Most Atlantic salmon production occurs in large sea cages, limiting individual growth recording opportunities except at stocking or harvest. This hampers management for monitoring mean weight, biomass, feed usage, and genetic selection. Integrating passive integrated transponder (PIT) identification with non-invasive diode frames allows continuous, minimally disruptive growth monitoring. However, diode frame accuracy for growth traits needs evaluation as the phenotypes can have some level or measurement error and missing phenotypes as not all fish are guaranteed to be recorded each day. Random regression genetic models enable predicting breeding values despite missing data. We stocked over 5000 individually PIT tagged Atlantic salmon post-smolts in a sea cage, monitoring growth using diode frames. Our study aims to assess diode frame accuracy, estimate genetic parameters for growth, and validate this approach through genetic correlations with harvest phenotypes.
The details of animal husbandry and phenotyping is published in detail elsewhere and briefly mentioned herein [1]. We stocked 5500 Atlantic salmon individually tagged with passive integrated transponders (PIT-tags) from the MOWI strain and transferred them to a sea cage. A diode frame (Biomass Daily, Vaki AS) and PIT tag antenna was installed. At the end of the trial all remaining 5133 Atlantic salmon were then manually recorded for weight and length for comparisons with diode frame data. Phenotypic validation was conducted between the latest measurements at sea and the manually recordings at harvest. Random regression models with second order polynomials were run using DMU v6 [2]. The last 5 measurements obtained within two weeks of the final harvest were averaged to obtain a single “corrected phenotype” and then run in bivariate with the harvest phenotypes to assess the genetic correlations and the effect of averaging diode frame phenotypes.
In total more than 35 000 reliable measurements were obtained on 4980 fish over 125 days [1]. A subset of 510 individuals with more than 10 measurements each were used for random regression models. At the population level diode frames were highly accurate with a mean difference of 0.002% for length and repeatable 0.35 with moderate concordance (CCC) of 0.52 for length. The random regression model for length was best fit by a second order Legendre polynomial on the additive genetic and permanent environmental effects. Heritability for length was low for diode frame measurements starting at 0.13 and increasing to 0.22 over the 125 day period and length at harvest was 0.38 (Figure1). The residual variance of diode frame measurements were proportionally higher, thus lowering heritability. The genetic correlation between harvest length and corrected length was 0.90 ± 0.01. Diode frame measurements hold promise for continuous growth measurements at sea, however they have higher residual and lower heritability, but are highly correlated to harvest length.
This work was supported by the research council of Norway project SmoltFieldGen (267650) and DigiFishent (334821).
1. Difford et al https://doi.org/10.1016/j.compag.2020.105411, 2. Jensen J, Madsen P. Calculation of Standard Errors of estimates of genetic and phenotypic parameters in DMU. 2005;