| title | Syllabus for "Spatial modelling of distance sampling data" |
|---|---|
| author | David L Miller and Eric Rexstad |
This document defines the syllabus for the "Spatial modelling of distance sampling data" course. I've attempted to separate the lectures into "chunks" that are 45 minutes long, followed by a practical that takes up a similar amount of time (I think it's important that we leave plenty of time to explore the data, get comfortable with Arc/R and so on).
- Perhaps worth having the confirmed participants go through a questionnaire beforehand to see what the level of the audience is.
- Data set names (in monospace font after reference) relate to data that is freely available and shipped with R packages.
- Have folks log into mozpad to post questions and answer questionnaires anonymously.
Day/Time Objective 0900-1030 1045-1215 1345-1515 1530-1700
Sunday Learn R Panic Panic R workshop R workshop
Monday Fit detection function, DS covariates, variance, Intro to DS via current Simulation engine Simulation engine estimate abundance with model checking/selection vignettes. Covars, Horvitz-Thompson. variance. model selection, variance Simulation engine.
Tuesday Fit and check a density surface model (DSM) What is a DSM? Intro to generalized Practical GAM/DSM Adding covariates: why, additive models (GAMs) how, what? Intro to environmental data
Wednesday Add environmental Practical covariates to the model Multiple smooths, model Fit and check multismooth Predictions & variance Catch-up period Produce maps of predicted selection DSMs theory and in practice abundance
Thursday Advanced topics/mrds mrds Practical Advanced DSM topics
mrds
- Line transect sampling
- General sampling setup
- Exact distances vs. grouped/binned distances
- Assumptions
- Detection functions
- What are the options?
- What do they look like?
- Fitting detection functions
- Brief R examples
- Explanation of general syntax
- Interpreting
summary
- Investigating the data, basic EDA in R
- Fitting a detection functions
- Interpreting
summaryresults - Some setup for covariate models (plot group size etc against
$\hat{p}$ )
- Model checking/selection
- Goodness-of-fit
- AIC
- Detection function criteria
- Assumptions
- Convergence
- Variance estimation
- Where does uncertainty come from? (Conceptually)
- Effects of survey design
- Finding the AIC, GoF test results
- Q-Q plotting
- Adding covariates
- When should we include covariates?
- Possible covariates (what data to collect)
- Group size issues
- What are detection functions good for?
- Estimating probability of detection
- Simple Horvitz-Thompson ideas -- in the limit, many strata leads to DSM
- Fitting covariate models
- Model checking from above, revisited
- Abundance estimation
- Stratification
- Variance estimation
- Survey/data setup
- What is a segment?
- Useful conventions for data management
- Offsets
- Count as response
- Calculation of offset
- Model formulation
- Estimated abundance as response
- Calculation of response/offset
- Model formulation
- Relative merits of each approach
- Segment-level observation covariates
- What is a GAM?
- Smoothers
- Penalties and bases
- What do they look like?
- Uni- vs. multivariate smoothers
- Response distributions
- Count distributions zoo
- How do we fit GAMs?
- Linear model vs. interpolation (what is a smoothing parameter?)
- Basis complexity and relation to penalty
- 5 minute guide to GCV/REML (conceptual)
- Model checking/discrimination/criticism
- Q-Q plots
- Residuals
- AIC/deviance/$R^2$/REML score/GCV score
- Syntax-maths/concepts translation
- Simple (perhaps non-distance sampling) examples
- 1D smooths
- 2D smooths
- Emphasis on plotting smooths, looking at model checking plots
- Getting data into R
- Fitting basic models with just spatial covariates
- Model checking as covered above
- Using environmental covariates
- Appropriate bases for particular measurements
- Extra shrinkage/penalty ideas
- Response distributions
- Using Tweedie and negative binomial distributions
- Model selection
- Follow up on deviance/AIC/REML score ideas
- Fitting many models with different covariates
- Looking at check plot output
- Which summaries to look at -- selection criteria
- Prediction
- What are predictions? (Scale and temporal relevance)
- Predictions over a grid vs for an area
- Offsets
- Variance estimation of predictions
- Where does uncertainty come from?
- Per cell vs. overall uncertainty
- Presenting predictions and their uncertainties
- Using
predict- Making prediction grid
- Plotting predictions
- "Overall" (whole study area) abundance
- Variance estimation
- Plotting uncertainty (with effort)
- Overall uncertainty
This list is very rough, may depend on what participants are interested in. At this point these are possible topics not a definitive programme.
One option is to review the landscape of what's going on and offer only a couple of slides on each with references so people know where to go.
- Spatial autocorrelation
- Checking
- Strategies for accounting for it
- Strange shaped study areas
- soap film smoothing
-
$g(0)$ problems- "Perception bias" -- double observer methods, mark-recapture distance sampling
- "Availability bias" -- fixed correction, hidden Markov models etc
- Combining surveys
- Different platforms at similar times
- Geographical/temporal "patching" (Jason)
- Temporal trends
- Movement/displacement