Skill: community-ecology-ordination

Outputs

Steps

1. Data Preparation

• Check for sites with zero species (remove or flag)
• Check for species observed at only one site (rare species handling: keep or remove)
• Standardise if needed (Hellinger, Wisconsin, presence/absence)
• Choose dissimilarity metric: Bray-Curtis (abundance), Jaccard (presence/absence), UniFrac (phylogenetic)

2. Alpha Diversity

• Species richness (S)
• Shannon index (H')
• Simpson index (1−D)
• Rarefaction curves to assess sampling adequacy
• Report metric ± SE per group

3. Ordination

NMDS:

• Run with k=2 (default) and k=3; choose lowest stress with acceptable fit
• Stress < 0.1 = excellent, < 0.2 = acceptable, > 0.2 = poor
• Run with ≥ 20 random starts; confirm convergence

PCA (for environmental gradients or species scores):

• Use Hellinger-transformed data or correlation matrix
• Report eigenvalues and % variance per axis

PCoA / MDS:

• For non-Euclidean dissimilarity matrices

4. Beta Diversity Partitioning

• Partition total beta diversity into nestedness and turnover components (betapart)
• Report contribution of each component per group comparison

5. Hypothesis Testing

• PERMANOVA (adonis2): test if group centroids differ
• PERMDISP: test if group dispersions (variances) differ (required before interpreting PERMANOVA)
• ANOSIM: alternative non-parametric test
• Report F/R statistic, R², p-value (permutation-based)

6. Species Contributions

• SIMPER: identify species driving dissimilarity between groups
• IndVal: identify indicator species per group
• Report top N contributing species per axis or group

7. Cluster Analysis

• Hierarchical clustering: Ward.D2 linkage preferred
• Cophenetic correlation to assess cluster quality
• k-means as alternative for large datasets
• Determine optimal k using silhouette or elbow method

Decision Points

Key Decisions to Document

• Dissimilarity metric and rationale
• Rare species handling
• Data transformation applied
• Number of NMDS dimensions
• Permutation count for PERMANOVA

Tools and Libraries

R: vegan, betapart, indicspecies, ape, dendextend, ggplot2
Python: skbio, scipy.cluster, sklearn.manifold

Resources

• resources/dissimilarity-metric-guide.md — which metric for which data type
• resources/nmds-interpretation-guide.md — how to read and report NMDS plots
• examples/ — worked NMDS and PERMANOVA example

Notes

• Always run PERMDISP before interpreting PERMANOVA; significant dispersion differences can inflate PERMANOVA results
• Stress value must be reported alongside all NMDS plots
• Rarefaction is mandatory when sites have very different sampling intensities