- A recent observation by an amateur naturalist of a fiddler crab species hundreds of kilometers north of its known range challenged the complementary strengths of computer vision and human expertise in mapping species distributions.
- The naturalist uploaded this record to the iNaturalist species database used by amateurs and experts to document sightings; expert input correctly identified the specimen after the platform’s computer vision algorithms did not acknowledge the species outside its documented range.
- Citizen naturalist observations can be used to document rapid changes in species distributions. They also can improve modeling and mapping work conducted by researchers and play an increasing prominent role in building environmental databases.
On a midsummer evening earlier this year, Tracy Pham was on a walk along Huntington Beach, California, an outing she usually made to photograph birds. This time, along the way, she noted a collection of fiddler crabs scuttling along the mud and determined they were worth recording as well.
After posting her crab photos to iNaturalist, a mobile app and website used to document sightings in nature, sudden interest picked up in these crabs. A crustacean expert identified them as large Mexican fiddler crabs (Uca princeps), and their appearance at Huntington Beach was well out of the known geographic range – 240 kilometers farther north than any had been observed before. We might wonder, with increased security at the southern border of the United States, how this Mexican crab made its way north almost to Los Angeles. Perhaps even more surprising is how a single observation by a birder quickly spread throughout the global crustacean research community.
Since the launch of iNaturalist in 2008, the database has registered numerous observations of organisms well out of their known range, such as the first sighting of a gecko species in Colombia, or the rare blue tiger butterfly appearing in Singapore despite no knowledge of its genus in the country. Although some such findings are chance encounters with a vagrant individual, other sightings may hint at potential range expansions of species.
“After looking at Tracy’s observation, I immediately realized that this was novel to California and the U.S.,” said Michael Rosenberg, the biologist who identified the crab from the photos. “It was when I pointed this out in the thread that the excitement built and others started looking for them, leading not only to additional confirmations of the species at that particular site, but now in others sites in California as well.”
Experts hypothesize that El Niño events led to the movement of the large Mexican fiddler crab so far north of its previous range. The intensity of the 2014-2016 El Niño led to higher temperatures and changed ocean current patterns in the eastern Pacific. This warmer climate may have enabled the crab to advance to new latitudes, or swift currents may have carried crabs up the coastline.
Although most iNaturalist users are non-specialists who identify organisms occasionally, the data amateurs collect still have potentially high scientific value. Managed by the California Academy of Sciences, iNaturalist has served as an effective platform for capturing information that conventional scientific data collection methods may miss, such as a chance encounter with a rare species. Citizen-collected data can also help fill in gaps of the recorded ranges of more abundant organisms when the published data are deficient. More data on a species’ presence improves modeling work by scientists to predict how species distributions will shift with climate change or habitat loss.
Uploading observations to iNaturalist also disseminates information quickly, which helps scientists track range expansions or invasions. Global trade networks can incidentally carry invasives to new regions suddenly and sporadically. Species documentation also needs to keep pace with local extinctions and catastrophic natural disasters.
Publishing findings in a scientific journal typically involves peer (expert) review and takes several months to more than a year. Sightings uploaded to citizen science databases such as iNaturalist or eBird can put observational data in front of taxonomic experts nearly immediately, who can then approve them as “research grade,” thereby conducting an expedited, less rigorous, form of peer review.
iNaturalist observations have been fully vetted and published in the peer-reviewed literature as well. Data used in a scientific paper are typically extensive and replicated. Yet occasionally a single observation, such as an organism found well out of its known range, is novel enough to be deemed worthy of publication. With the sheer number of iNaturalist users, some will witness such interesting moments that can be published as standalone discoveries, as Rosenberg did with Pham’s fiddler crab sighting.
“The kind of data iNaturalist is collecting is, in many ways, akin to the kind of specimen data museums have been collecting for hundreds of years,” Scott Loarie, co-director of iNaturalist, told Mongabay-Wildtech. “But whereas museums often take decades to catalog, digitize, and share the information they collect, iNaturalist observations are available immediately.”
Citizen science data hubs have begun to complement the repositories used by practicing scientists. Research-grade iNaturalist observations are uploaded to the Global Biodiversity Information Facility (GBIF), an platform for species distribution data from research and surveys but also from other sources like eBird and Encyclopedia of Life. Over a third of the entries to GBIF in the past few months have been from iNaturalist, and the majority of the database’s sightings are from citizen science sites.
iNaturalist benefits from sheer numbers of observations, but not all records are valid or even related to nature: spam posted by vendors seeking business is an issue the platform faces. Verifying genuine citizen naturalist data is still a demanding task that usually falls on the shoulders of experts.
To help address this issue, last year iNaturalist’s developer team released Computer Vision, an image classification model, to speed up identification of user observations. Computer Vision is composed of machine learning algorithms that draw upon existing iNaturalist data to suggest identifications for new uploads. The artificial intelligence (AI) uses pixel matching of existing and newly uploaded photos as well as the spatio-temporal data (i.e. where and when the observation was made) to inform its suggestions.
There are inherent limits to this application: while a crustacean expert can recognize a large Mexican fiddler crab found near Los Angeles, Computer Vision incorrectly suggested the crab was a different species, Uca crenulata, which is more common in the area.
“I think the [output of Computer Vision] is best thought of as a synthesis of the human expertise on the site,” Loarie said. “This means the AI will only work well in areas (geographic or taxonomic) where we have enough human expertise. But it also means that we need to take steps to make sure that the AI is educating people in a way that is complementary but not circular (e.g. if the AI is based on human expertise that’s based only on the AI, it’s circular).”
AI can more accurately classify images of species for which it has abundant training data, and the identification abilities of computer vision reflect those of the collective iNaturalist community. Common species are more likely to be spotted and therefore offer enough data to train the computer vision’s algorithms to identify such species, but rare organisms are more difficult for the AI to “recognize.” Computer vision therefore builds intelligence on the more common species of passerines, anoles, and nudibranchs, yet is not proficient with any particular group.
According to Loarie, iNaturalist relies upon taxonomic specialists to provide expert opinion, yet the scientific community lacks a method to incentivize scientists to participate more broadly.
“It would be great to find a way to give professionals ‘credit’ towards their professional careers for participating in citizen science,” Loarie said. “This could be like how many scientists volunteer time to peer review because they see it as part of their professional careers.”
Citizen science platforms may constantly need experts to identify specimens. But scientists, in turn, can benefit from the data available on these platforms, which can fill in the gaps of knowledge that limited research funding and manpower cannot.
In the ten years since iNaturalist’s launch, the platform has spread to become one of the most prominent global hubs for species observations. The maps show all observations from iNaturalist.org since 2008; each locations is from the coordinates where the observations were made and subsequently uploaded to the platform.
“I suspect that most scientists currently under-appreciate how much data these endeavors are creating and how useful they might be,” said Rosenberg, the crustacean expert. “If the site and its user-base persists long enough, and if its data is readily accessible through API, it could be an extremely good source for mapping range changes of at least commonly recorded species…[T]he data might just be waiting there to answer the right questions if we just think to pose them.”
Amateur observations may not be as reliable as data collected by professional wildlife researchers, but they expand on the efforts by researchers. For answering certain types of research questions, such as the distribution of a species in space and time, collaboration with citizen scientists can be valuable. Non-experts provide an abundance of data, which is verified by experts or algorithms and then passed into the hands of researchers. Researchers skilled in fusing datasets together can use the data to train powerful computer models and derive meaning. This pipeline has fueled new findings in conservation science and is starting to become a standard methodology in the scientific community.
Professional naturalists, citizens, and artificial intelligence can all play a role in environmental research. Their complementary strengths– expertise, abundance, and versatility – combine to more effectively map the species distributions of the world.
Jacox, M. G., et al. (2016) Impacts of the 2015-2016 El Niño on the California Current System: Early assessment and comparison to past events. Geophysical Research Letters, 43. https://doi.org/10.1002/2016GL069716
Climate Prediction Center Internet Team (2015). Background information: East Pacific hurricane season. National Oceanic and Atmospheric Administration. http://www.cpc.ncep.noaa.gov/products/Epac_hurr/background_information.html
Rosenberg, M. S. (2018) New record and range extension of the fiddler crab Uca princeps (Smith, 1870) (Brachyura, Ocypodidae) from California, USA. Journal of Crustacean Biology, 38. https://doi.org/10.1093/jcbiol/ruy071
Scyphers, S. B., et al. (2014) The role of citizens in detecting and responding to a rapid marine invasion. Conservation Letters, 8. 2015. https://doi.org/10.1111/conl.12127
McKinley, D. C., et al. (2017) Citizen science can improve conservation science, natural resource management, and environmental protection. Bioloical Conservation, 208. https://doi.org/10.1016/j.biocon.2016.05.015
FEEDBACK: Use this form to send a message to the editor of this post. If you want to post a public comment, you can do that at the bottom of the page.