Module 8: Time Series Modelling

Objectives

To use statistical and machine learning models to analyse high-frequency environmental time series, and to ask: how can we use time series models to understand and predict ecosystem conditions?

By the end of this module you should be able to:

  • explore relationships between ecological variables in environmental case studies
  • understand the structure of several time series models, including machine learning models
  • fit time series models to environmental data and assess the importance of driver variables
  • process an environmental dataset into the standard format needed to train and test a model
  • compare models on their out-of-sample predictions

Before you start

  • Revisit your Module 6 forecast — there you calibrated a process-based model; here you let the data pick the structure
  • Have a think about what “a good model” means when you can always add more predictors

Module Resources

This week you’ll be working through the Macrosystems EDDIE module Time Series Modeling and Prediction of Environmental Data, which runs in your browser — there is nothing to install.

Start on the Introduction tab and download the student handout — this is the worksheet you will complete as you go and submit at the end of the module.

The app saves your progress through Activity A, but not through Activities B and C, because those use uploaded data. Download your plots and fill in the handout as you go — if the session drops, that work is gone.

Exercises

Work through the three activities in order, answering the questions in the handout as you go.

Activity A: Visualise Data & Fit a Model

  1. Choose one of the environmental case studies and visualise its data — look at the response variable through time and its relationship with the candidate drivers.
  2. Fit a time series model to the case study, then assess it: how well does it reproduce the observations, and which driver variables matter?

Activity B: Bring Your Own Data

  1. Choose a second dataset from the module, or upload your own, and get it into the standardised format the app expects.
  2. Fit and assess a time series model for this new dataset.
Most of the work in time series modelling is getting the data into shape — consistent time steps, sensible units, and gaps handled explicitly. Note down anything you had to fix.

Activity C: Compare Models

  1. Fit several more models to your Activity B dataset, including at least one machine learning model.
  2. Compare the models on data they were not trained on, and decide which you would use to predict future conditions — and why.

Submission

Submit the completed student handout from the module:

  • All questions answered, for Activities A, B and C
  • The plots you generated in the app inserted at the points indicated in the handout
  • Your model comparison, with a short justification of the model you would choose