Big data of tree species distributions: how big and how good?

Abstract

Background: Trees play crucial roles in the biosphere and societies worldwide, with a total of 60,065 tree species currently identified. Increasingly, a large amount of data on tree species occurrences is being generated worldwide: from inventories to pressed plants. While many of these data are currently available in big databases, several challenges hamper their use, notably geolocation problems and taxonomic uncertainty. Further, we lack a complete picture of the data coverage and quality assessment for open/public databases of tree occurrences. Methods: We combined data from five major aggregators of occurrence data (e.g. Global Biodiversity Information Facility, Botanical Information and Ecological Network v.3, DRYFLOR, RAINBIO and Atlas of Living Australia) by creating a workflow to integrate, assess and control data quality of tree species occurrences for species distribution modeling. We further assessed the coverage -the extent of geographical data -of five economically important tree families (Arecaceae, Dipterocarpaceae, Fagaceae, Myrtaceae, Pinaceae). Results: Globally, we identified 49,206 tree species (84.69% of total tree species pool) with occurrence records. The total number of occurrence records was 36.69 M, among which 6.40 M could be considered high quality records for species distribution modeling. The results show that Europe, North America and Australia have a considerable spatial coverage of tree occurrence data. Conversely, key biodiverse regions such as South-East Asia and central Africa and parts of the Amazon are still characterized by geographical open-public data gaps. Such gaps are also found even for economically important families of trees, although their overall ranges are covered. Only 15,140 species (26.05%) had at least 20 records of high quality. Conclusions: Our geographical coverage analysis shows that a wealth of easily accessible data exist on tree species occurrences worldwide, but regional gaps and coordinate errors are abundant. Thus, assessment of tree distributions will need accurate occurrence quality control protocols and key collaborations and data aggregation, especially from national forest inventory programs, to improve the current publicly available data.

Mountain building, climate cooling and the richness of cold‐adapted plants in the Northern Hemisphere

“…More recently, initiatives such as the Global Biodiversity Information Facility seek to gather species distribution data from different sources globally under the same umbrella. This global species distribution information permits the mapping of biodiversity at large spatial scales (Serra‐Diaz, Enquist, Maitner, Merow, & Svenning, ). Together with the increase in computer processing power and the development of bioinformatic pipelines, researchers can synthesize this information and look for large‐scale patterns in species richness that were very difficult to investigate in the past (König et al, ; Šímová et al, ).…”

Half of the world’s tree biodiversity is unprotected and is increasingly threatened by human activities

“…In this study, we used the world tree species list and species occurrence data compiled and cleaned by (69). Briefly, they extracted the records from the world tree species checklist (GlobalTreeSearch, GTS; (70)), and further standardized the taxonomic names via the Taxonomic Name Resolution Service (TNRS) online tool (71).…”

“…"a woody plant with usually a single stem growing to a height of at least two meters, or if multi-stemmed, then at least one vertical stem five centimeters in diameter at breast height" (70). A dataset of 58,100 tree species was obtained (69).…”

Species' range model metadata standards: RMMS

“…Errors or inherent biases (i.e., spatial) in publicly available occurrence data are common (Serra‐Diaz, Enquist, Maitner, Merow, & Svenning, ) and may have serious consequences for modelling (Merow, Allen, Aiello‐Lammens, & Silander, ; Phillips et al, ). Common reasons for excluding coordinates include: coordinates not falling in the specified political division, coordinates reflecting non‐native or cultivated occurrences, coordinates representing centroids of a political division, duplicated coordinates or biased spatial clustering (Aiello‐Lammens, Boria, Radosavljevic, Vilela, & Anderson, ; Maitner et al, ; Robertson, Visser, & Hui, ; Serra‐Diaz et al, ). Valid points may also need to be removed if they constitute environmental outliers that may strongly bias a model (Soley‐Guardia, Radosavljevic, Rivera, & Anderson, ).…”