TY - JOUR
T1 - Otolith shape and size: The importance of age when determining indices for fish-stock separation
AU - Mapp, James
AU - Hunter, Ewan
AU - van der Kooij, Jeroen
AU - Songer, Sally
AU - Fisher, Mark
PY - 2017/6
Y1 - 2017/6
N2 - Stock-separation of highly mobile Clupeids (sprat – Sprattus sprattus and herring – Clupea harengus) using otolith morphometrics was explored. Analysis focused on three stock discrimination problems with the aim of reassigning individual otoliths to source populations using experiments undertaken using a machine learning environment known as \{WEKA\} (Waikato Environment for Knowledge Analysis). Six feature sets encoding combinations of size and shape together with nine learning algorithms were explored. To assess saliency of size/shape features half of the feature sets included size indices, the remainder encoded only shape. Otolith sample sets were partitioned by age so that the impact of age on classification accuracy could be assessed for each method. In total we performed 540 experiments, representing a comprehensive evaluation of otolith morphometrics and learning algorithms. Results show that for juveniles, methods encoding only shape performed well, but those that included size indices held more classification potential. However as fish age, shape encoding methods were more robust than those including size information. This study suggests that methods of stock discrimination based on early incremental growth are likely to be effective, and that automated classification techniques will show little benefit in supplementing early growth information with shape indices derived from mature outlines.
AB - Stock-separation of highly mobile Clupeids (sprat – Sprattus sprattus and herring – Clupea harengus) using otolith morphometrics was explored. Analysis focused on three stock discrimination problems with the aim of reassigning individual otoliths to source populations using experiments undertaken using a machine learning environment known as \{WEKA\} (Waikato Environment for Knowledge Analysis). Six feature sets encoding combinations of size and shape together with nine learning algorithms were explored. To assess saliency of size/shape features half of the feature sets included size indices, the remainder encoded only shape. Otolith sample sets were partitioned by age so that the impact of age on classification accuracy could be assessed for each method. In total we performed 540 experiments, representing a comprehensive evaluation of otolith morphometrics and learning algorithms. Results show that for juveniles, methods encoding only shape performed well, but those that included size indices held more classification potential. However as fish age, shape encoding methods were more robust than those including size information. This study suggests that methods of stock discrimination based on early incremental growth are likely to be effective, and that automated classification techniques will show little benefit in supplementing early growth information with shape indices derived from mature outlines.
KW - Fourier descriptors
KW - Curvature Scale Space
KW - Otolith shape
KW - Otolith size
KW - Stock discrimination
UR - http://www.sciencedirect.com/science/article/pii/S0165783617300267
U2 - 10.1016/j.fishres.2017.01.017
DO - 10.1016/j.fishres.2017.01.017
M3 - Article
VL - 190
SP - 43
EP - 52
JO - Fisheries Research
JF - Fisheries Research
SN - 0165-7836
ER -