Leveraging Public Data to Predict Global Niches and Distributions of Rhizostome Jellyfishes

Abstract

As climate change progresses rapidly, biodiversity declines, and ecosystems shift, it is becoming increasingly difficult to document dynamic populations, track fluctuations, and predict responses to climate change. Concurrently, publicly available databases and tools are improving scientific accessibility, increasing collaboration, and generating more data than ever before. One of the most successful projects is iNaturalist, an AI-driven social network doubling as a public database designed to allow citizen scientists to report personal biodiversity reports with accuracy. iNaturalist is especially useful for the research of rare, dangerous, and charismatic organisms, but requires better integration into the marine system. Despite their abundance and ecological relevance, there are few long-term, high-sample datasets for jellyfish, which makes management difficult. To provide some high-sample datasets and demonstrate the utility of publicly collected data, we synthesized two global datasets for ten genera of jellyfishes in the order Rhizostomeae containing 8412 curated datapoints from both iNaturalist (n = 7807) and the published literature (n = 605). We then used these reports in conjunction with publicly available environmental data to predict global niche partitioning and distributions. Initial niche models inferred that only two of ten genera have distinct niche spaces; however, the application of machine learning-based random forest models suggests genus-specific variation in the relevance of abiotic environmental variables used to predict jellyfish occurrence. Our approach to incorporating reports from the literature with iNaturalist data helped evaluate the quality of the models and, more importantly, the quality of the underlying data. We find that free, accessible online data is valuable, yet subject to biases through limited taxonomic, geographic, and environmental resolution. To improve data resolution, and in turn its informative power, we recommend increasing global participation through collaboration with experts, public figures, and hobbyists in underrepresented regions capable of implementing regionally coordinated projects.

Keywords: Cnidaria; Scyphozoa; citizen science; distribution modeling; iNaturalist; macroecology; marine ecology; niche modeling; online databases; random forests model.

Figure 1

(A) iNaturalist reports of genera within the order Rhizostomeae with sufficient sampling determined by rarefied sampling coverage (SC > 95%) (Figure S1); circles represent individual reports. (B) Density curves along the y-axis visualize the report density across latitudes. Lowercase letters next to genus names indicate groups of statistically similar distributions based on post-hoc Dunn tests (p < 0.05). (C–F) Photos are examples of ‘Research Grade’ reports assigned the following taxon names on iNaturalist (inaturalist.org accessed on 3 November 2022): (C) barrel jelly Rhizostoma pulmo from Mauguio, France by Pascal GIRARD in 2022; (D) fried egg jelly Cotylorhiza tuberculata from Thessaly, Greece by marieta55 in 2021; (E) hizen kurage Rhopilema hispidum from Sai Kung, Hong Kong by Ryan Yue Wah Chan in 2021; (F) blubber jelly Catostylus mosaicus from NSW, Australia by John Sear in 2022 (photos under CC-BY-NC licenses).

Figure 2

(A) Rhizostomeae niche space from generalized linear models visualized as a biplot with larger gaps between points indicating greater niche separation between genera. (B) Environmental variables (colors) and depths (shades) that best predict the occurrence of each genus as determined by random forest models. (C–I) Oceanic, environmental data used to build models in A and B: salinity, temperature, silicate, dissolved oxygen, oxygen saturation, nitrate, and phosphate. Lowercase letters in each row indicate statistically similar distributions based on post-hoc Dunn tests (p < 0.05).

Figure 3

Predicted distributions of several Rhizostomeae genera based on RFMs built from iNaturalist reports (pink squares) and legacy data from the literature (blue circles). Dark density-based polygons indicate regions corresponding to suitable environments predicted by random forest models (see Figure 2B and Figure S3) to support the existence of each genus: (A) Rhizostoma, (B) Cotylorhiza, (C) Lobonema, and (D) Lychnorhiza. Percentages in pink font reflect RFM model accuracy built from iNaturalist data, while percentages in blue font are the distribution model accuracy based on legacy data (see File S2); both metrics can be used to infer confidence in the model.