Regenerates the population and reruns the full pipeline.

Seeds

Landscape

Seed for the coastline shape and environmental fields (Dh, Ph, Rh)

Population

Seed for dolphin placement, survey effort, and sightings (fix landscape, vary population)

Population

N 60

Number of dolphins; only those with ≥ 4 sightings total are retained for analysis

G 4

Number of latent habitat clusters; each has a centroid in the 3D preference space (depth proxy, productivity, disturbance). Each dolphin draws its own preference vector from a Gaussian around its cluster's centroid.

Overlap 0.30

Dispersion of each latent cluster in preference space; 0 = tight clusters (near-discrete, easy to separate), 0.9 = widely dispersed Gaussians with heavy cross-cluster overlap

Specialist 0.50

Fraction of dolphins with strong habitat fidelity (guild affinity coefficient 10× vs 5× for generalists); specialists show lower habitat-type spread: their B* concentrates on a narrower slice of the habitat axes, independent of where those hexes sit in space

Transients 2

Transient visitors with no resident home range; mostly filtered out by the ≥ 4 sightings gate, but contribute to the raw sightings layer

Kgen 2

Number of peaks in habitat space for generalists (specialists fixed at 1).

κspec 0.08

Habitat peak width for specialists (lower = sharper preference).

κgen 0.15

Habitat peak width for generalists.

Survey

Years 6

Survey years; each year draws 60 independent sample points from the radial effort field, so more years increase both total detections and spatial coverage

Effort bias 0.70

Radial decay scale for survey effort; 1 = effort tightly clustered near port, 0 = effort spread across the full domain

Detection p 0.30

Per-visit detection probability for a dolphin in a hex (scaled by λ and effort)

Analysis

k (kNN) 8

Neighbours each dolphin retains in the kNN graph fed to Louvain community detection

Emin 0.5

Effort floor in B* = Y / max(E, Emin); prevents inflation in rarely-surveyed hexes

Resolution 1.0

Louvain resolution γ; lower → fewer large communities, higher → more small ones

Domain

Hex size 30

Spatial resolution; the arena stays the same size — smaller hexes = more cells and finer grain, larger = fewer cells and coarser grain (see MAUP)

Retained: Communities: Q: ARI: W/B ratio:

Specialist / generalist

x = niche breadth (hexes with B* > 5% of peak)  ·  y = Pielou's J evenness of B* across hexes (low = concentrated, high = even)  ·  size = √(total sightings); small dots flag low-data dolphins whose position may be noise  ·  colour = detected community  ·  dark ring = specialist in the generative model (sharper θ around guild centroid plus higher guild affinity coefficient: 10 vs 5 for generalists)  ·  transients excluded