The exploitation of wild species globally: exploiting the internet for data

<p dir="ltr">The hunting, trapping trade and consumption of wild species, known as wildlife exploitation, has been identified as the second most important driver of terrestrial biodiversity loss. However, data on the location and intensity of exploitation is lacking in many regions, due to the challenging and resource intensive nature of field research. Consequently, we have limited understanding of the large-scale patterns of this threat, which impedes our ability to efficiently allocate conservation resources and interventions to where they are most needed. The increasing availability and global reach of data from online platforms, such as social media, news and blogs, could present a unique opportunity to fill in knowledge gaps and monitor wildlife exploitation at regional to global scales. Relevant content can be complex to identify from online searches, is susceptible to bias, and its reliability in representing on the ground trends has not been comprehensively assessed. Therefore, this thesis aimed to investigate if and how data from online sources could be used to understand large scale patterns of wildlife exploitation, using the case study of bats (Chiroptera). Bats are a globally distributed mammal order and play important ecological roles, such as pollination and pest suppression. However, their slow life histories and colony roosting behaviour make them particularly vulnerable to overexploitation. Despite being widely hunted, consumed and traded for bushmeat, medicine and other uses, our knowledge of the distribution and intensity of bat exploitation is very limited in some regions. Collecting online sources relevant to bats can be complex because the word ‘bat’ has many alternative meanings, and the implication of bats in high profile disease outbreaks has generated vast amounts of online commentary. Therefore, I first developed and evaluated the utility of a novel hierarchical text classification pipeline in identifying online sources relevant to bat exploitation from large scale searches of social media and search engines. I demonstrated that text classification performance was influenced by the machinelearning model used, the size and stratification of training data, and the complexity of different classification tasks. The high performance of the pipeline developed through this work emphasises the value of automated tools in facilitating online analyses in conservation science. Using the dataset collated through my automated pipeline, I then generated a global database of online bat exploitation records, and compared their spatial and temporal distribution to existing knowledge documented in academic literature and the IUCN Red List. The number of online bat exploitation records per country was predicted by the number of academic studies per country. However, online records also provided new locations and more recent records of bat exploitation, including 22 additional countries. Using a subset of my global database, along with records from a supplementary online search in Indonesian, I then applied statistical methods typically used in species distribution modelling to predict the spatial occurrence of bat exploitation across Indonesia, based on environmental predictors, such as human population density. This allowed me to develop spatially explicit threat maps that can now be ground-truthed and updated as new data become available. In addition to the documentation of exploitation in social media posts or web articles, wildlife trade is carried out online, using e-commerce platforms. Therefore, I also characterised the online trade in bat taxidermy and specimens, through automated monitoring of Ebay and Etsy over 18-months. This was facilitated by automated text classification and extraction of taxonomic and location information via automated named entity recognition. The data collected in this chapter provides new evidence for the bat species traded online, as well as potential sourcing hotspots, derived from their ranges. The findings of this thesis add to the increasing body of literature demonstrating the value of online platforms in providing data to inform conservation. I show that records of bat exploitation from social media and web articles both reliably expand our knowledge of the global extent of this threat, and can be used to generate spatially explicit threat maps for use in conservation decision making. Throughout my thesis, I evaluate and apply a suite of custom and off-the-shelf text analysis tools, which can be applied to future monitoring of threats to bats or other taxa. The various ways in which I applied text analysis demonstrate that the suitability and complexity of automated data collection approaches is highly dependent on the specific textual sources being collected and the research question. Importantly, this work also highlights the urgent need to address bat exploitation globally, and better understand the online, international trade in bats, which is likely an emerging threat.</p>