576 Revista de Biología Tropical, ISSN: 2215-2075 Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

Satellite and historical data, and statistical modeling to predict

potential fishing zones for dolphinfish, Coryphaena hippurus

(Perciformes: Coryphaenidae) in Colombian Pacific

Adriana Martínez Arias1; https://orcid.org/0000-0002-2457-4047

Luis Octavio González Salcedo1; https://orcid.org/0000-0003-2460-6106

Iván Felipe Benavides Martínez2; https://orcid.org/0000-0002-1139-3909

John Josephraj Selvaraj1; https://orcid.org/0000-0002-9195-4883

1. Universidad Nacional de Colombia – sede – Palmira, Facultad de Ingeniería y Administración, Recursos

Hidrobiológicos ; GEAL & GIMMA. Carrera 32 # 12 - 00, Palmira, 763533, Colombia;

amartinezar@unal.edu.co, logonzalezsa@unal.edu.co, jojselvaraj@unal.edu.co

2. Universidad Nacional de Colombia – sede – Tumaco, Instituto de Estudios del Pacífico, San Andrés de Tumaco, 52835

Colombia; ibenavidesm@unal.edu.co

Received 11-VI-2021. Corrected 10-VIII-2022. Accepted 17-VIII-2022.

ABSTRACT

Introduction: The prediction of potential fishing areas is considered one of the most immediate and practical

approaches in fisheries and is an essential technique for decision-making in managing fishery resources. It helps

fishermen reduce their fuel costs and the uncertainty of their fish catches; this technique allows to contribute to

national and international food security. In this study, we build different combinations of predictive statistical

models such as Generalized Linear Models and Generalized Additive Models.

Objective: To predict the spatial distribution of PFZs of the dolphinfish (Coryphaena hippurus L.) in the

Colombian Pacific Ocean.

Methods: We built different combinations of Generalized Linear Models and Generalized Additive Models to

predict the Catch Per Unit Effort of C. hippurus captured from 2002 to 2015 as a function of sea surface tem-

perature, chlorophyll-a concentration, sea level anomaly, and bathymetry.

Results: A Generalized Additive Model with Gaussian error distribution obtained the best performance for

predicting PFZs for C. hipurus. Model validation was performed by calculating the Root Mean Square Error

through a cross-validation approach. The R2 of this model was 50 %, which was considered suitable for the type

of data used. January and March were the months with the highest Catch per Unit Effort values, while November

and December showed the lower values.

Conclusion: The predicted PFZs of C. hippurus with Generalized Additive Models satisfactorily with the results

of previous research, suggesting that our model can be explored as a tool for the assessment, decision making,

and sustainable use of this species in the Colombian Pacific Ocean.

Key words: potential fishing zones; General Additive Models (GAM); Geographic Information System (GIS);

prediction models.

https://doi.org/10.15517/rev.biol.trop.2022.47375

AQUATIC ECOLOGY

Pelagic fishes such as the common dol-

phinfish (Coryphaena hippurus L) are highly

migratory in tropical and subtropical waters

(Zúñiga-Flores, 2009). However, similarly to

other species, their mobility is restricted by

oceanographic variability, which determines

their survival, growth, and development (Bel-

lido et al., 2008; Blaxter & Hunter, 1982).

577

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

Generally, changes in environmental conditions

trigger fish migration, which can last for sev-

eral months across big areas, generating good

conditions for fishing (Bellido et al., 2008).

Several studies have indicated that the

environmental variables in oceans influence the

distribution of pelagic fish species such as the

dolphinfish (Hsu et al., 2021). The physiology

of this species restricts populations to occur

in areas where the sea surface temperature

(SST) varies between 21.5 °C and 27.5 °C, and

chlorophyll-a (Chl-a) between 0.01 mg/m3 and

9.5 mg/m3 (Guzmán et al., 2010). It has been

demonstrated that SST is an accurate predictor

of habitat distribution, because it holds a close

relationship with species richness and plays

a vital role in the physiology of pelagic fish.

On the other hand, Chl–a is recognized as an

important oceanographic parameter as it deter-

mines primary production and biological pro-

ductivity in oceans. Several findings suggest

that SST and Chl–a are key factors influencing

patterns of spatial distribution and abundance

variability of pelagic fish like tuna, swordfish,

and the common dolphinfish. However, for this

last species, the spatial distribution of potential

fishing zones (PFZ) in the tropical region (par-

ticularly in the Colombian Pacific Ocean), and

their controlling factors are not clear enough

(Zainuddin et al., 2013).

A PFZ is a prediction of fish aggregation

in specific areas across the ocean, which is

made based on the relationship of fish biology

and the environmental context. PFZs are gener-

ally made using statistical parameters of SST

and Chl-a (Natteshan & Kumar, 2016). Differ-

ent researchers have applied and discussed a

wide range of techniques for the prediction of

PFZs, including remote sensing of a variety of

oceanographic variables, geographic informa-

tion systems (GIS) and statistical approaches

such as Generalized Linear Models (GLM),

Generalized Additive Models (GAM) and Clas-

sification and Regression Trees (CART). These

approaches ultimately find the functional links

between the environmental information and

fish abundance data to produce a PFZ predic-

tion (Marín-Enríquez et al., 2018).

Research published by Salvadeo et al.

(2020) and Marín-Enríquez et al. (2018) show

the application of these techniques to identify

the distribution of dolphinfish along Southern

California and the Mexican coast Herrera-

Montiel et al. (2019) adopted this approach

to identify and evaluate the PFZs of seerfish

(Scomberomorus sierra) in the CPO. The iden-

tification of PFZs has several challenges associ-

ated with the requirements of habitat modeling,

because, beyond the relationship between fish

abundance and their environment, other fac-

tors such as management, the spatiotemporal

variation of population sizes, the existence of

regulatory zones for fishing and the damage to

marine ecosystems have also an influence on

fish distribution (Selvaraj et al., 2009).

Colombia is a country where fisheries

contribute significantly to the economic growth

and food security (FAO, 2021), hence the appli-

cation of habitat models to determine patterns

of abundance, spatial distribution, and the

relationship of these patterns with the charac-

teristics of the habitat and the fishing areas,

is of crucial importance (Salazar et al., 2021;

Selvaraj et al., 2009). Furthermore, consider-

ing the rapid changes of fish populations due

to climate change in the oceans, it is highly

important to determine their PFZs, specially

for commercial fishes such as C. hippurus.

So far, there is no yet available information

on this issue.

Consequently, this research aimed at build-

ing, running, and comparing different models

to predict the PFZs of C. hippurus in the CPO,

considering the influence of determinant envi-

ronment conditions for the species suitability.

MATERIALS AND METHODS

Study area: The CPO belongs to the East-

ern Pacific Ocean, which is located along the

West coast of Colombia. It has a coast length of

1 300 km and borders to the North with the isth-

mus of Panama and to the South with Ecuador

(Fig. 1). The oceanographic conditions of CPO

are subject to the seasonal changes in the posi-

tion of the Inter-Tropical Convergence Zone

578 Revista de Biología Tropical, ISSN: 2215-2075 Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

(ITCZ) and to the strong inter-annual changes

due to El Niño South Oscillation (ENSO), both

irregular fluctuations that involve changes in

the general circulation of winds, insolation

and hydrography across the Tropical Eastern

Pacific (Rodríguez-Rubio et al., 2003). ENSO

implies an unstable interaction between the

SST and the atmospheric pressure, causing

strong variability in the winds, precipitation

and the depth of the thermocline, affecting the

biological productivity and the reproduction

and survival of marine species (Beltrán, 2011;

Fiedler, 2002) (Fig. 1).

Data sources and structure: Catch data

of C. hippurus was provided by the fishery

Sepulveda Rodgers LTDA and contains the

historical catch data of semi-industrial fishing

boats in the study area during the species’s sea-

son (November to March), from years 2000 to

2010, 2012 and 2015. These catches were per-

formed with longlines of 1 100 circular hooks.

The number of captured fish divided by the

1 100 total hooks is reported here as the Catch

per Unit of Effort (Selvaraj, 2010), which is a

proportion ranging from 0 to 1. Each CPUE

value results from a particular fishing set, from

fishing trips that last between 2 and 10 days at

the sea. Each particular CPUE value represents

the catch efficiency, and indirectly, the fish

abundance during each of the 150 total fishing

that occurred during the studied period. All the

CPUE values were positive, implying no zero

catches. Additionally, each CPUE is linked to a

geographical position where each fish trip was

performed along the 800 km of the CPO.

Oceanographic variables: Satellite data

of SST and Chl-a was obtained from the

MODIS sensor (Moderate Resolution Imaging

Spectroradiometer) at the website http://www.

oceancolor.gsfc.nasa.gov/. We downloaded

daily L3 level data (4 km) that matched the

catch dates for the period 2002-2015. Data

of Sea Level Anomalies (SLA) was obtained

from the AVISO project (Archiving Valida-

tion and Interpretation of Satellite Ocean-

ographic data), which is supported by the

Centre National D’Etudes Spatiales (CNES) in

France (Piedrahíta et al., 2013) (http://www.las.

aviso.oceanobs.com/las/getUI.do). SLA data

was combined to available information from

satellites ENVISAT, ERS1/2, GFO, JASON1,

JASON 2 and TOPEX/POSEIDON, with a

Fig. 1. Location of the study area showing its Exclusive Economic Zone (EEZ).

579

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

spatial resolution of 0.25 decimal degrees. The

bathymetric information was acquired from

the website: https://www.ngdc.noaa.gov/mgg/

global/ with a spatial resolution of 1 minute of

arc (1 km). This resolution was modified to 4

km by using a bilinear resampling algorithm in

ArcGIS version 10.5. This oceanographic data

set was processed together with the fishing data

in order to build the models for PFZ’s predic-

tion of C. hippurus in the CPO (Table 1).

Statistical modeling: The prediction of

the spatial distribution of PFZ’s of C. hippurus

was performed using parametric and semi-

parametric statistical models such as GLMs

and GAMs respectively. GLMs were used as

they allow the use of specific probability distri-

butions appropriate for normal errors (Gauss-

ian), asymmetric errors (Gamma) and errors

coming from proportion data (Beta); plus, link

functions to account for nonlinear relationships

between predictors and responses (Myers et al.,

2012). GAMs were used because besides speci-

fying specific probability distributions and link

functions, they offer the flexibility of fitting

nonlinear relationships via smoothed functions

(spline or kernel) combined with parametric

functions. Both models perform parameter esti-

mation via Maximum Likelihood Estimation

(MLE), (Wood, 2017).

GAM:

Where g is a link function, μi is the expect-

ed value of the response variable (CPUE), α0

is a model constant for every oceanographic

predictor, Sn is a smoothing function for each

predictor and ε is a random error (Wood,

2017) with a specific probability distribution

(Table 2).

TABLE 1

Data sets used to build the predictive statistical models and their sources

Data sets Source Period Spatial Reference Resolution

Spatial Temporal

C. hippurus captures Records from Sepúlveda Rodger Ltda 2000-2010

2012 and 2015

GCS_WGS_1984 NA Daily

Chl-a MODIS sensor 2002-2015 GCS_WGS_1984 4 km Daily

TSM MODIS sensor 2002-2015 GCS_WGS_1984 4 km Daily

SLA ENVISAT, ERS1/2, GFO, JASON1,

JASON 2 y TOPEX/POSEIDON

2002-2015 ITRF2008 4 km Daily

Bathymetry ETOPE 1 2002-2015 GCS_WGS_1984 1 km NA

TABLE 2

Results of the modeling and validation procedures showing the type of model, probability distribution of errors, link-

functions and the metrics used for model selection and validation

Modeling Validation

Model Probability distribution Link-function Df AICc Adjusted R2D RMSE

GAM Gaussian Log 14 -525.5 0.50 54.2 % 0.037

GAM Gamma Log 10 -526.2 0.38 34.9 % 0.040

GAM Beta Logit 10 -522.4 0.39 41.1 % 0.040

GLM Beta Logit 4 -525.4 0.28 25.2 % 0.044

GLM Gaussian Log 6 -494.2 0.27 30.1 % 0.045

GLM Gamma Log 6 -519.3 0.17 27.4 % 0.052

Df = degrees of freedom.

580 Revista de Biología Tropical, ISSN: 2215-2075 Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

GLM:

Where is a link function, Yi is the ith obser-

vation of CPUE, X1i, X2i, Xki are the ith observa-

tions of the k oceanographic predictors, μi is

the ith term of the random error with a specific

probability distribution and b1, b2 bk are the

effect parameters for each predictor (Espinoza-

Morriberón, 2010).

These models were built with all the

oceanographic variables as predictors using a

random data subset with 70 % of the dataset (N

= 105). The remaining 30 % (N = 45) was kept

for model validation (see next section). The

Akaike Information Criterion (AIC) (Akaike,

1973; García et al., 2014), the coefficient of

determination (R2) and the explained deviance

(D) were used as the criteria to select the best

model among all candidate models that were

built. To identify differences in CPUE between

months of the year, we used a Kruskal-Wallis

test plus a-posteriori pairwise multiple com-

parisons (Zar, 2010), where P values lower than

0.05 were considered as statistically significant.

To identify spatial and temporal autocor-

relation patterns in the CPUE data that would

risk the reliability of the model predictions,

we performed Moran tests (Legendre, 1993)

and Partial Autocorrelation Functions (PACF)

(Tsitsika et al., 2007) respectively on the

residuals of the best model selected after model

validation. Moran test showed no significant

spatial autocorrelation between residual values

and the matrix of spatial distances (P = 0.69)

between catch points, and the PACF showed

only a weak autocorrelation in lag 1 (R =

0.22). For Gaussian GAM and GLM models,

normal distribution and homoscedasticity were

checked by performing QQ-plots and residuals

vs fitted plots. All these assumptions were cor-

rectly fulfilled to ensure model validation and

applicability for PFZs prediction.

Model validation: We used a cross-vali-

dation approach for the predictions of CPUE

(Pérez-Planells et al., 2015). This approach

was performed with a random subset contain-

ing 30 % of the data that was not involved

in modeling. We used the Root Mean Square

Error (RMSE) as a metric to evaluate predic-

tion accuracy validate and model performance.

This metric quantifies the quality of predictions

through the difference between the observed

and predicted values (Martínez, 2016). RMSE

is calculated as:

Where is the actual observed value, is

the predicted value and is the number of

observations. All the statistical analyses were

performed in R-studio (ver 1.1.442) with pack-

ages base, MuMin, ggplot2, mgcv, ape, metrics,

betareg, boot and raster.

RESULTS

Oceanographic variables: SST varied

between 25.13 and 29.06 ºC, being higher in

January and lower in March. Higher values of

CPUE occurred in temperatures between 25.5

and 27.5 ºC. Above 27.5 °C and below 25.5 °C,

CPUE decreases. Chl–a varied between 0.15

and 2.2 mg/m3, being higher in December and

February. Higher values of CPUE occurred in

places where Chl–a varied between 0.5 and

2.2 mg/m3, Below and above these values

respectively, CPUE decreased. SLA varied

between -21.18 and 8.9 cm. Higher CPUE val-

ues occurred between -10 and 0.1 cm. Below

and above these values respectively, CPUE

decreased. Bathymetry varied between -24 and

-400 m, with no optimal ranks for CPUE.

Model selection and validation: Accord-

ing to the values shown in Table 2, the Gaussian

GAM, Beta GAM, and Gamma GAM were the

best candidate models. Despite Gamma GAM

had the lowest AICc (slightly lower than the

Gaussian GAM), the R2 and D values were con-

siderably higher for the Gaussian GAM. Simi-

larly, the RMSE values for the cross-validation

581

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

procedure showed that the predictions resulting

from Gaussian GAM were the best fitted to

the observed CPUE data. When plotting CPUE

predictions over the years (Fig. 2A), Gaussian

GAM was able to capture higher time vari-

ability than other models. This was particularly

conspicuous over periods of abrupt oscillations

in fish captures like those between 2001-2003,

2006-2009 and 2010-2015. When comparing

the CPUE predictions of every model to the

observed CPUE (Fig. 2B) and fitting simple

linear regressions on the data, we found again a

higher performance for Gaussian GAM, which

had a higher slope (0.7) and lower intercept

(0.03) in comparison to Gamma GAM and

Beta GAM.

Despite all these criteria in favor of Gauss-

ian GAM, it is clear that this model still

underestimates CPUE and its R2 cannot be con-

sidered high. However, due to the high variabil-

ity of these kinds of data both in the spatial and

temporal dimensions and to the multiplicity of

factors not able to be included in our models,

we considered that the performance of Gauss-

ian GAM was enough as a first approach to

the prediction of PFZ’s for C. hippurus in the

CPO. The results of the Gaussian GAM are

shown in Table 3. For this model, the intercept

as well as the effect of Chl-a, SST, and SLA

were statistically significant, yet based on the

F-values, Chl-a had the highest effect on the

behavior of CPUE.

Prediction of PFZs: Using the CPUE pre-

dictions of the Gaussian GAM, we generated a

series of maps according to the monthly season-

ality of C. hippurus (November-March) for all

the capture years (2002-2015) (Fig. 3). These

maps show the probability of aggregation of C.

hippurus, and consequently, reflect the occur-

rence of PFZs across the CPO. The spatial dis-

tribution of PFZs during November expanded

uniformly from coordinates 1°00’ N & 79°00’

W to 8°00’ N & 83°00’ W along the Pacific

Fig. 2. A. Total observed and predicted CPUE values over capture years for the whole study area according to the three best

candidate models: Gaussian GAM, Gamma GAM, and Beta GAM. Predictions are shown in the original CPUE scale after

exponential transformation of fitted values. B. Relationship between observed and predicted CPUE values according to the

three best candidate models. Lines represent a linear regression between the observed and predicted CPUE data for every

model. Intercept and slope parameters are shown in the upper boxes for every model.

582 Revista de Biología Tropical, ISSN: 2215-2075 Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

coast of Colombia. Meanwhile, the PFZ’s dis-

tribution in the oceanic area was not uniform

in comparison to the coast. However, impor-

tant fish aggregation was observed between

coordinates 6°00’ N & 80°00’ W (coastal area)

and 4°00’ N & 82°00’ W (oceanic area), espe-

cially to the South (border with Ecuador) in the

Nariño region, with CPUE values around 0.24.

In December, the PFZs are mainly grouped in

front of the Nariño region, similarly to the pat-

tern of November, with CPUE values ranging

from 0.17 to 0.19 between coordinates 6°00’

N & 77°05’ W to 1°00’ N & 79°05’ W near to

the coast. Two oceanic conglomerations were

observed at coordinates 3°00’ N & 81°02’ W to

2°00’N & 82°00’ W and 2°08’ N & 82°03’ W

to 1°05’ N & 82°03’ W approximately. Mean

CPUE values of December are lower in com-

parison to November, with maximum values

up to 0.234.

For January, PFZs show a persistent occur-

rence in front of the Colombian coast from

coordinates 6°00’ N & 77°05’ W to 8°00’ N

& 78°5’W to 1°00’ N & 83°1’ W, with mean

CPUE values around 0.24. This zone starts on

the Panama coast and extends southwards to

the Chocó and Valle del Cauca regions in the

Colombian coast, with a mean CPUE of 0.24,

which gradually decreases towards the Cauca

and Nariño regions. CPUE values in January

were significantly higher than in December

and November for all the study years. During

February there were two conspicuous PZFs,

one between coordinates 1°17’ N & 79°00’ W

to 6°53’ N & 77°42’ W and the other at the

oceanic area of the Panama basin from coor-

dinates 7°18’5’’ N & 79°46’ W to 3°20°N &

80°83’ W. Mean CPUE values in these zones

are around 0.18.

Finally, the spatial distribution of PFZs for

March showed higher CPUE values. Locations

of PFZs persisted near the coast at coordinates

6°00’ N & 77°5’ W to 1°00’ N & 79°05’ W and

at an oceanic area with coordinates 3°00’ N &

81°02’ W to 2°00’ N & 82°00’ W, 2°08’ N &

82°03’ W and 1°05’ N & 82°03’ W approxi-

mately. Generally, the PFZs were spatially

persistent over the months and years. Neverthe-

less, the PFZs with larger sizes occurred dur-

ing January, February, and March, with mean

CPUE values around 0.25. The difference of

CPUE values between all pairs of months were

highly significant according to Kruskal-Wallis

tests (P < 0.001 for every a-posteriori pairwise

multiple comparison).

DISCUSSION

The GIS approach used in this study was

successful to capture enough variability in the

oceanographic variables, which helped to under-

stand the spatiotemporal dynamics of the PFZs

of C. hippurus. This is in accordance with the

results described by Devis et al. (2002), where

the variation in Intertropical Convergence Zone

(ITCZ) triggers changes in SST and Chl-a. This

phenomenon creates a gradual seasonality in

climate and oceanic conditions, allowing the

presence of different temperatures along the

surface water. Concerning the spatial distribu-

tion of SST, our results resemble the findings

of Rodríguez-Rubio (2003), who mentioned

that this variable evolves seasonally between

January and March, when surface waters of low

temperature (25-26 °C) extend from the Gulf of

Panama to the center of the basin in a South-

to-Southwest orientation. From April to June

the temperature increases gradually eastwards

and by July-September, the cold-water plume

TABLE 3

Parametric coefficients and approximate significance

smooth terms resulted from Gaussian GAM model

Parametric coefficients

Estimate SE t-value P-value

Intercept 0.08 0.04 -59.45 < 0.001

Approximate significance of smooth terms

EDF RDF F-value P-value

Chl-a2.84 3.47 17.56 < 0.001

SST 4.86 5.60 2.08 0.05

SLA 3.50 4.28 3.62 0.006

Bathymetry 1.00 1.00 0.01 0.91

EDF = Effective degrees of freedom; SE = Standard error;

RDF = Reference degrees of freedom.

583

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

disappears altogether with 3 °C increase to the

North and 1 °C decrease to the South. By the

end of the year (October-December), the spatial

pattern of temperature stays, though with a ~1

°C decrease in across the CPO.

On the other hand, the results obtained

with the GAM model with Gaussian distribu-

tion are in accordance with the findings made

by Bigelow et al. (1999), Rodríguez-Marín et

al. (2003), Zagaglia et al. (2004), Zainuddin et

al. (2013), and Setiawati et al. (2015). These

authors indicated that this modeling approach

is highly efficient to predict PFZs of pelagic

species in the Pacific and Atlantic oceans.

This efficiency relies on the fact that GAMs

offer high flexibility regarding the type of data

Fig. 3. Assemble maps of monthly predicted PFZs for C.

hippurus during 2002 to 2015 in the Colombian Pacific

Ocean.

584 Revista de Biología Tropical, ISSN: 2215-2075 Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

and the link between predictors and responses,

allowing non-linear and non-monotonous rela-

tionships to make predictions in the response

variable (França & Cabral, 2015; Guisan et

al., 2002; Yee & Mitchell, 1991). Maunder

and Punt (2004), compared the performance

of GAMs, GLMs and GLMMs (Generalized

Linear Mixed Model) to predict the spatial dis-

tribution of valuable fish species such as tuna

and sharks, remarking the benefits of GAMs.

Though we consider that the Gaussian

GAM model used in this study was success-

ful as a first approach to predict the spatial

distribution and PFZs of C. hippurus in the

CPO, our results suggest that the precision of

the model still needs to be increased for better

results. When observing the model predictions

over the capture years in (Fig. 3), it is clear that

for many capture months the model underes-

timates or overestimates the observed CPUE,

with maximum difference values up to 9.6

%. We suggest that these underperformances

can be solved by including other biological

variables not considered in this study, such as

physiologic traits, life-history traits, genetic

structuring, predation, and migration rates of

C. hippurus populations. When including some

of these variables in the model, a higher CPUE

variability could be captured, hence increasing

the chance of making more accurate predic-

tions. This would help the developing of more

powerful frameworks for studies that consider

fish management and decision making under

climate change scenarios.

Our model detected Chl-a as the most effi-

cient to predict CPUE, agreeing with Sund et

al. (1981), who considers it as the most restric-

tive oceanic variable for the horizontal and

vertical distribution of pelagic species. Further-

more, it has been a successful proxy to estimate

the available food in pelagic ecosystems, since

this pigment is present in most of the phyto-

plankton species and forms the basis of trophic

chains (Martínez-Rincón, 2012; Sartimbul et

al., 2010). Loukos et al. (2003) suggested

specific large-scale changes on the fish habi-

tat in the equatorial Pacific by means of SST

effects, which could explain the distribution of

available habitats for pelagic fishes on higher

latitudes due to the increase of temperature by

global warming (Cheung et al., 2009; Mugo

et al., 2010). Despite SLA and Bathymetry

showed a lower effect on the CPUE, they were

included in the model because they are related

to the spatial distribution and movement of

pelagic species (Maravelias, 1999). SLA is

related to mesoscale cyclonic movements that

occur at the beginning of the year (Rodríguez-

Rubio, 2003), bringing oceanic upwelling, low

temperatures, and high chlorophyll concentra-

tions. According to Sabarros et al. (2009), it

is highly probable that these mesoscale pro-

cesses are related to the presence of PFZs of

pelagic fishes.

The predicted PFZs for C. hippurus and

their spatial distribution was similar to the

findings drawn by Lasso and Zapata (1999),

who mentioned that the higher fish abundances

occur during February and March. Further-

more, this season could be associated with

phenomena that trigger fish migration, since

it coincides with the lowest SST and higher

Chl–a concentrations in comparison to the rest

of the year. Giraldo et al. (2010) and Guzmán

et al. (2010) demonstrated that the aggregation

of C. hippurus are linked to mesoscale oceanic

processes, such as the oceanic thermal fronts.

Lasso and Zapata (1999) suggested that the

migratory path of C. hippurus could be associ-

ated with surface ocean currents in the eastern

Pacific, as a response to the changes driven by

ENSO, which cause fish populations to migrate

southwards in search for warmer waters (Caeta-

no-Nunes, 2013; Lasso & Zapata, 1999).

The predicted PFZs shown by our model

from February and March agree with the

reports of the highest catch rates in Colom-

bia (Lasso & Zapata, 1999) an also, with the

isotherm changes from January to April, and

the salinity increase in the Pacific Ocean that

brings an upsurge of nutrients in front of the

Panama Bay (Valencia et al., 2013). Gener-

ally, these results show that the distribution

patterns of PFZs for C. hippurus are associ-

ated with changes in SST, which influence fish

585

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

aggregation, growth, reproduction, and survival

(Boyce et al., 2008; Martínez-Rincón, 2012).

As shown in this study, the identifica-

tion of PFZs of fish species emerges from the

combination of remote sensing, geographic

information systems, historical registers of in

situ data and statistical modeling tools, which

can be used as a robust package to improve

the management and decision making of fish

resources. As stated by Selvaraj et al. (2009),

the maps resulting from this multidisciplinary

approach can be used for conservation o extrac-

tion purposes. However, to achieve these pur-

poses, we still need a standardization of data

sharing, data depuration and analysis and train-

ing of the entities and people in charge of its

administration and distribution.

The development of these prediction tech-

niques has created new alternatives that allow

understanding the behavior and distribution of

fish populations. However, the variability of

the prediction performances among different

models results in contrasting predictions. This

implies that a predictive model must be select-

ed very carefully, especially if it is intended

for practical applications (Zhang et al., 2017).

The advantage of using GAMs relies to their

robustness to outliers and its ability to capture

complex interactions between predictors. These

advances in the model’s predictive power allow

a more reliable understanding of potential fish-

ing areas, especially in those areas at the local

level (Venables & Dichmont, 2004).

Ethical statement: the authors declare

that they all agree with this publication and

made significant contributions; that there is no

conflict of interest of any kind; and that we fol-

lowed all pertinent ethical and legal procedures

and requirements. All financial sources are

fully and clearly stated in the acknowledge-

ments section. A signed document has been

filed in the journal archives.

ACKNOWLEDGMENTS

We are grateful to Ocean color, AVISO,

NOAA and Sepulveda Rodgers Ltda for

providing the logbook fishing data for this

study; to COLCIENCIAS for providing the

funding through the project 067-2016, and to

the Universidad Nacional de Colombia.

RESUMEN

Datos satelitales e históricos, y modelado estadístico,

para predecir zonas de pesca potencial del pez dorado,

Coryphaena hippurus (Perciformes: Coryphaenidae)

en el Pacífico colombiano

Introducción: La predicción de zonas potenciales de pesca

se considera uno de los enfoques más inmediatos y efec-

tivos en las pesquerías, es una técnica importante para la

toma de decisiones en el manejo de los recursos pesqueros.

Ayuda a los pescadores a reducir su costo de combustible

y también a disminuir la incertidumbre de sus capturas,

esta técnica permite contribuir a la seguridad alimentaria

nacional e internacional. En este estudio, se construyeron

diferentes combinaciones de modelos estadísticos pre-

dictivos como modelos lineales generalizados y modelos

aditivos generalizados.

Objetivo: predecir la distribución espacial de las zonas

potenciales de pesca del pez dorado (Coryphaena hippurus

L.) en el Pacífico colombiano.

Métodos: La variable de respuesta se expresó en escala

de captura por unidad de esfuerzo, es decir, el número de

individuos de C. hippurus capturados por un número total

de anzuelos disponibles entre 2002 y 2015. Temperatura de

la superficie del mar, concentración de clorofila, anomalía

del nivel del mar y batimetría, se utilizaron como varia-

bles explicativas para los meses de estacionalidad de C.

hippurus (noviembre - marzo).

Resultados: El modelo con mejor rendimiento para la

predicción de zonas potenciales de pesca fue un modelo

aditivo generalizado con distribución de error gaussiana y

función de enlace de registro, que se seleccionó en función

del criterio de información de Akaike, el R2 y la desviación

explicada. La validación del modelo se realizó calculando

el error cuadrático medio a través de un enfoque de vali-

dación cruzada. El ajuste de este modelo fue del 50 %,

lo que puede considerarse adecuado para el tipo de datos

utilizados. Enero y marzo fueron los meses con mayor

captura por unidad de esfuerzo y noviembre-diciembre los

meses con menor.

Conclusión: Las zonas potenciales de pesca previstas coin-

cidieron satisfactoriamente con investigaciones anteriores,

lo que sugiere que nuestro modelo es una herramienta

poderosa para la evaluación, toma de decisiones y uso

sostenible de los recursos pesqueros de C. hippurus en el

Pacífico colombiano.

Palabras clave: zona potencial de pesca; Modelo Aditivo

Generalizado (GAM); sistema de información geográfica;

modelos de predictivos.

586 Revista de Biología Tropical, ISSN: 2215-2075 Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

REFERENCES

Akaike, H. (1973). Information theory and an extension of

the maximum likelihood principle. In B. N. Petrov, &

F. Csaki (Eds.), Second International Symposium on

Information Theory (pp. 267–281). Akademiai Kiado.

Bellido, J. M., Brown, A. M., Valavanis, V. D., Giráldez,

A., Pierce, G. J., Iglesias, M., & Palialexis, A. (2008).

Identifying essential fish habitat for small pelagic

species in spanish Mediterranean waters. Hydrobio-

logia, 612(1), 171–184.

Beltrán, F M. C. (2011). Evaluación de las pesquerías

de los bancos de pesca artesanal en el Pacífico sur

colombiano, Colombia (Tesis de pregrado). Universi-

dad Nacional de Colombia, Colombia.

Blaxter, J. H. S., & Hunter, J. R. (1982). The biology of the

clupeoid fishes. Academic Press.

Bigelow, K., Boggs, C., & He, X. (1999). Environmental

effects on swordfish and blue shark catch rates in the

US North Pacific longline fishery. Fisheries Oceano-

graphy, 8(3), 178–198.

Boyce, D., Tittensor, D., & Worm, B. (2008). Effects of

temperature on global patterns of tuna and billfish

richness. Marine Ecology Progress Series, 355(5),

267–276.

Caetano-Nunes, B. R. (2013). Caracterización genética y

ambiental del dorado (Coryphaena hippurus L.) en

el Pacífico colombiano (Tesis doctoral). Universidad

Nacional de Colombia, Colombia.

Cheung, W. W., Lam, V. W., Sarmiento, J., Kearney, K.,

Watson, R., & Pauly, D. (2009). Projecting global

marine biodiversity impacts under climate change

scenarios. Fish and Fisheries, 10(3), 235–251.

Devis, A., García, I., Málikov, I., & Villegas, N. (2002).

Compilación oceanográfica de la cuenca pacífica

colombiana. Centro Control Contaminación del Pací-

fico, Tumaco, Colombia.

Espinoza-Morriberón, D. (2010). Estandarización de la

CPUE de la flota industrial de cerco del stock norte-

centro de anchoveta peruana (Engraulis ringens

Jenyns 1842) entre 1996 y el 2008 (Tesis de pregra-

do). Universidad Nacional Mayor de San Marcos,

Perú.

FAO. (2021). Visión general del sector acuícola nacional-

Colombia. National Aquaculture Sector Overview

fact sheets (Sitio web). https://www.fao.org/fishery/

en/countrysector/co/es

Fiedler, P. C. (2002). Environmental change in the eas-

tern tropical Pacific Ocean: Review of ENSO and

decadal variability. Marine Ecology Progress Series,

244(2002), 265–283.

França, S., & Cabral, H. N. (2015). Predicting fish species

richness in estuaries: Which modelling technique to

use? Environmental Modelling & Software, 66(2015),

17–26.

García, M. D. C., Castellana, N., Rapelli, C., Koegel, L., &

Catalano, M. (2014). Criterios de información y pre-

dictivos para la selección de un modelo lineal mixto.

SaberEs, 6(2), 61–76.

Giraldo, A., Payán, L., & Chirima, H. (2010). Variabilidad

térmica en el ambiente pelágico de la Isla Gorgona

entre marzo 2009-febrero 2010 (Presentación oral).

XIV Semanario Nacional de Ciencia y Tecnologías

del Mar, Calí, Colombia.

Guisan, A., Edwards, T. C., & Hastie, T. (2002). Generali-

zed linear and generalized additive models in studies

of species distributions: setting the scene. Ecological

Modelling, 157(2), 89–100.

Guzmán, A. I., Selvaraj, J. J., & Martínez, A. (2010).

Variación de la temperatura superficial del mar,

clorofila “a” y su relación con capturas de dorado

(Coryphaena hippurus) en la costa pacífica colom-

biana (Presentación oral). III Congreso Brasileiro de

Oceanografía, Fortaleza, Brasil.

Herrera-Montiel, S. A., Coronado-Franco, K. V., &

Selvaraj, J. J. (2019). Predicted Changes in the

Potential Distribution of Seerfish (Scomberomorus

sierra) under multiple climate change scenarios in the

Colombian Pacific Ocean. Ecological Informatics,

53(2019), 100985.

Hsu, T. Y., Chang, Y., Lee, M. A., Wu, R. F., & Hsiao, S. C.

(2021). Predicting skipjack tuna fishing grounds in

the western and central Pacific Ocean based on high-

spatial- temporal-resolution satellite data. Remote

Sensing, 13(5), 1–17.

Lasso, J., & Zapata, L. (1999). Fisheries and biology of

Coryphaena hippurus (Pisces: Coryphaenidae) in

the Pacific coast of Colombia and Panama. Scientia

Marina, 63(3-4), 387–399.

Legendre, P. (1993). Spatial autocorrelation: trouble or new

paradigm? Ecology, 74(6), 1659–1673.

Loukos, H., Monfray, P., Bopp, L., & Lehodey, P. (2003).

Potential changes in skipjack tuna (Katsuwonus pela-

mis) habitat from a global warming scenario: mode-

ling approach and preliminary results. Fisheries

Oceanography, 12(4-5), 474–482.

Maravelias, C. D. (1999). Habitat selection and clustering

of a pelagic fish: effects of topography and bathyme-

try on species dynamics. Canadian Journal of Fishe-

ries and Aquatic Sciences, 56(3), 437–450.

Marín-Enríquez, E., Seoane, J., & Muhlia-Melo, A. (2018).

Environmental modeling of occurrence of dolphin-

fish (Coryphaena spp.) in the Pacific Ocean off

Mexico reveals seasonality in abundance, hot spots

587

Revista de Biología Tropical, ISSN: 2215-2075, Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

and migration patterns. Fisheries Oceanography,

27(1), 28–40.

Martínez, D. N. (2016). Predicción y Análisis de los Retra-

sos en los Vuelos (Tesis de pregrado). Universidad

Autónoma de Barcelona, España.

Martínez-Rincón, R. O. (2012). Efecto de la variabilidad

ambiental en la distribución de las capturas inciden-

tales de pelágicos mayores en el Océano Pacífico

Oriental (Tesis doctoral). Instituto Politécnico Nacio-

nal, México.

Maunder, M. N., & Punt, A. E. (2004). Standardizing catch

and effort data: a review of recent approaches. Fishe-

ries Research, 70(2), 141–159.

Mugo, R., Saitoh, S. I., Nihira, A., & Kuroyama, T. (2010).

Habitat characteristics of skipjack tuna (Katsuwonus

pelamis) in the Western North Pacific. Fisheries

Oceanography, 19(5), 382–396.

Myers, R. H., Montgomery, D. C., Vining, G. G., & Rob-

inson, T. J. (2012). Generalized linear models: with

applications in engineering and the sciences (2nd

Ed.). John Wiley & Sons Inc.

Natteshan, N. V. S., & Kumar, N. S. (2016). A survey of

various methods of potential fishing zone identifi-

cation. International Journal of Control Theory and

Applications, 9(40), 697–703.

Pérez-Planells, L., Delegido, J., Rivera-Caicedo, J. P., &

Verrelst, J. (2015). Análisis de métodos de validación

cruzada para la obtención robusta de parámetros bio-

físicos. Revista Española de Teledetección, 4(2015),

55–65.

Piedrahíta, C., Selvaraj, J. J., & Guzmán, A. I. (2013).

Detección de afloramientos de algas usando datos

modis de fluorescencia en el pacífico colombiano.

En L. M. Marina, & C. García (Eds.), Investigación

en ciencias del mar (pp. 57–69). Aportes de la Uni-

versidad Nacional de Colombia.

Rodríguez-Marín, E., Arrizabalaga, H., Ortiz, M., Rodrí-

guez-Cabello, C., Moreno, G., & Kell, L. T. (2003).

Standardization of bluefin tuna, Thunnus thynnus,

catch per unit effort in the baitboat fishery of the Bay

of Biscay (Eastern Atlantic). ICES Journal of Marine

Science, 60(6), 1216–1231.

Rodríguez-Rubio, E. (2003). Variabilidad de la circulación

de mesoescala en la cuenca oceánica del Pacífico

colombiano (Tesis doctoral). Universidad de Con-

cepción, Chile.

Rodríguez-Rubio, E., Schneider, W., & Abarca del Río, R.

(2003). On the seasonal circulation within the Pana-

ma Bight derived from satellite observations of wind,

altimetry and sea surface temperature. Geophysical

Research Letters, 30(7), 631–634.

Sabarros, P. S., Ménard, F., Lévénez, J. J., Tew-Kai, E., &

Ternon, J. F. (2009). Mesoscale eddies influence dis-

tribution and aggregation patterns of micronekton in

the Mozambique Channel. Marine Ecology Progress

Series, 395(2009), 101–107.

Salazar, J. E., Benavides, I. F., Cabrera, C. V. P., Guzmán,

A. I., & Selvaraj, J. J. (2021). Generalized additive

models with delayed effects and spatial autocorrela-

tion patterns to improve the spatiotemporal prediction

of the skipjack (Katsuwonus pelamis) distribution in

the Colombian Pacific Ocean. Regional Studies in

Marine Science, 45(2021), 101829.

Salvadeo, C., Auliz-Ortiz, D. M., Petatán-Ramírez, D.,

Reyes-Bonilla, H., Ivanova-Bonchera, A., & Juárez-

León, E. (2020). Potential poleward distribution shift

of dolphinfish (Coryphaena hippurus) along the

southern California current system. Environmental

Biology of Fishes, 103(8), 973–984.

Sartimbul, A., Nakata, H., Rohadi, E., Yusuf, B., &

Kadarisman, H. (2010). Variations in chlorophyll-a

concentration and the impact on Sardinella lemuru

catches in Bali Strait, Indonesia. Progress in Oceano-

graphy, 87(1-4), 168–174.

Selvaraj, J. J., Martínez, A., Guzmán, A. I. (2010). Envi-

ronmental preference of dolphinfish (Coryphaena

hippurus) derived from remotely sensed data along

the Pacific coast of Colombia (Oral presentation).

ICES Annual Science Conference, Nantes, France.

Selvaraj, J. J., Rajasekharan, M., & Guzmán, Á. I. (2009).

Aplicaciones de los sistemas de información geo-

gráfica y sensores remotos al manejo de pesquerías

marinas y desafíos para su desarrollo en Colombia.

Boletín de Investigaciones Marinas y Costeras, 38(1),

105–120.

Setiawati, M. D., Sambah, A. B., Miura, F., Tanaka, T.,

& As-Syakur, A. R. (2015). Characterization of

bigeye tuna habitat in the Southern waters off Java-

Bali using remote sensing data. Advances in Space

Research, 55(2), 732–746.

Sund, P. N., Blackburn, M., & Williams, F. (1981). Tunas

and their environment in the Pacific Ocean: a review.

Oceanography and Marine Biology an Annual

Review, 1981(19), 443–512.

Tsitsika, E. V., Maravelias, C. D., & Haralabous, J. (2007).

Modeling and forecasting pelagic fish production

using univariate and multivariate ARIMA models.

Fisheries Science, 73(5), 979–988.

Valencia, B., Lavaniegos, B., Giraldo, A., & Rodríguez-

Rubio, E. (2013). Temporal and spatial variation of

hyperiid amphipod assemblages in response to hydro-

graphic processes in the Panama Bight, eastern tropi-

cal Pacific. Deep-Sea Research I, 73(2013), 46–61.

Venables, W. N., & Dichmont, C. M. (2004). GLMs, GAMs

and GLMMs: an overview of theory for applications

588 Revista de Biología Tropical, ISSN: 2215-2075 Vol. 70: 576-588, Enero-Diciembre 2022 (Publicado Ago. 22, 2022)

in fisheries research. Fisheries Research, 70(2-3),

319–337.

Wood, S. M. (2017). Generalized additive models, an

introduction with R (2nd Ed.). Chapman and Hall/

CRC Press.

Yee, T. W., & Mitchell, N. D. (1991). Generalized additive

models in plant ecology. Journal of Vegetation Scien-

ce, 2(5),587–602.

Zagaglia, C., Lorenzzetti, J., & Stech, J. (2004). Remote

sensing data and longline catches of yellowfin tuna

(Thunnus albacares) in the equatorial Atlantic. Remo-

te Sensing of Environment, 93(1), 267–281.

Zainuddin, M., Nelwan, A., Farhum, S. A., Hajar, M. A. I.,

& Kurnia, M. (2013). Characterizing potential fishing

zone of skipjack tuna during the southeast monsoon

in the Bone Bay-Flores Sea using remotely sensed

oceanographic data. International Journal of Geos-

ciences, 4(1), 259–266.

Zar, J. H. (2010). Biostatistical analysis (5th Ed.). Prentice

Hall.

Zhang, X., Saitoh, S. I., & Hirawake, T. (2017). Predicting

potential fishing zones of Japanese common squid

(Todarodes pacificus) using remotely sensed images

in coastal waters of south-western Hokkaido, Japan.

International Journal of Remote Sensing, 38(21),

6129–6146.

Zúñiga-Flores, M. S. (2009). Dinámica poblacional del

dorado (Coryphaena hippurus) en Baja California

Sur, México: implicaciones para su manejo (Tesis

doctoral). Instituto Politécnico Nacional, México.