54 Priyantha I. Hathurusingha and Kenneth R. Davey
They reported, generally very good agreement between observed and
predicted taint concentration in the range 0 to 2, μg kg-^1 , and especially below
the important consumer rejection threshold (< ~ 0.7 μg kg-^1 ). Despite a very
good correspondence between model prediction and the extensive
experimentally determined data, the model in its present form was shown to
over-predict taste-taint chemical accumulation in fish-flesh. Particularly, this
general over-prediction of chemical was about a half of the N = 706
simulations in the range 0 to 12, μg kg-^1. Predictions were reported to be
conservative therefore i.e., on the ‘safe’ side. The possible reasons for this
include dissimilar (exponential) growth constants for smaller and larger fish,
and the oscillatory RAS environment. However, it proposed to include in a
revised model two growth constants, one for smaller (juvenile) fish and one
for the fish as they approach harvest. Significant additional validation studies
would be required however in this case. They reported that this theoretical
development was currently being planned (Hathurusingha and Davey, 2016 b).
Findings highlighted (Hathurusingha, 2015) that the work could be
meaningfully applied to RAS systems to develop protocols to limit taste-taint
in harvested fish. Significantly, the results are the first for RAS farmed-fish
covering an entire production cycle from fingerlings to harvest and will
therefore be of immediate benefit and interest to RAS farmers, selling agents
and researchers.
This validated model of Hathurusingh and Davey (2014) provided insights
into taste-taint modelling in RAS and demonstrated new ways to predict taint
accurately in fish-growth.
It is worth noting that modelling is ongoing process - and it is expected
will continue in time to supplant older ones with new.
RISK ANALYSES OF PREDICTIVE MODELS
A major drawback of current predictive models is they are based on
‘average’ or ‘mean’ deterministic values for input. There will however be
naturally occurring, chance fluctuations in biological input parameters that
could impact outcomes. It is uncertain how these can impact on predictions.
To be able to quantify the impact of naturally occurring random
fluctuations in key parameters in otherwise well-operated systems, Davey and
co-workers developed a new probabilistic quantitative methodology of risk
analyses known as ‘Fr 13’. Their hypothesis is that these random fluctuations
can unexpectedly and suddenly accumulate and combine in one direction and