1. The use of detection dogs to effectively monitor bird and bat fatalities at wind farms is becoming increasingly popular. All studies to date agree that dogs outperform human searchers at finding bird and bat carcasses around wind turbines; however, it remains unclear how particular conditions during the search may influence carcass detection.

2. We investigate the effect of carcass size, habitat characteristics and weather conditions on carcass detection probability, for both dogs and humans, using data from the monitoring program of a wind farm in Spain.

3. A generalized linear model reveals a high performance of dogs (~80% detection probability), with no clear influence of any of the variables analysed. Humans, on the contrary, were markedly affected by the size of the carcass and to some extent, by the vegetation structure. Humans performed poorly at detecting small carcasses (~20% detection probability), more so in dense vegetation.

4. Synthesis and applications. Our results provide evidence that dogs perform at a high level under a wide range of environmental conditions. They are particularly well-suited for the monitoring of fatalities of small, rare or inconspicuous species in cluttered environments. Humans, by contrast, are very poor at detecting all but the largest carcasses.16-Jun-2020

We compared the effectiveness of dogs and humans searchers at finding bird and bat carcasses under wind turbines. We used two Belgian Malinois and one Border Collie, between 2 and 4 years old. We also used two human searchers (alternatively) who were experienced in searching for birds and bats at wind farms for the human detection trials.

When a dog-handler team was to search at a turbine, the handler would place 1-3 carcasses at allocated random locations. Carcasses would be thrown from a distance to avoid leaving a track on the ground that the dog could follow. To prevent the handler from giving any clues to the dog as to where the carcasses were placed, the handler always walked in a fixed search pattern during searches, at a constant pace and only communicated with the dogs to ensure they stayed within the search band. Handlers only approached and rewarded the dogs after they displayed a proper detection alert. The search pattern of the dog-handler team consisted of four parallel line transects on each search plot. The handler followed straight line transects traversing the search area, while the dog was allowed to move freely within approximately 10 meters at either side of the handler. Searches were conducted perpendicular to the wind direction to maximize the probability of coming in contact with the scent cone produced by the carcasses. An average of five wind turbines were surveyed per day. Detection trials were performed weekly from 24th January 2012 through to 30th July 2013, in days without precipitations or strong winds. We conducted 589 detection trials, using 50 different species.

Human trials followed the same procedure as dog trials, although in this case it was a helper who placed the carcasses prior to the search. Humans searched along 8-10 parallel line transects aiming to cover a 5 m wide band at each side of the transect. Trials for human searchers were conducted on a selection of 15 wind turbines out of the turbines searched by the dogs. Therefore, all turbines searched by humans were also searched by dogs but not the other way around. The sample of turbines searched by humans covered a similar proportion of habitats as those searched by the dogs. Humans surveyed 5 wind turbines per day. Fewer detection trials were conducted for human searchers than for dogs; between the 16th October 2012 through to the 30th July 2013, on days without precipitations or strong winds, we conducted 121 detection trials using 28 different species.  We used the same range of carcass sizes, and in a similar proportion, for human and for dog detection trials.

We used carcass species wingspan, as a measure of carcass size. To characterize the vegetation in each search plot, we measured six variables related to vegetation height and cover in four non-overlapping circular areas of 5 m radius. The location of the centre of these areas was defined randomly using a random point generator. For each measured variable, we took the average value of the four areas and used it as a covariate in the analysis. We used the daily mean wind speed (km/h) and mean temperature (°C) for the Masegoso municipality (one kilometre away from the nearest wind turbine), obtained from the meteorological web site www.eltiempo.es. 

To reduce the dimensionality of the vegetation descriptors and to prevent multicollinearity in the models, we performed a principal component analysis (PCA)

Each row in the dataset corresponds to the search of one carcass. The dataset contains the following variables:

- id: unique identifier of the search event.

- detect: 1 - carcass detected, 2 - carcass not detected

- date: date of the search

- type: type of searcher - dog, human

- ind: individual searcher: d1, d2, d3, h1, h2

- size: carcass species wingspan (cm)

- turb: turbine at which the search was conducted

- cobground: bare ground cover (%)

- cobveg0_10: cover of vegetation between 0 and 10 cm high (%)

- cobveg10_30: cover of vegetation between 10 and 30 cm high (%)

- cobveg30_100: cover of vegetation between 30 and 100 cm high (%)

- cobveg100_300: cover of vegetation between 100 and 300 cm high (%)

- cobveg300_: cover of vegetation higher than 300 cm (%)

- temp: mean day temperature (degrees celsius)

- wind: mean day wind speed (km/h)

- pc1: vegetation structure principal component 1

- pc2: vegetation structure principal component 1

- pc3: vegetation structure principal component 1