Power Through or Keep Looking? Comparing Species‐Area Relationships of Habitat Fragments and Their Drivers in Different Ecoregions

Our study aimed to (1) determine how the shape varies and mechanisms influence species-area relationships within the same taxon but between different ecoregions and (2) determine how slope (<i>z</i>) and intercept (<i>c</i>) values of the linearized power model were influenced by ecoregion. Location: Madagascar. Taxon: Arboreal mammals (lemurs). We surveyed arboreal mammals (lemurs) in 42 tropical dry deciduous forest fragments in Ambanjabe Field Site in Ankarafantsika National Park in Western Madagascar and 27 primary mid-elevation raiforest fragments in the Tsinjoarivo-Ambalaomby new protected area in Eastern Madagascar using line-transect methods. We determined which of 20 species-area models were the most likely using the '<i>sars</i>' R package and AICc in each ecoregion. We compared <i>z</i> and <i>c</i> values of the power model in each ecoregion using ANCOVA. We assessed what drove the shape of the SARs using the Measurement of Biodiversity framework. We found that SAR models differed between ecoregions, with the power model (AICc = 89.04) as the most likely in the west and the Monod model (AICc = 86.98) followed by three candidate models (Kobayashi, AICc = 87.02; logarithmic, AICc = 87.6; and negative exponential, AICc = 88.61) in the east. We found no significant difference in <i>z</i> values between ecoregions (<i>F</i> _1,66 = 2.991, <i>p</i> = 0.088) and a non-significant trend in <i>c</i> values between ecoregions (<i>F</i> _1,65 = 3.938, <i>p</i> = 0.051). Spatial aggregation of species drove species richness patterns in the west, and species diversity and evenness drove species richness patterns in the east. Our study demonstrates that while the power and negative exponential model are good starting points, other models are also likely models to describe SARs in arboreal mammals such as primates. These patterns can reflect different mechanisms driving SARs. Ecoregion was not strongly related to differences in either <i>z</i> or <i>c</i> values of the power model.