Fourier transform near infrared spectroscopy of otoliths coupled with deep learning improves age prediction for long-lived northern rockfish

Abstract

The northern rockfish (Sebastes polyspinis) is an economically valuable, long-lived species distributed over the continental shelf of the North Pacific Ocean. Ages for this species which can be in excess of 80 years comprise an essential component of models for assessing population status and are crucial for fisheries management. Traditional microscope-based methods of estimating age using otoliths can be time-intensive and prone to reader variability. We explored the application of Fourier transform near infrared (FT-NIR) spectroscopy coupled with multimodal convolutional neural networks (MMCNN) for age prediction. Our study included 2613 FT-NIR scans and associated ages of northern rockfish otoliths from years 2013–2019, with ages ranging from 3 to 66 years. The optimal MMCNN model demonstrated strong performance, yielding an R² of 0.92 and an RMSE of 3.38 for the training set and an R² of 0.89 and an RMSE of 3.74 for the test set. Spectral information in the 11,500 to 4000 cm⁻¹ wavenumber range, otolith weight, and other biological/geospatial data contributed to age predictions that were comparable to traditional age estimates. Despite challenges, FT-NIR spectroscopy coupled with MMCNN emerged as a promising alternative for age estimation in long-lived species. This approach, while demonstrating effectiveness for northern rockfish, could be a valuable tool for diverse fish species, supporting sustainable fisheries practices and population monitoring.