Machine learning

What is Machine Learning in Ecology?

The broad goal of machine learning is to provide machines with ways to improve at tasks as they get more experience of those tasks. In many cases the task the machine is doing might take humans many months, or it could be something a machine can do more accurately. In our work, this means developing and applying methods that perform various kinds of ecological classification and prediction using a wide range of input data types. The data we use with machine learning methods include: species observations recorded by structured surveys or citizen science projects; images from cameras, sonar devices, drones, or satellites; acoustic recordings; video recordings; and telemetry data, among others.

One research theme is to develop computer-assisted approaches to assist or automate the labelling of ecological datasets, and to integrate this with other approaches in statistical ecology. Speeding up this process allows for more efficient and responsive monitoring and management. For example, we have developed machine learning models for identifying gibbon calls in acoustic files obtained by placing recorders in the forest; for identifying seals in images produced by sonar devices in rivers; for identifying diving behaviour in penguins from videos attached to them; for counting animals in aerial images; and to match photographs of the same individual. Often the outputs of this labelling process are used as inputs to statistical models, for example density estimation or spatial capture re-capture. Other areas where we use machine learning are: classification or prediction of animal behaviour from telemetry, identifying boats engaging in fishing activity, and estimating trends in bird populations from citizen science data.

What species are these methods used for?

With the wide range of input data types that can be accommodated, machine learning methods can be used with almost any species that data is collected on. A selection of the species we have worked includes crickets, dolphins, gibbons, salmon, seals, toads, ungulates, many bird species, and whales.

two pictures of the same dolphin dorsal fin, taken 4 years apart — Identifying that these two images are of the same bottlenose dolphin (photos taken four years apart) is important for accurate estimation of population sizes.
Credit: Sea Mammal Research Unit, University of St Andrews

Who in CREEM works on these methods?

A few relevant publications by CREEM staff

Acoustic classification

Wang, Y., Ye, J. and Borchers, D.L. (2022) Automated call detection for acoustic surveys with hierarchically structured calls of varying length. Methods in Ecology and Evolution 13: 1552-1567.

Dufourq, E., Batist, C., Foquet, R., & Durbach, I. (2022). Passive acoustic monitoring of animal populations with transfer learning. Ecological Informatics, 70, 101688.

Dufourq, E., Durbach, I., Hansford, J. P., Hoepfner, A., Ma, H., Bryant, J. V., … & Turvey, S. T. (2021). Automated detection of Hainan gibbon calls for passive acoustic monitoring. Remote Sensing in Ecology and Conservation, 7(3), 475-487.

Image classification

Dȩbicki, I. T., Mittell, E. A., Kristjánsson, B. K., Leblanc, C. A., Morrissey, M. B., & Terzić, K. (2021). Re-identification of individuals from images using spot constellations: a case study in Arctic charr (Salvelinus alpinus). Royal Society Open Science, 8(7), 201768.

Conway, A. M., Durbach, I. N., McInnes, A., & Harris, R. N. (2021). Frame‐by‐frame annotation of video recordings using deep neural networks. Ecosphere, 12(3), e03384.

Fell, C., Mohammadi, M., Morrison, D., Arandjelović, O., Syed, S., Konanahalli, P., … & Harris-Birtill, D. (2023). Detection of malignancy in whole slide images of endometrial cancer biopsies using artificial intelligence. Plos one, 18(3), e0282577.

Other data types

Mendo, T., Smout, S., Photopoulou, T., & James, M. (2019). Identifying fishing grounds from vessel tracks: model-based inference for small scale fisheries. Royal Society Open Science, 6(10), 191161.

Fink, D., Johnston, A., Strimas-Mackey, M., Auer, T., Hochachka, W. M., Ligocki, S., … & Rodewald, A. D. (2022). A Double Machine Learning Trend Model for Citizen Science Data. arXiv preprint arXiv:2210.15524.

Murgatroyd, M., Photopoulou, T., Underhill, L. G., Bouten, W., & Amar, A. (2018). Where eagles soar: Fine‐resolution tracking reveals the spatiotemporal use of differential soaring modes in a large raptor. Ecology and Evolution, 8(13), 6788-6799

six harbour seals lying on seaweed and rocks — Counting and individually identifying harbour seals hauled out on the Scottish coast.

spectrograph of animal calls, with frequency on y-axis and time on x-axis. — Identifying Hainan gibbon calls in spectrograms, a visual representation of an audio segment. The Hainan gibbon call is in the blue box; other animals and noise sources are in red.