Seabird Watch: testing out of the box image analysis methods and citizen science data for large-scale monitoring of seabirds.

Seabird Watch: testing out of the box image analysis methods and citizen science data for large-scale monitoring of seabirds.

Can we use count data, nest-based or burrow monitoring to compare reproductive strategies within species across space and time? Secondly, can such approaches be automated to allow for large scale comparisons across taxa?

Background

Seabirds worldwide are in rapid decline, yet basic breeding, attendance and survival data is missing between sites to understand differing threats across regions. In particular, we need to understand how widespread species react and possibly adapt to environmental perturbations and anthropogenic stressors.

Recent advances in video camera and data logging technology have led to a burgeoning of applications in a wide variety of fields. Web-cams and video data loggers are now being widely used to remotely observe animals in their natural habitats, providing crucial new insights for understanding their biology and conservation.

At present, we can collect large amounts of imagery, but the cross-utility of analytical techniques mean that tools need to be redeveloped for each new study system. By combining human annotations using citizen science data from SeabirdWatch, we intend to produce generic statistics for nest or burrow-nesting seabirds and test these against ‘out of the box’ visual analytical methods.

Project Description

Many large-scale questions in ecology are confounded by whether local, regional or global scales are studied, with contrasting population trajectories and behaviours, and confounding effects of population structure or barriers. At the small scale, effects may have a strong signal but be anecdotal or specific to a colony, compared to noisy, confounded signals at a larger scale. While seabirds are in global decline, they show strong ecological differences in breeding success and phenology between regions and years. We therefore need to integrate high resolution data at the local level with global, representative monitoring.

Camera technology gives us the opportunity to simultaneously monitor many colonies and to understand how behaviours and strategies differ over space and time. The ability to scale up this form of monitoring depends on the ability to extract data from imagery on the same scale and as fast as it is collected. Using Penguin Watch citizen science platform, we have already done this for penguins and have had success in automating the process. We recently launched Seabird Watch with the same aims over a much wider geographical range and taxa.

Using previously collected data from seven study species (Australasian gannet, black-legged kittiwake, northern gannet, black guillemot, Sandwich tern and puffin), this project will develop machine-learning techniques to automate analysis of breeding chronology, nest attendance, parental investment, chick growth and survivorship of colonial seabirds from video recordings. The study will provide the suitable candidate the opportunity to work at the cutting-edge of video data analysis to address a variety of pressing behavioural ecology and conservation outcomes. Data from Australasian gannets will come from a unique colony (~150 nests) located on an artificial structure (Popes Eye) in Port Phillip Bay, south-eastern Australia, that has been the focus of research for several decades. The nesting individuals are habituated to humans (enabling easy capture/handling and the deployment of animal-borne data loggers), the majority are of known age and sex, and information on their foraging ecology and diet has already been (and continues to be) obtained. In addition, the entire colony is constantly monitored by web-cam during daylight hours. The project will use these features to develop automated video analysis tools to be used in conjunction with basic biological information and remote-sensed data to investigate intrinsic and extrinsic factors influencing the timing of breeding, parental investment, fledging success, potential for extra-pair copulations, and disease/parasite prevalence.

Data from puffin, kittiwake and guillemot species will come from timelapse cameras set up or coordinated by Seabird Watch, Deakin University and the RSPB and include many colonies where sex and age structure is unknown, but which have annual timelapses from which breeding phenology and survivorship can be calculated.

This project would suit someone who is passionate about complex conservation projects and data driven large-scale ecology. The ideal candidate would already be proficient in R or python and have some field experience (not essential and training will be given).

Aims of the Project

To develop machine-learning or automated parameter estimation from camera data on seabirds.

Methods to be used

Take citizen science aggregated data and use this to automate data extraction from counts to measures of breeding success and foraging effort.

Specialised skills required

The student will need to be proficient in R and preferably in machine-learning. The student will also need to be capable of carrying out fieldwork in remote areas although this will be with substantial support and training.

Please contact tom.hart@zoo.ox.ac.uk if you are interested in this project.

Associated Pages