314 research outputs found
New Analytical Methods for Camera Trap Data
No full textDensity estimation of terrestrial mammals has become increasingly important in ecology, and robust analytical tools are required to provide results that will guide wildlife management. This thesis concerns modelling encounters between unmarked animals and camera traps for density estimation. We explore Rowcliffe et al. (2008) Random Encounter Model (REM) developed for estimating density of species that cannot be identified to the individual level from camera trap data. We demonstrate how REM can be used within a maximum likelihood framework to estimate density of unmarked animals, motivated by the analysis of a data set from Whipsnade Wild Animal Park (WWAP), Bedfordshire, south England. The remainder of the thesis focuses on developing and evaluating extended Random Encounter Models, which describe the data in an integrated population modelling framework. We present a variety of approaches for modelling population abundance in an integrated Random Encounter Model (iREM), where complicating features are the variation in the encounters and animal species. An iREM is a more flexible and robust parametric model compared with a nonparametric REM, which produces novel and meaningful parameters relating to density, accounting for the sampling variability in the parameters required for density estimation. The iREM model we propose can describe how abundance changes with diverse factors such as habitat type and climatic conditions. We develop models to account for induced-bias in the density from faster moving animals, which are more likely to encounter camera traps, and address the independence assumption in integrated population models. The models we propose consider a functional relationship between a camera index and animal density and represent a step forward with respect to the current simplistic modelling
approaches for abundance estimation of unmarked animals from camera trap data. We illustrate the application of the models proposed to a community of terrestrial mammals from a tropical moist forest at Barro Colorado Island (BCI), Panama
Data‐driven counterfactual evaluation of management outcomes to improve emergency conservation decisions
No full textMonitoring is needed to assess conservation success and improve management, but naïve or simplistic interpretation of monitoring data can lead to poor decisions. We illustrate how to counter this risk by combining decision-support tools and quantitative counterfactual analysis. We analyzed 20 years of egg rescue for tara iti (Sternula nereis davisae) in Aotearoa New Zealand. Survival is lower for rescued eggs; however, only eggs perceived as imminently threatened by predators or weather are rescued, so concluding that rescue is ineffective would be biased. Equally, simply assuming all rescued eggswould have died if left in situ is likely to be simplistic. Instead, we used the monitoring data itself to estimate statistical support for a wide space of uncertain counterfactuals about decisions and fate of rescued eggs. Results suggest under past management, rescuing and leaving eggs would have led to approximately the same overall fledging rate, because of likely imperfect threat assessment and low survival of rescued eggs to fledging. Managers are currently working to improve both parameters. Our approach avoids both naïve interpretation of observed outcomes and simplistic assumptions thatmanagement is always justified, using the same data to obtain unbiased quantitative estimates of counterfactual support
Activity level estimation data
No full text<p>Camera-trapping data from Barro Colorado Island, Panama, used in Rowcliffe, J.M., Kays, R., Kranstauber, B., Carbone, C. & Jansen, P.A. Quantifying levels of animal activity using camera-trap data, Methods in Ecology and Evolution, accepted August 2014.</p
Economical crowdsourcing for camera trap image classification
No full textCamera trapping is widely used to monitor mammalian wildlife but creates large image datasets that must be classified. In response, there is a trend towards crowdsourcing image classification. For high‐profile studies of charismatic faunas, many classifications can be obtained per image, enabling consensus assessments of the image contents. For more local‐scale or less charismatic communities, however, demand may outstrip the supply of crowdsourced classifications. Here, we consider MammalWeb, a local‐scale project in North East England, which involves citizen scientists in both the capture and classification of sequences of camera trap images. We show that, for our global pool of image sequences, the probability of correct classification exceeds 99% with about nine concordant crowdsourced classifications per sequence. However, there is high variation among species. For highly recognizable species, species‐specific consensus algorithms could be even more efficient; for difficult to spot or easily confused taxa, expert classifications might be preferable. We show that two types of incorrect classifications – misidentification of species and overlooking the presence of animals – have different impacts on the confidence of consensus classifications, depending on the true species pictured. Our results have implications for data capture and classification in increasingly numerous, local‐scale citizen science projects. The species‐specific nature of our findings suggests that the performance of crowdsourcing projects is likely to be highly sensitive to the local fauna and context. The generality of consensus algorithms will, thus, be an important consideration for ecologists interested in harnessing the power of the crowd to assist with camera trapping studies
Statistical Development of Animal Density Estimation Using Random Encounter Modelling
No full textCamera trapping is widely used in ecological studies to estimate animal density, although these studies are largely restricted to animals that can be identified to the individual level. The random encounter model, developed by Rowcliffe et al. (J Anal Ecol 45(4):1228–1236, 2008), estimates animal density from camera-trap data without the need to identify animals. Although the REM can provide reliable density estimates, it lacks the potential to account for the multiple sources of variance in the modelling process. The density estimator in REM is a ratio, and since the variance of a ratio estimator is intractable, we examine and compare the finite sample performance of many approaches for obtaining confidence intervals via simulation studies. We also propose an integrated random encounter model as a parametric alternative, which is flexible and can incorporate covariates and random effects. A data example from Whipsnade Wild Animal Park, Bedfordshire, south England, is used to demonstrate the application of these methods
) in an African savanna at high temporal resolution
No full textDetermining the drivers of aboveground net primary production (ANPP), a key ecosystem process, is an important goal of ecosystem ecology. However, accurate estimation of ANPP across larger areas remains challenging, especially for savanna ecosystems that are characterized by large spatiotemporal heterogeneity in ANPP. Satellite remote sensing methods are frequently used to estimate productivity at the landscape scale but generally lack the spatial and temporal resolution to capture the determinants of productivity variation. Here, we developed and tested methods for estimating herbaceous productivity as an alternative to labour-intensive repeated biomass clipping and caging of small plots. We compared measures of three spectral greenness indices, normalized difference vegetation index derived from Sentinel-2 (NDVIs) and a handheld radiometer (NDVIg), and green chromatic coordinate derived from digital repeat cameras (GCC) and tested their relationship to biweekly field-measured herbaceous ANPP using movable exclosures. We found that a satellite-based model including average NDVIs and its rate of change (ΔNDVIs) over the biweekly productivity measurement interval predicted herbaceous ANPP reasonably well (Jackknife R2 = 0.26). However, the predictive accuracy doubled (Jackknife R2 = 0.52) when including the sum of day to day increases in camera trap-derived vegetation greenness (tGCC). This result can be considered promising, given the current lack of productivity estimation methods at comparable spatiotemporal resolution. We furthermore found that the fine (daily) temporal resolution of GCC time series captured fast vegetation responses to rainfall events that were missed when using a coarser temporal resolution (>2 days). These findings demonstrate the importance of measuring at a fine temporal resolution for predicting herbaceous ANPP in savanna ecosystems. We conclude that camera traps are promising in offering a reliable and cost-effective method to estimate productivity in savannas and contribute to a better understanding of ecosystem functioning and its drivers
Building skills in remote wildlife monitoring techniques
No full textRelatório de estágio de mestrado, Bioestatística, 2022, Universidade de Lisboa, Faculdade de CiênciasCamera traps have become a standard tool in wildlife management and conservation as they enable the
monitoring of unmarked populations. Methods that allow estimating animal density, such as the Random
Encounter Model (REM), require the estimation of three parameters i) encounter rate (between moving
animals and static cameras), ii) day range (average daily distance travelled), and iii) detection zone (effective area in which the cameras detect animals). To estimate the animal’s speed and detection zone,
we rely on the animal’s position data measured using a computer vision model that maps image pixel
position to position on the ground.
The model’s accuracy depends on the camera’s fixed position and the acquisition of calibration images
from its initial position. If the camera shifts, it may change the detection zone, which breaks the model
and makes animal positions in subsequent images unreliable. On the other hand, excluding images after the first camera movement may result in a significant data loss in the analysis. There is a lack of
information about how to proceed in this situation. In addition, data processing pipelines and camera
trap imagery software used in these tasks are under active development, raising questions about the most
effective way to apply them.
In this context, this report compares three methods used to deal with data when cameras move during
deployments and focuses on questions about the sensitivity of estimates in terms of accuracy and precision. It documents all the steps of generating, processing, and analysis of camera trap data for REM.
Our findings did not reveal significant differences concerning the density values estimated by the three
methods. The results presented in this report provide insights for future REM applications and encourage
users to share how they process their imagery and data
Building skills in remote wildlife monitoring techniques
No full textRelatório de estágio de mestrado, Bioestatística, 2022, Universidade de Lisboa, Faculdade de CiênciasCamera traps have become a standard tool in wildlife management and conservation as they enable the
monitoring of unmarked populations. Methods that allow estimating animal density, such as the Random
Encounter Model (REM), require the estimation of three parameters i) encounter rate (between moving
animals and static cameras), ii) day range (average daily distance travelled), and iii) detection zone (effective area in which the cameras detect animals). To estimate the animal’s speed and detection zone,
we rely on the animal’s position data measured using a computer vision model that maps image pixel
position to position on the ground.
The model’s accuracy depends on the camera’s fixed position and the acquisition of calibration images
from its initial position. If the camera shifts, it may change the detection zone, which breaks the model
and makes animal positions in subsequent images unreliable. On the other hand, excluding images after the first camera movement may result in a significant data loss in the analysis. There is a lack of
information about how to proceed in this situation. In addition, data processing pipelines and camera
trap imagery software used in these tasks are under active development, raising questions about the most
effective way to apply them.
In this context, this report compares three methods used to deal with data when cameras move during
deployments and focuses on questions about the sensitivity of estimates in terms of accuracy and precision. It documents all the steps of generating, processing, and analysis of camera trap data for REM.
Our findings did not reveal significant differences concerning the density values estimated by the three
methods. The results presented in this report provide insights for future REM applications and encourage
users to share how they process their imagery and data
The evolutionary ecology of animal information use and social dominance
No full textOrganisms are frequently faced with uncertainty regarding how best to exploit vital resources, and may benefit from collecting information about their distribution through space and time. However, the ways in which competition over resources might systematically facilitate or constrain an individual's ability to use information has been largely overlooked. In this thesis, I develop a conceptual framework for considering how the distribution of limited resources might underpin interdependencies between competition and information use. I focus on the evolutionary ecology of relationships between social dominance and social information use. I begin with an observational study of wild chacma baboons (Papio ursinus) suggesting that, when resources can be monopolised, individuals with low competitive ability are limited in their ability to use social information. Building on these findings, I then develop a general model exploring selection on social information use in a competitive context across three axes of 'resource ecology' (scarcity, depletion rate, monopolisability). This study makes predictions regarding the resource conditions under which competitive ability might constrain social information use, and the potential importance of social information use in the evolution of social dominance. I go on to test these predictions in chacma baboons using a field experiment. This experiment also explores whether the predictability of resource distribution might facilitate the decoupling of social information use from the competitive context in which it was collected. Taken together, these findings provide general insights into the combinations of ecological conditions and behavioural mechanisms that should underpin the benefits of social dominance. I end by building a simple population matrix model to study social dominance using an eco-evolutionary approach, in which feedback loops between ecological and evolutionary processes are considered. By modelling relationships between dominance rank and survival, reproduction, inheritance, and development, I am able to derive estimates of long-term fitness associated with dominance. Using these estimates, I generate predictions regarding how dominance hierarchies should impact the dynamics of group stability, viability, and fission.</p
Food aquisition and predator avoidance in a Neotropical rodent
No full textForaging activity in animals reflects a compromise between acquiring food and avoiding predation. The Risk Allocation Hypothesis predicts that prey animals optimize this balance by concentrating their foraging activity at times of relatively low predation risk, as much as their energy status permits, but empirical evidence is scarce. We used a unique combination of automated telemetry, manual radio telemetry and camera trapping to test whether activity at high-risk times declined with food availability, as predicted, in a Neotropical forest rodent, the Central American agouti (Dasyprocta punctata). We found that the relative risk of predation by the main predator, the Ocelot (Leopardus pardalis), estimated as the ratio of ocelot to agouti activity on camera trap footage, was up to four orders of magnitude higher between sunset and sunrise than during the rest of the day. Kills of radio-tracked agoutis by ocelots during this high-risk period far exceeded expectations given agouti activity. Both telemetric monitoring of radio-tagged agoutis and camera monitoring of burrow entrances indicated that agoutis exited their burrows later at dawn, entered their burrows earlier at dusk, and had lower overall activity levels, as they lived in areas with higher food abundance. Thus, agoutis avoided activity during the high-risk period more strongly as access to food was higher. Our study provides quantitative empirical evidence of prey animals concentrating their activity at times of relatively low predation risk
- …
