Using Relative Body Measurements of Nassau Grouper to Predict Total Length

Fish species that form spawning aggregations are particularly vulnerable to overexploitation. The case of Nassau Grouper in the Caribbean is a prime example, in which overfishing spawning aggregations has caused dramatic population declines. Collecting length distribution data at spawning aggregations is an effective method of monitoring populations, but requires significant time and resources. The Grouper Moon Project has collected 15 years of length data via diver-operated laser calipers at the Nassau Grouper spawning site off Little Cayman Island, resulting in 9,131 images with accurate length estimates. In this study, we leverage this image database to test how well relative body measurements can predict total length from single images. We used ImageJ to measure five distinct body parts of each fish in images that were perpendicular to the camera, and then tested three methods to predict total length. First, we found that the head height to eye diameter ratio was a significant, but weak, predictor of total length in a linear regression (p < 0.0001, R² = 0.05). Second, we fit a random forest model including all five measurements, which explained 16% of the variance in total length. Finally, we used a random forest classification model to predict 4 size classes of Nassau Grouper. The classification model accuracy was 0.5386, and the mean true lengths of the predicted size classes were significantly different (p < 2.551e-06). Thus, while the relative body measurements were not able to accurately estimate lengths of individual fish, they may be able to distinguish between populations with smaller or larger fish. Future research could explore using deep learning packages such as Keras or VIAME to develop algorithms to scan each image and select the best body measurements to predict total length.