Controls of plant diversity and composition on a desert archipelago

Abstract

Aim: With the most robust floristic data set for any arid archipelago, we use statistical modeling to determine the underlying controls of plant diversity and species composition. Location: The study was undertaken in the Midriff Islands of the Gulf of California, Mexico. Methods: Using the area–diversity relationship we estimate the power coefficient z with generalized linear models (GLM). We tested eight predictors (area, human presence, habitat diversity, topography, distance to mainland, island type, precipitation, and seabird dynamics) using a step-wise process on the same GLM procedure. Plant species composition was assessed by conducting a non-standardized principal component analysis on a presence-absence matrix of the 476 (plant species) 14 (islands). Finally, families were tested for over or under representation with a X2 analysis subjected to a Bonferroni correction. Results: The classic species-area model explained 85% of the variance in island plant diversity and yielded a slope (z) of 0.303 (±0.01). When the effect of area is removed, four additional factors were shown to account for observed variation; habitat diversity (34%), seabird dynamics (23%), island type (21%), topography (14%). Human presence and distance to mainland were not predictors of species richness. Species composition varies significantly with island area; small islands have a particular flora where certain families are overrepresented, such as Cactaceae, while the flora of larger islands is strongly dependent on the continental source. Main conclusions: The factors that control diversity levels are expressions of geology, landscape heterogeneity, and land-sea connections. Species assemblages in small islands are governed by copious marine nutrients in the form of guano that depress species diversity. Distance to mainland and human presence hold no predictive power on diversity. The results show these islands to be isolated arid ecosystems with functioning ecological networks.