Instructor Notes
Concepts
How (generally) do species distribution models work?
- Model relationship between species occurrence and environment and use to predict species presence at unsampled places and times
What is the distinction between environmental & geographic space?
- Environmental space shows how locations are related to one another in terms of similarity in environmental conditions
What is the distinction between using SDMs for explanation vs. interpolation vs. extrapolation
- Explanation: understanding why a species occurs in some areas and not others
- Interpolation: understanding where is species should occur within a region at points that haven’t been surveyed
- Extrapolation: understanding where/when is species should occur outside of the bounds of current sampling
Predictors
How do we select predictors for SDMs (or other ecological models)?
- Ideally: the variables that directly drive species distributions
- Practically: what data is available
What should we do if we don’t have measurements of our preferred variables?
- Try to find proxies that are likely to be well correlated with preferred variables
Modeling approaches
What are some of the major categories of modeling approaches for SDMs and what are their characteristics?
- Statistical
- Machine learning
- Process based
How do differences in these approaches relate to their use of explanation vs. prediction?
- Machine learning mostly useful for prediction due to difficulty of interpretation
- Process based best for explanation because they parameters of direct biological meaning
How are presence only data dealt with when building models?
- Pseudo absences
Challenges
Why do you think SDMs are so popular in attempts to forecast ecological change?
What are the challenges for making SDM based forecasts?
How might we determine how well these forecasts perform?
- Hindcasting
Key Processes
Should SDMs model the influence of other species and do they currently do this?
- Yes if possible
- JSDM
What about other key biological processes like density dependence and dispersal?
- Biological process are good, if you can successfully measure/model them
Uncertainty
How is uncertainty addressed in SDMs and are there areas for improvement?
- Probabilistic forecasts
- But often not evaluated
- Often converted into presence/absence maps
Future
What should be the next big steps in species distribution modeling?
- JSDMS
- More process
- Active evaluation