Using morphometric relationships for total length prediction of Atlantic cod (Gadus morhua) in electronic monitoring videos

Effective fisheries management relies on accurate data on catch compositions to ensure sustainable fishing practices and protect marine ecosystems. While traditional data collection methods effectively document the wanted catch components, the data for the unwanted catch component often remain elusive, leaving a significant gap in fisheries management. Electronic monitoring (EM) systems have shown potential as a tool that can help in enhancing legal compliance and improve data accuracy by remotely monitoring catches. However, analysis of EM videos remains challenging for both trained video observers and computer vision algorithms due to issues such as occlusion from overlapping fish or obstructed field of view. This study explores the use of morphometric relationships specifically for Atlantic cod (Gadus morhua), aiming to establish a proof-of-concept to estimate the total lengths of partially occluded fish in EM videos. An exploratory analysis of 34 morphometric relationships found seven morphometric parameters to be particularly suited for length estimation, namely, head length (HL), pre-orbital length (PROL), post-orbital length (POOL), horizontal eye diameter (HED), pectoral fin length (PCFL), pectoral fin width (PCFW), and caudal peduncle height (CPH). Based on bi- and multi-variate ordinary least squares (OLS) regressions with heteroskedasticity-consistent standard error (HCSE) estimators, prediction equations to support trained video observers and computer vision algorithms are proposed for 127 occlusion scenarios. Our results show that morphometric parameters can be used to accurately predict total lengths in catch compositions with partly occluded individuals. The predictive accuracy of the regressions improves as the number of morphometric parameters used for length prediction increases.