Fig 1.
A) Distribution of rainforest types in mainland eastern Australia. B) Bioregions defined by similar lizard and frog lineage composition based on hierarchical clustering of the lineage distribution models. C) A dendrogram of the affinity between lineages showing deep splits between the AWT and MEQ regions, and a second major break at the Border Ranges, dividing the SEQ and MEQ regions to the north, from CEC and SE to the South. The horizontal axis represents Bray-Curtis dissimilarity of lineage composition from identical composition (left edge) to no shared lineages at the root.
Table 1.
Lineages of rainforest specialist lizards and frogs included in the study.
Fig 2.
Steps in the lineage distribution model (LDM) method.
First a species distribution model is fitted, using established methods (step 1). Then, a cost distance calculation is used to determine, for each pixel, its connection through suitable habitat to known locations of each lineage (step 2). Finally, the cost distance is turned into a weight, and scaled so that for each pixel, the weights for the lineages sum to the species model suitability (step 3). In effect, the species model is partitioned among the known lineages in proportion to their likely level of connectivity to that location.
Fig 3.
Example of lineage distribution models for 6 lineages of Saproscincus rosei.
A) Phylogenetic relationships between the lineages. B) Known locations of each lineage (green symbols) are supplemented by locations of specimens (black dots) where the presence of the species is known, but not the lineage identity. For each lineage, the relative likelihood of occurrence is shown with colors matching the corresponding tree branch. C) The likelihood of occurrence of each lineage along the dashed transect in B is shown in the same colors. In this stacked plot, the total value (black line) is the predicted suitability for the species, partitioned between the lineages. Values for several lineages at the same location represent uncertainty of lineage occupation, not co-occurrence.
Fig 4.
A) Paleo habitat stability of rainforest since 120 ka using the dynamic stability model allowing dispersal at 10 m yr-1. B) Model-weighted lineage endemism peaks in the AWT just north of Cairns, and shows distinct highlights in a small subset of each rainforest region, such as Cape Melville in CYP, Conway Range and Eungella in MEQ, and the Border Ranges just south of Brisbane. C) Model-weighted species richness. D) Lineage endemism minus species endemism. Red areas have far greater lineage than species endemism, identifying concentrations of spatially restricted diversity, which would be missed in a species-level analysis. Colors by quantile classes for each map. Note that for compact map presentation, north is offset by 20° as indicated.
Fig 5.
Lineage endemism (left column), species endemism (middle column), and the difference between them (right column), for three areas.
Places labelled on panels A-C refer to areas of high endemism. Colours by quantile classes, separately for each panel.
Fig 6.
Influence of current rainforest and historical rainforest stability on lineage endemism as predicted by the boosted regression tree model.
Values of two predictors are shown: current rainforest habitat suitability, and dynamic stability of that habitat. Other predictors are held to their means. Lineage endemism increases with current rainforest habitat suitability and dynamic stability, but endemism is low in historically unstable areas (dynamic stability < 0.45) regardless of their current habitat.