Deep Learning–Based Estimation of Fish Abundance and Temporal Patterns in Agricultural Drainage Canals for Sustainable Ecosystem Monitoring

Agricultural drainage canals provide critical habitats for fish species whose populations are sensitive to agricultural practices, highlighting the need for effective and non-invasive monitoring techniques. In this study, we developed a practical method using the YOLOv8n deep learning model to automatically detect and quantify fish occurrence in underwater images captured from an agricultural canal in Ibaraki Prefecture, Japan. Two fish species, Pseudorasbora parva (topmouth gudgeon) and Cyprinus carpio (common carp), which represent small- and large-bodied fish, respectively, were targeted. The trained YOLOv8n model achieved a high validation performance (F1-score = 91.6%, Precision = 95.1%, Recall = 88.4%), although external inference testing on 12,600 images indicated a reduced performance under practical field conditions (F1-score = 61.6%, Precision = 88.9%, Recall = 47.1%). By correcting for detection errors and repeated detections, we estimated that approximately 7,300 individuals of P. parva and 80 individuals of C. carpio passed the observation point over a seven-hour monitoring period. This study demonstrates the potential of deep-learning-based automated monitoring methods for sustainable aquatic ecosystem management in agricultural landscapes. Future research should focus on improving the detection Recall under challenging environmental conditions, such as turbidity and partial visibility, to enhance practical applicability.