On the Effectiveness of UAS for Anti-Poaching in the African Arid Savanna

This paper describes a field study that examined the effectiveness of unmanned aerial vehicles (UAV) in anti-poaching enforcement in parks and game reserves. In the field study, a UAV attempted to spot mock poachers while the mock poachers tried to spot the UAV. The field study was conducted at N/a’an ku sê, an operational game reserve in the central region of Namibia. In total, 118 trials were completed, providing 236 UAV-poacher interdiction scenarios. Of these, 198 were during the day, 152 with a quadcopter and 46 with a fixed-wing. Live spotting success during the day varied due to the hiding behavior of the mock poachers, with the highest and lowest success rates of spotting being 86% for poachers in the open and 25% for poachers hiding under canopy cover. The UAVs were demonstrated to be a potentially effective tool for anti-poaching patrol and interdiction, in part, because of their ability to spot poachers. The pursuit of integrating the UAV into current anti-poaching patrol and interdiction efforts in arid savanna landscapes is strongly recommended.


14
The high rate of animal extinction of such high-value African species, such as rhinoceros and elephant, is 15 credited to poaching (Western 1985;Douglas 1987). They are illegally hunted for their horns and tusks, 16 and sold for the production of traditional medicines and for items of cultural status. From 1990 to the 17 present, the African elephant population decreased by 90% and between 1960 and 1990 the black rhino 18 population decreased by 95% (Kamminga, 2018). To combat the decline of high-value African species, 19 effective anti-poaching policies require strong support by local communities, strong anti-poaching efforts 20 by law enforcement, and strong prosecution by legal systems. In law enforcement, the focus is on best 21 practices in patrol and interdiction. Today's tools include the camera trap, ground surveillance (walking 22 and driving), and aircraft surveillance with the latter often being unaffordable. The advent of the 23 unmanned aerial vehicle (UAV) offers law enforcement with a new, potentially powerful tool. To date, published research on the effectiveness of the UAV in anti-poaching enforcement is scant. 26 Exceptions include anti-poaching UAV research on 13 farms across South Africa with 20 total trials 27 (Mulero-Pazmany, 2014). Their research focused on image quality of three types of cameras mounted on 28 a fixed-wing UAV for day and nighttime trials. They found some viability in using UAVs as an anti-29 poaching tool to spot both rhinos and humans. Also, a security group implemented an anti-poaching UAV 30 program using a fixed-wing UAV as a nighttime anti-poaching tool (Air Shepherd, 2018). They reported 31 statistics regarding their successes, noting that in one area in which 19 rhinoceros were once killed, there 32 were no killings after deploying their UAVs. Information on their methodology, however, was not 33 released. Ultimately, the role of the UAV in enforcement, despite early signs of great promise, is not yet 34 settled. How will the UAV compliment camera traps, walking, and driving at parks and game reserves? 35 What overall level of training and effort will be required? At the root of these questions is the more basic 36 question, which this paper examines, of the relative abilities of the UAVs and the poachers to sense, by 37 sight and sound, each other's presence.

39
The method section describes a field study in which a UAV attempts to spot mock poachers while the 40 mock poachers try to spot the UAV. The different factors that impact the ability to spot one another are 41 discussed in the results section and the implications of the study to patrol and interdiction are described in 42 the summary and conclusions section. The objective of this study was to examine the anti-poaching UAV and the poacher's abilities to sense, by 46 sight and sound, each other's presence. These abilities, once understood, could be applied to the patrol 47 scenario, where the focus is on protecting an area, and to the interdiction scenario, where the focus is on 48 3 apprehending poachers. Indeed, the parameters that characterize the abilities and constraints of sensing 49 each other's presence will drive how to best integrate the UAV into patrol and interdiction. Currently, the 50 three most popular methods employ (1) camera traps, (2) walking, and (3) driving. Today, camera traps 51 are most commonly used as a way to collect information after-the-fact, to use in prosecution and to learn 52 poacher habits. Such countries as Tanzania and Borneo have used them effectively. Collecting real-time 53 data with them, on the other hand, has been difficult due to technical hurdles and manpower requirements 54 to process data. Also, camera traps must be hidden from the view of the poachers and covering large, open 55 savannas with camera traps is problematic. Walking and driving are currently the two most popular 56 methods of patrolling and interdicting poachers. A pair of well-trained security personnel can walk 3 to 5 57 km in a day without being detected. The areas in which they walk are unrestricted by roads and they can 58 exploit all of their senses. On the other hand, security personnel can cover much less area on foot than by 59 ground vehicle and walking is particularly dangerous when confronting a poacher. In a ground vehicle, a 60 pair of security personnel routinely covers 75 km in a day. In contrast with walking, the ground vehicle is 61 readily detected by the poacher so the ground vehicle serves as a deterrent and is most commonly deployed 62 around perimeters to areas. In interdiction, it serves as a rapid means of getting close to an area and as a 63 means to herd poachers to desirable locations. In light of these considerations, the UAV has the potential 64 of complimenting the walking and driving approaches. In patrol, it can cover an area relatively fast. Before 65 needing its battery changed, a simple, low-weight radio controlled or autonomous multi-rotor UAV can 66 cover about 8 km in twenty minutes and a comparable fixed-wing UAV can cover 15 km to 20 km in 30 67 minutes. Like the ground vehicle, the UAV collects visual data but, unlike walking, it does not capitalize 68 on the other senses. In patrol, the UAV is well-suited to following perimeters. It can cover a much wider 69 swath than the ground vehicle can and its vantage point from the air is much better than the vantage point 70 from the ground. As a result, the UAV can protect a perimeter much better than a ground vehicle can. In information about the poachers, such as their number, arms, and heading. Since the UAV is remote, it also 73 brings greater safety to both patrol and interdiction.

75
The basic question is therefore this: What are the practical constraints of the UAV with regard to the 76 comparative abilities of the UAV and the poacher to sense each other's presence? These abilities, once 77 understood, could then be applied to the patrol scenario, where the focus is on protecting an area, and 78 could be applied to the interdiction scenario, where the focus is on apprehending poachers. Note that this   As shown, the test area was divided into six study zones. Each had differences in terrain and vegetation 88 cover. Each measured 500m by 500m which was sufficiently large to challenge the UAVs and the mock 89 poachers and which was sufficiently small to enable the sighting to be accomplished in an appropriately 90 small amount of time. As shown, each of the zones was further subdivided into a grid of nine sectors, each 91 sector serving as a hiding area for the mock poachers. Three hiding behaviors were defined as follows: The tests were performed over a 10-week period in 2018. Each test was accomplished by a six-person 97 field unit consisting of two mock poachers, a pilot, a co-pilot, a note-taker and an observer. The roles were 98 randomized after each trial to reduce biases resulting from variations in skill level. The pilot was 99 responsible for setting up the radio control equipment, launching and landing the UAV, and video 100 observation. The co-pilot was responsible for the initial UAV setup and video observation. The note-taker 101 was responsible for recording all data, before, during and after the test. The observer maintained situational 102 awareness and visual contact with the UAV and operated the radio for safety purposes. In each test, the 103 mock poachers were assigned to a random sector and to a random hiding behavior.

105
Before the beginning of each test, role assignments, wind speed and direction, cloud cover, flight altitude, 106 flight speed, UAV camera angle, and GPS takeoff location were recorded. Next, at the beginning of each 107 test, the field unit set up its equipment beside the takeoff location while the mock poachers went to their 108 assigned random hiding sectors and assumed their hiding behaviors. Then, upon takeoff, the note taker 109 started a stopwatch and the observer told the poachers via radio to start their stopwatch. The two synced 110 stopwatches were later used to correlate UAV and mock poacher observations. After takeoff, the UAV 111 autonomously flew a pre-programmed search pattern, with the pilot and co-pilot continuously monitoring 112 a live video feed. Any time the pilot and co-pilot thought they spotted a mock poacher, the time was 113 recorded. During the test, the mock poachers recorded the GPS coordinates of their hiding spot and their 114 assigned hiding behavior and the times they first heard the UAV, saw the UAV, and when the UAV was 115 directly overhead. The note-taker entered this data into the dataset at the conclusion of the test.

117
A post processing review of photos and recorded video allowed the sightings to be either confirmed or 118 marked as false and it enabled sightings that the pilot and the co-pilot missed to be identified. Three UAV systems were chosen: two quadcopter systems and a fixed-wing UAV system. They were 121 meant to be representative of the different systems that would be employed in anti-poaching efforts. flown in the north-south directions to prevent flying into and away from the sun. The ability to adjust 128 parameters throughout the flights was also necessary to obtain the best results. The adjusted parameters 129 were camera angle, flight speed, altitude, and recording mode (stills and video). For the fixed-wing system, 130 a modified version of the parallel track search method was flown to account for its limited turning radius.

131
The difference in flight paths is shown in Fig. 2 below. The quadcopter systems required two batteries to complete the search pattern, returning to the takeoff 137 location to change batteries half way through the trial. The fixed-wing system was capable of completing 138 two search patterns per battery, and, to minimize wear and tear on the UAV, was allowed to circle, rather 139 than land between tests. In the night tests using the quadcopter system with a FLIR camera, the onboard 140 lights were blacked out except for the navigation lights, which were allowed to remain turned on during 141 takeoff and landing. Several color pallets were available for the thermal imagery however, after testing,    Flight speeds for all of the trials ranged from 4.7 m/s to 9.5 m/s, as shown in Table 3. These were divided 180 into three ranges with the majority of the trials falling into the lower two of the three ranges. This is   Table 4 below shows the success rates for spotting poachers in the field during the daytime using both the 188 fixed-wing and quad platforms. Overall the quadcopter system was more successful than the fixed-wing   The observers watching the live video noted that spotting with the fixed-wing was more difficult because    Table 5).   In all of the cases, day or night, the mock poacher heard the UAV well before spotting it or being spotted.   In this study, the quadcopter proved more successful than the fixed-wing, primarily due to its gimbaled 321 camera, higher quality live video transmission, and slower flight speeds. With further improvements the 322 fixed-wing would be expected to provide similar success rates to the quad during the day while also 323 covering more ground and staying aloft longer. At night the fixed-wing would be more difficult to 324 implement due to the need to take off and land on a clearly lit runway.

326
While the vehicles in this study proved capable of spotting poachers, for them to act as a better covert 327 tool, the noise signature could be reduced. Choosing quieter propulsion systems or flying at higher 328 altitudes would achieve the necessary noise reduction, and higher altitudes would also allow each transect 329 to cover a wider strip of land. However, flying at higher altitudes would also compound the issue of 330 spotting small figures live. To assist in spotting and positively identifying poachers, the use of larger 331 screens and higher quality video for live viewing would help observers be more effective. At night, the 332 slow frame rate and resolution of the thermal camera was a hindrance so a higher resolution thermal 333 camera, with higher frame rate would also be an improvement. During the day, image quality was not the