A new baseline for countrywide α‐diversity and species distributions: illustration using >6,000 plant species in Panama

Abstract

Estimating α‐diversity and species distributions provides baseline information to understand factors such as biodiversity loss and erosion of ecosystem services. Yet, species surveys typically cover a small portion of any country's landmass. Public, global databases could help, but contain biases. Thus, the magnitude of bias should be identified and ameliorated, the value of integration determined, and application to current policy issues illustrated. The ideal integrative approach should be powerful, flexible, efficient, and conceptually straightforward. We estimated distributions for >6,000 species, integrating species sightings (S) from the Global Biodiversity Information Facility (GBIF), systematic survey data (S2), and “bias‐adjustment kernels” (BaK) using spatial and species trait databases (S2BaK). We validated our approach using both locational and species holdout sets, and then applied our predictive model to Panama. Using sightings alone (the most common approach) discriminated relative probabilities of occurrences well (area under the curve [AUC] = 0.88), but underestimated actual probabilities by ~4,000%, while using survey data alone omitted over three‐quarters of the >6,000 species. Comparatively, S2BaK had no systematic underestimation, and substantially stronger discrimination (AUC = 0.96) and predictive power (deviance explained = 47%). Our model suggested high diversity (~200% countrywide mean) where urban development is projected to occur (the Panama Canal watershed) and also suggested this is not due to higher sampling intensity. However, portions of the Caribbean coast and eastern Panama (the Darién Gap) were even higher, both for total plant biodiversity (~250% countrywide mean), and CITES listed species. Finally, indigenous territories appeared half as diverse as other regions, based on survey observations. However, our model suggested this was largely due to site selection, and that richness in and out of indigenous territories was roughly equal. In brief, we provide arguably the best estimate of countrywide plant α‐diversity and species distributions in the Neotropics, and make >6,000 species distributions available. We identify regions of overlap between development and high biodiversity, and improve interpretation of biodiversity patterns, including for policy‐relevant CITES species, and locations with limited access (i.e., indigenous territories). We derive a powerful, flexible, efficient and simple estimation approach for biodiversity science.

Climate and seasonality drive the richness and composition of tropical fungal endophytes at a landscape scale

“…Our remaining sites were chosen to represent major forest types from the Pacific to the Caribbean, spanning a known range of climate and seasonality factors (Table 1), land use history 47 , first-hand observations of forest composition (Ibáñez, unpubl. data), and estimates of plant species richness based on 48 . Broadly, we anticipated similar tree species richness in Bastimentos to that of the well-characterized forests at the Caribbean end of the transisthmian gradient of the Panama Canal watershed 48,49 ; in Parida, similar to that of the Pacific side of the Canal 49 ; and lower diversity in the montane forests of Baru, Price, and Copete, consistent with the decline tree species richness that occurs at higher elevations in the region 50 .…”

Optimizing ecological surveys for conservation

“…We created a preliminary dataset containing detections and nondetections for each species (hereafter, occurrence dataset; Figure S5). Although observation records and survey data often have different accuracies because they are collected by individuals with different skill levels—for example, observation data may be collected by citizen scientists and survey data by trained experts (Leung et al, 2019; Mengersen et al, 2017)—all of our observation and survey data were collected and verified by skilled botanists (McCune, 2016). As such, by treating observation and survey data as a series of independent detections and nondetections, we created a dataset for each species (separately) containing detections—places where botanists found and reported a particular species as a survey result or as an observation record—and nondetections—places where botanists conducted a survey and could not find a particular species—across the study area and surrounding counties.…”

“…To achieve this, conservation planning exercises require information on the locations of multiple species (Polasky & Solow, 2001). Although ecological surveys and statistical models can provide such information (Leung et al, 2019; Mengersen et al, 2017), uncertainty in collecting survey data and model prediction errors can undermine prioritization procedures (Foody, 2011; Raymond et al, 2020). Thus there has been growing interest in designing survey plans to directly inform and improve prioritizations (Chadès et al, 2008; Raymond et al, 2020).…”

Invasion dynamics and potential future spread of sea spurge across Australia’s coastal dunes

“…Second, for several areas of the European and Australian coast, we only had access to presence data from GBIF, with no presence-absence data from vegetation plots. Because vegetation surveys are generally considered less biased than occurrence-only data (Leung et al, 2019), we lowered the influence of observations from the GBIF dataset in our model. To do this, we created a predictor variable with a value of 0 for pixels including only observations from the GBIF dataset and a value of 1 for pixels including presence or absence from a vegetation plot.…”

“…To do this, we created a predictor variable with a value of 0 for pixels including only observations from the GBIF dataset and a value of 1 for pixels including presence or absence from a vegetation plot. For subsequent model predictions, we then set the value of this variable to 1 for all pixels in the reference Australian coastal buffer used for predictions (following recommendations by Leung et al, 2019).…”