Professional Documents
Culture Documents
1 s2.0 S0144860915300376 Main
1 s2.0 S0144860915300376 Main
1 s2.0 S0144860915300376 Main
Aquacultural Engineering
journal homepage: www.elsevier.com/locate/aqua-online
a r t i c l e i n f o a b s t r a c t
Article history: Partial harvesting is an alternative for managing cash flow of aquaculture farms. We use deterministic and
Received 7 July 2015 stochastic models to analyze zootechnical, water quality and management factors influencing intensive
Received in revised form production of shrimp when incorporating partial harvesting strategies. Data from a commercial farm in
24 September 2015
the State of Nayarit, Mexico were used for modeling. The main factors affecting shrimp production and
Accepted 30 November 2015
its variability were: final weight and growth rate of shrimp, water temperature, pond size, length of daily
Available online 4 December 2015
aeration, and the time when the first partial harvest is conducted. Using the largest pond size (4.0 ha),
minimum length of aeration (7.5 h), and first harvesting at 8.5 weeks resulted in a minimum total harvest
Keywords:
Partial harvesting
of 2690 kg ha−1 (partial and final harvests of 643, 269, 1075, and 703 kg ha−1 ). Using the smallest pond size
Shrimp (1.0 ha), maximum length of aeration (7.9 h), and first harvesting at 11.5 weeks resulted in a maximum
Modeling production total harvest of 3524 kg ha−1 (partial and final harvests of 1111, 234, 997, and 1182 kg ha−1 ). The increase
in shrimp production from improved management was 31%. The stochastic model showed that increasing
farm size from 1 to 40 ha diminished the variability of shrimp production by 84.0%, meaning a reduction
of 2.2% per hectare as size increased. Sensitivity analysis indicated that, overall, final weight of shrimp
and length of aeration are the most important factors determining production. The models can be used
to determine, in future research, the optimum harvesting strategy, using a bioeconomic approach.
© 2015 Elsevier B.V. All rights reserved.
1. Introduction Despite the relevance that this practice entails from an eco-
nomic perspective, the number of studies analyzing the advantages
Cash flow problems lead to more aquaculture business failures and management of partial harvesting is relatively scarce. Forsberg
than any other problem (Engle, 2010). Partial harvesting is a strat- (1999), using a bioeconomic approach, determined that it is more
egy for managing cash flow of aquaculture farms. Selling off part profitable to size-grade salmon prior to harvest compared to har-
of the inventory reduces stocking densities on the farm, resulting vesting a batch of fish with similar size distribution to that of the
in faster growth of the remaining fish and greater turnover of the standing stock. Brummett (2002) compared three typical partial
crop. Revenue from selling off a portion of the crop and the higher harvesting systems and an unharvested control for tilapia in terms
turnover of the crop often improves cash flow and reduces cash of gross yield and observed that significantly higher yields were
deficits (Engle, 2010). obtained in ponds that were partly harvested by hook and line. Yu
and Leung (2006) used impulsive control theory to develop a par-
tial harvesting model capable of addressing discrete and another
partial harvesting strategies for shrimp. Yu et al. (2009) developed
∗ Corresponding author. Tel.: +52 612 123 8416; fax: +52 612 125 3625. a model of partial harvesting of shrimp, using the network-flow
E-mail address: ahllamas04@cibnor.mx (A. Hernández-Llamas). approach.
http://dx.doi.org/10.1016/j.aquaeng.2015.11.003
0144-8609/© 2015 Elsevier B.V. All rights reserved.
A. Estrada-Pérez et al. / Aquacultural Engineering 70 (2016) 56–62 57
Table 1
Mean, minimum and maximum values of the variables in the database used for analysis, together with dates
and means weight of shrimp at each partial harvest.
and from 1 (Zar, 2010). After obtaining a not significant result, an equiv-
alence test was conducted to protect against falsely accepting the
F2 (t) = N0,2 exp (−z2 t) (5) corresponding null hypothesis (Type-II statistical error; Hauck and
where N0,1 , N0,2 , z1 and z2 are the initial populations and the mor- Anderson, 1986). For these tests, procedures in Stata 10 were used,
tality rates corresponding to the first and second phases. setting significance at P < 0.05 and tolerance error at 10%.
The following curve is introduced to link both survival functions:
2.5. Management strategies
H (t) = F1 (t) + G (t) × (F2 (t) − F1 (t)) (6)
where H(t) is the transition function used to simulate shrimp Alternative management strategies were defined using combi-
biomass collected at each partial harvest, and F1 (t), F2 (t) are, as nations of management variables that minimized (worst strategy)
defined above (Eqs. (4) and (5)), and G(t) is: and maximized (best strategy) total harvested biomass. Intermedi-
ate strategies were also defined and analyzed using intermediate
1 values of the management variables.
G (t) = ; S > 0 (7)
1 + exp (P − t) /S
2.6. Stochastic model
where G(t) is the link function, P is the time corresponding to
the central point of the transition between the phases (midpoint
Stochastic elements were incorporated into the stock model
between the beginning and the end of the partial harvest), and S
using the “envelope” method described in Vose (2002). For this,
is the shape factor of the transition curve. The link function G(t)
the deterministic value of Q in Eq. (8) was modified:
varies asymptotically between 0 and 1, and G(t) = 0.5 when t = P, so
that for t < P, the function resembles F1 (t) and for t > P, the curve is Qs = Q + r (9)
similar to F2 (t).
where Qs is the stochastic value of the parameter and r is a resid-
ual value calculated from a normal distribution fitted to the residual
2.3. Correlation and multiple regression analyses
values resulting from the corresponding multiple linear regression.
Probability distributions were also fitted to values of the water
A correlation analysis was conducted, using Statistica 6.0 (Stat-
quality variables that were significant in the regression analysis.
Soft, Tulsa, OK), to determine possible relationships between the
Monte Carlo simulations were conducted, using the proba-
parameters of the stock model and water quality and management
bility distributions of the residuals and water quality variables
variables, setting significance at P < 0.05. Complementarily, a mul-
as inputs. Output probability distributions of the total harvested
tiple regression analysis was conducted, using datasets of the 29
biomass were inferred from the simulations corresponding to the
cases, to calculate parameters of the stock model as a function of
alternative management strategies. The coefficient of variation
water quality and management variables. Accordingly:
(CV = standard deviation/mean) was used to compare the variabil-
Q = a0 + a1 T + a2 DO + a3 PS + a4 LA + a5 TP1 + a6 TP2 + a7 TP3 ity in production between management strategies. The effect of
farm size on the variability of production was also estimated for
+ a8 PH1 + a9 PH2 (8) farms operating 1 and 40 ponds (1 ha each pond). @Risk software
where Q is any of the parameters in the stock model (except wi , was used for Monte Carlo simulations.
n0 , and S), a0 , a1 ,. . ., a9 are regression coefficients, T is mean water
temperature measured during the cultivation period, DO is mean 2.7. Sensitivity analysis
dissolved oxygen measured during the cultivation period, PS is
pond size, LA is mean length of aeration used during the cultivation The relative importance that management variables have in
period, TP1, TP2, and TP3 are the times when the partial harvests are determining total shrimp production was calculated by sensitivity
conducted, and PH1, and PH2 are the percentages of the standing analysis, using the deterministic model, where
stock that was collected during the first and second partial harvests.
percentage of change in production
For each parameter of the stock model, the multiple regression AS = (10)
percentage of change in the variable
analysis was conducted testing only water quality and manage-
ment variables that previously resulted significant from correlation where AS is the absolute value of sensitivity attributable to the vari-
analysis. The percentage of the standing stock that was collected able, and the percentage of change in shrimp production and the
during the third partial harvests was not significant and was not management variable correspond to the changes calculated when
considered for regression analysis. The multiple linear regression passing from the worst management strategy to the best manage-
procedure in Stata 10 (StataCorp, College Station, TX) was used ment strategy. A high value of AS indicate high importance of the
with P < 0.1 to accept or reject independent variables. This regres- corresponding variable.
sion procedure automatically handles collinearity according to the The importance that the parameters of the stock model and
method described in Rencher (2002). water quality variables have on total harvested biomass was calcu-
lated by sensitivity analysis using the stochastic model. Regression
2.4. Model fitting procedures in @Risk were used for this analysis, where a high abso-
lute value of a regression coefficient indicates high importance of
The fitting of the stock model and the multiple regression equa- the corresponding variable.
tions to the total harvested biomass as registered (“observed”)
in the data base was tested as follows. Total harvested biomass 3. Results
was calculated for each case using the regression equations and
the stock model and compared with the corresponding observed The growth and survival models showed flexibility to fit the
total yield in the database. A linear regression analysis between growth and survival data in the database (Fig. 1a and b). Correla-
observed and calculated biomass was then performed, setting the tion analysis showed that the parameters of the stock model were
intercept at zero (Poole, 1974), and the Student’s t-test was applied significantly correlated with water quality and management vari-
to determine whether the regression slope differed significantly ables (Table 2). Final weight of shrimp was positively correlated
A. Estrada-Pérez et al. / Aquacultural Engineering 70 (2016) 56–62 59
Table 3
Equations resulting from multiple regression analysis to calculate parameters
of the stock model as a function of water quality and management variables.
Equation P
Fig. 2. Results from the equivalence test used for fitting the stock model and the
multiple regression equations.
Fig. 1. Sample-fitted growth curve using Eq. (2) (a), and sample-fitted survival curve
using Eqs. (4)–(7) (b).
Results from multiple regression analyses showed that, with the
exception of z2 , the parameters of the stock model were signifi-
with water temperature and length of aeration, and inversely cor- cantly related to water quality and management variables (Table 3).
related with the time of the second and third partial harvests. The Direct or inverse relationships between the parameters and water
growth coefficient was negatively correlated with dissolved oxygen quality and management variables were established, coinciding
content and positively correlated with pond size, length of aeration, with the results from the correlation analysis, although not all of
the week of the first partial harvest, and the biomass collected in the the variables appearing as significant in the correlation test were
first harvest. The mortality rate during the first phase was positively necessary to satisfactorily calculate parameter values using the
correlated with pond size and inversely correlated with length of regression equations. The fitting of the stock model, using these
aeration. The mortality rate during the second phase was inversely equations, proved satisfactory as indicated by the confidence inter-
correlated with pond size and the biomass collected during the first val of the corresponding equivalence test (−0.048, 0.011; within
harvest. Mortality rate during the third phase was negatively corre- −0.05, 0.05; Fig. 2).
lated with temperature and dissolved oxygen content, and directly The stock model was used to calculate, as a function of time,
correlated with pond size and the weeks of the second and third partial and final harvests yielding the minimum and maximum
harvests. Mortality during the last phase was negatively correlated total harvested biomass of shrimp. Minimum total biomass was
with pond size and the weeks of the three partial harvests. calculated using the largest pond size (4.0 ha), the minimum length
Table 2
Correlations between parameters of the stock model and water quality and management variables.
Fig. 3. Biomass calculated for the worst and best management strategies using the
stock model as a function of cultivation time.
of aeration (7.5 h), and having the first partial harvest at 8.5
weeks (the worst management strategy). Total biomass calcu-
lated was 2690 kg ha−1 , resulting from successive partial harvests
of 643, 269, and 1075 kg ha−1 and a final harvest of 703 kg ha−1
(Fig. 3a). Maximum total biomass was calculated using the smallest
ponds (1.0 ha), longest aeration (7.9 h), and having the first par-
tial harvests at 11.5 weeks (the best management strategy). Total
calculated biomass was 3524 kg ha−1 , resulting from successive
partial harvests of 1111, 234, and 997 kg ha−1 and a final harvest
of 1182 kg ha−1 (Fig. 3b). Based on these figures, the increase in
total harvested biomass derived from improved management was
31%. From results obtained with the worst and best management
strategies, intermediate strategies were defined (Table 4).
Sensitivity analysis, using Eq. (10) indicated that changing the
Fig. 4. Output probability distributions of shrimp production for a 1 ha pond. (a)
values of the management variables as described above, resulted in Worst management strategy; (b) best management strategy.
increases in production of 70.0 kg ha−1 (pond size), 235.0 kg ha−1
(length of aeration), and 385.0 kg ha−1 (week of first partial har-
vest), representing increases of 2.6, 8.7, and 14.3%, respectively.
Yet, the corresponding changes in the values of the variables were
75.0, 5.3, and 35.3%, resulting in absolute values of sensitivity of
0.03, 1.63, and 0.4, indicating that length of aeration is the most
important management variable.
Results from the Monte Carlo simulation showed that improving
management resulted in increased shrimp production, as reflected
in higher mean production obtained with the best management
strategy, but also in relatively lower variability of production, as
indicated by lower values of the coefficient of variation (Figs. 4a
and b). The simulation also showed that improving management
and increasing farm size diminished the variability of production,
as indicated by progressively lower values of the coefficient of
variation (Fig. 5). Improving management diminished variability
by 8.0–10.0%, while increasing farm size diminished variability by
84.0%.
Strategy PS (ha) AT (h day−1 ) TH1 (weeks) Results from sensitivity analysis (Table 5) showed that, regard-
N1 (worst) 4 7.5 8.5
less of the management strategies, the stochastic elements
N2 3 7.63 9.5 associated with growth parameters contributed more importantly
N3 2 7.76 10.5 to total variability of shrimp production, followed by water temper-
N4 (best) 1 7.9 11.5 ature. Mortality related to the successive phases of the cultivation
PS, pond size; AT, length of aeration; TH1, time of first partial harvest. process were less important.
A. Estrada-Pérez et al. / Aquacultural Engineering 70 (2016) 56–62 61
The production models developed in this study can be used to Hernández-Llamas, A., Villarreal-Colmenares, H., 1999. TEMA: a software reference
determine, in future research and using bio-economic analysis, the to shrimp Litopenaeus vannamei farming practices. Aquacult. Econ. Manage. 3
(3), 267–280.
best management strategy. For such analysis, it must be taken into Hernandez-Llamas, A., Zarain-Herzberg, M., 2011. Bioeconomic modeling and risk
account seasonality of shrimp prices, and investment and produc- analysis of raising shrimp Litopenaeus vannamei in floating cages in
tion costs associated with alternative management strategies of northwestern Mexico: assessment of hurricane hazard, stochastic variability of
shrimp and feed prices, and zootechnical parameters. Aquaculture 314,
pond size, length of culture period, length of aeration, and the time 261–268.
when the first partial harvest is conducted. Hernández-Llamas, A., Ruiz-Velazco, J.M.J., Gomez-Muñoz, V.M., 2013. Economic
risk associated with white spot disease and stochastic variability in economic,
zootechnical and water quality parameters for intensive production of
Acknowledgements Litopenaeus vannamei. Rev. Aquacult. 5, 121–131.
Magallon, F.J., 2006. Desarrollo y aplicación de una metodología, para evaluar la
variabilidad de la capacidad de carga de la acuicultura de camarón, en la región
The authors are grateful to the shrimp farmers in the State of del Golfo de California. Universidad de La Habana, La Habana, Cuba, Doctoral
Nayarit for their cooperation. Ira Fogel of CIBNOR provided valuable Thesis.
editorial services. A.E.P. is a recipient of a student fellowship from Milstein, A., Islam, M.S., Wahab, M.A., Kamal, A.H.M., 2005. Characterization of
water quality in shrimp ponds of different sizes and with different
Consejo Nacional de Ciencia y Tecnología of Mexico. management regimes using multivariate statistical analysis. Aquacult. Int. 13,
501–518.
Poole, R.W., 1974. An Introduction to Quantitative Ecology. McGraw-Hill, New
References York, NY.
Quinn, G.P., Keough, M.J., 2002. Experimental Design and Data Analysis for
Brummett, R.E., 2002. Comparison of African tilapia partial harvesting systems. Biologists. Cambridge University Press, London.
Aquaculture 214, 103–114. Rencher, A.A., 2002. Methods of Multivariate Analysis. John Wiley & Sons, USA.
Engle, C., 2010. Aquaculture Economics and Financing. Willey-Blackwell, Ames, IA. Ruiz-Velazco, J.M.J., Hernández-Llamas, A., Gómez-Muñoz, V.M., Magallón, F.J.,
FIRA, 2009. Situación Actual y Perspectivas del Camarón en México. Banco de 2010. Dynamics of intensive production of shrimp Litopenaeus vannamei
Mexico, 2009 (3). affected by white spot disease. Aquaculture 300, 113–119.
Forsberg, O.I., 1999. Optimal harvesting of farmed Atlantic salmon at two cohort Ruiz-Velazco, J.M.J., Hernández-Llamas, A., Gómez-Muñoz, V.M., 2014. A
management strategies and different harvest operation restrictions. Aquacult. continuous diphasic model for prediction of survival of cultivated populations
Econ. Manage. 3, 143–158. when affected by disease: the case of shrimp and white spot disease. Aquacult.
González-Romero, M.A., Hernández-Llamas, A., Ruiz-Velazco, J.M.J., Res., http://dx.doi.org/10.1111/are. 12454.
Plascencia-Cuevas, T.N., Nieto-Navarro, J.T., 2014. Stochastic bio-economic Vose, D., 2002. Risk Analysis: A Quantitative Guide. John Wiley & Sons, Chichester,
optimization of pond size for intensive commercial production of whiteleg West Sussex, UK.
shrimp Litopenaeus vannamei. Aquaculture 433, 496–503. Yu, R., Leung, P.S., 2006. Optimal partial harvesting strategies for a single
Gulland, J.A., 1969. Manual of Methods for Fish Stock Assessment: Part 1, Fish production cycle in aquaculture operations. Mar. Resour. Econ. 21 (3),
Population Analysis. Food and Agriculture Organization of the United Nations, 301–315.
Rome. Yu, R., Leung, P., Bienfang, P., 2009. Modeling partial harvesting in intensive shrimp
Hauck, W.W., Anderson, S., 1986. A proposal for interpreting and reporting culture: a network-flow approach. Eur. J. Oper. Res. 193, 262–271.
negative studies. Stat. Med. 5, 203–209. Zar, J.H., 2010. Biostatistical Analysis. Prentice Hall, Upper Saddle River, NJ.