CANSSI OTN Hidden Markov Model Workshop 2023
This workshop was developed by Fanny Dupont, Natasha Klappstein, Arturo Esquivel, Marco Gallegos Herrada, Vinky Wang, Eric Ste-Marie, Ron Togunov, Vianey Leos Barajas, and Marie Auger-Méthé, for the 2023 CANSSI meeting and OTN ECR workshop. The goal of the workshop is to illustrate how hidden Markov models (HMMs) can be used to classify behaviours and identify behaviour-specific habitat associations using a range of movement and biologging data. It is divided in two parts: morning and afternoon.
- Intermediate R coding
- Familiarity with animal movement/telemetry/accelerometer data
- Before the workshop, please read the instruction in one of the Installing instruction files (e.g., 0_Installing_packages) files. Some of the packages can be tricky to install, so please install them before the workshop.
- Make sure all packages are up-to-date as older versions may not work
- Download and unzip workshop zip file from Github (press on the green "< > Code" button on upper right corner)
The tutorial will be completed together, and can be found in the activity folder and will begin with a 30-minute introduction that will provide an overview of the tutorial objectives, and will go through the statistical background required. The tutorial has some guided activities.
- Understand the statistical framework for HMMs and their application to animal movement data
- Select appropriate temporal resolution for HMM analysis
- Interpolate missing locations (linear, crw, path segmentation, and multiple imputation)
- Fit HMMs to animal movement data to identify behaviours
- Incorporate covariates on state transition probability to identify conditions that promote different behaviours
- Work through the activity file activity
- Follow the slides by opening slides
- If you get lost, you can follow along with the completed tutorial html file.
The session will begin with a 20-minute introduction to accelerometer data then move into a live coding demonstration of how a basic HMM can be fit to acceleration data. There will be exercises at the end and the solutions will be posted the following day.
- Fit a basic HMM to accelerometer data using
momentuHMM
- Incorporate and interprete covariates on behaviour transition probabilities
- Visualize the depth time series with decoded states
- Use the Akaike Information Criteria for model selection