Step-by-Step Guide to Community Occupancy Modeling
- Species Selection
- First, select which species to include in your community model
- Review the species table showing detections and number of sites for each species
- Use the filtering options to select species based on minimum number of detections or sites
- You can manually select/deselect individual species by clicking the rows
- Consider removing rare species with too few detections for reliable parameter estimation
- Model Configuration
- Choose model type (Occupancy or Royle-Nichols model)
- Set occasion length (in days) to create detection histories
- Configure detection and occupancy covariates:
- Fixed Effects: Effect is same for all species
- Species Random Effects: Species-specific effects with shared variance
- Independent Effects: Completely independent effects for each species
- Configure how species intercepts are modeled (fixed, random, or independent)
- Optional & recommended: Include survey effort as detection covariate
- Optional: Add species-site random effects
- Model Fitting
- Set MCMC parameters (chains, iterations, burn-in, thinning)
- Run model
- Monitor progress (currently R console output, not in the dashboard)
- Results Inspection
- Review parameter estimates and their uncertainties
- Check convergence diagnostics (Gelman-Rubin statistics)
- Examine trace plots for key parameters
- Optional: Run Goodness-of-Fit tests (not yet implemented)
- Effect Visualization
- View response curves for covariates
- Compare effect sizes across species
- You can:
- Select subsets of species to display
- Order effects by size
- Adjust confidence intervals
- Scale plot sizes for better visibility
- Spatial Predictions
- Generate species-specific occupancy/abundance maps
- Create species richness predictions
- Calculate percentage of area occupied
- Maps include uncertainty estimates
Important Notes:
- More species and covariates increase computation time
- Check convergence diagnostics before interpreting results
- Covariates are scaled to mean = 0 and standard deviation = 1 automatically.
- Prediction rasters shuold be provided in original scale. They are scaled automatically to match model covariatesy
- Species with very few detections may have unreliable estimates
- Save your model objects to avoid rerunning long computations
Tips for Model Convergence:
- Start with simpler models and gradually add complexity
- Covariates are standardized automatically
- Increase iterations if chains haven't converged
- Check for highly correlated covariates
- Consider removing rare species or combining similar species