Roe deer is a species that hides their neonates as an anti-predator strategy. This may prove efficient against mammalian predators, such as the red fox; however, it might be an ecological trap as large numbers of fawns are killed by tractors with harvesters each year during grass harvest. Here, we evaluate a low-tech, low-cost method, using scaring flags placed in the field the evening before grass harvest, intended to stimulate the roe deer doe to remove any fawns from the field. We evaluate the effects of the scaring flags by counting the number of fawns in the fields using a multi-rotor drone equipped with a thermal camera. The drone proved to be an efficient tool for detecting roe deer fawns. However, contrary to other studies, we found limited or no effect of placing the scaring flags in the field. This discrepancy highlights the fallacy of generalizing based on single case studies. There is still a need to develop low-cost, easily applicable tools to reduce roe deer fawn mortality during grass harvest. However, with the increasing technological development, drones with thermal cameras are a promising tool for wildlife management and monitoring in the future.

The study was conducted in Nærøy Municipality, Nord-Trøndelag County (64° 48’ N, 11° 16’ E,). The landscape features a mixture of agricultural fields (mainly grass production for livestock feed), and forest patches. The field size of the area is rather small, with an average surveyed field size of 2.4 ha.

Experimental setup

In mid-June 2017 and end of May to mid-June 2018 we surveyed fields prior to grass harvest, using a drone equipped with a thermal camera (see “Drone setup” below). In 2017 we surveyed 38 fields and in 2018 we surveyed 69 fields. Roe deer fawns leave clearly visible heat signatures in the field in the early morning before the sun warms up the grass and reduces heat contrast.

We utilized an experimental design where we first surveyed fields using drones in the morning (07:00-10:00), then randomly selected fields for the treatment and control groups. We surveyed the same fields with drones a second time the following morning (07:00- 10:00). The 2nd flight represents the harvest itself during which any remaining roe deer fawns could potentially be harmed. This resulted in a total of 52 control fields and 55 treatment fields. In the treatment fields, scaring flags (plastic shopping bags, approximately 30x40 cm) on 2m sticks were set up in the evening (18:00-20:00), control fields were not visited. As such, the scaring flags themselves, as well as the action of placing the flags in the field can be viewed as part of the experimental treatment. The scaring flags were set up 10- 20m from the field edge assuring maximum distance of 50m between flags. The scaring flags were left to stand until the second drone survey the following morning (07-10:00). This design simulates a grass harvest timeline where scaring devices are deployed, the evening before harvest.

Drone setup

We used a DJI Phantom 3 Professional multi-rotor drone with an effective flight time of < 15 minutes, depending on weather conditions. The drone was equipped with a FLIR TAU 2 640 long-wave infrared (LWIR) thermal camera (resolution: 640×512 pixels; field of view: 90° horizontal, 69° vertical; frame rate: 9 Hz; spectral range: 7.5–13.5 μm). Thermal images were recorded onboard via the TeAx module and downloaded post-flight, as no live image stream was available. Images were manually analysed in ThermoViewer 2.1.6. To avoid double counting, we used visible landmarks (e.g. field edges, trees, vehicle tracks, and other features) in neighbouring images in combination with the drone’s GPS data to accurately identify individual fawns.

Quality control

We utilized an experimental approach to our study, thus controlling for any unwanted variation in our data caused by e.g. location of the surveyed fields (high versus low roe deer use/abundance) or weather conditions that could affect e.g. 1) fawn detectability, 2) roe deer activity or 3) scaring flag flapping. However, as an additional quality control of the classification of roe deer fawns versus other potential objects/animals from thermal images we: 1) Analysed a subset of the survey images immediately after the second flight and visited positions at which fawns were detected to confirm the image analysis. 2) In cases where we discovered fawns while putting out the scaring flags in the evening after the first flight – we viewed the thermal images from the prior morning to check if the fawns were present in the thermal image. We only did this quality control if we were able to identify the exact location of the potential object with reference to other structures in the image such as shape of the field, powerlines, paths, roads, boulders etc. We defined confirmed fawns as either the presence of fawns or distinctive fresh bedsites, characterised by completely flattened grass. These bedsites are unique to roe deer fawns and differ from resting places of other animals, such as hares or birds, commonly found in these fields. The surveyed fields would also contain multiple old fawn bedsites, however these would 1) look different as grass would gradually right itself after the fawn has left, 2) would not provide any heat signature on the thermal images as we fly in the morning while these bedsites are still shaded by grass and therefore not warmed by the sun. Thus, the identified bedsites will represent a fawn that was present at the time of the drone flight, but that moved during the hours it took to download and review the mages and the subsequent visit to the site. In 25 out of 26 such checks, we found 1) either a fawn or a fresh bedsite of a fawn when it first was identified in the thermal image, or 2) in the case of discovering a fawn in a field, we were able to go back to the thermal image and identify the observed fawn.