Lab 11 · November 13, 2026

Spatial Analysis of Community Composition

How can we quantify change in community composition as a function of geography and environment?

From aggregated occurrence records to spatial community turnover, how do we get there?

BIO597 Spatial Analysis of Biodiversity
Maine 20 kilometer cells colored by the first two principal components of all 19 GDM-transformed bioclimatic predictors. Similar colors summarize similar transformed environments in the displayed plane.

Amphibian community similarity as a function of environment (similar colors indicate more similar communities)

How do communities change across the landscape?

Well, how do we quantify community change? Richness is one dimension. Community composition another.

Four hypothetical ponds each contain three species. Two contain A B C, and two contain D E F; richness is identical while identities differ.

A richness map can remain constant while community composition changes completely.

From observations to communities

How do we delineate communities?

The Maine state boundary with 14,466 original amphibian occurrence records shown as teal points, without grid cells or assigned community boundaries.

The points locate observations; we still need to define the sites we compare.

Original Maine data · amphibian occurrence records used in Lab11.

If we have spatially continuous samples, how do we determine 'communities'?

The same points can form different communities

Identical invented occurrence points in the same 80 kilometer landscape, divided into 10, 20, and 40 kilometer cells.

A simple way is to partition the landscape into grid cells with a given grain (grid cell size)

The size of the gridcell will determine community composition.

What aggregation does

A larger cell pool more species

Four 10 kilometer cells have species lists AB, BC, CD and D. Their mean richness is 1.75, while the enclosing 20 kilometer cell has pooled richness four.

Increasing grain may (or may not) increase species richness of a grid cell.

This may (or may not) have consequences for downstream analysis, and this is one of the things we will explore in the lab today.

Species richness as a function of grid cell size for ME amphibians

How do richness patterns change as a function of spatial grain?

Maine recorded-richness maps at 10, 20, and 40 kilometers share the same color scale and show 641, 223, and 69 sampled sites.

Larger cells contain more pooled species.

Are these patterns largely concordant or discordant as grain size increases?

Quantifying community difference

Two patterns can determine community difference

AB compared with ABCD is a subset pattern, while ABC compared with ACD shows replacement. Both have Jaccard dissimilarity 0.5.

Jaccard dissimilarity (difference in community composition) can include replacement and nestedness.

Nestedness means one community is a subset of another

Pairwise dissimilarity for all communities

Each comparison needs two species lists

Sites A, B, C contain ABC, ABD, DEF. Distances AB, BC, AC are 30, 40, 50 kilometers; Jaccard dissimilarities are 0.5, 0.8, 1.0.

Keep site order aligned when pairing species differences with geographic distances.

Worked example · Jaccard = 1 − shared / union; schematic distances, not to scale.

Distance decay of similarity

Do distant sites share fewer recorded species?

Actual Maine 20 kilometer Jaccard similarity versus distance, with 50 kilometer band means, alongside band curves at all three grains.

Distance decay summarizes a pattern. Its strength and shape can change with grain.

Original Maine data · points = pairs; lines = means in 50 km distance bands.

Community change over the landscape

Several processes can produce distance decay

Three hypothetical scenarios show changing environments, a dispersal barrier between similar habitats, and reduced reporting of the same underlying community.

Environment, dispersal, history, and observation can act together.

Neighborhood similarity

Which recorded communities differ nearby?

A focal ABC cell is compared with seven sampled neighbors: three identical lists and four disjoint DEF lists. The mean dissimilarity is four sevenths; an unsampled gray cell is excluded.

Mean neighbor Jaccard describes local compositional difference.

Worked example · same-radius comparisons are made within each grain.

Richness and difference answer different questions

A rich site need not be the most distinctive

Two actual Maine 20 kilometer maps compare recorded richness with mean neighboring-site Jaccard dissimilarity on separate appropriate scales.

Use the species lists and record totals to interpret a high local difference.

Original Maine data · gray indicates no mapped value, not zero difference.

Add environmental context

A site has an environmental setting

Maine 20 kilometer site means for annual temperature and annual precipitation illustrate two gradients; the full analysis uses all 19 bioclimatic variables.

All 19 bioclimatic variables enter GDM; these two maps orient us to the landscape.

Original Maine data · WorldClim variable definitions.

Generalized dissimilarity modeling

Relate community differences to site differences

Two hypothetical sites have different species lists, temperature and precipitation values plus 17 other predictors, and a geographic separation. The observed pairwise Jaccard response is 0.5.

GDM predicts the expected difference between communities.

Illustrated inputs · method: Ferrier et al. (2007).

The main nonlinear intuition

Equal climate steps need not mean equal change

A hypothetical monotone temperature transformation compares 3–5 and 8–10 degrees Celsius. The equal two-degree intervals have different transformed distances.

GDM stretches parts of a climate gradient where communities change more.

Illustrated transformation · this curve is not a fitted lab result.

Environmental correlates of community turnover

Steep segments identify intervals of modeled change

Actual Maine fitted transformations for the three climate predictors with highest 20 kilometer transformation heights, showing each grain on the same predictor panel.

Read the slope and climate interval, then compare grains.

Original Maine data · all curves come from models using all 19 predictors.

Projecting estimated community turnover across the landscape

Summarize the transformed environments

A workflow transforms 19 site climate variables with GDM, sums weighted spline components per predictor, runs PCA, and assigns the same colors to map and biplot.

Transform first, then use PCA to display the major axes of expected compositional variation.

Method: PCA · implementation: gdmbayes.

One color key in two spaces

Nearby colors summarize similar expected composition

Actual Maine 20 kilometer map and PCA scatter use identical RGB colors for each modeled site. Two arrows identify the longest climate loading vectors in PC1 and PC2.

PC1 + PC2 retain 91.7% of the GDM-transformed variation.

Original Maine data · first two PCs; geographic-distance effects are excluded from the site colors.

Connect the ordination to environmental biology

Which gradients align with the displayed change?

Maps of the two original climate variables selected by longest loading-vector length in the actual Maine PC1–PC2 biplot.

Two longest arrows: BIO5 and BIO17.

Original Maine data · select by length in the PC1–PC2 plane; display only two arrows.

Choose the map for the biological question

Three maps, three different kinds of evidence

Maine 20 kilometer maps show recorded richness, neighbor Jaccard differences, and the PCA summary of climate-associated expected composition.

Which map would help locate a richness hotspot, a local contrast, or an environmental transition?

Original Maine data · each panel answers a different question; colors are not interchangeable.

Readings and figure provenance

Methods behind the pictures