AI-assisted data extraction helps uncover spatiotemporal patterns and socioeconomic drivers of wildlife crime involving sea turtles

Abstract

Crimes associated with an ever-increasing demand for wildlife products are one of the most notable threats to marine and freshwater ecosystems. To combat such crimes, it is crucial to identify their spatiotemporal patterns and hotspots, which have largely been overlooked in previous research. However, especially after the emergence of large language models (LLMs), this process has become more time-efficient and accurate. In this study, we analyzed spatiotemporal patterns and socioeconomic drivers of wildlife crime in sea turtles, using Deepseek to extract data from 247 court verdicts. DeepSeek data extraction reached an accuracy of over 99 % in extracting 25 items from each verdict. We found that most individual sea turtles and products were seized in southeastern coastal cities of China and identified two main trafficking hotspots. First, nearly 73 % (450/613) of the hawksbill turtles and 84 % (325/386) of the green turtles were seized or originated from Hainan province, China. Second, nearly 98 % (207/211) of the loggerhead turtles were seized from Zhoushan, Zhejiang province. Moreover, nearly all the manufactured products (over 99 %, mainly made of tortoiseshell) were seized or originated from Hainan. Destinations of trafficking tended to be northern inland cities, with one main hotspot: 5.5 % (68/1236) of individuals and 30.3 % (8896/29,323) of the products were seized in Xuzhou, Jiangsu province, which originated from Hainan. Our study highlights how AI tools can boost biodiversity conservation research by leveraging large datasets. In doing so, we were able to identify major hotspots of wildlife crime, as well as main trafficking routes. These findings might be relevant for law enforcement efforts and help to enhance sea turtle conservation.