Report of the 2015 Herring Age Reading Exchange
The PGCCDBS (2014) proposed a full scale herring otolith exchange for the Atlantic and Baltic Seas to take place in 2015. The current exchange was initiated in 2015 and followed a small calibration exercise where only 3 institutes participated in reading otoliths from the North Sea and Irish Sea areas. It includes samples from the North Sea, Celtic Sea, Irish Sea and VIa (North and South) areas and was completed by 13 readers from 9 institutes. The aim of this combined exchange was to assess the accuracy of the age readings i.e. the proximity of the estimated ages to the modal age which is determined by an index of average percentage error (APE), percentage agreement and relative bias values, and to assess the precision i.e. the reproducibility of age estimates between readers which is determined using the coefficients of variation (CV). In addition, growth curves were compiled based on the distance data between annotations made on the otolith images hosted on the online annotation tool, WebGR. The growth curves allow for detailed examination of where the main problems with age interpretation are. Finally, Age Error Matrices were compiled for each area; these provide a measure of accuracy of the age readings and will be provided to HAWG 2016.
For the North Sea area (based on expert readers only) the overall APE is 14.8%. Bias in age estimates were found between the German and Dutch readers who are overestimating the ages in comparison to the modal age. Overall CV was 21.1 % and overall percentage agreement 73.6%.
For the Celtic Sea area (based on expert readers only) the overall APE is 14.2%. Bias in age estimates were found between the German and Dutch readers who are overestimating the ages in comparison to the modal age and to a lesser extent the Northern Ireland reader who is underestimating the ages in comparison to the modal age. Overall CV was 19.6 % and overall percentage agreement 75.2%.
For the Irish Sea area (based on expert readers only) the overall APE is 11.6%. Bias in age estimates were found between the German and Dutch readers who are overestimating the ages in comparison to the modal age and to a lesser extent the Northern Ireland reader and one reader from Norway who are underestimating the ages in comparison to the modal age. Overall CV was 16 % and overall percentage agreement 77.7%.
For the West of Scotland Sea area (based on expert readers only) the overall APE is 13.6%. Bias in age estimates were found between the German and Dutch readers and to a lesser extent two readers from Norway who are overestimating the ages in comparison to the modal age. Overall CV was 18.8 % and overall percentage agreement 69.1%.
The combined results show that 3 of the readers (2 of which are experts) are showing significant bias in their age readings. This maybe partly due to the differences which arise in age estimates when fish are aged in terms of “rings” versus “years”. The third reader is repeatedly omitting the first winter ring in the count of age. The age error matrices show that, in most cases, ages are overestimated my more than one year and this indicates that there is more than one ageing problem. The results of the growth curve analyses confirm this but annotation standardisation problems are apparent which can confound the results. Bias tests and plots give a more detailed description of reader performance.
