Getting Started with Single-Species Occupancy Models
This dashboard supports two workflows for fitting occupancy models: Basic and Advanced. Both use the same underlying statistical methods but offer different levels of control over model specification.
Choosing Your Workflow
- Basic Workflow: Best for standard analyses and newcomers to occupancy modeling
- Simple linear effects of covariates
- Streamlined model specification
- Quick model comparison
- Advanced Workflow: For complex analyses requiring:
- Quadratic effects
- Interaction terms
- Random effects
- Fine control over model structure
Basic Workflow Guide
- Model Setup
- Choose between 'unmarked' (faster) and 'ubms' (Bayesian) packages
- Select model type: Occupancy or Royle-Nichols
- Optionally scale covariates (recommended)
- Select detection covariates (including effort if desired)
- Select occupancy covariates
- Model Selection
- Run multiple models with different covariate combinations
- Add models to model selection table
- Compare using AIC (unmarked) or elpd (expected log pointwise predictive density, in ubms)
- Response Plots
- Visualize relationships between covariates and detection/occupancy
- Adjust confidence levels
- Compare detection vs. occupancy effects
- Spatial Predictions
- Generate prediction maps using extracted covariates or custom rasters
- View uncertainty in predictions
- Export results for further analysis or visualization in GIS software
Advanced Workflow Guide
- Model Setup
- Same package and model type options as Basic workflow
- Add effects one at a time with more options:
- Linear effects (standard relationship)
- Quadratic effects (unimodal relationships)
- Interactions (combined effects of two covariates)
- Random effects (for grouped or nested data)
- Preview formula before fitting
- Results & Diagnostics
- Detailed model summaries
- MCMC diagnostics for ubms models
- Trace plots for convergence checking
- Effect Visualization
- Response curves for each effect type
- Separate plots for detection and occupancy
Tips and Best Practices
- General Tips
- Start with the Basic workflow unless you need Advanced features
- Scale covariates to improve model convergence (especially with large absolute covariate values)
- Include effort as a detection covariate
- Check for correlations between covariates before modeling (section "Data Processing" > "Covariate Correlation")
- Model Selection
- Start with simple models and gradually add complexity
- Use AIC/elpd to compare models formally
- Consider biological significance alongside statistical significance
- MCMC Settings (ubms)
- Increase iterations if chains haven't converged
- Use multiple chains to assess convergence
- Check trace plots for mixing
- Adjust thinning if autocorrelation is high (recommended value for thinning is 1 though, see ?stan_occu)
Common Issues
- Model Convergence
- Non-convergence often indicates insufficient data for complex models
- Ensure adequate detections for reliable parameter estimation
- Prediction Issues
- Ensure prediction rasters match the scale of model covariates (if not using extracted rasters created by dashboard - dashboard handles scaling of prediction rasters automatically)
- Check for missing values in prediction surfaces
- Be cautious about predicting beyond the range of observed covariates