Species richness variation in marine and terrestrial fauna across widespread, fragmented territories: assessing inherent challenges of data scarcity at local and regional scales.

The ongoing biodiversity crisis calls for a complete biodiversity inventory of marine and terrestrial ecosystems. The task is particularly challenging for fragmented island territories, where baseline biodiversity information is often difficult to procure. By centralising information from different sources (museums, research institutions, citizen scientists), 'big-data' platforms provide an opportunity to evaluate species biodiversity information of understudied regions. Using data primarily sourced from the Global Biodiversity Information Facility (GBIF), and complemented by a review of 56 potential data sources-of which nine provided unique, non-redundant records-we curated the first biogeographic dataset for both marine and terrestrial animal species in French Polynesia, a large territory composed of 124 islands and atolls that belongs to the Central Pacific region, a marine biodiversity hotspot facing conservation challenges. The dataset revealed heterogeneous species richness across archipelagos and islands, prompting an investigation into potential sampling biases (institutional, taxonomic, spatial) as well as an assessment of island-specific accessibility biases. We estimated that the archipelagos and islands had an inventory completeness rate that ranges from 1.9 to 98.4%, suggesting that a large proportion of the studied area remains poorly documented. Spatial and temporal sampling biases were partly explained by accessibility constraints (proximity to airports, roads or ports), and inventory completeness was higher for marine than terrestrial species. The biases quantified here challenge our ability to conduct biogeographic analyses that integrate the land-sea meta-ecosystem. Our database allows identifying taxa and sampling locations that require urgent attention, as well as comprehensively recorded species that can serve as indicators for environmental degradation. Explicitly acknowledging the inherent biases of biodiversity datasets is the first step towards a more comprehensive characterization of species diversity across fragmented territories. This information is crucial for guiding sound adaptive-management and conservation planning strategies.