Animal Wildlife Population Estimation Using Social Media Images

While tracking of wildlife populations is critical for protecting endangered species, the operational and financial burdens of large-scale censuses sharply limit their use. A promising solution to tracking such populations is to turn to an opportunistic form of citizen science: mining publicly available social media photos of animals. Estimation of wildlife population using such a solution is not straight-forward because of complexities in patterns of sharing due to biases inherent in social media data. In this paper, we aim to explicitly capture the biases of social media photo posts of wildlife. Specifically, we build a learning model that can predict the likelihood that any given safari photo will be shared. In our experiments, we find the classifiers showed promising prediction accuracies and other metrics. It is evident from the results that we can train classifiers that can model preferences in sharing safari photos on social media very accurately. Furthermore, we investigate the predictive features to explore the feasibility of embedding this approach in a population estimation application.