See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/308018122

Experimental validation of a time-dependent

model for chemical taste taint accumulation as
geosmin (GSM) and 2-methylisoborneol (MIB)
in commercial RAS farmed barramundi (Lates
cal...

Article in Ecological Modelling · November 2016

DOI: 10.1016/j.ecolmodel.2016.08.017

CITATIONS READS

0 3

2 authors:

Hathurusingha Arachchige Priyantha Indraj… Kenneth R Davey

University of Adelaide University of Adelaide
5 PUBLICATIONS 9 CITATIONS 126 PUBLICATIONS 974 CITATIONS

SEE PROFILE SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate, Available from: Hathurusingha Arachchige Priyantha Indrajith
letting you access and read them immediately. Retrieved on: 19 October 2016
 Ecological Modelling 340 (2016) 17–27

Contents lists available at ScienceDirect

Ecological Modelling
journal homepage: www.elsevier.com/locate/ecolmodel

Experimental validation of a time-dependent model for chemical

taste taint accumulation as geosmin (GSM) and 2-methylisoborneol
(MIB) in commercial RAS farmed barramundi (Lates calcarifer)
Priyantha I. Hathurusingha, Kenneth R. Davey (PhD, FIChemE, CEng, FAIFST, CSci) ∗
School of Chemical Engineering, The University of Adelaide, 5005, Australia

a r t i c l e i n f o a b s t r a c t

Article history: The accumulation of naturally-derived taste taint chemicals, geosmin (GSM) and 2-methylisoborneol
Received 18 March 2016 (MIB), impair the flavour of Recirculating Aquaculture System (RAS) farmed-fish. Quantification and con-
Received in revised form 22 August 2016 trol of these by analytical or sensory means are not presently practical for RAS farmers. To forecast taste
Accepted 30 August 2016
taint in RAS, a time-dependent model was synthesised (Ecological Modelling 291 (2014) 242–249). Here
we report for the first time an extensive two-year validation study of this model with commercial RAS
Chemical compounds studied in this article:
farmed barramundi (Lates calcarifer) in which fingerlings (∼0.01 kg) were grown to harvest (∼0.85 kg) at
(−)-Geosmin (PubChem CID: 29746)
245 days. The concentration of GSM and MIB in the growth water and fish-flesh was determined (weekly)
(+)-Geosmin (PubChem CID: 15559490)
2-Methylisoborneol (PubChem CID: 16913)
using Headspace Solid-Phase Micro-Extraction, followed by Gas Chromatography Mass-Spectroscopy
(and microwave-mediated distillation). The concentration of both taint chemicals in the RAS growth
Keywords: water was controlled using continuous dosing of hydrogen peroxide (2.5 mg L−1 ) as a benign biocide. A
2-Methylisoborneol (MIB) special dosing apparatus was developed for this purpose. Results showed, generally, very good agree-
Geosmin (GSM) ment between observed and predicted taint values in the range 0–2 ␮g kg−1 , and especially below the
Barramundi (Lates calcarifer)
important consumer rejection threshold (<∼0.7 ␮g kg−1 ). A minor anomaly was a general over-prediction
Recirculating aquaculture system (RAS)
of chemical in about a half of the N = 706 simulations in the range 0–12 ␮g kg−1 . Predictions were con-
Experimental studies
Model validation servative therefore i.e. on the ‘safe’ side. This is attributed, largely, to dissimilar (exponential) growth
constant () for smaller and larger fish, and the fact that the RAS environment is oscillatory. Findings
highlight that the work could be meaningfully applied to RAS systems to develop protocols to limit taste
taint in harvested fish. Significantly, these results are the first for RAS farmed-fish covering an entire
production cycle from fingerlings to harvest. The work will be of immediate benefit and interest to RAS
farmers, selling agents and researchers.
Crown Copyright © 2016 Published by Elsevier B.V. All rights reserved.

1. Introduction A number of approaches to remove or inactivate these taste

Farming of fish in Recirculating Aquaculture Systems (RAS) is
becoming widespread to fill the demand gap due to diminishing
wild-caught supplies. Barramundi (Lates calcarifer) is grown as an
RAS farmed-fish because of high demand as a premium food.
However, the natural accumulation in RAS fish-flesh of ‘earthy’
or ‘muddy’ off-flavours due to taint accumulation of chemicals as
geosmin (GSM) or 2-methylisoborneol (MIB) is a major concern.
Inconsistent taste quality of farmed barramundi has been identified
as a major issue in buyer resistance.
∗ Corresponding author.
E-mail address: kenneth.davey@adelaide.edu.au (K.R. Davey).
taint chemicals have been investigated − but effective practices
to control taste taint have not gained acceptance (Howgate, 2004;
Schrader et al., 2010). Presently, barramundi farmers rely on post-
harvest purging with clean water as a taste taint abatement
strategy. However, a drawback is fish are not fed during this
purge-stage and lose mass (weight), resulting in a financial penalty
(Palmeri et al., 2008).
The consumer rejection threshold concentration for these taste
taint chemicals in RAS fish-flesh is ∼0.7 ␮g kg−1 (Robertson et al.,
2005). Fish exceeding this threshold concentration are regarded
as un-marketable. The limiting concentration of GSM and MIB in
the RAS growth water that will impart this level of unwanted
off-flavour to the harvested fish-flesh is 0.015 and 0018 ␮g L−1
respectively (Persson and York, 1978; Persson, 1980). That is, where
the concentration in the growth water of the taint chemical exceeds

http://dx.doi.org/10.1016/j.ecolmodel.2016.08.017
0304-3800/Crown Copyright © 2016 Published by Elsevier B.V. All rights reserved.
18 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27

used to predict taste taint in fish-flesh with time and assist farmers
Nomenclature to harvest fish with taste taint lower than the rejection threshold.
This has led to a number of modelling approaches. Until recently
a Taint elimination coefficient, day−1 (5) however these models were based on assumptions of steady-state.
b taint accumulation coefficient, ␮g kg−1 day−1 (4) Because there was no evidence to show that net chemical
COX concentration of dissolved oxygen, mg L−1 (12) exchange was in fact zero across the RAS water phase and fish body,
CW Taint concentration (as GSM or MIB) in water, ␮g L−1 Hathurusingha and Davey (2014) synthesised a quantitative model
(2) that predicted the time-dependent concentration of taste taint
dy
dt
Rate of change of taint in fish-flesh, ␮g kg−1 day−1 chemicals as GSM and MIB in RAS fish-flesh. The model was based
(1) on conservation of mass and energy, and thermodynamic pro-
e Lipid mass ratio, dimensionless fraction (16) cesses established in bio-chemical engineering. Chemical uptake
EW Chemical uptake efficiency across the gill, dimen- and elimination routes into and from the fish were integrated. The
sionless fraction (9) model was illustrated with independent (but fragmented) litera-
GSM Geosmin ture data for farmed barramundi
GV Gill ventilation rate, l day −1 (9) (n ≥ 14). The applicability of a generalized form of the
k1 Taint accumulation through gills, l kg−1 day−1 (2) model to other aquaculture species, in particular rainbow trout
k2 Taint elimination through gills, day−1 (2) (Oncorhynchus mykiss), was demonstrated with independent data
kg Growth dilution plus metabolic transformation, (n ≥ 15) (Davey and Hathurusingha, 2014; Hathurusingha and
day−1 (2) Davey, 2013).
kG Rate constant for growth dilution, day−1 (17) There has been however no detailed study reported (or as far as
km Rate constant for metabolic transformation, day−1 can be determined even undertaken) to analyse GSM and MIB in
(17) RAS growth water and accumulated concentrations in fish-flesh of
KOW Octanol- water partition coefficient, dimensionless barramundi, especially covering an entire commercial production
(10) cycle (245 days). This gap emphasized the need for an extensive
mf Mass of the fish at t, kg (2) and dedicated sampling of the RAS growth water and fish-flesh for
MIB 2-methylisoborneol the model validation.
n Number of data
N Number of all data
QW Rate of chemical transport in the aqueous phase, 1.1. This study
l day−1 (13)
Here, the model of Hathurusingha and Davey (2014) is exper-
QL Rate of chemical transport in the lipid phase, l day−1
imentally tested against extensive and original data obtained
(13)
from commercial-scale RAS farmed barramundi from fingerlings
RAS Recirculating aquaculture system
(∼0.01 kg) to harvested fish (∼0.85 kg) for the first time.
R2 Correlation coefficient
Both GSM and MIB in growth water and barramundi fish-flesh
T temperature, ◦ C (12)
were sampled weekly. A number of experimental and analytical
t Growth time, day (1)
protocols were expressly adapted for this purpose. The concentra-
VL Lipid mass (weight), kg (13)
tion of taint chemicals in the RAS growth water was controlled
y Taint (as GSM or MIB) concentration, ␮g kg−1 (1)
using continuous dosing of hydrogen peroxide (2.5 mg L−1 ) as a
benign biocide via a specially-developed apparatus. This consisted
Greek symbols
of a controller, H2 O2 sensor (ProMinent Fluid Controls Pty Ltd,
ˇ Pre-exponential growth constant, kg (3)
Germany) and a metering pump (Global Pumps Pty Ltd, Australia).
 Exponential growth constant, day−1 (3)
It was operated automatically in an on/off mode in which dosing of
H2 O2 to the growth water was cut off when the bulk water residual
concentration reached a set-point. Dosing resumed when the con-
this limiting value, the fish-flesh will consequently become unac- centration dropped below the set-point. The apparatus and its safe
ceptably tainted. and Standard Operating Procedures (SOPs) are described in detail
A major difficulty is quantification of GSM and MIB taste in Hathurusingha (2015). Further, this study is the first full-scale
taint chemicals in RAS water and fish-flesh because of the costs experimental investigation of the accumulation of GSM and MIB
involved, experimental time and the few research institutes that in any RAS fish species. Rigorous analyses of observed against pre-
have the necessary facilities (G. Vandenberg, Faculté des sciences dicted taste taint data are presented. Results are discussed with a
de l’agriculture et de l’alimentation, Université Laval, pers. comm.). view to development of RAS growth protocols to limit GSM and MIB
As an alternative, sensory (taste and smell) assessments have been in the flesh of fish.
adopted to determine whether a fish is tainted beyond a threshold Findings will be of immediate benefit and interest to RAS farm-
(Grimm et al., 2004; Percival et al., 2008). This approach however is ers, selling agents and researchers.
based on human perception and requires ‘experts’ to do this accu-
rately. It is therefore subjective, and; it is questionable whether the
2. Materials and methods
experts are representative of the consumer population (Howgate,
2004). Moreover, it is suggested, successive testing of MIB, for 2.1. Model for taste taint
example, influences the taste adaptation in experts that can lead
to a reduction in sensitivity to taste taint (Brett and Johnsen, 1996; The transient model for chemical taste taint accumulation of
Johnsen and Brett, 1996). Both approaches therefore are not practi- Hathurusingha and Davey (2014) can be summarised as follows:
cal or feasible for use by farmers and farming management during The governing route for taste taint uptake into the fish-flesh
RAS fish growth. is assumed to be the gills, and elimination, from the gills,
Predictive models, widely used in chemical and bio-chemical bio-transformation and fish growth dilution. The optimum growth-
engineering however, provide a quantitative basis for evaluating temperature of the RAS water for barramundi is considered to be
toxicology and risk assessments in fish. An adequate model could be 28 ◦ C.
 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27 19

The uptake of taint (y) in the fish-flesh with time (t) was given where T = RAS growth water temperature (◦ C).
by The assumption of oxygen saturated RAS growth water was
justified by Hathurusingha and Davey (2014) because, in conjunc-
dy
= uptake rate − elimination rate (1) tion with temperature control, aerator-mixers are continuously
dt
employed to ensure optimum fish growth (Southern Barramundi
where all notation used throughout is carefully defined in the Farmer’s Association, Clarendon, Australia, pers. comm.).
Nomenclature. The chemical elimination rate from the fish gills (k2 ) was corre-
Substitution for the uptake of taint chemical as a rate constant lated with the chemical transport rate in aqueous and lipid phases
(k1 , L kg−1 day−1 ), fish mass (mf , kg) and taint concentration in the of the fish, lipid content of the fish and octanol water partition coef-
RAS water (CW ) – together with elimination through the gills (k2 , ficient of the taint chemical based on the work of (Gobas, 1993)
day−1 ) and bio-transformation and growth dilution (kg , day−1 ) gave as
dy 1 VL VL
= k1 mf CW − (k2 + kg )y (2) =( )KOW + ( ) (13)
dt k2 QW QL
where ( dy
dt
) was the rate of change of taint in the fish-flesh where QW is rate of chemical transport in the aqueous phase
(␮g kg−1 day−1 ). (L day−1 ), QL is the rate of chemical transport in the lipid phase
The mass of fish (mf ) grown for harvest was known to be a (L day−1 ) and VL is lipid weight (mass). The relationship derived by
function of growth-time in RAS. Published growth data as mass of Gobas and Mackay (1987) based on experimental data between QW
fish (kg) vs growth time (day) was highly exponentially correlated and mf was
(Glenn et al., 2007) and fish mass was given by QW = 88.3mf 0.6(±0.2) (14)
mf = ˇexpt (3) The chemical transport rate in the aqueous phase compared
Two constants were introduced such that for uptake with the lipid phase reported by Gobas and Mackay (1987) was
used i.e.
b = ˇk1 CW (4)
QL = 0.01QW (15)
and for elimination
Lipid mass of the fish was correlated to the lipid mass ratio (e)
a = (k2 + kg ) (5) (dimensionless) and conveniently expressed as
Substitution for b and a into Eq. (2) and rearranging gave VL = emf (16)
dy
+ ay = bexpt (6) The rate constant for combined metabolic transformation of the
dt taint chemical and growth dilution rate of the taint chemical was
Eq. (6) was integrated by parts to give given by
bexpt b kg = (kG + km ) (17)
y= − (7)
(a + ) (a + )expat
The rate constant for growth dilution was computed using the
Eq. (7) was rearranged to give a predictive model for taint chem- equation of Thomann et al. (1992)
ical in the fish-flesh
  kG = 0.00251mf −0.2 (18)
b
y=( ) expt − exp−at (8)
a+ A value of rate constant for metabolic transformation of the
taint chemical km = 0.00063 day−1 was taken from the refereed lit-
The derived model (Eq. (8)) is seen to be the difference of two
erature to cover a wide range of fish species. A typical growth
exponential terms with time in RAS growth tanks, namely, uptake
curve was presented by Hathurusingha and Davey (2014) for bar-
and elimination of taste taint chemical(s).
ramundi (Lates calcarifer) based on independent data (n = 9) (Glenn
Based on the work of Arnot and Gobas (2004) the gill uptake
et al., 2007) from which it was found Eq. (3) gave a very good fit
rate constant (k1 ) was expressed as
(R2 = 0.96) (Snedecor and Cochran, 1989) with ˇ = 0.0519 kg and
EW GV  = 0.0133 day−1 .
k1 = (9)
mf Eqs. (1) through (18) plus the values for km , ˇ and  define the
where EW was the gill chemical uptake efficiency (dimensionless model of Hathurusingha and Davey (2014) for chemical taste taint
fraction) and GV was the gill ventilation rate (L day−1 ). The chemical accumulation as GSM and MIB in RAS barramundi. An advantage of
uptake efficiency was assumed based on Gobas (1998) to be a func- the model was that it could be conveniently set-up and solved in
tion of an octanol-water partition coefficient (KOW ) of the chemical standard spread sheeting tools.
of interest that was expressed through the following relationship
−1
2.2. Commercial site
EW = (1.85 + 155/KOW ) (10)

The relationship between gill ventilation rate and oxygen con- The experimental study was carried out at Barra Fresh Farm,
sumption rate of the fish was based on empirical data reported by Gawler, Australia. The commercial farm has four (4) separate rect-
Arnot and Gobas (2004) angular RAS growth-tanks in a large, insulated indoor shed. Each,
contained fresh water with a volume of 120,000 L and each was
1400mf 0.65 operated in a fully closed-loop system – with some water replen-
GV = (11)
COX ished periodically to replace evaporation losses and to remove
where COX was the concentration of dissolved oxygen (mg L−1 ). This fish-waste.
was considered as a function of temperature and computed from Each of the four separate growth-tanks (1, 2, 3 and 4) was parti-
the equation given by Neely (1979) tioned into four (4) rectangular sub-tanks (30,000 L) with nets for
convenient managing, rearing and stocking fish of different age-
COX = 14.45 − 0.413T + 0.00556T 2 (12) groups. The sub-tanks of each of the growth-tanks are identified as
20 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27
1.0
0.8
Mass (kg)
0.6
0.4
0.2
0.0
0 30 60 90 120 150 180 210 240
Day
Fig. 1. Fit of growth data determined for barramundi from the treated tank (2B)
at measured water temperature ∼27 ◦ C (mf = 0.0226e0.0169t , R2 = 0.92). Error bars
indicate standard deviation on 20 fish.
A, B, C and D. Each of the sub-tanks were physically segregated by
nets but shared a common RAS water matrix in which the growth
water was recirculated through common biological and mechanical
filters.
The commercial stocking density of barramundi is set at around
80 kg m−3 . These are fed at a rate of 1.5% of body mass day−1 . Live
barramundi, when farmed and harvested, are supplied to the local
market with an average mass of 0.8–1.0 kg obtained in a (typical)
growth of 245 days.
The bulk temperature of the growth water is maintained at
27 ± 2 ◦ C for optimum growth of the fish. The required heat energy
is generated through external heaters (in a separate room with the
heated water being pumped back to the growth-tanks). In addition,
submerged mechanical aerators are used to increase the dissolved
oxygen concentration in the RAS growth water to maintain it at
5–7 mg L−1 .
This commercial RAS consisted of typical industry components
with biological and mechanical filters, disease control elements,
pumps, plumbing, environmental controls and, backup-power sup-
ply.
2.3. Preliminary developments
Significant preliminary developmental work was carried out
to test and establish a controlled dosing apparatus to administer
hydrogen-peroxide (2.5 mg L−1 ) as a benign biocide to limit con-
centration of GSM and MIB in the commercial RAS growth water.
Two isolated sub-tanks (2B and 4D) were selected for the
245 days experimental validation: 4D was used as a ‘control’
(untreated) and 2B as ‘treated’ with hydrogen-peroxide. This
arrangement fitted with the commercial management of the Farm.
2.4. Introduction of fingerlings to commercial RAS
Some 3326 and 3318 fingerlings, respectively, were introduced
into commercial tanks 2B and 4D. The mass of these fingerlings
ranged from 0.010 to 0.020, kg.
Each fingerling was injected with 0.1 mL of vaccine (Streptococ-
cus iniae) prior to release (by Barra Fresh Farm staff). The vaccination
of fingerling is usual practice at the Farm to increase immunity of
the fish, and to minimise fish mortality.
1.0
0.8
0.6
Mass (kg)
0.4
0.2
0.0
0 30 60 90 120 150 180 210 240
Day
Fig. 2. Fit of growth data determined for barramundi from the control tank (4D)
at measured water temperature ∼ 27 ◦ C (mf = 0.0229e.00161t , R2 = 0.98). Error bars
indicate standard deviation on 20 fish.
2.5. Water and fish analyses
Samples were collected and data obtained concurrent with
the Farm’s commercial production. RAS growth water was sam-
pled for analyses of GSM and MIB weekly over the 245-day
growth period. This coincided with growth of the fingerlings to
commercially harvestable fish (Sampling commenced in the first
week (November 2014) and was completed in the last (June
2015)).
The growth water temperature and dissolved oxygen concen-
tration (COX ) were measured in situ daily (using calibrated probes
at the Farm).
Water samples for GSM and MIB analyses were collected into 5 L
Low-density polyethylene (LDPE) containers without headspace.
These were transported to the Laboratory, School of Chemical Engi-
neering, in a cooler box at 4 ◦ C. Samples were taken at the same
point on each occasion at one side of the RAS tanks. Water samples
were drawn from 0.01 m below the surface and 0.25 m away from
the tank wall.
Two (2) sets of fish samples from 2 B (treated) and 4D (control)
were taken for analyses of taste taint chemicals and fish growth.
Fish for flesh tissue(s) was sampled weekly in parallel with the
water samples. Hand-nets (with the help of Farm staff) were used
for random sampling. The fish sampled gave sufficient tissue to be
analysed. They were killed at the Farm by ice-water immersion.‘ The
collected fish were weighed, filleted and enclosed in polythene bags
with labels for later analyses. They were frozen and transported to
the Laboratory in a cooler box. Fish samples for fish growth data
were taken with a minimum number of 20 fish for each set of
analyses.
The RAS growth water samples were analysed using the most
advanced robust Headspace Solid-Phase Micro-Extractor (HSPME)
followed by GC–MS (known as HSPME-GC–MS) as detailed in
Watson et al. (2000) with some optimizations. These sample prepa-
ration methods were optimised to improve the sensitivity of GSM
and MIB extraction using large volume of RAS growth water (1 L)
with a manual extraction technique.
Microwave mediated Distillation with Solid Phase Micro
Extraction (MD-SPME) as detailed in Lloyd and Grimm (1999)
was modified and adopted for GSM and MIB analyses in fish-
flesh.
 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27 21

0.045

0.040

0.035
Concentration (µg L-1)

Limiting concentration in RAS growth water

0.030

0.025

0.020

0.015

0.010

0.005

0.000
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

Week

Fig. 3. Variation of GSM concentration (CW ) in RAS growth water in treated (2B) ( ) and control (4D) ( ) tanks over a 35 week period. Results presented are the average
of (n = 3) replicates. Error bars indicate standard deviation on the triplicates. The horizontal dashed-line indicates the limiting concentration for GSM (0.015 ␮g L−1 ) in RAS
growth water.

0.18

0.16

0.14
Concentration (µg L-1)

Limiting concentration in RAS growth water

0.12

0.10

0.08

0.06

0.04

0.02

0.00
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
Week

Fig. 4. Variation of MIB concentration (CW ) in RAS growth water in treated (2B) ( ) and control (4D) ( ) tank over a 35 week period. Results presented are the average
of (n = 3) replicates. Error bars indicate standard deviation on the triplicates. The horizontal dashed-line indicates the limiting concentration for MIB (0.018 ␮g L−1 ).

3. Results Importantly, it is concluded therefore there was no difference in

the growth curve for treated (2B) and control (4D), fish populations.
3.1. Fish growth curve
3.2. GSM and MIB in RAS water
The detailed fish growth curve for tanks 2B (treated) and 4D
(control) over 245 days is presented in Figs. 1 and 2, respectively.
The variation of GSM and MIB concentration of growth water in
The number of fish used for each sample was 20 and the average
the treated (2B) and control tanks (4D) over the 35 week period is
value of mass is shown. As can be seen in the figures, the fish from
shown in Figs. 3 and 4, respectively. The results are presented as
both tanks exhibit typical exponential growth with time. The cor-
the average concentration of (n = 3) three samples. The threshold
relation coefficients (R2 = 0.92 and 0.98) indicate a very good fit
concentrations of 0.015 ␮g L−1 for GSM and 0.018 ␮g L−1 for MIB are
(Snedecor and Cochran, 1989).
highlighted as horizontal dashed-lines in the respective figures.
The pre-exponential growth constants (ˇ) for treated and con-
The variation of GSM concentration over the 35 week period is
trol groups was 0.0226 and 0.0229, kg respectively. There is no
illustrated in Fig. 3. As can be seen, there is no clear, or distinctive,
meaningful difference between these. Similarly, the exponential
pattern in the variation of GSM in either the treated or control tank.
growth constants () of the fish were very similar at 0.0169 and
Notably, the concentration of GSM in the growth water in both
0.0161 day−1 .
the treated and control tanks (2B) and (4D) respectively seems
to vary concurrently, increasing and decreasing in concentration
22 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27

over the 35 week growth period. The variation in concentration of

GSM in the water in the treated (2B) tank ranged from 0.0016 to
0.0364 ␮g L−1 and 0.0014 to 0.0385 ␮g L−1 in the control (4D) tank.
The computed average-concentration of GSM in the treated
tank over the 35 week period was 0.0161 ␮g L−1 and for the
control, 0.0147 ␮g L−1 . It is important to note that this average
concentration was close to the limiting concentration for GSM i.e.
0.015 ␮g L−1 (Persson, 1980; Persson and York, 1978).
The variation in MIB concentration in both tanks over the 35
week period is illustrated in Fig. 4. The pattern of variation of MIB
in both tanks is seen to be dissimilar. The variation in the treated
tank ranged from 0.0187 to 0.154 ␮g L−1 and 0.0114 to 0.170 ␮g L−1
in the control tank.
It is seen from Fig. 4 that the MIB concentration in both tanks
is above the limiting concentration (0.018 ␮g L−1 ) for most of the
time. For example the control tank has a concentration that is below
the limiting concentration on two (2) occasions, whereas that in the
treated tank was none. Fig. 5. Variability in lipid concentration (w/w%) within the fillet of 0.45 kg barra-
The calculated average concentration of MIB in the treated tank mundi. The lipid concentrations (w/w%) shown were determined from sample taken
from the area marked by the respective dashed-lines.
over the 35 week period was 0.0659 ␮g L−1 , and that in the con-
trol tank, 0.0716 ␮g L−1 . The concentration in the treated tank was
Table 1
therefore 4.09 times greater than that of average GSM concentra-
Weekly average dissolved oxygen concentration (COX ) and RAS growth water tem-
tion in the same tank. Similarly, the concentration in the treated perature (T) of the 2 B and 4D tanks over the 35 week period. Results are the average
tank was 4.86 times greater than that of average GSM in the same value of 7-day readings. The standard error is shown in parentheses.
tank.
Week Tank
It is concluded therefore that these data demonstrate that MIB is
the dominant taste taint chemical in both treated (2B) and control 2B 4D
(4D) tanks. Temp (◦ C) COX (mg L−1 ) Temp (◦ C) COX (mg L−1 )

1 26.95 (0.21) 5.00 (0.42) 26.34 (0.28) 5.08 (0.47)

3.3. GSM and MIB in fish-flesh 2 27.22 (0.14) 4.80 (0.17) 26.87 (0.22) 4.09 (0.38)
3 27.12 (0.20) 4.62 (0.19) 26.81 (0.55) 5.12 (0.52)
4 27.44 (0.19) 4.64 (0.15) 27.44 (0.20) 4.84 (0.29)
The concentration of GSM in fish-flesh of the treated (2B) tank 5 27.77 (0.17) 5.24 (0.69) 27.92 (0.17) 4.98 (0.48)
ranged from 0.03 to 6.5 ␮g kg−1 and in the control (4D) tank from 6 27.72 (0.13) 4.72 (0.57) 28.04 (0.24) 4.35 (0.16)
0.01 to 7.00 ␮g kg−1 . The correlation coefficient between the taint 7 27.41 (0.33) 5.07 (0.68) 27.80 (0.33) 4.88 (0.42)
as GSM concentration in fish-flesh and the mass of the fish of the 8 27.28 (0.30) 4.71 (0.73) 27.61 (0.21) 5.07 (0.49)
9 27.22 (0.24) 4.40 (0.19) 27.47 (0.13) 4.78 (0.42)
treated and control fish tanks was R2 = 0.45 and R2 = 0.53, respec- 10 27.84 (0.25) 4.31(0.37) 27.97 (0.26) 4.90 (0.40)
tively. In general however, these data show greater concentration 11 27.98 (0.26) 4.08 (0.20) 27.92 (0.44) 5.17 (0.77)
of MIB in larger fish compared with smaller ones. 12 28.08 (0.40) 3.95 (0.39) 27.81 (0.21) 4.55 (0.17)
The MIB concentration in fish-flesh of the treated tank ranged 13 27.27 (0.26) 4.68 (0.39) 27.28 (0.32) 4.75 (0.54)
14 27.02 (0.21) 4.52 (0.34) 27.35 (0.42) 4.45 (0.51)
from 0.06 to 6.00 ␮g kg−1 and in the control tank from 0.01 to
15 28.05 (0.31) 5.02 (0.38) 28.32 (0.33) 4.74 (0.80
7.00 ␮g kg−1 . The MIB accumulation of fish-flesh is seen to be 16 28.74 (0.31) 4.47 (0.51) 29.10 (0.21) 4.80 (0.57)
moderately exponential for fish in both tanks. The correlation coef- 17 28.02 (0.59) 4.78 (0.39) 28.35 (0.58) 5.24 (0.23)
ficients between the taint as MIB concentration in fish-flesh and the 18 27.50 (0.14) 5.02 (0.34) 27.84 (0.13) 4.61 (0.70)
mass of the fish of the treated and control tanks fish was R2 = 0.50 19 27.62 (0.16) 5.25 (0.36) 27.95 (0.33) 5.17 (0.34)
20 27.84 (0.19) 4.80 (0.73) 27.92 (0.34) 5.20 (1.12)
and R2 = 0.59, respectively. 21 27.11 (0.38) 5.01 (0.40) 26.88 (0.59) 5.45 (0.15)
The value of these correlation coefficients for both taint concen- 22 27.10 (0.26) 4.72 (0.12) 26.90 (0.26) 4.74 (0.81)
trations from the treated and control tanks is in the range described 23 27.38 (0.24) 5.08 (0.60) 27.07 (0.33) 5.10 (0.66)
generally as ‘moderate’. The values are however typical for (com- 24 28.05 (0.22) 5.01 (0.57) 27.78 (0.31) 5.04 (0.49)
25 27.92 (0.17) 4.77 (0.54) 27.47 (0.36) 5.38 (0.30)
plex) biological systems. Interestingly, it is not clear why the value
26 27.57 (0.17) 4.67 (0.64) 27.28 (0.25) 5.27 (0.39)
appears systematically lower for the control tank. 27 27.62 (0.12) 4.55 (0.69) 27.22 (0.18) 5.21 (0.27)
28 27.92 (0.14) 4.77 (0.43) 27.57 (0.56) 5.38 (0.22)
29 27.94 (0.26) 3.98 (0.44) 27.74 (0.37) 4.90 (0.55)
3.4. Lipid in fish-fillets 30 27.65 (0.43) 3.65 (0.45) 27.17 (0.24) 5.14 (0.40)
31 27.52 (0.14) 4.40 (0.54) 27.17 (0.24) 5.14 (0.40)
The variation of lipid concentration (as% wet basis) in differ- 32 27.44 (0.22) 5.21 (0.38) 26.98 (0.30) 5.75 (0.29)
ent fish tissues is shown in Fig. 5. The results are based on a 33 27.17 (0.28) 5.78 (0.28) 27.17 (0.12) 4.81 (0.31)
34 26.71 (0.03) 6.02 (0.12) 26.94 (0.07) 5.14 (0.55)
medium-sized fish of mass of 0.450 kg i.e. 180 days growth, to be
35 26.70 (0.29) 5.80 (0.30) 27.00 (0.20) 5.02 (0.47)
representative overall of RAS growth. The tissue areas taken for
lipid analyses were: front, dorsal, lateral, belly and tail.
As can be seen in Fig. 5, the greatest lipid concentration was in 3.5. Dissolved oxygen (COX ) and bulk growth temperature
the belly (8.75% w/w), whereas the lowest (3.42% w/w), was in the
tail. The calculated average concentration for all regions was found The dissolved oxygen (COX ) concentration and the bulk temper-
to be 6% w/w. This compared with that for fingerlings, mass range ature of the RAS growth water of the two (2) tanks over the 35 week
from 0.015 to 0.030, kg i.e. 30 days growth, of 5.8% w/w. This dif- period are given in Table 1.
ference is not thought meaningful, suggesting lipid concentration The data are presented as the average value of 7-day readings.
did not vary significantly with growth time. The dissolved oxygen concentration in the treated tank can be seen
 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27 23

to have varied from 3.95 to 6.02 mg L−1 and in the control tank from Table 2
Example model simulation showing inputs, calculations and output for predicted
4.09 to 5.75 mg L−1 .
taste taint (y) as both GSM and MIB in barramundi fish-flesh from an average fish
The growth water temperature of the treated tank varied from (0.435 kg) from the commercial-scale RAS studies.
26.7 to 28.7 ◦ C and that in the control from 26.3 to 29.1 ◦ C.
Parameter and units Taint taste as
In summary, the measured average dissolved oxygen concen-
tration in the treated and control tanks over the experimental GSM MIB
period was 5.00 and 4.90 mg L−1 , respectively. The measured aver- Row
age growth water temperature of both tanks was 27 ◦ C, for the Inputs
commercial growth of 35 weeks. 1 t day 169 169
2 kOW dimensionless 3.57 3.31
3 km day−1 0.00063 0.00063
4. Discussion 4 CW ␮g L−1 0.0121 0.0548
◦
5 T C 27 27
4.1. Estimating taint parameters 6 e dimensionless 0.1 0.1

Calculations
Substitution of the growth constants found for the barramundi 7 COX mg L−1 5.04 5.04
8 GV L day−1 161.70 (Eq. (11)) 161.70 (Eq. (11))
(Figs. 1 and 2) into Eq. (3) gives for fish in the treated (2B) tank
9 EW dimensionless 0.528 (Eq. (10)) 0.519 (Eq. (10))
10 k1 L kg−1 day−1 196.50 (Eq. (9)) 193.00 (Eq. (9))
mf = 0.0226exp0.0169t (19)
11 QW day−1 53.58 (Eq. (14)) 53.58 (Eq. (14))
12 QL day−1 0.53 (Eq. (15)) 0.53 (Eq. (15))
and for fish in the control (4D) tank
13 VL kg 0.026 (Eq. (16)) 0.026 (Eq. (16))
14 k2 day−1 0.538 (Eq. (13)) 0.959 (Eq. (13))
mf = 0.0229exp0.0161t (20)
15 kG day−1 0.0029 (Eq. (18)) 0.0029 (Eq. (18))
16 kg day−1 0.0036 (Eq. (17)) 0.0036 (Eq. (17))
It is concluded that these are not meaningfully different. The
17 mf kg 0.435 (Eq. (19)) 0.435 (Eq. (19))
implication, importantly, is that treatment of the RAS water with 18 a day−1 0.5417 (Eq. (5)) 0.9621 (Eq. (5))
hydrogen peroxide at 2.5 mg L−1 did not significantly influence typ- 19 b ␮g kg−1 day−1 0.0544 (Eq. (4)) 0.2422 (Eq. (4))
ical commercial-growth of the fish. 20 γ dimensionless 0.0161 (Eq. (19)) 0.0161 (Eq. (19))
The bulk temperature of the growth water over the growth Output
period was 27 ± 2 ◦ C. This value indicates that the measures taken 21 y ␮g kg−1 1.3876 (Eq. (8)) 3.0057 (Eq. (8))
to regulate the temperature of the commercial RAS growth water
were successful. The experimental temperature values used for the
model validation are therefore highly appropriate.
In the model synthesis, the dissolved oxygen concentration of
the RAS growth water was assumed at 7.24 mg L−1 at 28 ◦ C. This
was based on the assumption of saturated conditions at the RAS
water bulk temperature and the equation of Neely (1979).
However, experimental data collected daily in the commercial-
scale trials showed that dissolved oxygen (COX ) concentration
varied over time. Therefore, in the model validation, daily experi-
mental values averaged weekly (Table 1) were used to compute gill
ventilation rates, GV (Eq. (11)).
The chemical uptake efficiencies (EW ) for GSM = 0.528 and
MIB = 0.519 remain constant and independent of RAS (see Eq. (10))
because these depend only on the octanol-water partition coef-
ficient (KOW ). The uptake rate constant (k1 ) was calculated by
substituting for EW, GV and mf values into Eq. (9).
Eqs.(13)–(16) were used sequentially to compute k2 . Eq. (18)
was used to compute the input kG . Similarly, Eq. (17) was used to
calculate kg .
The model was set-up as a Microsoft ExcelTM spread sheet.
Model input parameters CW for GSM and MIB were taken from the
weekly growth water analyses over the 35 week period. Calculated
model inputs derived from the experimental data k1 , k2 , kg were
fed to the spread sheet together with growth time (t).

4.2. Example model validation
An overall example simulation of the new model with the exper-
imentally determined data from the commercial-scale RAS studies
with barramundi is conveniently summarised as Table 2.
An (average) fish (0.435 kg) from the commercial-scale control
(4D) tank is used to demonstrate the model. The input parameters
t, kOW , km , CW , T and e are shown in rows 1–6 of Table 2. The calcu-
lated parameters COX , GV , EW , k1 , QW , QL , VL , k2 , kG , kg , mf , a, b and γ
are shown in rows 7–20. Presented this way the model can be read
logically down columns 4 and 5, respectively, for GSM and MIB. The
output as predicted level of taint, y, is shown in row 21, respectively
Fig. 6. Predicted vs observed taste taint (y) as GSM in fish-flesh from RAS ‘treated’
(2B) tank. Results are presented for n = 167 data from 121 fish. The number of fish
harvested in the 30,000 L was 3326. The box (LL) indicates those data below the
consumer rejection threshold for GSM (0.74 ␮g kg−1 ).
columns 4 and 5 for GSM and MIB. This tabulated presentation pro-
vides a convenient (and prudent) check on the consistency of units
used in the model.
A comparison between model simulations for taste taint as GSM
and experimentally determined data is presented in Figs. 6 and 7,
respectively for fish from the treated (2B) and control (4D) RAS
24 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27
Fig. 7. Predicted vs observed taste taint (y) as GSM in fish-flesh from RAS ‘control’
(4D) tank. Results are presented for n = 186 data from 145 fish. The number of fish
harvested in the 30,000 L was 3318. The box (LL) indicates those data below the
consumer rejection threshold for GSM (0.74 ␮g kg−1 ).
Fig. 8. Predicted vs observed taste taint (y) as MIB in fish-flesh from RAS ‘treated’
(2B) tank. Results are presented for n = 167 data from 121 fish. The number of
fish harvested in 30,000 L was 3326. The box (LL) indicates those data below the
consumer rejection threshold for MIB (0.7 ␮g kg−1 ).
tanks. Fig. 6 summarizes GSM in fish-flesh for n = 167 data (from
121 fish) and Fig. 7 for n = 186 data (from 145 fish).
Similarly, a comparison between model simulations for taste
taint as MIB and experimentally determined data is presented in
Figs. 8 and 9, respectively, for fish from the treated (2B) and control
(4D) RAS tanks. Fig. 8 summarizes MIB in fish-flesh for n = 167 data
(from 121 fish) and Fig. 9 for n = 186 data (from 145 fish).
Fig. 9. Predicted vs observed taste taint (y) as MIB in fish-flesh from RAS ‘con-
trol’ (4D) tank. Results are presented for n = 186 data from 145 fish. The number
of fish harvested in 30,000 L was 3318. The box (LL) indicates those data below the
consumer rejection threshold for MIB (0.7 ␮g kg−1 ).
More than one sample were obtained and analysed from a single
fish whenever possible to improve data. Smaller fish (<0.030 kg)
generally however provided only one sample due to reduced mass.
Importantly, all experimental data that were determined are
presented in Figs. 6–9, that is, no outliers have been removed, nor
has the data been mathematically ‘smoothed’ in any way. More-
over, It is can be readily determined that the total number of
simulations used for model validation
 (N) is a sum of each of all
of Figs. 6–9 to give N = 706 (i.e. n = 167 + 186 + 167 + 186). This
number is considered more than sufficient for the purpose of model
validation.
4.3. Comparison of predictions with observed data
Overall, it is satisfying that the predicted and experimentally
determined data for taste taint as GSM and MIB (Figs. 6 through 9)
covers the same range of 0–12 ␮g kg−1 .
Importantly, for the data of Figs. 6 and 7 (treated and control
tanks) the range of observed taint as GSM can be seen to be nearly
10 fold that of the consumer rejection threshold (i.e. from Fig. 6,
GSM range ∼6.6/0.7 ␮g kg−1 = 9.4). From the data of Figs. 8 and 9
(treated and control tank) the range of observed taint as MIB can
be seen to be ∼10 fold that of the consumer rejection threshold also
(i.e. for Fig. 8, GSM range ∼6/0.7 ␮g kg−1 = 8.6).
However, the area of interest is fish-flesh with taste taint chem-
ical below the consumer rejection threshold. To clearly visualize
these data an expanded view that covers the range 0 ≤ y ≤ 2 ␮g kg−1
from, respectively, Figs. 6 through 9 are presented in Figs. 10–13.
The consumer rejection threshold concentration (∼0.7 ␮g kg−1 ) is
bounded (as a solid line) in the LL of each figure. The remaining
shaded area of the figures indicates that fish-flesh concentration
exceeded the consumer rejection threshold.
Figs. 10 and 11 show, respectively, the predicted versus
observed concentration of GSM for fish from treated (2B) and con-
trol (4D) RAS tanks. Fig. 10 summarizes GSM in fish-flesh for n = 137
data (from 80 fish) and Fig. 11 for n = 160 data (from 121 fish).
 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27
Fig. 10. Expanded area of Fig. 6 for taste taint concentration (y) (covering
0–2 ␮g kg−1 ) as GSM in fish-flesh from RAS ‘treated’ (2B) tank. Results are for n = 137
data from 80 fish. The shaded area shows fish-flesh exceeding the consumer rejec-
tion threshold for GSM (0.74 ␮g kg−1 ).
Fig. 11. Expanded area of Fig. 7 for taste taint concentration (y) (covering
0–2 ␮g kg−1 ) as GSM in fish-flesh from RAS ‘control’ (4D) tank. Results are for n = 160
data from 121 fish. The shaded area shows fish-flesh exceeding consumer rejection
threshold for GSM (0.74 ␮g kg−1 ).
Similarly, Figs. 12 and 13 show, respectively, the predicted
versus observed concentration of MIB for fish from the treated (2B)
and control (4D) RAS tanks. Fig. 12 summarizes MIB in fish-flesh
for n = 111 data (from 85 fish) and Fig. 13 for n = 107 data (from 91
fish).
It can be seen from the data of Figs. 10 through 13 that model
predictions for both GSM and MIB are more or less evenly dis-
tributed about the Y = X line. Especially, those data below the
consumer rejection threshold concentration, LL of the figures. It is
25
Fig. 12. Expanded area of Fig. 8 for taste taint concentrations (y) (covering
0–2 ␮g kg−1 ) for as MIB in fish-flesh from RAS ‘treated’ (2B) tank. Results are for
n = 111 data from 85 fish. The shaded area shows fish-flesh exceeding the consumer
rejection threshold for MIB (0.7 ␮g kg−1 ).
Fig. 13. Expanded area of Fig. 9 for taste taint concentration (y) (covering
0–2 ␮g kg−1 ) as MIB in fish-flesh from RAS ‘control’ (4D) tank. Results are for n = 107
data from 91 fish. The shaded area shows fish-flesh exceeding the consumer rejec-
tion threshold for MIB (0.7 ␮g kg−1 ).
believed this finding augurs well for development of possible proto-
cols to help farmers control taste taint in fish-flesh of the harvested
fish.
The accuracy of model predictions irrespective however of range
was investigated. The data which deviated from Y = X line were
counted as a percentage of the total and summarised. The findings
are presented as Table 3.
All data in Figs. 6–9, and four (4) criteria were used in the inves-
tigation. These criteria were those data points that: 1. fell on the
26 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27
Table 3
Comparison of deviations from observed data versus model predictions about the
Y = X line for GSM and MIB, treated and controls, presented as a percentage of all
data (N = 706) from Figs. 6–9.
Percentage of data
Deviation Treated Control
Criteria GSM MIB GSM MIB
±<10% 21 18 24 28
± 10–20% 17 18 15 18
>20% above Y = X 47 56 50 45
>20% below Y = X 16 8 11 8 10.8
Y = X line or deviated <±10%, 2. deviated ±10–20%, 3. those that
deviated >+20% above the Y = X line, and; those that deviated
>−20% below the Y = X line. For e.g. from the table those data
that fell on the Y = X line or deviated ±<10% are summarised in row
1 of Table 3.
These tabulated data provide significant insight, namely, that
some 39.8% (=22.8 + 17.0%) of model predictions will be within
±20% of the observed value. However, some 49.5% are seen to be
consistently greater than the observed value for taste taint chemi-
cals as GSM and MIB. That is, in about a half of cases the actual taste
taint concentration will in fact be lower than that predicted. This
provides a welcomed level of conservatism in predictions.
4.4. Explaining variance in predictions
One reason for the difference in predicted versus observed taint
concentration in the RAS fish-flesh is that the model in its present
form uses a mean value for the (exponential) growth constant
for the fish of  = 0.0161 day−1 (Table 2); this almost certainly
changes with growth-time with larger fish growing more quickly
than smaller. This difference in growth constant with fish mass can
be gleaned from Fig. 1 with some two (2) slopes evident, approxi-
mately, 0–140 day and 140–240 day, respectively.
Further, RAS is oscillatory in taste taint chemicals, despite mea-
sures taken to control the aqueous growth environment. These
changes impact model predictions as the model makes predictions
based on an ‘equilibrium’ taste taint chemical concentration. There
is no evidence however that there is chemical equilibrium between
the matrices of water and fish-flesh. For example. the average GSM
concentration in growth water of the treated (2B) tank in week 8
was 0.027 ␮g L−1 and this concentration in the growth water fell to
as low as 0.002 ␮g L−1 by week 9. The predicted GSM concentration
in fish-flesh of treated (2B) tank fish at that time was 0.04 ␮g kg−1
but the measured concentration was 0.15 ␮g kg−1 .
A potential drawback of the predictive model of Hathurusingha
and Davey (2014) (and others) is therefore that it is based on ‘aver-
age’ or ‘mean’ deterministic values for input. There are however
these naturally occurring, oscillatory fluctuations in parameters −
these could significantly impact outcomes under particular condi-
tions. To estimate the possible significance of the impact of these
fluctuations in concentration of taste taint chemicals in the RAS
water on accumulation in fish-flesh Hathurusingha and Davey
(2016) applied the emerging risk methodology of Davey and co-
workers (see for e.g. Chandrakash et al., 2015; Abdul-Halim and
Davey, 2016) and found that some 10.10% of all RAS barramundi
harvests over the long term will result in taint as GSM above the
threshold in the flesh of the harvested fish (and 10.56% for MIB).
Further, it was found that the time to harvest impacted most sig-
nificantly, and not the extent of concentration change of the taste
chemicals in the RAS water. This finding was important as the time
to harvest is readily controlled by the fish farmers – thereby min-
imising the accumulative impact of the oscillatory affects.
Because validation of the model predictions is based directly on
chemical analyses it is important to obtain representative samples
of fish-flesh from particular fish. This is because GSM and MIB are
not homogeneously distributed in the fish-flesh but are based on
Mean
lipid concentrations (Percival et al., 2008). The lipid-rich matrix can
make extraction tedious and become a potential source of error.
4.5. Modifying a future model
22.8
17.0
49.5 This research is the first reported work to validate experimen-
tally a time-dependent model for predicting the accumulation
of taste taint chemicals as GSM and MIB in fish-flesh from
commercial-scale RAS operations. The data are extensive and cover
the entire growth from fingerlings to harvested adult fish. However,
despite a very good correspondence between model prediction and
the extensive experimentally determined data, the model in its
present form tends to over-predict taste taint chemical concen-
tration accumulation in fish-flesh above the consumer threshold
concentration.
It is proposed, an expedient modification of the model could
be introduction of an empirical ‘correction factor’ to improve pre-
diction for the period approaching harvest. However, it is more
theoretically satisfying to include in a revised model two growth
constants, one for smaller (juvenile) fish and one for the larger fish
as they approach harvest. Significant additional validation studies
will be required however.
This theoretical development is currently being planned.
5. Conclusions
The model predictions of Hathurusingha and Davey (2014) and
commercially-derived experimental data exhibited the same range
of concentration for both GSM and MIB taste taint chemicals in the
flesh (0–12 ␮g kg−1 ) of RAS farmed barramundi.
Importantly, the model gave a very good fit to data for low taste
taint concentration (0–2 ␮g kg−1 ), this range covers the consumer
rejection threshold (∼0.7 ␮g kg−1 ).
In general, model predictions for GSM showed better agreement
with the experimental values than those for MIB at low values
(<2 ␮g kg−1 ). This is attributed to a lower variation of GSM con-
centration in the RAS growth water with time.
The model predicts greater taste taint chemical concentration
for both GSM and MIB than were determined experimentally. Pre-
dictions are therefore conservative for both chemicals and can be
thought of as ‘safe’. A major reason for this is attributed to dissim-
ilar (exponential) growth constants () for smaller and larger fish,
and less, to the fact that RAS environment is naturally oscillatory
in concentration of the taste taint chemicals.
The findings of this research could be used to develop quanti-
tative protocols to limit taste taint in fish-flesh of harvested fish in
RAS farming.
Acknowledgments
The authors are indebted to the CRC Seafood, Australia for finan-
cial assistance to carry out the work and to BarraFresh, Gawler,
Australia, for generously providing commercial facilities for the
experimental study, and help and advice during the work. P.I. H. is
grateful to The University of Adelaide and the CRC Seafood, Australia,
for provision of a research scholarship and financial assistance to
carry out this research.
References
Abdul-Halim, N., Davey, K.R., 2016. Impact of suspended solids on Fr 13 failure of
UV irradiation for inactivation of Escherichia coli in potable water production
 P.I. Hathurusingha, K.R. Davey / Ecological Modelling 340 (2016) 17–27
with turbulent flow in an annular reactor. Chem. Eng. Sci. 143, 55–62, http://
dx.doi.org/10.1016/j.ces.2015.12.017.
Arnot, J.A., Gobas, F.A.P.C., 2004. A food web bioaccumulation model for organic
chemicals in aquatic ecosystems. Environ. Toxicol. Chem. 23 (10), 2343–2355,
http://dx.doi.org/10.1897/03-438.
Brett, K.L., Johnsen, P.B., 1996. Challenges of evaluating sensory attributes in the
presence of off-flavours (geosmin and 2- methylisoborneol). J. Sens. Stud. 11,
1–17.
Chandrakash, S., Davey, K.R., O’Neill, B.K., 2015. An Fr 13 risk analysis of failure in a
global food process – illustration with milk processing. Asia-Pac. J. Chem. Eng.
10 (4), 526–541, http://dx.doi.org/10.1002/apj.1887.
Davey, K.R., Hathurusingha, P.I., 2014. A new transient predictive model to
quantify taint as either geosmin (GSM) and 2-methylisoborneol (MIB) in
Rainbow Trout (Oncorhynchus mykiss) farmed in Recirculating Aquaculture
Systems (RAS). In: Proc. 26th European Modeling and Simulation Symposium
(EMSS), Bordeaux, France, Sept. 10–12, pp. 490–497 (ISBN 978-88-9799-38-6).
Glenn, S., Bosmans, J., Humphrey, J., 2007. Northern Territory Barramundi Farming
Handbook. Department of Primary Industries and Mines, Australia, pp. 38,
(viewed 06.03.15) http://www.nt.gov.au/d/Primary Industry/.
Gobas, F.A.P.C., Mackay, D., 1987. Dynamic of Hydrophobic (sic) organic chemical
bio- concentration in fish. J. Environ. Toxicol. Chem. 6, 495–504.
Gobas, F.A.P.C., 1993. A model for predicting the bio-concentration of hydrophobic
organic chemicals in aquatic food-webs: application to Lake Ontario. Ecol.
Model. 69, 1–17.
Gobas, F.A.P.C., 1998. Bioaccumulation of Hydrophobic Organic Chemicals in Fish
(PhD Thesis). University of Toronto, ON, Canada.
Grimm, C.C., Lloyd, S.W., Zimba, P.V., 2004. Instrumental versus sensory detection
of off-flavors in farm-raised channel catfish. Aquaculture 236, 309–319, http://
dx.doi.org/10.1016/j.aquaculture.2004.02.020.
Hathurusingha, P.I., Davey, K.R., 2013. A new transient-state model for quantitative
prediction of taint in farmed barramundi fish (Lates calcarifer). In: Proc. 34th
International Conference on Marine Science and Aquaculture–ICMSA 2013,
Amsterdam, The Netherlands, May 15–16, pp. 380–382 (eISBN 2010–3778).
Hathurusingha, P.I., Davey, K.R., 2014. A predictive model for taste taint
accumulation in Recirculating Aquaculture Systems (RAS) farmed-fish –
demonstrated with geosmin (GSM) and 2-methylisoborneol (MIB). Ecol.
Model. 291, 242–249, http://dx.doi.org/10.1016/j.ecolmodel.2014.08.009.
Hathurusingha, P.I., Davey, K.R., 2016. Chemical taste taint accumulation in RAS
farmed fish – a Fr 13 risk assessment demonstrated with geosmin (GSM) and
2-methylisoborneol (MIB) in barramundi (Lates calcarifer). Food Control 60,
309–319, http://dx.doi.org/10.1016/j.foodcont.2015.08.014.
Hathurusingha, P.I., 2015. Predictive Modelling and Experimental Studies on
Taste-taint as Geosmin (GSM) and 2-methylisoborneol (MIB) in Farmed
Barramundi (Lates Calcarifer). (PhD Thesis), Chpt. 4. The University of Adelaide,
Australia, pp. 253, Appendix D.
27
Howgate, P., 2004. Tainting of farmed fish by geosmin and 2-methylisoborneol: a
review of sensory aspects and of uptake/depuration. Aquaculture 234,
155–181, http://dx.doi.org/10.1016/j.aquaculture.2003.09.032.
Johnsen, P.B., Brett, K.L., 1996. Sensory evaluation of the off-flavors geosmin and
2-methyliosborneol (MIB) in farmed-raised catfish. J. Appl. Aquacult. 6, 21–37.
Lloyd, S.W., Grimm, C.C., 1999. Analysis of 2-methyliosborneol and gesomin in
catfish by microwave -distillation solid-phase micro-extraction. J. Agric. Food
Chem. 47, 164–169, http://dx.doi.org/10.1021/jf980419x.
Neely, W.B., 1979. Estimating rate constant for the uptake and clearance of
chemicals by fish. J. Environ. Technol. 13, 1506–1510.
Palmeri, G., Turchini, G.M., Caprino, F., Keast, R., Moretti, V.M., Silva, S.S.De., 2008.
Biometric, nutritional and sensory changes in intensity farmed Murray cod
(Maccullochella peelii, Mitchell) following different purging times. Food Chem.
107, 1605–1615, http://dx.doi.org/10.1016/j.foodchem.2007.09.079.
Percival, S., Drabsch, P., Glencross, B., 2008. Determining factors affecting muddy
flavour taint in farmed barramundi (Lates calcarifer). Aquaculture 284,
136–143, http://dx.doi.org/10.1016/j.aquaculture.2008.07.056.
Persson, P.E., York, R.K., 1978. Notes on muddy odour: 1. Sensory properties and
analysis of 2-methylisoborneol in water and fish. J. Aqua Fenn. 8, 83–88.
Persson, P.E., 1980. Sensory properties and analysis of two muddy odour
compounds geosmin and 2-methylisoborneols in water and fish. Water Res.
14, 1113–1118.
Robertson, R.F., Jauncey, K., Beveridge, M.C.M., Lawton, L.A., 2005. Depuration rates
and the sensory threshold of geosmin responsible for earthy-musty taint in
rainbow trout, Oncorhynchus mykiss. Aquaculture 245, 89–99, http://dx.doi.
org/10.1016/j.aquaculture.2004.11.045.
Schrader, K.K., Davidson, J.W., Rimando, A.M., Summerfelt, S.T., 2010. Evaluation of
ozonation on levels of the off-flavor compounds geosmin and
2-methylisoborneol in water and rainbow trout Oncorhynchus mykiss from
recirculating aquaculture systems. Aquacult. Eng. 43, 46–50, http://dx.doi.org/
10.1016/j.aquaeng.2010.05.003.
Snedecor, G.W., Cochran, W.G., 1989. Statistical Methods, eighth ed. Iowa State
University Press, Ames, lowa, USA, ISBN 0-8138-1561-6.
Thomann, R.V., Connolly, J.P., Parkerton, T., 1992. Modelling accumulation of
organic chemicals in aquatic food-webs. In: Gobas, F.A.P.C., McCorquodale, J.A.
(Eds.), Chemical Dynamics in Aquatic Ecosystems. Lewis Publishers, Chelsea,
USA, pp. 153–186.
Watson, S.B., Brownlee, B., Satchwill, T., Hargesheimer, E.E., 2000. Quantitative
analysis of trace levels of geosmin and MIB in source and drinking water using
headspace SPME. Water Res. 34 (10), 2818–2828, http://dx.doi.org/10.1016/
S0043-1354(00)00027-0.