Rhythms in the Landscape: Bayesian Circular Models Reveal Spatial Structure in Tapir Activity Patterns

This repository contains all data, scripts, and outputs supporting the manuscript:

Rhythms in the Landscape: Bayesian Circular Models Reveal Spatial Structure in Tapir Activity Patterns
Álviz et al. — submitted to Journal of Animal Ecology (2026)

All analyses are conducted in R using Bayesian two-component von Mises mixture models, with a focus on diel activity timing, temporal concentration, and landscape covariates across six sites in the Colombian Orinoquia.

---

Repository Structure

activitypatterns/
│
├── README.md                  ← project overview and reproducibility guide
│
├── data/
│   ├── raw/                   ← original detection data (never modified)
│   └── processed/             ← filtered 317 independent detections (30-min threshold)
│
├── covariates/                ← landscape covariate extractions per camera (1-km buffer)
│
├── scripts/
│   ├── 01_data_preparation.R  ← independence filter, solar time conversion to radians
│   ├── 02_model_comparison.R  ← GAMM, GLMM, LMM cross-validation (count-based frameworks)
│   ├── 03_mixture_models.R    ← base, covariate, and site+covariate von Mises mixture models
│   ├── 04_model_diagnostics.R ← R-hat, ESS, trace plots, LOO/PSIS-LOO diagnostics
│   └── 05_figures.R           ← all manuscript figures (Figures 2–5 and supplementary)
│
└── outputs/
    ├── model_fits/            ← saved brms model objects (.rds)
    └── figures/               ← final figures as exported

---

Description

This project applies a Bayesian two-component von Mises mixture modeling framework to camera-trap detection times of the lowland tapir (Tapirus terrestris) recorded at 111 camera stations across six sites in the Colombian Orinoquia (2018–2023). The workflow:

1. Quantifies bimodal diel activity patterns — estimating separate peak times (μ), temporal concentration (κ), and mixing weights (θ) for evening and pre-dawn activity components.
2. Evaluates competing statistical frameworks — comparing count-based hierarchical models (GAMM, GLMM, LMM) and a unimodal circular model against the mixture approach using cross-validation.
3. Models landscape drivers of activity timing — assessing the effects of dense forest, gallery/riparian forest, open forest, secondary vegetation, and cropland cover on pre-dawn peak timing across sites.
4. Provides site-specific conservation insights — quantifying site-level variation in temporal concentration (ρ) and activity timing across a gradient of habitat types and anthropogenic pressure.

---

Requirements

• R ≥ 4.5.0 (analyses conducted in R 4.5.0; earlier versions may work but are untested)
• Stan (required by brms; install via install.packages("rstan"))

R packages

Package	Version used	Purpose
brms	≥ 2.21	Bayesian mixture model fitting via Stan
circular	≥ 0.5	Circular statistics and Watson's U² tests
loo	≥ 2.7	PSIS-LOO model comparison
ggplot2	≥ 3.5	Visualization
dplyr	≥ 1.1	Data manipulation
tidyr	≥ 1.3	Data reshaping
readr	≥ 2.1	Data import
sf	≥ 1.0	Spatial data and covariate extraction
terra	≥ 1.7	Raster processing for landscape covariates

Install all packages

install.packages(c("brms", "circular", "loo", "ggplot2", 
                   "dplyr", "tidyr", "readr", "sf", "terra"))

Note: Installing brms will prompt installation of rstan. Follow the RStan Getting Started guide if you encounter compilation issues.

---

How to Reproduce

Run scripts in numerical order:

source("scripts/01_data_preparation.R")
source("scripts/02_model_comparison.R")
source("scripts/03_mixture_models.R")
source("scripts/04_model_diagnostics.R")
source("scripts/05_figures.R")

⚠️ Script 03_mixture_models.R is computationally intensive (8 chains × 16,000 iterations per model). Pre-fitted model objects are available in outputs/model_fits/ as .rds files and can be loaded directly to skip refitting.

---

Outputs

• Posterior parameter estimates (μ, κ, θ) for all three mixture model variants
• Conditional effect plots for five landscape covariates on pre-dawn peak timing (Figure 4)
• Site-specific diel activity curves (Figure 3)
• Model comparison table (LOOIC, ΔELPD, SE; Table in manuscript)
• Pairwise Watson's U² and temporal overlap coefficients (Δ̂) across sites (Figure S3)
• Session info for full reproducibility (outputs/sessionInfo.txt)

---

Citation

If you use this code or data, please cite:

Álviz Á, Salazar-Bravo J, van Gestel N, Stevens RD (2026). Rhythms in the Landscape: Bayesian Circular Models Reveal Spatial Structure in Tapir Activity Patterns. Journal of Animal Ecology (under review).

---

License

This repository is licensed under CC BY 4.0 — you are free to share and adapt the material for any purpose, provided appropriate credit is given.

---

Contact

For questions about the data or analyses, please open a GitHub Issue or contact the corresponding author via the journal submission.