et al. 2013), where animals smaller than elephants could not be easily identified flying with a 10 MP camera at the same altitude, maybe because they just made a rapid naked-eye image analysis. Species behavior and habitat characteristics also affect detectability by means of UAS. Bayliss & Yeomans (1989) noted that the main source of aerial survey bias of feral livestock is obstructive vegetation cover. We addressed this problem in our study by using correction visibility factors adequate for the present land covers. This factor, estimated from random location of points, assumes that animals are also randomly distributed with respect to tree cover, but if the animals select for cover, then the UAS density estimates could be biased low, or the opposite if individuals selected otherwise. Besides, selection for cover may vary by species, season and time of day (in our case all the flights were performed in the late afternoon and in summer). Equipping UAS with thermal cameras allows distinguishing animals in dense vegetation areas or at night, but it has been proven that detectability with thermal cameras is low for daylight conditions 88 and in dense vegetation habitats (Mulero-Pázmány et al. 2014b). Admittedly, behavioral responses and habitat characteristics are less critical when data are obtained through bio-logging. Assuming a suitable detection rate for UAS, one of the main advantages of this method versus bio-logging is that it provides the researcher with an image of the animals that are present in the area, permitting to include group influence or interspecific aggregation as variables of the ecological studies. Data accuracy, diversity and frequency Spatial accuracy of the animal locations obtained by UAS after processing is estimated between 1-3 m. This constitutes a major advantage for UAS in spatial distribution studies against bio-logging that provides less accuracy (e.g. 26 m for the GPS collars we used). The use of specific sensors in bio-logging tags is developing fast, allowing to measure individual parameters (e.g. physiological, behavioral, movement speed and range), which is information that could not be obtained with UAS approach. On the other hand, UAS have the capacity to provide real time information on habitat characteristics, which is especially interesting in highly dynamic landscapes (Rodríguez et al. 2012a), where short term changes affecting animals’ movements (i.e. produced by fires, human interventions, rainfalls) may not be reflected on satellite or GIS resources available with proper spatial-temporal resolution. This temporal accuracy is important, as obtaining animal information and environmental variables at the same level of detail and reliability would significantly improve ecology studies (Gaillard et al. 2010). While trapping animals may be complex, once the animals are bio-logged they produce enormous volumes of data for a long period of time. Long-term data with UAS requires additional flight field campaigns and it is difficult to associate data to specific individuals. UAS flights are subjected to favorable meteorological conditions that also constraint the period on which data collection is possible. Impact Bio-logging requires capture and handling of the animals that might affect their behavior and survival (Silvy et al. 2012) and obtaining permits are factors that can complicate the use of bio-logging (Cooke et al. 2004). A point in favor of the use of UAS is their low impact on the surveyed animals. Due to the small size and the reduced noise UAS produce, animal response is very low (at least not visually noticeable in our case) so that the method does not disturb the study subjects. Electric UAS are also zeroemission vehicles and this is an aspect particularly important when surveying nature 95 SUPPLEMENTARY MATERIAL Table S1. Environmental covariates, descriptions, mean values (X) and standard deviations (SD) of UAS track grids and GPS locations buffers versus MA3 and total study area grids used in the analysis of cattle spatial abundance patterns in Doñana Nature Reserve (DNR). Code Variable UAS grid (X±SD) GPS location buffer used and available (X±SD) Total MA3 (X±SD) Total study area (X±SD) DW Distance to nearest water point (km) 0.33±0.19 0.55±0.38 0.94±0.79 0.97±0.73 DE Distance to nearest marshshrub ecotone (km) 1.22±0.90 1.50±1.49 4.06±2.76 2.47±2.17 GA Exacted UAS grid area (ha) 1.25±0.85 0.21±0 1±0 1±0 LT1 Dense scrub (%) 24.66±35.20 20.95±36.1 18.50±31.54 11.09±26.28 LT2 Low-clear shrub (%) 27.54±34.37 31.14±42.15 48.59±41.54 32.04±39.89 LT3 Herbaceous grassland (%) 14.25±26.79 28.67±40.94 10.92±26.23 12.34±26.82 LT4 Woodland (%) 18.40±34.24 4.16±18.27 11.14±26.28 19.84±34.77 LT5 Bare land (%) 8.82±24.41 2.55±12.93 4.58±15.71 11.30±26.88 LT6 Watercourse vegetation (%) 6.33±19.95 12.28±30.03 4.50±17.61 11.28±28.24 MA Cattle management area (categorical 1-5) - - - - 96 General discussion The recent remarkable development of UAS has led to a decrease in prices and a large variety of equipment in the market, which has favored the incorporation of these systems to environmental research. The novelty of the technology explains that there is almost no scientific literature up to only ten years ago, but the explosion of published papers and news in the last three years confirms that the use of UAS is being explored by numerous research teams worldwide. This Ph.D. attempts to fill the gap of knowledge in the use of UAS in conservation biology. It describes for the first time the use of these systems in an immediately applicable way in impact assessment of infrastructures for wildlife and protection of endangered species. Furthermore, it presents UAS as a tool for obtaining high-resolution spatiotemporal information which helps to understand habitat use in rapidly changing human dominated areas and demonstrates that these systems can provide information as valid as the obtained by conventional techniques on the spatial distribution of species in protected areas. What do UAS bring in conservation biology? Aerial perspective offered by UAS offers certain advantages over data collection observers from the ground (as far as target are detectable by the system) as it allows covering larger areas, which saves time and effort on tasks that can be tedious by other means. Even low cost UAS equipped with a basic payload have proved useful for routine tasks such as power lines characterization (Chapter 1) where the savings in logistics allows diverting resources towards the installation of mitigation measures, which is ultimately the goal of the impact assessment study. In this line, UAS can also contribute to the impact assessment of other infrastructures that present problems in terms of conservation, such as roads, water channels or wind farms (Crockford 1992; Forman and Alexander 1998; Kingsford 2000). Aerial perspective allows studying the surroundings of these infrastructures: death records, distribution of wildlife, vegetation gradients or the appearance of invasive species that use them to expand, providing useful information for conservation and management. The most interesting UAS capability for environmental applications is the high spatial and temporal resolution information they can provide by means of the embarked 97 sensors (Jones et al. 2006; Anderson and Gaston 2013). Spatial resolution depends on the quality of onboard sensors and flight height, but even low cost UAS equipped with commercial cameras already offer sufficient quality to identify designs of electric poles (Chapter 1), locating animals (chapter 2, (Watts et al. 2010; Israel 2011; Sardà- Palomera et al. 2012; Vermeulen et al. 2013)) and habitat characterization (Chapter 3 and 4, (Koh and Wich 2012; Getzin et al. 2012; Chabot et al. 2014) ) with a resolution of centimeters, much higher and at a lower cost than the achievable with other available technologies (satellite, manned aircraft). In chapter 4 we demonstrate that UAS are able to provide distribution models that show equivalent patterns than with biologging. Therefore, if the species is detectable, UAS may constitute an alternative to more invasive (biologging) or expensive (manned flight) techniques. Besides, working at a high resolution scale permits the researcher to gather information from all the animals detectable in the surveyed areas, allowing to include inter-individual or interspecific relations in the study and to compare the animals distributions with the environmental variables, which are recurrent ecological questions that may also affect human activities and conservation efforts (Morrison et al. 2006). Temporal accuracy is important, as obtaining animal information and environmental variables at the same level of detail and reliability would significantly improve ecology studies (Gaillard et al. 2010). Microelectronics revolution has allowed gathering high accuracy and frequency animals position data animals, but sample the environment in which they move with the desired frequency for the researcher is not possible with conventional technology. This is especially important when the environment changes rapidly, especially if the changes that occur affect short-term movements of the species studied, as demonstrated in chapter 3 and may be extended to other situations (i.e. human interventions, fires, floods). Moreover, thanks to their easy deploy and to autopilots capability UAS flights are easily repeteable which allows to revisit sites and to perform systematic studies (Watts et al. 2008; Anderson and Gaston 2013). We foresee that the role of UAS in this aspect will be revolutionary, as it can help us to understand habitat selection at a fine scale and contribute to movement ecology. Most small electric UAS like the ones used in the majority of environmental research produce reduced noise and are visually discrete. We did not see any negative reactions on the fauna during any of the performed flights (nor are recorded in the scientific literature), which in combination with being zero-emission make UAS a low 98 impact method. This constitutes a major advantage for their use in fauna surveys, as a basic principle of any researcher is not to interfere in the object of study, but it is also fundamental in reconnaissance operations where the desirable is not to be detected by the “intruder”. This makes UAS a very convenient tool for endangered species or protected areas monitoring (i.e. anti-poaching, see chapter 2, (Lisein et al. 2013; Vermeulen et al. 2013)). Current UAS limitations in conservation biology Although UAS have proved to be useful to perform aerial surveys (Watts et al. 2010; Chabot and Bird 2012; Sardà-Palomera et al. 2012; Vermeulen et al. 2013), or to do surveillance tasks (chapter 2) the systems that are affordable for a common research group (<100,000 !) have a standard range of 10-30 km, limiting the current scope of UAS in field biology to missions within that range. Therefore it is not realistic within the current market scenario to consider substituting a manned aircraft for a UAS in typical wildlife survey campaigns (i.e. periodic aerial censuses in large protected areas) or to monitor large areas, but to complement them. The decrease in UAS prices and payload sensors is one of the favorable circumstances for their integration in ecological research. Particularly in the last two years there is an emergence of inexpensive (< 3,000 !) UAS and experienced model aircraft hobbyists may be able to built their own systems for even less than 1,000 !. Very low cost systems may be capable to perform basic missions, but from our experience, reliability of cheap components is low and their failure may cause serious problems producing dangerous situations or diminishing data quality. Usability and quality have a price (Watts et al. 2010) and researchers that want to perform scientific UAS campaigns need to invest a considerable budget to acquire functional and reliable systems, which may be a limiting factor. UAS present some operating constraints (Jones et al. 2006; Anderson and Gaston 2013): i) UAS usability is constrained by favorable weather conditions (less than 15– 20 km/h wind speed for the platform safety and preferably clear days for good quality images), ii) The legal status of UAS is complicated, a controversial topic in which different agencies (mainly FAA in US and EUROCONTROL in Europe) are working for a long time. Currently, he situation varies between countries from a clear prohibition, the possibility of authorized flights with altitude and range restrictions up to a total absence of regulation, a unclear situation that poses significant limitations to 99 entry for scientific users, iii) there is a need of specialized personnel to operate the systems, some knowledge for maintenance (Jones et al. 2006) and some expertise for data processing. As stated by Ellwood et al. 2007 who analyzed technology integration in conservation biology “failure to understand technology limitations can have serious consequences”. Most of ecology studies using UAS are based on image analysis, therefore, experimental design must be carefully matched to avoid “drowning” into thousands of images and so that researchers understand the significance of missing data. Animal detectability depends on their size, shape, color, shadow, contrast against background, their response to the aircraft (listed by Fleming & Tracey 2008 for manned aerial surveys), but also on the UAS camera characteristics, flight altitude and stability of the aircraft and the environmental conditions. Habitat characteristics such as vegetation cover (Bayliss and Yeomans 1989) also affect animal’s detectability, but besides, to select or avoid for cover may vary by species, season and time of day. The inclusion of detectability coefficients (chapter 4) or using automatic pattern recognition techniques into data-processing procedures may improve the process (Abd-elrahman et al. 2005) but still, a word of caution should be raised regarding the cost/benefit ratio of some of these systems and their applicability. Future prospects of UAS in conservation biology The present Ph.D. tried to contribute to analyze UAS integration in conservation biology, however all the experiments were performed with small UAS that, as mentioned above, have range and autonomy limitations that constrain the scope of the research. To gain access to larger UAS, which is currently limited to a few international agencies, would allow to address issues of great importance in conservation in a global context, such as the study of climate change, deforestation and habitat fragmentation, which are major causes of biodiversity loss. Although this possibility seems complicated in the short term, it may be feasible through agreements between research and military agencies, at least for sharing some data or to embark scientific equipment in the training missions of large military UAS. One of the UAS capabilities that has hardly been exploited in small UAS is the inclusion of other sensors different from cameras as payload. Meteorological sensors, water probes and sampling devices can provide another insight on vertical habitat 100 characteristics and contribute to understand animal movement patterns, such as flight dynamics and migration. As pointed out in chapter 3 and 4, UAS can be combined with other available technologies such as biologging, and as suggested in chapter 2 it is possible to achieve better results combining UAS with static surveillance cameras and movement detectors. A further step would be to communicate other sensors (static or mobile) that could work together with UAS forming an heterogeneous cooperating objects network for sensitive areas surveillance. 101 Conclusions This Ph.D. describes for the first time the use of UAS for impact assessment of infrastructures, protection of endangered species, habitat selection and to determine animal spatial distribution patterns, demonstrating that these systems can provide information in conservation biology. The main benefits that UAS bring to conservation biology are: 1) Aerial perspective that offers advantages over data collection observers from the ground at lower costs than other available technologies. 2) High spatial resolution images that allow to assess environmental impact of infrastructures, wildlife and habitat characterization. 3) High temporal resolution information that permits to monitor the environment at the researcher’s desired frequency that allows revisiting sites to perform systematic studies. The main limitations for UAS integration in conservation biology are: 1) Scope of the missions is limited by range and autonomy of the affordable UAS to short range (<30 km radio) 2) Although UAS prices are decreasing, to safely perform professional missions budget is still limiting factor. 3) Operating constrains: weather conditions, legal limitations and specialized personnel for operating the systems. 4) Experimental design must be carefully addressed to avoid “drowning” into thousands of images and to consider the significance of missing data (mainly related with detectability problems). 102 Literature cited ABC (2014) UAS to Monitor Marshlands in New South Wales. UAS news 1–4. Abd-elrahman A, Pearlstine L, Percival F (2005) Development of Pattern Recognition Algorithm for Automatic Bird Detection from Unmanned Aerial Vehicle Imagery. Surv L Inf Sci 65:37–45. Acevedo P, Quirós-Fernández F, Casal J, Vicente J (2014) Spatial distribution of wild boar population abundance: Basic information for spatial epidemiology and wildlife management. Ecol Indic 36:594–600. doi: 10.1016/j.ecolind.2013.09.019 AeryonLabs (2014) Monitoring sensitive salmon populations on Snake River is safer with Aeryon Scout. In: Appl. Case Stud. http://www.aeryon.com/applications/casestudies/588-monitoring-sensitive-salmon-populations-on-snake-river-is-safer-with- aeryon-scout.html. Accessed 27 Nov 2014 Akaike H (1974) A new look at the statistical model identification. IEEE Trans Autom. Contr AC. pp 19:716–723 Alcaide M, Negro JJ, Serrano D, et al. (2010) Captive breeding and reintroduction of the lesser kestrel Falco naumanni: a genetic analysis using microsatellites. Conserv Genet 11:331–338. doi: 10.1007/s10592-009-9810-7 Anderson K, Gaston KJ (2013) Lightweight unmanned aerial vehicles will revolutionize spatial ecology. Front Ecol Environ 11:138–146. doi: 10.1890/120150 Arjomandi M (2007) Classification of Unmanned Aerial Vehicles. Mechanical Engineering. 3016. Aeronaut Eng 49. Averett N (2014) Drones Take Off as Wildlife Conservation Tool. Audubon Mag 1–8. Bailey DW, Gross JE, Laca EA, et al. (1996) Mechanisms that result in large herbivore grazing distribution patterns. J Range Manag 49:386–400. Bayliss P, Yeomans K (1989) Correcting bias in aerial survey population estimates of feral livestock in northern Australia using the double-count technique. J Appl Ecol 26:925–933. Berni JAJ, Zarco-Tejada P, Suárez L, Fereres E (2009) Thermal and narrowband multispectral remote sensing for vegetation monitoring from an unmanned aerial vehicle. IEEE Trans Geosci Remote Sens 47:722–738. BirdLife International (2011) Species factsheet: Falco naumanni. Blackburn T, Gaston K (1994) The distribution of body sizes of the worlds bird species. Oikos 127–130. 103 Bluff L a, Rutz C (2008) A quick guide to video-tracking birds. Biol Lett 4:319–22. doi: 10.1098/rsbl.2008.0075 Blyenburgh P Van (1999) UAVs - Current situation and considerations for the way forward. Dev. Oper. UAVs Mil. Civ. Appl. Bologna F, Mahatho N, Hoch D (2002) Infra-red and ultra-violet imaging techniques applied to the inspection of outdoor transmission voltage insulators. IEEE Africon. Electrotech. Serv. Africa. IEEE, 345 E 47 Th St. NY 10017 USA, Pretoria, South Africa, pp 593–598 Braza F, Alvarez F (1987) Habitat use by red deer and fallow deer in Doñana National Park. Miscellània Zoològica 11:363–367. Bugalho MN, Milne J a (2003) The composition of the diet of red deer (Cervus elaphus) in a Mediterranean environment: a case of summer nutritional constraint? For Ecol Manage 181:23–29. doi: 10.1016/S0378-1127(03)00125-7 Campoy, P., García, P.J., Barrientos, A., Cerro, J.D., Aguirre, I., Roa, A., García, R., Muñoz JM (2001) An stereoscopic vision system guiding an autonomous helicopter for overhead power cable inspection. Int. Work. Rob. Vis. Auckland, New Zealand, p pp 115–124 Cano M, Granados J, Castillo A, et al. (2007) Nuevas tecnologías aplicadas al seguimiento de ungulados silvestres en Sierra Nevada: Collares GPS-GSM. In: SGHG (ed) Biodivers. y Conserv. fauna y flora en Ambient. mediterráneos. Granada, pp 691–705 Casasús I, Riedel JL, Blanco M, Bernués A (2012) Extensive livestock production systems and the environment. In Animal farming and environmental interactions in the Mediterranean region. 81–88. Chabot D, Bird DM (2012) Evaluation of an off-the-shelf Unmanned Aircraft System for Surveying Flocks of Geese. Waterbirds 35:170–174. Chabot D, Bird DM (2013) Small unmanned aircraft: precise and convenient new tools for surveying wetlands. Jounal Unmanned Veh Syst 24:15–24. Chabot D, Carignan V, Bird DM (2014) Measuring Habitat Quality for Least Bitterns in a Created Wetland with Use of a Small Unmanned Aircraft. Wetlands 34:527–533. doi: 10.1007/s13157-014-0518-1 Chapman A (2014) It ’ s okay to call them drones. J Unmanned Veh Syst 2:1–3. CLAVE S.L. (1992) Análisis de impactos de líneas eléctricas sobre la avifauna de espacios naturales protegidos. Seville, Spain Collinge S (2010) Spatial Ecology and Conservation. Nat Educ Knowl 3:69. 104 Consejería de Medio Ambiente y Ordenación del Territorio. (2013) Cartografía y evaluación de la vegetación y flora de los ecosistemas forestales de Andalucía a escala de detalle (1:10.000). Spain Cooke S, Hinch S, Wikelski M, et al. (2004) Biotelemetry: a mechanistic approach to ecology. Trends Ecol Evol 19:334– 343. Coulombe ML, Massé A, Côté SD (2006) Telemetry Quantification and Accuracy of Activity Data Measured with VHF and GPS Telemetry. Wildl Soc Bull 34:81–92. Cox TH, Nagy CJ, Skoog MA, Somers IA (2004) Civil UAV capability assessment. 103. Cramp S, Simmons K (1980) Hawk to Bustards. Handb. Birds Eur. Middle East North Africa.Vol 2. Crivelli, A.J., Jerrentrup, H., Mitchev T (1988) Electric power lines: a cause of mortality in Pelecanus crispus Bruch, a world endangered bird species. Waterbirds 11:301–305. Crockford NJ (1992) A review of the possible impacts of wind farms on birds and other wildlife. Peterborough (United Kingdom) Dewar H (2013) Terrapin 1 drone soars, protecting rhinos from poachers. In: News. https://cmns.umd.edu/news-events/features/1101. Accessed 24 Nov 2014 Van Diermen J, van Manen W, Edwin B (2009) Terreingebruik en activiteitspatroon van Wespendieven Pernis apivorus op de Veluwe. Takkeling 109–133. Donázar J, Negro J, Hiraldo F (1993) Foraging habitat selection, land-use changes and population decline in the lesser kestrel Falco naumanni. J Appl Ecol 30:515–522. Dublin H (2011) African Elephant Specialist Group Report. Pachyderm 49:1–5. Duerr AE, Miller T a, Lanzone M, et al. (2012) Testing an emerging paradigm in migration ecology shows surprising differences in efficiency between flight modes. PLoS One 7:e35548. doi: 10.1371/journal.pone.0035548 Dunford R, Michel K, Gagnage M, et al. (2009) Potential and constraints of Unmanned Aerial Vehicle technology for the characterization of Mediterranean riparian forest. Int J Remote Sens 30:4915–4935. doi: 10.1080/01431160903023025 Eléctrica R (2005) Red Eléctrica y la avifauna: 15 años de investigación aplicada. 38. Elliot JM (1977) Statistical analysis of samples of benthic invertebrates. Ambleside Ellwood SA, Wilson RP, Addison AC (2007) Technology in conservation: a boon but with small print. In: Macdonals DW, Service K (eds) Key Top. Conserv. Biol. Blackwell Publishing, Padstow, UK, p 329 111 Pomarol M (1993) Lesser Kestrel recovery project in Catalonia. In: Nicholls M, Clarke R (eds) Biol. Conserv. small falcons Proc. 1991 Hawk Owl Trust Conf. The Hawk and Owl Trust, London, pp 24–28 Potapov, Utekhina I.G., McGrady M.J. RD (2013) Usage of UAV for Surveying Steller ’ s Sea Eagle Nests. Raptors Conserv 27:253–260. Pyper W (2008) Population survey pilots an unmanned aircraft. ECOS 17. Qamar FM, Ali H, Ashraf S, et al. (2011) Distribution and habitat seletion mapping of key fauna species in selected areas of Western Himalaya, Pakistan. J Anim Plant Sci 21:396–399. QGIS Development Team (2012) QGIS Geographic Information System Open Source Geospatial Foundation Project. R Foundation for Statistical Computing, Vienna A (2013) R: A Language and Environment for Statistical Computing. Redfern J V., Viljoen PCP, Kruger JMM, Getz W. M (2002) Biases in estimating population size from an aerial census: a case study in the Kruger National Park, South Africa. S Afr J Sci 98:455–461. Ribeiro E (2007) Seasonal variation in foraging habitat preference in Lesser Kestrel Falco naumanni. 30. Rodríguez A, Negro JJ, Bustamante J, et al. (2009) Geolocators map the wintering grounds of threatened Lesser Kestrels in Africa. Divers Distrib 15:1010–1016. doi: 10.1111/j.1472-4642.2009.00600.x Rodríguez A, Negro JJ, Mulero M, et al. (2012a) The eye in the sky: combined use of unmanned aerial systems and GPS data loggers for ecological research and conservation of small birds. PLoS One 7:e50336. doi: 10.1371/journal.pone.0050336 Rodríguez A, Rodríguez B (2009) Attraction of petrels to artificial lights in the Canary Islands: effects of the moon phase and age class. Ibis (Lond 1859) 151:299–310. doi: 10.1111/j.1474-919X.2009.00925.x Rodríguez A, Rodríguez B, Curbelo ÁJ, et al. (2012b) Factors affecting mortality of shearwaters stranded by light pollution. Anim Conserv 15:519–526. doi: 10.1111/j.1469-1795.2012.00544.x Rodríguez C, Tapia L, Kieny F, Bustamante J (2010) Temporal changes in lesser kestrel (Falco naumanni) diet during the breeding season in southern Spain. J Raptor Res 120–128. Rodríguez C, Wiegand K (2009) Evaluating the trade-off between machinery efficiency and loss of biodiversity-friendly habitats in arable landscapes: The role of field size. Agric Ecosyst Environ 129:361–366. doi: 10.1016/j.agee.2008.10.010 112 Room MHM, Ahmad a (2014) Mapping of a river using close range photogrammetry technique and unmanned aerial vehicle system. IOP Conf Ser Earth Environ Sci 18:012061. doi: 10.1088/1755-1315/18/1/012061 Ropert-Coudert Y, Wilson R (2005) Trends and perspectives in animalattached remote sensing. Front Ecol Env 3:437–444. Rutz C, Bluff L (2008) Animal borne imaging takes wing, or the dawn of “wildlife video-tracking.” Trends Ecol Evol 292–294. Rutz C, Bluff L, Weir A, Kacelnik A (2007) Video cameras on wild birds. Science (80- ) 765. Rutz C, Hays GC (2009) New frontiers in biologging science. Biol Lett 5:289–92. doi: 10.1098/rsbl.2009.0089 SACAA (2008) Interim policy for civil Unmanned Aircraft Systems in South Africa. Second draft. South Africa Sakamoto KQ, Takahashi A, Iwata T, Trathan PN (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. doi: 10.1371/journal.pone.0007322 SAMAA (2001) Manual of procedure for model aircraft and remote-piloted air vehicle operation. South Africa Sardà-Palomera F, Bota G, Viñolo C, et al. (2012) Fine-scale bird monitoring from light unmanned aircraft systems. Ibis (Lond 1859) 154:177–183. doi: 10.1111/j.1474- 919X.2011.01177.x SCB (2014) Society for Conservation Biology. http://www.conbio.org. Accessed 25 Nov 2014 Schiffman R (2014) Drones Flying High as New Tool For Field Biologists. Science (80- ) 344:2014. Schwarzbach M, Laiacker M, Mulero-Pazmany M, Kondak K (2014) Remote water sampling using flying robots. 2014 Int Conf Unmanned Aircr Syst 72–76. doi: 10.1109/ICUAS.2014.6842240 Shepard E, Lambertucci S, Vallmitjana D, Wilson R (2011) Energy Beyond Food: Foraging Theory Informs Time Spent in Thermals by a Large Soaring Bird. PLoS One e27375. Silvy NJ, Lopez RR, Peterson MJ (2012) Wildlife Marking Techniques. In: N.J.Silvy (ed) Wildl. Tech. Manual., The Johns . Baltimore, MD, USA., pp 230–257 113 Soriano P, Caballero F, Ollero A (2009) RF-based Particle Filter localization for Wildlife Tracking by using an UAV. 40 th Int. Symp. Robot. Barcelona, España., pp 239–244 Soulé ME (2007) What is Conservation Biology? Bioscience 35:727–734. Stolper R, Hart J, Mahatho N (2009) The design and evaluation of a multi-spectral imaging camera for the inspection of transmission lines and substation equipment. Insul News Mag Rep 1–14. Teal Group (2014) Teal Group Predicts Worldwide UAV Market Will Total $91 Billion in Its 2014 UAV Market Profile and Forecast. Paris, France Tella J, Forero M, Hiraldo F, Donázar J (1998) Conflicts between Lesser Kestrel conservation and European Agricultural Policies as identified by habitat analyses. Cons Biol 593–604. Tellería JL (1986) Manual para el censo de los vertebrados terrestres. Fac Biol Univ … 1–32. Tesla N (1898) Method of and apparatus for controlling mechanism of moving vessels or vehicles. Tracey JP, Fleming PJS, Melville GJ (2008) Accuracy of some aerial survey estimators: contrasts with known numbers. Wildl Res 35:377. doi: 10.1071/WR07105 U.S. Army (2010) Eyes of the Army – Unmanned Aircraft Systems. Roadmap 2010– 2035. 140. UK Ministry of Defence (2010) Unmanned Aircraft Systems: Terminology, Definitions and Classification. Joint Doctrine Note 3/10. 21. UNESCO (2014) World heritage list. http://whc.unesco.org/en/ list/685. Accessed 5 Feb 2014 Ursua E, Serrano D, Tella J (2005) Does land irrigation actually reduce foraging habitat for breeding lesser kestrels? The role of crop types. Biol Conserv 122:643–648. doi: 10.1016/j.biocon.2004.10.002 US Department of Defense (2014) US DoD. In: UAV Unmanned Aer. Veh. http://www.defense.gov/specials/uav2002/. Accessed 23 Nov 1BC Valavanis KP (2008) Advances in Unmanned Aerial Vehicles: State of the Art and the Road to Autonomy.Volume 33 of Intelligent Systems, Control and Automation: Science and Engineering. 568. Vermeulen C, Lejeune P, Lisein J, et al. (2013) Unmanned Aerial Survey of Elephants. PLoS One 8:e54700. doi: 10.1371/Citation 114 Viitala J, Korplmäki E, Palokangas P, Koivula M (1995) Attraction of kestrels to vole scent marks visible in ultraviolet light. Nature 373:425–427. doi: 10.1038/373425a0 Watts AC, Ambrosia VG, Hinkley E a. (2012) Unmanned Aircraft Systems in Remote Sensing and Scientific Research: Classification and Considerations of Use. Remote Sens 4:1671–1692. doi: 10.3390/rs4061671 Watts AC, Bowman WS, Abd-Elrahman AH, et al. (2008) Unmanned Aircraft Systems (UASs) for ecological research and natural-resource monitoring (Florida). Ecol Restor 26:10–15. Watts AC, Perry JH, Smith SE, et al. (2010) Small Unmanned Aircraft Systems for Low-Altitude Aerial Surveys. J Wildl Manage 74:1614–1619. doi: 10.2193/2009- 425 WcUAVC (2013) Wildlife Conservation UAV Challenge. http://www.wcuavc.com/. Accessed 20 Dec 2013 Whitworth C, Duller A, Jones D, Earp G (2001) Aerial video inspection of overhead power lines. Power Eng J 15:25–32. Wild S (2013) Drones now part of anti-poaching arsenal. In: Bussines day. http://www.bdlive.co.za/national/science/2013/02/08/drones-now-part-of-anti- poaching-arsenal. Accessed 10 Mar 2013 Wilkinson BE (2007) The design of the georeferencing techniques for unmanned autonomous aerial vehicle video for use with wildlife inventory surveys: a case study of the National Bison Range, Montana. 95. Williamson SP (2011) Utilization of Unmanned Aircraft Systems for Environmental Monitoring. 36. WWF (2013) Google Helps WWF Stop Wildlife Crime. http://worldwildlife.org/stories/google-helps-wwf-stop-wildlife-crime. Accessed 3 May 2013 Zavalaga CB, Dell’Omo G, Becciu P, Yoda K (2011) Patterns of GPS tracks suggest nocturnal foraging by incubating Peruvian pelicans (Pelecanus thagus). PLoS One 6:e19966. doi: 10.1371/journal.pone.0019966 Zenko M (2013) Reforming U . S . Drone Strike Policies. Council Special Report No. 65. Washington, USA 115 116 Acknowledgements The overall objective of this Ph.D. is to evaluate the use of UAS in conservation biology, identifying their capacities and limitations in the following applications t )PXDBO6"4DPOUSJCVUFUPFOWJSPONFOUBMJNQBDU assessment of infrastructures? t )PXDBO6"4DPOUSJCVUFUPNBOBHFNFOUPGFOEBOHFSFE species? t $POTFSWBUJPOJOBIVNBOEPNJOBUFEMBOETDBQF$BO6"4 constitute a useful tool for obtaining high-resolution spatiotemporal information on animals habitat use? t $POTFSWBUJPOJO B QSPUFDUFEBSFB"SF6"4DBQBCMF of providing information as valid as the obtained by conventional techniques on the spatial distribution of species in protected areas?