The lake sediment and tree cores have gotten us out to the field and demonstrated the types of methods used to collect paleodata. They have also provided us with new paleodata for the course. However, although going to the field to collect data is fun and important, it is the Bayesian statistics and ecosystem modeling that are the focus for camp. Without the statistical and ecosystem modeling tools, our analysis of the data would be very limited.
Our course is set up to introduce topics and then build on them throughout the week. For example, over Day 1 and 2 we have covered introductions to R, probability and Bayesian statistics, as well as dendrochronology, and ecosystem modeling. Day 3 was a big day for starting to put the pieces together with a statistical discussion in the morning by Chris Paciorek on running MCMC using JAGS and followed by an ecosystem model discussion by Mike Dietze about using PEcAn.
The following are a few nuggets that were covered on Day 3:
1. Bayes theorem, priors and posteriors.
2. Bayesian credible intervals vs. 95% confidence intervals – the Bayesian credible interval is what many ecologists typically think their 95% confident interval is!
3. Bayesian approaches are really nice for common issues in paleoecology – sampling that is not orthogonal, data that is not evenly spaced through time.
4. Acknowledging uncertainty and including it in statistical models is important.
5. When creating ecosystem modeling, it is important to quantify BOTH uncertainty in the data and the uncertainty in the models.
6. Think of models as scaffolds for data synthesis. Models are working theory on how things work. Models represent different spatial and temporal scales. Model acts like one giant covariance matrix.
7. Observations inform models but models can also inform what is measured in the field.
Lastly, it’s important to have fun and collaborate with your peers!