Citizen science and online data: Opportunities and challenges for snake ecology and action against snakebite

An artificial intelligence model to identify snakes from across the world: Opportunities and challenges for global health and herpetology

“…First, the snake photo dataset used to train the model does not yet cover the whole species diversity of some highly biodiverse regions: the mean percentage of species covered per country in Asia is 57%, 44% for Oceania, 41% for Latin America/Caribbean, and 34% for Africa. We have shown that online communities of snake enthusiasts and herpetologists can contribute large volumes of geo-tagged snake observations [ 19 ] and that aggregating photos among online data sources brought our team to the leading edge of global snake image datasets, with almost 80% of the world’s species represented by at least one image [ 20 ]. Yet, only 20% of the world’s species met our 10-image threshold for inclusion in the training dataset, highlighting the extremely long-tailed distribution of photos per species.…”

“…private WhatsApp snake identification chat groups, private or natural history museum image collections) communities will allow us to progress in snake identification. The species under-represented in our snake photo dataset are shown in S1 Fig and S3 Table and the most-wanted species globally in [ 20 ]. To help improve the model accuracy and coverage, we encourage professional herpetologists and snake enthusiasts with photos of these species or missing species to submit them to The Reptile Database [ 4 ] or the citizen science biodiversity platforms iNaturalist and HerpMapper.…”

“…We used a subset of the world’s largest snake photo dataset, described in detail in Durso et al, 2021 [ 20 ], which we provided within the snake species identification challenge SnakeCLEF2021. This challenge is part of LifeCLEF21, the Conference and Labs of the Evaluation Forum (CLEF) that proposes data-oriented challenges related to the identification and prediction of biodiversity [ 23 ].…”

“…The vast majority (77%) of photos came from the United States and Canada, with 9% from Latin America and the Caribbean, 5.7% from Europe, 4.5% from Asia, 1.8% from Africa, and 1.5% from Australia/Oceania. Bias at smaller spatial scales undoubtedly exists as well [ 20 ], largely due to where participants in citizen science projects and other snake photographers are concentrated. Nevertheless, snake species from nearly every country were represented, with 46/215 (21%) of countries having all of their snake species represented, mostly in Europe.…”

“…Computer vision techniques have been developed to identify birds (e.g., Merlin Bird ID app recognizes over 7,500 species) and other animals like fish and butterflies [ 15 ], yet few initiatives seek to identify snakes and they are limited to certain taxonomic groups or geographic areas (e.g., [ 16 – 18 ]). In this study, we developed and tested the performance of a computer vision model to classify a large diversity of snakes using thousands of snake photos from across the world publicly available on open biodiversity platforms ( iNaturalist and HerpMapper ) and another online platform (Flickr)[ 19 , 20 ]. We further showed the high average per-country accuracy of this algorithm and its capability to distinguish sympatric lookalike species.…”