CLASSIFYING ENVIRONMENTAL IMPACTS WITH A COUPLED MANUAL-AUTOMATED LITERATURE REVIEW

Anthropogenic drivers of global environmental change are occurring at unprecedented rates. Rapid ecosystem changes have resulted in a rapidly growing body of scientific literature. Harnessing the power of existing literature through literature synthesis can provide important insights on data gaps and management needs. Manual literature synthesis remains an arduous and costly task, but advances in machine learning provide opportunities for automated text classification using natural language processing. This study applies coupled manual and machine learning methods to an expansive literature set for major global inland fisheries and explores opportunities for improving user efficiency for linking anthropogenic drivers of environmental change to direct impacts. We highlight driver-impact-response relationships and demonstrate the use of natural language processing text classification for environmental science literature. Results from this study will inform the relative influence of threats in the development of a global inland fisheries assessment, a necessary step towards improving fisheries conservation and sustainable management. More broadly, results provide an example of coupled manual-automated approaches for improved efficiency of preliminary classification steps in literature synthesis. Findings also points to the need for improved refinement of machine learning processes for fields of study (e.g., environmental sciences) where syntax may be less structured or standardized.