Abstract
Presence‐only data are typical occurrence information used in species distribution modelling. Data may be originated from different sources, and their integration is a challenging exercise in spatial ecology as detection biases are rarely fully considered. We propose a new protocol for presence‐only data fusion, where information sources include social media platforms, to investigate several possible solutions to reduce uncertainty in the modelling outputs. As a case study, we use spatial data on two dolphin species with different ecological characteristics and distribution, collected in central Tyrrhenian through traditional research campaigns and derived from a careful selection of social media images and videos. We built a spatial log‐Gaussian cox process that incorporates different detection functions and thinning for each data source. To finalize the model in a Bayesian framework, we specified priors for all model parameters. We used slightly informative priors to avoid identifiability issues when estimating both the animal intensity and the observation process. We compared different types of detection function and accessibility explanations. We showed how the detection function's variation affects ecological findings on two species representatives for different habitats and with different spatial distribution. Our findings allow for a sound understanding of the species distribution in the study area, confirming the proposed approach's appropriateness. Besides, the straightforward implementation in the R software, and the provision of examples' code with simulated data, consistently facilitate broader applicability of the method and allow for further validations. The proposed approach is widely functional and can be considered with different species and ecological contexts.
Highlights
Progress of ecological science is more and more reliant on combining data from diverse sources (Fletcher et al 2019)
For striped dolphin model with varying detection functions (Supporting information), the effects of categorized depth were fairly in agreement with the species distribution ranges: it is generally not found in very shallow waters, observed at 100–200 m depth, and more often encountered at depths over 200 m
This study demonstrates that methods of spatial data integration able to carefully consider and minimize datasets’ biases can be efficiently used to predict species’ distribution
Summary
Progress of ecological science is more and more reliant on combining data from diverse sources (Fletcher et al 2019). Data availability to model species distribution, for example, is rapidly expanding thanks to the fast development of new technologies (Soranno and Schimel 2014), the growth of citizen science initiatives (SicachaParada et al 2020, Matutini et al 2021) and the opportunity of exploiting huge information harvested from social media platforms (Mikula and Tryjanowski 2016, Pace et al 2019). The latter data types can be intrinsically challenging to merge in with existing, valued and validated data collected via standard research protocols. A simple data pooling (Fletcher et al 2019) with data gathered under conventional research methodologies is not enough to reliably model the presence of a species considering different explanatory variables both environmental and anthropogenic and to define its distribution over multiple spatial and temporal scales
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
