Drone-Based Thermal Imaging in the Detection of Wildlife Carcasses and Disease Management

Because animal carcasses often serve as reservoirs for pathogens


Introduction
In recent years, the increasing spread of infectious diseases by wildlife has raised concerns regarding the related threats to humans and domestic animals [1,2].Such rising pathogen emergence can be driven by growing anthropogenic impacts on nature, leading to increasing human-wildlife interactions [1,3].Contact between humans and wildlife can lead to the transmission of zoonotic diseases, while contact between wildlife and livestock threatens the health of domestic animals and, consequently, food production [4].Developing strategies aimed at preventing the spread of diseases originating in wildlife is therefore of utmost importance.As animal carcasses play a key role in pathogen transmission [5], their timely removal is recommended to prevent disease spread, especially in the case of highly infectious diseases such as African swine fever [6,7].
African swine fever (ASF) is a viral disease that originated in wild suids in Africa.However, it can be transmitted to other suids, including domestic pigs and wild boar (Sus scrofa), and has recently spread to Europe, Asia, and the Caribbean [8,9], causing a pandemic among suids.Te severity of the disease, as illustrated by the high case fatality ratio, and the transmission dynamics have led to massive economic losses in pig production and drastic reductions in the size of afected wild boar populations [7,10].Direct transmission of the ASF virus among wild boar can result from direct contact with infected animals or with the carcass of an infected wild boar (e.g., during scavenging or chewing on bones); indirect transmission is possible as well, such as by rooting in contaminated soil at the carcass site [6,11,12].Because the carcasses of wild boar that have died from ASF can remain infectious for several months [13][14][15], they are considered to be one of the main drivers of ASF transmission [10,16,17].
To contain the disease, it is crucial that wild boar carcasses are found and removed quickly [15,18,19].Te active search for wild boar carcasses in high-risk areas has been proposed as the most efcient way to detect and counteract the disease [7,20,21].However, the problem is that infected wild boars tend to hide and rest in more dense vegetation and choose those locations, especially for their deathbeds, which complicates subsequent discovery of their carcasses [22][23][24].Developing and improving strategies for the rapid detection of wild boar carcasses is therefore crucial for a successful control of ASF [19,20,25].
Although a carcass cools down quickly after the death of the animal, its temperature rises again during decomposition.Tis is mainly due to the activities of necrophagous insects, whose closely packed larvae hatch within a short time period to form so-called maggot masses.Metabolic heat is generated by maggots feeding on body tissue, with the magnitude of the temperature rise increasing with increasing mass size [26].Since maggot eggs are oviposited during the early stages of carcass decomposition, maggots often accelerate the decomposition process to a stage of active decay, during which the level of metabolic heat is highest [27][28][29].Insect activity and the formation of maggot masses depend on the ambient temperature, as both are only possible above certain temperature thresholds [26,30,31].Te temperature characteristics and dependency of maggots can thus be exploited to improve carcass detection by thermal imaging [7].
Forward-looking infrared (FLIR) cameras (hereafter, thermal cameras) which can visualize heat-emitting objects have been used in the remote monitoring of wildlife [32], especially for the tracking of ungulates [33][34][35][36][37].While thermal cameras were frst mounted on helicopters [28,29], with reductions in sensor weight they can now be used with unmanned aerial vehicles (hereafter, drones), which are relatively low cost and more fexible in their applications [27].Temperature diferences detected by thermal cameras have formed the basis of several forensic studies testing the feasibility of thermal imaging in the detection of human remains, with pig cadavers often used as human surrogates [27,29,38].However, the application of thermal sensors is limited by several interfering factors, such as a dense forest canopy and the presence of other heat-emitting sources in the surrounding environment [39].
In this study, we examined the use of drone-based infrared imagery in fnding wild boar carcasses, as part of the efort to contain diseases such as African swine fever.Specifcally, we (1) tested the accuracy of a thermal camera in measuring carcass temperature and the biotic and abiotic factors that infuence measurement accuracy.We then identifed (2) the environment-and (3) carcass-related variables that determine the probability of carcass detection.

Study Area.
Carcass exposure and drone fights were conducted in the Bavarian Forest National Park (BFNP, 249 km 2 , 49 °3′ 19″N, 13 °12′ 9″E), located in southeastern Germany.Te elevation in the park ranges from 600 m.a.s.l to 1453 m.a.s.l on mountain peaks.Te annual air temperature varies between 3.5 and 7.2 °C, and the annual precipitation ranges from 830 mm to 1820 mm.Snow cover can persist from October to May, with the greatest depth from January to March [40,41].Te area is covered by temperate forests composed mainly of Norway spruce (Picea abies), European beech (Fagus sylvatica), silver fr (Abies alba), common rowan (Sorbus aucuparia), and sycamore maple (Acer pseudoplatanus) [42].Te BFNP provides habitat to large mammals, including wild boar (Sus scrofa), red deer (Cervus elaphus), roe deer (Capreolus capreolus), and large carnivores, such as Eurasian lynx (Lynx lynx) and gray wolf (Canis lupus).Mesocarnivores, such as red fox (Vulpes vulpes) and badger (Meles meles), as well as diferent bird species, such as common buzzard (Buteo buteo) and raven (Corvus corax), are frequent scavengers in the study area.Flyovers took place directly from the nearest forest road in a straight line over the known carcass site, with a starting altitude above the carcass of 40 m.Te drone was then navigated back and stopped directly over the location of the carcass.If the underlying vegetation allowed it, we descended the fight altitude to 25 m vertically towards the carcass.If a carcass was detected at 40 m, the fight altitude was raised vertically to 60 m to further evaluate the detection capability of the thermal camera.Using the HeatTrack function, the camera recognizes the hottest object in the camera's view and indicates it by a red point on the monitor of the remote controller, which provides a live feed of the thermal camera.Accordingly, a carcass was designated as found when the drone detected it as the warmest point, and if that point remained stable.If a point switched between the carcass and the surroundings, we did not consider this carcass as detected at that given fight altitude.At all fight heights, the carcass temperature (if measurable), air temperature, and coldest spot measured by the infrared camera of the drone were noted.Te body temperature of each carcass was manually recorded based on measurements in the rectum, mouth, and on the carcass' surface and underside immediately before or after the drone fight, using a handheld thermometer (Voltcraft DT-300).Te temperature of the maggot masses was also determined.Heat is not evenly distributed over the carcass, as it often develops frst at natural body openings, where fies preferentially lay their eggs (Figure 2).As the thermal camera can detect local heat sources, the maximum measured temperature was used for the analysis.Temperature values were recorded based on stable values, i.e., those displayed within 10 s.

Environmental and Carcass
Conditions.Te temperature diference was calculated between the maximum temperature measured on or in the carcass using a handheld thermometer and the drone-based measurements.Habitat types were classifed as meadow, deadwood area, mixed forest, coniferous forest, and deciduous forest based on Silveyra Gonzalez et al. [43].Te soil condition was either wet or dry, based on our observations.Canopy openness was calculated from hemispherical photographs taken with a 180 °fsheye lens, and Gap Light Analyzer software [44] was then used to extract the percentages of openness (higher percentages represent greater canopy openness).Sky conditions were characterized as sunny or cloudy; dawn referred to fights that took place during nautical and civil dawn, i.e., in the early morning hours on a clear day with the sun still below the horizon.Te air temperature was calculated as the mean of the air temperature measurements obtained by the thermal camera at the three fight altitudes: 25 m, 40 m, and 60 m.Te number of fyovers for each identifed carcass was also recorded.Te decomposition stage was defned as fresh, putrefaction, bloated, post-bloated, advanced decay, and dry remains based on Lee Gof [46] and Anderson and Van Laerhoven [70] (Table 1 and Figure 3).
Further binary variables were the presence of maggot masses and the weight of the carcass.Te diference between the maximum carcass temperature measured by a handheld thermometer and the air temperature was also calculated (Table 2).Transboundary and Emerging Diseases 2.4.Statistical Analysis.First, the accuracy of the carcass temperature measurements obtained by the thermal camera was assessed by comparison with the maximum carcass temperatures measured on the ground.For this, we ftted a generalized linear model regressing drone-based over handheld measurements.Measurement accuracy was evaluated using the coefcient of determination (R 2 ) and rootmean-square error (RMSE, given in °C, range of the data Table 1: Te six decomposition stages used in the statistical analysis, based on Anderson and Van Laerhoven [45] and Lee Gof [46].Two GLMMs with binomial distributions were then established to determine (model 2) which of the environmental parameters and (model 3) which of the carcass parameters infuenced the detection probability of a wild boar carcass.Environmental variables were diferentiated from carcass-specifc factors because information on the former is available in advance to a drone pilot and allows the fight day or time to be adjusted accordingly, unlike carcass parameters, which are not known prior to carcass detection.Detection success served as the response variable for both models (1 � found; 0 � not found).Te carcass ID number was included as a random intercept, and for model 2, the number of fights for each carcass was added as a random slope to account for diferences in the number of fight days.Model 2 also included habitat type, soil condition, canopy openness, air temperature, and sky condition as independent variables (Table 2).Canopy openness and air temperature were included with a second-order orthogonal polynomial.In model 3, the explanatory variables were decomposition stage, presence of maggot masses, and the diferences between the air temperature and the carcass temperature (second-order polynomial) (Table 2).Te GLMMs were constructed using the R package glmmTMB [47].
Finally, two conditional inference trees (CTrees) were ftted to derive thresholds for the environmental and carcass-related factors that determine the probability of carcass detection, using the same variables as in models 2 and 3. CTrees are decision trees based on unbiased recursive partitioning using a signifcance test to select input variables.Te package partykit [48] was used to build the CTrees [49].A 27% proportion of observations was used to establish a terminal node; an increased minimum sum of weights in a terminal node was applied to calculate the infuence of environmental variables; and a 46% proportion of observations was used to establish terminal nodes for possible predictor variables of the carcass.
All statistical analyses were performed using R 4.0.2[50].Statistical signifcance was assumed for P values <0.05.
In the test of the accuracy of the thermal camera, the generalized linear model yielded a value for R 2 of 0.75 and the RMSE calculation value of 5.9 °C (Figure 4).According to model 1, the accuracy of the thermal camera increased signifcantly and linearly with greater canopy openness (51.789 ± 8.382, t � −6.18, P < 0.001, orthogonal polynomial: coefcients estimated are not on the response scale) (Table 3).Sky conditions signifcantly increased measurement accuracy, with a higher accuracy obtained in fyovers conducted on days with an overcast sky (4.666 ± 1.269, t � 368, P < 0.001) or at dawn (8.337 ± 2.234, t � 3.73, P < 0.001).Model 1 also indicated a lower accuracy at a fight altitude of 60 m (1.242 ± 0.206, t � 6.02, P < 0.001) and a higher accuracy at a fight altitude of 25 m (−1.267 ± 0.206, t � −6.14, P < 0.001) than at a fight altitude of 40 m.
In model 2, the probability of carcass detection was signifcantly higher in meadows than in forests (3.920 ± 1.549, t � 2.53, P � 0.011).Te probability of carcass detection also increased signifcantly and linearly with greater canopy openness (28.331 ± 7.410, t � 3.82, P < 0.001) (Table 4) and was signifcantly higher on days with cloudy sky conditions than on days with sunshine (2.546 ± 0.486, t � 5.23, P < 0.001).In addition, on days with a clear sky, the carcass detection probability was higher in the early morning hours, when the sun was not yet above the horizon, than later in the day, under full sun (1.514 ± 0.677, t � 2.24, P � 0.025).Also, the probability of carcass detection increased at warmer air temperatures and followed a curvilinear (second-order) relationship at higher temperatures (linear term: 15.530 ± 5.224, t � 2.97, P � 0.003, quadratic term: −10.332 ± 3.991, t � −2.59, P � 0.010).
Similar to model 2, the CTree for environmental variables provided signifcant thresholds (P < 0.01) for four variables: habitat, canopy openness, sky conditions, and air temperature (Figure 5).Te frst split of the decision tree indicated habitat as the main factor afecting the probability of carcass detection using thermal imaging, diferentiating between forest types (mixed, coniferous, and deciduous) and open habitats (deadwood areas and meadows).Within forests, canopy openness was the next most relevant factor.According to our data, for a canopy openness of <29.3%, the probability of carcass detection was <25%.When the canopy openness exceeded 29.3% greater detection success was consistent with the infuence of sky conditions, as fights conducted in cloudy weather or during early morning hours (dawn) resulted in a higher detection probability (<70%) than fights conducted in sunshine (<40%).Te highest detection probability (>70%) was in open areas such as deadwood or meadows and at air temperatures >3.0 °C; at colder temperatures, the detection probability was <25%.
Similar to model 3, the CTree provided signifcant thresholds (P < 0.01) for decomposition stage, presence of maggot masses, and the temperature diference between the carcass and the air.Te latter was determined by the CTree algorithm as the most infuential variable, with a threshold of 6.4 °C (Figure 6).Temperature diferences >6.4 °C resulted in the highest probabilities (≤80%) of carcass detection, together with the presence of maggot masses in the decomposition stages post-bloated and advanced decay.Te probability was still >60% for carcasses in the dry remains stage but <30% for those in the inner decay stages such as fresh, putrefaction, and bloated.

Discussion
Tis study assessed the potential advantages and limitations of drone-based infrared technology to locate wild boar carcasses and examined the factors that infuence the detection probability.Te rapid detection of carcasses is critical to the successful containment of emerging wildlife diseases such as ASF.Our results demonstrate that a drone equipped with a thermal camera can detect carcasses and accurately measure carcass temperatures.Greater canopy openness, cloudy sky conditions, and a lower fight altitude positively infuenced the accuracy of the carcass temperature measurements, while open habitats such as meadows positively infuenced the detection success.In forests, the detection probability was higher when canopy openness exceeded 30%, with higher success rates achieved during cloudy sky conditions or fights at dawn.An ambient temperature >3 °C was shown to be necessary for carcasses to develop enough detectable heat.Detection success was also infuenced by a diference between the carcass temperature and the ambient air temperature of >6.4 °C.Higher success rates were possible for active carcass decay stages, especially postbloating and advanced decay.Tese stages are characterized by the development of maggot masses, which was the most important determinant of detection success.
Our results revealed a fair match between the carcass temperature and the temperature measured by the thermal camera, which forms the basis of carcass detection.Te observed accuracy was previously defned as the estimated average obtained from the small thermal sensors ftted on drones [51], and it corresponded with the accuracy of the Zenmuse XT2 thermal sensor, which is ±5 °C (https://www.dji.com/zenmuse-xt2).Te slight diferences in accuracy can  Air temperature was included using a second-order orthogonal polynomial.Transboundary and Emerging Diseases be attributed to surface emissivity and environmental factors such as air temperature, humidity, and sky conditions [52,53].Tese same factors might also explain the wider spread in the data at higher temperatures (Figure 4).Previous studies that tested the accuracy of thermal cameras reported slightly better values (e.g., R 2 � 0.70-0.96and RMSE � 3.18-5.45)because the tests were conducted either indoors in a temperature-controlled room or outside within controlled settings [54][55][56][57].Cooled thermal cameras would be an alternative, as they provide greater sensitivity, higher spatial resolution, and faster frame rates, but they need to be carried by larger vehicles, such as helicopters, which are more costly and not readily available [53].
Our temperature measurements were infuenced by fight altitude, as the accuracy decreased with increasing elevation.Tis can be attributed to atmospheric attenuation but also to the fact that a lower fight altitude allows the object of interest to fll more of the camera's feld of view.

Transboundary and Emerging Diseases
Te same dependency of fight height was reported in previous studies [27,56,58,59].However, Bodnar et al. [58] recommended fights at higher altitudes (50-100 m) because they allow the observation of a larger area at a time [58].In the study of Lee et al. [28], carcasses were detected from a 1 km fight height using a thermal camera on a helicopter [28].In our study, a 25 m fight altitude was not always possible because of tall trees, while at a 60 m fight altitude, both the measurement accuracy and the detection probability declined.Terefore, a fight height of 40 m ofers a good compromise between the coverage of larger areas and a higher detection probability.Canopy cover also infuenced the temperature measurements, with a denser canopy resulting in less accurate drone measurements.If a carcass was covered by vegetation, the thermal camera was not able to accurately measure its temperature, whereas for a carcass detected in habitats with less canopy cover, the temperature measurements were highly reliable.Within carcass temperatures ranging from −1 °C up to 60 °C, a RMSE of 5.89 °C was deemed acceptable.
Our results showed that canopy openness, sky conditions, and habitat type infuenced the detection probability of wild boar carcasses.Te carcass detection probability was higher in meadows than in forests.Within the latter, the detection probability increased with greater canopy openness.A threshold of 29.3% openness was determined below which carcasses were unlikely to be found (detection probability <25%).Tis fnding is in accordance with those of other studies in which the detection of wildlife in diferent forest types became increasingly limited with denser canopy cover [34,60].Terefore, drone-based searching should preferably be conducted in more open areas, given the very high probability of missing carcasses in densely forested areas.Tis limitation is a drawback of drone use in eforts to prevent or reduce the spread of ASF, as infested wild boars are often found in continuous forests [61] and typically withdraw to denser, mostly young forests, for their deathbeds [7,22,24].In denser forests, a fight design with overlapping overfights might increase the detection probability.Furthermore, carcass searches with specially trained dogs might help to increase the detection rate.Several studies have tested the ability of this approach in the search for remains under diferent ecological conditions and were able to show that, in the detection of carcasses, dogs were more efcient, with a higher detection rate and a shorter time span between detections, than humans alone [62][63][64].However, whether the use of dogs to detect wild boar carcasses will have a sufcient impact on the containment of ASF still needs to be tested.
Sky conditions also infuenced the detection probability.Flyovers were successful when the sky was cloudy, as the lack of sunshine prevented heating of the surrounding vegetation.Based on our experience, the latter can happen within minutes after direct sunlight exposure.Especially treetops or tree stumps can heat up quickly.Termal cameras do not diferentiate between emitted or refected heat [38], such that refected heat is detected with a similar signature as the heat emitted from a carcass [39].Masking efects due to sunshine and refected solar radiation are a well-known problem in the application of thermal cameras in ecology [27,28,38,39].Direct solar radiation can also raise the temperature of a carcass, by 10-20 °C compared to a carcass in the shade [39].Within our study, the thermal camera measured carcass temperatures up to 65.5 °C, which were likely reached by the absorption of direct sunlight (open canopy, ambient temperature ≈ 30 °C).We therefore recommend that, in searching for carcasses using thermal imaging, the searches should be conducted on overcast days or in the early morning hours of clear days, when the sun is still below the horizon and does not heat up the carcass or the surrounding vegetation.Ambient air temperature also signifcantly afected the carcass detection probability.Te warmer the air temperature during the fyover, the greater the likelihood of carcass detection.Our analysis showed that the probability of carcass detection <25% when the ambient air temperature was <3 °C.At these lower temperatures, the slow rate of metabolic decomposition and the absence of insects [65] hinder the rise in carcass temperature.During winter, carcass decomposition is much slower [59].In the study of Hohmann et al. [39], at temperatures <10 °C, there were no signs of carcass decomposition or heat development.Based on our observations, carcass detection with thermal cameras on colder days was only possible if the carcass was already in an advanced stage of decomposition.At this stage, maggots had most likely already invaded the carcass during warmer temperatures while at colder temperatures microbial activity continued to emit heat.By contrast, wild boars that die in late autumn or winter often decompose very slowly, with no noticeable signs of decomposition for several weeks or months [66].Terefore, the applicability of thermal imaging to detect wild boar carcasses in winter is rather limited.
Our analysis of carcass parameters identifed the variable temperature diference between the carcass and the ambient temperature as the most important, with a threshold difference of 6.4 °C needed for detection.Tis result is in accordance with those of Lee et al. [28], who determined that a threshold of ≥8 °C is sufcient for carcass detection by thermal imaging [28].A relatively large temperature difference was previously shown to be crucial for the use of thermal devices in carcass location [59].Te presence of maggot masses also signifcantly increased the detection probability.When maggot masses were present on a carcass, our analysis revealed with up to 80% the highest chance for detection (Figure 6).Tis is in accordance with results from Butters et al. [27], who found that heat detection peaked with insect activity [27] while Lee et al. [28] showed that, in the absence of maggots, the detection probability decreased [28].Maggot masses develop in larger amounts when the air temperature is warmer [67] and emit proportionally more heat [68].Temperatures in the range of 22-25 °C favor the development of maggot masses [26], which in turn can cause a carcass to become much warmer than the surrounding air.On overcast days, the largest diferences recorded between the manually determined carcass temperature and air temperature and between the carcass temperature measured by the thermal camera and the air temperature were 36.3 °C and 17.9 °C, respectively (carcass with maggots present, advanced decay, 62 days post-mortem, air temperature: −1.7 °C, max.carcass temperature � 34.5 °C, 04.05.2021, 07: 50).During fyovers on sunny days, the largest diference between the carcass temperature as recorded by the thermal camera and the air temperature was 31 °C.Studies measuring ground-based carcass temperatures reported that these peak at ∼26 °C above air temperature [29,39].In comparisons of carcass temperature measured by thermal cameras and the surrounding ground area temperature, the diferences were in the range of ∼8-12 °C [27,28].Amendt et al. [29] recorded a maximum carcass temperature of ∼40 °C, reached in the presence of maggot masses.In our study, in the presence of maggot masses and relatively large temperature diferences, carcass detection was at times possible even in forests with increased canopy closure.
In the usual process of decomposition, fy maggots develop on a carcass during active decay stages and form masses mostly during the post-bloating stage, which enables further the advanced decay stage [69,70].In our study, carcasses in the fresh and putrefaction stages of decomposition had a lower detection probability (<30%, 4 out of 28) than those in the advanced decay stage.During the stages of active decay, when maggots are present, the detection probability was ≥75%.Fresh refers to the frst stage after death, with cooling of the body, until the frst signs of inner putrefaction appear [70].Following the death of an animal, the body cools quickly [29] such that the probability of detection of a still-warm body by thermal imaging is possible only within a very limited period.Most of the carcasses detected in our successful fyovers were in the advanced decay stage, which was for our study also the longest decomposition stage.Although, during this stage, maggots start to abandon the carcass in order to pupate, the residual emitted heat is usually sufcient to be detected by a thermal camera.Lee et al. [28] observed similar detectability and highlighted the usage of thermal imaging for searching purposes in the advanced decay stage [28].Te detection rate in carcass searches will be even higher when fyovers are conducted during a period of warmer ambient temperatures, when the likelihood of the presence of maggot masses is also higher [28,59].
Our analysis revealed a probability of over 60% to detect carcasses in the stage of dry remains.Tere is evidence that microbially mediated decomposition can raise carcass temperatures even in the absence of maggots or solar radiation [30,58].Accordingly, the successful detection in this study of carcasses in the dry remains stage (detectable even at colder temperatures) can therefore most likely be explained by microbial processes.Des Marais [59] tested thermal imaging on pig carcasses and found no diference in the temperature of carcasses in the dry remains stage vs. that of the surroundings [59], whereas in the study of Hohmann Transboundary and Emerging Diseases et al. [39], carcasses were detected as late as after day 50 [39].Amendt et al. [29] and Lee et al. [28] also detected carcasses several weeks post-mortem, but those still showed the presence of maggots or insect activity and were in the advanced decay stage [28,29].Te detectability of older carcasses with thermal imaging may be useful in the management of ASF [28], especially during colder seasons, as older carcasses in late stages of decomposition may still be an active source of ASF infection.It has been shown that ASFV remains infectious at colder temperatures for several months, and that wild boars have contact with dead conspecifcs weeks after death [7,10,14].In general, wild boars seem to approach carcass sites containing carcasses of their conspecifcs on average after a week, with direct contact after around 2 weeks and often in later stages of decomposition [71].Terefore, the detection of older carcasses can contribute to the successful management of ASF and other diseases.

Conclusion
Using drones for the rapid detection of wild boar carcasses to prevent the spread of diseases, in particular ASF, is gaining increasing interest.Tis study demonstrated the feasibility of drone-based thermal imaging to detect large-animal carcasses, but the success of this approach depends on environmental and carcass factors.Our results provide guidelines for the optimization of search eforts, by identifying the conditions leading to the highest probability of detection: cloudy days, early morning hours, ambient temperatures >3.0 °C, and canopy openness >30%.Under these conditions, the benefts of drones, i.e., their rapid coverage of large areas, including early succession stages of deadwood areas with difcult access, can best be exploited.In other habitats and conditions, complementary methods, such as the use of dogs, will still be needed.Further research should examine whether the search for infected carcasses can be improved using thermal cameras with higher sensitivities during more favorable conditions, for example, at night.It should also be systematically tested up to which fight speeds thermal cameras can detect carcasses.Te fundamental knowledge presented herein will contribute to the development of measures aimed at preventing the spread of wildlife diseases, such as ASF in wild boar.
Design.Between September 2020 and July 2021, 42 wild boar carcasses were regularly deployed at a rate of four carcasses every 3 weeks.Due to high snow cover, cold temperatures, and thus the expected lack of visible carcass decomposition, deployment and drone fights were not performed between 23 rd November 2020 and 3 rd March 2021.Te carcasses were placed in open habitats (meadows, deadwood areas afected by bark beetle disturbance) and forests.At each deployment, two carcasses were placed in habitats with an open canopy (n � 22) and two others in relatively closed forests (n � 20), with one carcass placed on dry and the other on wet soil (dry: n � 21, wet: n � 21) (Figure 1).Te setup used in carcass detection consisted of a DJI Matrice 200 V2 drone equipped with a Zenmuse XT2 thermal camera ftted with a FLIR Tau temperature sensor (longwave infrared thermal camera, Tau 2 thermal core, 2 Transboundary and Emerging Diseases spectral range: 7.5-13.5 μm, heat sensitivity <50 mK at f/1.0, 30 Hz max.frame rates, 4 K visual sensor, 13 mm lens, and feld of view: 25 °× 19 °) (https://www.dji.com/zenmuse-xt2,SZ DJI Technology Co., Shenzhen, China).

Figure 1 :
Figure 1: Two forests with diferent canopy openness.Percentage of canopy openness was used as a predictor variable to analyze its infuence on the probability of fnding carcasses using thermal imaging from drones.

Figure 2 :
Figure 2: Termal infrared image of a 10-day old wild boar carcass in autumn, showing the typical pattern of heat emission from the snout and behind the animal's neck, due to the frst appearance of feeding fy maggots.Small white spots indicate insects on the carcass.

Figure 3 :
Figure 3: Typical decomposition of a wild boar carcass placed in a forest with wet soil and a closed canopy in summer 2020.Te decomposition stage after deployment: (a) bloated (7 days); (b) post-bloated (14 days); (c) dry remains (42 days).

Figure 4 :
Figure 4: Temperature measurements by drone-based thermal imaging compared to measurements on the ground.R 2 and RMSE in model 1 were 0.75 and 5.9 ( °C) (within a data range of −1 °C to 60 °C), respectively.Te red line represents a 1 : 1 line (indicating a perfect correlation).

Figure 5 :
Figure5: Conditional inference tree (CTree) predicting the infuence of environmental factors on the probability of wild boar carcass detection using thermal imaging from drones.Each tree node represents a split of the data into signifcant subsets according to the contribution of variable importance.Split nodes include the number of observations and the signifcance level.Te x-axis shows the predicted probability of carcass detection using drone-based thermal imaging (not found � 0, found � 1).Treshold values and factors responsible for splitting of the dataset are labeled on all branches.

Figure 6 :
Figure 6: CTree used in the analysis of the infuence of wild boar carcass factors on the probability of wild boar carcass detection using drone-based thermal imaging.For a detailed description of the CTree structure, see Figure 5.

Table 2 :
Characteristics of the variables used in the statistical analysis.°Cto 60 °C).Te variables that infuenced the accuracy of the temperature measurements were determined in a generalized linear mixed model (GLMM) (model 1) within which the diference in the carcass temperature as measured with the thermal camera vs. manually was the dependent variable and fight altitude(25 m, 40 m, and

Table 3 :
Estimated regression parameters, standard errors, t values, and P values for generalized linear mixed-efect model 1, evaluating the efect of environmental variables and diferent fight altitudes on the accuracy of temperature measurements of the drone-based thermal camera.

Table 4 :
Parameter estimates of generalized linear mixed-efect model 2, assessing the infuence of environmental parameters on the probability of wild boar carcass detection using drone-based thermal imaging.

Table 5 :
Estimated regression parameters, standard errors, t values, and P values for generalized linear mixed-efect model 3, evaluating the efect of wild boar carcass factors on the probability of wild boar carcass detection using drone-based thermal imaging.