Modeling sensitive parrotfish (Labridae: Scarini) habitats along the Brazilian coast

Accepted Manuscript Modeling sensitive parrotfish (Labridae: Scarini) habitats along the Brazilian coast Natalia C. Roos, Adriana R. Caravalho, Priscila F.M. Lopes, M.Grazia Pennino PII:

S0141-1136(15)30033-7

DOI:

10.1016/j.marenvres.2015.08.005

Reference:

MERE 4053

To appear in:

Marine Environmental Research

Received Date: 3 June 2015 Revised Date:

31 July 2015

Accepted Date: 10 August 2015

Please cite this article as: Roos, N.C., Caravalho, A.R., Lopes, P.F.M., Pennino, M.G., Modeling sensitive parrotfish (Labridae: Scarini) habitats along the Brazilian coast, Marine Environmental Research (2015), doi: 10.1016/j.marenvres.2015.08.005. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

1

Modeling sensitive parrotfish (Labridae: Scarini) habitats along the Brazilian coast.

2

NATALIA C. ROOS 1*, ADRIANA R. CARAVALHO1, PRISCILA F. M. LOPES1, M. GRAZIA

3

PENNINO1

4

1

5

Nova. CEP 59.098-970 Natal (RN), Brazil.

Universidade Federal do Rio Grande do Norte - UFRN. Dept. of Ecology, Campus Universitário s/n, Lagoa

7

RI PT

6

* email: [email protected] / telephone number: +5584988116227

8 9

Abstract

In coral reef environments, there is an increasing concern over parrotfish (Labridae: Scarini) due

11

to their rising exploitation by commercial small-scale fisheries, which is leading to significant

12

changes in the reefs’ community structure. Three species, Scarus trispinosus (Valenciennes,

13

1840), Sparisoma frondosum (Agassiz, 1831) and Sparisoma axillare (Steindachner, 1878),

14

currently labeled as threatened, have been intensively targeted in Brazil, mostly on the

15

northeastern coast. Despite their economic importance, ecological interest and worrisome

16

conservation status, not much is known about which variables determine their occurrence. In this

17

study, we adopted a hierarchical Bayesian spatial-temporal approach to map the distribution of

18

these three species along the Brazilian coast, using landing data from three different gears

19

(gillnets, spear guns, and handlines) and environmental variables (bathymetry, shore distance,

20

seabed slope, Sea Surface Temperature and Net Primary Productivity). Our results identify

21

sensitive habitats for parrotfish along the Brazilian coast that would be more suitable to the

22

implementation of spatial-temporal closure measures, which along with the social component

23

fishers could benefit the management and conservation of these species.

AC C

EP

TE D

M AN U

SC

10

24 25

Keywords: Bayesian models, coral reef species, conservation, small-scale fishery, fisheries

26

management.

1

27

ACCEPTED MANUSCRIPT

Introduction

An ecosystem approach to reef fisheries should include the study of current and expected

29

fishing impacts on exploited habitats in order to prepare efficient strategies for the conservation of

30

the marine environment and, in particular, of its living resources. The approach should also include

31

careful marine spatial planning of reef use to ensure the protection of the relevant habitats of key

32

species (Katsanevakis et al. 2011). Therefore, it is a requirement to have a solid knowledge of

33

species-environment relationships and to identify priority areas for conservation or management

34

using a robust analysis based on the available information. In this regard, habitat and species

35

mapping are important tools for conservation programs because they provide a clear picture of the

36

distribution and the extent of these marine resources and their key species and thus facilitate

37

management (Pennino et al. 2013).

M AN U

SC

RI PT

28

Key species are the species that play important functional roles with significant consequences

39

in the food web, including the persistence of other species (Bond 1994). They are not restricted to

40

one level of the food chain, including predators and herbivores alike (Mills et al. 1993). In fisheries,

41

top predators, some of which are considered key species, tend to get most of the attention, first from

42

fishers and then consequently from researchers (Myers and Worm 2003; but see cases where

43

fisheries start at lower trophic levels, especially in places where the targets are highly valued

44

invertebrates; Tewfik and Béné, 2004). Usually other large-bodied species at lower trophic levels,

45

such as reef herbivores, become new targets (Bender et al. 2014) when apex predatory fish species

46

become economically unfeasible (Pauly et al. 1998).

AC C

EP

TE D

38

47

It is then important to understand the consequences of not having this type of species in the

48

environment anymore, especially if they are also key species, and to address management solutions

49

for their exploitation. Herbivores play an important functional role in the ecosystem, especially in

50

reef environments, and their role can be easily disrupted if their numbers go slightly down (Cheal et

51

al. 2010), which is achievable even by relatively low fishing pressure. The overfishing of

2

ACCEPTED MANUSCRIPT

52

herbivores, a more extreme situation, has been shown to simultaneously impact multiple ecosystem

53

functions (Bellwood et al. 2012), resulting in several negative consequences for coral reef

54

ecosystems (Lokrantz et al. 2009), including phase shifts and loss of resilience (Hughes et al. 2007;

55

Mumby et al. 2007). Parrotfish (Labridae: Scarini) are among the most notable and dominant group of herbivore

57

fish in tropical and subtropical reef habitats around the world both in numbers of individual fish and

58

in biomass (Choat and Randall, 1986; Francini-Filho and Moura 2008). This group of fish is

59

classified into three functional groups, excavators, scrapers, and grazers (Streelman et al. 2002;

60

Francini-Filho et al. 2008), and plays key functional roles in coral reef ecosystems (Hughes 1994;

61

Bellwood et al. 2003). As consumers of benthic algae, parrotfish directly affect the structure and

62

composition of benthic communities through the top-down control of macroalgae, which supports

63

coral survival, growth, and recruitment (Mumby et al. 2006).

M AN U

SC

RI PT

56

Fishing pressure on parrotfish has grown in the last decades around the world (Edwards et al.

65

2014). This increasing exploitation has taken place mainly in areas of decreasing catches of top

66

predators (Ferreira and Gonçalves 1999), such as the Caribbean (Hughes 1994) and the

67

southwestern Atlantic (Francini-Filho and Moura 2008; Bender et al. 2014). In Brazil, parrotfish

68

have also been increasingly exploited (Francini-Filho et al. 2008) with many of them already

69

showing signs of depletion (Bender et al. 2014). Some are emblematic species, such as Scarus

70

trispinosus (Valenciennes, 1840), which in addition to being the largest Brazilian parrotfish is also

71

an endemic species (Padovani-Ferreira et al. 2012). These large-bodied and long-living fish have

72

become a favorite spearfishing target in the last decade, which explains its sharp population decline

73

along the Brazilian coast (Floeter et al. 2006; Francini-Filho et al. 2008; Bender et al. 2014). The

74

same fate has occurred with other large-bodied parrotfish in Brazil, such as the Scarus zelindae,

75

Sparisoma amplum, Sparisoma axillare, and Sparisoma frondosum (Francini-Filho et al. 2008).

76

Four out of these five species have recently been classified as vulnerable, including Sparisoma

AC C

EP

TE D

64

3

ACCEPTED MANUSCRIPT

77

axillare and S. frondosum, while Scarus trispinosus has been listed as endangered in Brazil.

78

Therefore, the economic interest in their exploitation associated with the recent conservation

79

concerns highlight the need to have basic information on parrotfish to guide conservation and

80

management actions. Our study aimed to identify sensitive habitats for three parrotfish species, the greenback

82

parrotfish (Scarus trispinosus, Valenciennes, 1840), the agassiz’s parrotfish (Sparisoma frondosum,

83

Agassiz, 1831), and the gray parrotfish (Sparisoma axillare, Steindachner, 1878), along the

84

Brazilian coast and to develop probabilistic spatial scenarios as effective tools to support decision

85

making within the conservation framework. Therefore, we analyzed data on the presence of these

86

parrotfish in commercial fish landings in the Brazilian northeast. Using the hierarchical Bayesian

87

spatial-temporal models with a Species Distribution Model approach (models containing biotic or

88

accessibility predictors and/or being spatially limited) (Soberon and Peterson 2005; Soberón 2010),

89

we predicted the presence of these species for the entire Brazilian coast.

M AN U

SC

RI PT

81

Because S. axillare and S. frondosum are browser/scraper species (Ferreira and Gonçalves

91

2006) that have strategies that allow the use of different types of habitats (Ferreira and Gonçalves

92

1999), we expected them to be distributed along a wide area of the Brazilian coast. On the other

93

hand, we expected a narrower distribution of Scarus trispinosus, which is an excavator (Ferreira

94

and Gonçalves 2006) and hence requires more specific food items. This approach underlines the

95

importance of having marine-protected areas that consider not only species diversity but also

96

functional biodiversity, reducing overall fish assemblage vulnerability (Parravicini et al. 2014).

97

Also, the approach applied in this study improves the understanding of species distribution and

98

demonstrates the biological needs for the management of any biologically and economically

99

important reef species under some sort of threat.

AC C

EP

TE D

90

100 101

Material and Methods

102

Study area 4

ACCEPTED MANUSCRIPT

Parrotfish were sampled in fish landings with fishing activities that took place up to three

104

miles offshore within a Regional Protected Reef Area, hereafter referred to as APARC (from

105

Portuguese, Área de Proteção Ambiental dos Recifes de Corais, see Fig.1). This protected area was

106

established in 2001 in the northern part of the oriental coast of the Rio Grande do Norte state, Brazil

107

(5º00’5º30’S – 35º10’35º30W). The entire area covers 180.000ha and is characterized by complex

108

rocky reefs formed by various types of benthic environments (rhodolith, prairies of phanerogams,

109

corals, algae, and sandy fund) (Amaral et al. 2005).

RI PT

103

The small-scale fishing activities in this area have a high heterogeneity of gear and targets and

111

are performed by about 700 fishers. Most of these fishers use small sail vessels (4-6m) that are also

112

powered by small engines. These vessels employ various types of fishing gears, but bottom-set

113

panel gill nets with stretched mesh sizes that range from 30 to 60mm (knot-to-knot), handlines, and

114

spear guns are primarily used.

M AN U

SC

110

The fishing gear changes throughout the year according to the seasonality of the main target

116

species. Red snapper (Lutjanus analis), dog snapper (Lutjanus jacu), black grouper (Mycteroperca

117

bonaci), king mackerel (Acanthocybium solandri), Spanish mackerel (Scomberomorus brasiliensis),

118

and Atlantic little tunny (Euthynnus alletteratus) are caught year-round mainly using handlines.

119

Gillnets are primarily used to catch ballyhoo (Hemiramphus brasiliensis), manjuba (Anchoviella

120

lepidentostole), lane snapper (Lutjanus synagris), and yellowtail snapper (Ocyurus chrysurus),

121

usually from December to April. Octopus (Octopus insularis) is also targeted yearlong, although

122

usually between December and May, by free divers using a long hook.

124

EP

AC C

123

TE D

115

Data sampling

125

Parrotfish landings were registered monthly in the Maracajaú and the Rio do Fogo harbors

126

from September 2013 to August 2014. These locations were previously identified as the most

127

frequent places for parrotfish fishing. The sampling was performed for two consecutive days (from

128

approximately 10:00am to 4:00pm) in each place every month. Harbor observers recorded 5

ACCEPTED MANUSCRIPT

129

information about fishing grounds, fishing gear, crew size, date, duration of the fishing operation,

130

and the species caught. Fishers provided information regarding the fishing grounds using a geo-referred spatial grid

132

(1°x1°) of the area. The latitude and longitude of each fishing operation was extracted a posteriori

133

using the “Feature to Points” tools of the ArcGis 10.2.2 version (ESRI, 2014, link:

134

http://www.esri.com/).

RI PT

131

Fishing trips that resulted in no parrotfish being caught were not available in the original

136

dataset because only fishery operations that landed parrotfish were sampled. In order to generate a

137

presence/absence dataset, pseudo-absences were chosen at random for the entire area, considering

138

both expert and bibliographic knowledge (Bonaldo et al. 2006; Francini-Filho et al. 2008; Francini-

139

Filho and Moura 2008) of the habitat range of the species. Such information was used to restrain the

140

areas in which to generate these absences. We generated the set of pseudo-absences 100 times using

141

the srswor function (simple random sampling without replacement) from the “Sampling” package in

142

R (R Development Team 2014). In each case, the number of generated pseudo-absences was the

143

same as the number of real presences. Pseudo-absences were then combined with real presences

144

into a single presence-absence dataset to be used in a binomial model.

TE D

M AN U

SC

135

Although the generation of pseudo-absence datasets is an open hot research topic, (e.g. Hastie

146

and Fithian 2013), subjected to some debate, we preferred to use such a binomial distribution with a

147

Bayesian spatial model instead of a less accurate model that allows the use of presence-only data

148

(e.g. BIOCLIM, DOMAIN). The advantage of using a Bayesian approach, instead of other models,

149

is, among other advantages mentioned in the next section, the capacity to include a spatial effect to

150

deal with spatial autocorrelation. Spatial autocorrelation should be taken into account in the species

151

distribution models, even if the data were collected in a standardized sampling, since the

152

observations are often close to each other and, therefore, subjected to similar environmental

153

features.

AC C

EP

145

6

ACCEPTED MANUSCRIPT

154 155

Environmental variables Potential fixed-effects have been considered for each of the models for each species (the

157

modeling approach is described in the modeling section). As environmental variables, we included

158

bathymetry (mean depth of each fishing operation), distance to shore (km.), slope of the seabed (m.),

159

monthly mean of Sea Surface Temperature (SST in oC), and monthly mean of Net Primary

160

Productivity (NPP in mg m-3).

RI PT

156

The sea surface temperature (SST) was extracted from NODS_WOA94 long-term monthly

162

mean climatology provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA (at

163

http://www.esrl.noaa.gov/psd/).

M AN U

SC

161

164

The variable net primary productivity (NPP) was retrieved from 920x1680 global grids of NPP

165

and calculated as a function of chlorophyll, available light, and photosynthetic efficiency using the

166

entire

167

(http://www.science.oregonstate.edu/ocean.productivity/index.php).

chlorophyll

record

on

the

Ocean

Productivity

website

TE D

SeaWIFS

The bathymetry was retrieved from the NOAA-NGDC ETOPO1 Global Relief Model (Amante

169

and Eakins 2009), a 1 arc-minute global relief model of Earth’s surface that integrates land

170

topography and ocean bathymetry. In addition, fishers provided the bathymetry information at the

171

moment of the landing sampling. When the data differed between the NOAA and the fishers’

172

information, a mean of the two datasets was used. We opted for the mean because we did not know

173

a priori which of the two datasets was a more reliable source of data. However, only 15 points out

174

of 54 differed between the datasets and this difference was only about 4-5 meters.

AC C

EP

168

175

The distance to the coast and slope information were derived from the bathymetry map using the

176

Near (World Equidistant Cylindrical coordinate system) and Slope Spatial Analyst tools of ArcGis

177

9.2 (ESRI Inc., 2008; Redlands, California, USA).

7

ACCEPTED MANUSCRIPT

The remaining potential source of variation on the species occurrence dataset could be due to the

179

fishers sampled, who could have been sampled by us multiple times if they happened to land more

180

than once during the sampling year. The individual behavior of fishers caused by random aspects,

181

such as experience, age, and social needs, and/or unobserved gear characteristics could have caused

182

some of the variation in the data. Ignoring such non-independence of the data may lead to an invalid

183

statistical inference. To remove any bias caused by fishers-specific differences in the fishing

184

operation, a fisher effect was included. This variable was included as a random effect because there

185

was no interest in knowing the specific nature of the observed fishers.

SC

RI PT

178

Environmental variables were standardized and collinearity between explanatory environmental

187

variables was checked using a draftsman’s plot and the Pearson correlation index. The variables

188

were not highly correlated (r < 0.5), so all variables were considered in further analyses.

189 190

Modeling species occurrence

M AN U

186

Spatial distribution models use the range of sampled environments and the same general time

192

frame of the sampling to predict quantities of interest at un-sampled locations based on measured

193

values at nearby sampled locations. Many spatial distribution model algorithms can be used to

194

predict the spatial distribution of species; however, these algorithms do not always provide accurate

195

results if they are run using traditional prediction methods (frequentist inference) because there is

196

often a large amount of variability in the measurements of response and environmental variables.

197

To solve this problem, we used the hierarchical Bayesian spatial-temporal models.

AC C

EP

TE D

191

198

Bayesian methods have several advantages over traditional ones and are increasingly used in

199

fisheries (Colloca et al. 2009, Muñoz et al. 2012, Pennino et al. 2014). In fact, they provide a more

200

realistic and accurate estimation of uncertainty due to the possibility of using both the observed data

201

and the model parameters as random variables (Banerjee et al. 2004). Bayesian methods also allow

202

for the incorporation of the spatial component as a random-effect term, thereby reducing its 8

ACCEPTED MANUSCRIPT

203

influence on estimates of the effects of geographical variables (Gelfand and Silander 2006).

204

Specifically, by treating the spatial effect as a variable of interest, hierarchical Bayesian spatial

205

models can identify additional covariates that may improve the model fit or the existence of area

206

effects that may affect the density of recruits. We used the hierarchical Bayesian spatial-temporal approach, specifically a point-reference

208

spatial model, to estimate the probability of occurrence of the three parrotfish caught at the sample

209

site (Sparisoma axillare, S. frondosum and Scarus trispinosus). These models are highly suitable for

210

cases (as presented here) in which data are observed at continuous locations occurring within a

211

defined spatial domain. These models can also be considered to be a spatial extension of

212

Generalized Linear Models (GLMs) because the modeling process describes the variability in the

213

response variable as a function of the explanatory variables with the addition of a stochastic spatial

214

effect to model the residual spatial autocorrelation.

M AN U

SC

RI PT

207

Specifically, once we have generated the presence-pseudo-absence dataset, the response variable

216

Zij is a binary variable that represents the presence (1) or absence (0) of the species in each fishing

217

location identified in the fishing landings. Consequently, the conditional distribution of the data is

218

Zij ~ Ber(πij), which is the probability of occurrence at location i (i = 1,....n) and month j (j =

219

1,....12), assuming that observations are conditionally independent given πij. At the first stage of the

220

hierarchical model, the observed data (occurrence of species) was modeled as a GLM (Generalized

221

Linear Model) by using the logit link function for binary data but also incorporating one spatial and

222

one possible temporal effect. That is,

EP

AC C

223

TE D

215

Zi j~ Ber (πij)

224

logit (πij) = Xijβ + Yj + Zk + Wi

225

where β is the vector of regression parameters, Xij is the vector of the explanatory covariates at the

226

month j and at the location i, Yj is the component of the temporal random effect at the month j, Zk 9

ACCEPTED MANUSCRIPT

is the random effect of the fishers, and Wi represents the spatially structured random effect at the

228

location i. This model assumes independence between the data; however, the geographical location

229

introduces some correlation since the occurrence of a species at nearby locations is influenced by

230

similar environmental factors. Hence, nearby locations should show similar occurrences of the

231

studied species. The spatial effect, Wi, should account for this influence. Note that this modeling

232

process describes the variability in the response variable as a function of the explanatory variables

233

with the addition of a stochastic spatial effect, which models the residual spatial autocorrelation.

RI PT

227

For all of the parameters involved in the fixed-effects, a vague Gaussian prior distribution (β ~ N

235

(0.0.01)) was used, and thus the posterior distributions for all parameters were derived from the

236

data. For the spatial effect, a prior Gaussian distribution with a zero mean and covariance matrix

237

was assumed depending on the hyperparameters k and τ, which determined the range of the spatial

238

effect and the total variance, respectively (see Muñoz et al., 2013 for more detailed information

239

about spatial effects). Moreover, for the temporal component, a LogGamma prior distribution was

240

assumed on the log-precision λy (a = 1, b = 5e-05).

TE D

M AN U

SC

234

As usual in this context, the resulting hierarchical Bayesian model has no closed expression for

242

the posterior distribution of all of the parameters, and numerical approximations are needed. In this

243

study, an Integrated Nested Laplace Approximation (INLA) methodology (Rue et al. 2009) and

244

software (http:\\www.r-inla.org) were used as alternatives to the Markov chain Monte Carlo

245

methods. The main reason for this choice is the speed of calculation: MCMC simulations require

246

much more time to run, and performing predictions are so far practically unfeasible. In contrast,

247

INLA produces almost immediate accurate approximations to posterior distributions even in

248

complex models. Another advantage of this approach is its generality, which makes it possible to

249

perform Bayesian analysis in a straightforward way and to compute model comparison criteria and

250

various predictive measures so that models can be compared easily (Rue et al. 2009). INLA's

251

performance has been compared with MCMC and has shown a similar reliability (Held et al. 2010).

AC C

EP

241

10

ACCEPTED MANUSCRIPT

Both inference and prediction were performed simultaneously using the Stochastic Partial

253

Differential Equation module (Lindgren et al. 2011), which allows for fitting the particular case of

254

continuously indexed Gaussian fields with INLA, as is the case with this study’s spatial component.

255

The variable selection was performed beginning with all possible interaction terms based on the

256

Deviance Information Criterion (DIC) (Spiegelhalter et al. 2002) and on the cross validated

257

logarithmic score (LCPO) measure (Roos and Held 2011). Specifically, DIC was used as a measure

258

for goodness-of-fit, while LCPO was used as a measure of the predictive quality of the models. DIC

259

and LCPO are inversely related to the compromise between fit, parsimony, and predictive quality.

SC

RI PT

252

In addition, in order to avoid extrapolating the predictions to outside the range of the

261

environmental conditions encountered in the two datasets used for the models, we only predicted

262

probability of presence in areas for which the environmental values were contained in the 90%

263

quantile interval represented by the original datasets.

M AN U

260

264

Model validation

TE D

265

The model validation was performed through an internal 10-fold cross validation based on

267

randomly selected training and test datasets created by a random selection of 80% and 20% of each

268

dataset, respectively (Fielding and Bell 1997), using the ‘PresenceAbsence’ package in R (Freeman

269

and Moisen 2008). The model performance was assessed using the area under the receiver-

270

operating characteristic curve (AUC) (Fielding and Bell 1997) and the “True Skill Statistic” (TSS)

271

(Allouche et al. 2006).

AC C

EP

266

272

AUC has been widely used in the species distribution modeling literature (Elith et al. 2006). It

273

measures the model’s ability to discriminate between sites in which the species is present and those

274

in which the species is absent. The values for AUC range from 0 to 1, where 0.5 indicates a

275

performance no better than random, values between 0.7-0.9 are useful to indicate results of

276

presence/absence different from random, and values > 0.9 are excellent to ensure that the results are 11

ACCEPTED MANUSCRIPT

277

different from random. TSS corrects AUC for the dependence of prevalence on specificity (i.e.,

278

ability to correctly predict absences) and sensitivity (i.e., ability to correctly predict

279

presence)(Allouche et al. 2006). In addition, the information available on the presence of the parrotfish species for all the

281

prediction area was extracted from the AQUAMAPS website (Kaschner et al. 2013) in order to

282

verify whether or not the Bayesian models’ predictions matched theirs. In particular, the datasets

283

extracted from AQUAMAPS were used as secondary test datasets to compute the AUC and TSS

284

measures. The sampled datasets were restricted to the coastal Northeastern area of Brazil, while the

285

Bayesian models predicted a broader and more complete view of the distribution of this species on a

286

macro scale (Fig. 2). Hence, the dataset from AQUAMAPS worked as a control for the predictive

287

ability of the Bayesian approach.

288 289

Results

M AN U

SC

RI PT

280

Fifty-four parrotfish landings were sampled in the two studied locations, although most of them

291

landed in Rio do Fogo (N =40). Both Sparisoma frondosum and S. axillare were recorded in 37

292

landings comprising 1.3 tons caught. Almost one ton of Scarus trispinosus was recorded in 26 of

293

these landings. The annual total catch was estimated to be 9.4 tons of S. trispinosus and 15.4 tons of

294

S. frondosum and S. axillare. This estimation was computed as an extrapolation of the recored

295

landings for the fishing days and the number of fishers in the locations.

AC C

EP

TE D

290

296

At the time of the study, there were 60 parrotfish fishers using gillnets, spear guns, or

297

handlines. Most of them (60%) used gillnets and handlines to catch Sparisoma axillare and S.

298

frondosum. Spearfishing was used by 40% of the fishers specifically to catch Scarus trispinosus.

299

Parrotfish fishing operations were done in sailboats in reef environments ranging from 4 to 12m.

300

The fishing trips lasted from 4 to 8 hours beginning early in the morning, mainly between December

301

and May. The main predictors for the occurrence of parrotfish in the Northeastern Brazilian coast

12

ACCEPTED MANUSCRIPT

302

were bathymetry, sea surface temperature (SST), and distance to the coast. The higher net primary

303

production was only relevant to explain the distribution of Scarus trispinosus (Table 1). The random spatial effect and the random effect representing the behavior of fishers were

305

relevant for all models and species, while slope and month did not affect the variability of parrotfish

306

distribution.

RI PT

304

The best fitted model of the distribution for Scarus trispinosus showed a positive relationship

308

with bathymetry (posterior mean = 4.66; 95% CI = [2.75, 6.99]), sea surface temperature (posterior

309

mean = 2.74; 95% CI = [1.92, 5.13]), and net primary productivity (posterior mean = 2.53; 95% CI

310

= [1.28, 3.95]; Table 1). It was also observed that the median posterior probability of the occurrence

311

of Scarus trispinosus on the Brazilian coast was higher in lower latitudes where higher values of sea

312

surface temperature and net primary productivity were found (Fig. 3A). Conversely, the expected

313

presence of this species decreased as the distance to the coast increased (posterior mean = -1.22;

314

95% CI = [-1.20, -0.05]).

M AN U

SC

307

The spatial effect indicating the intrinsic variability of the distribution of Scarus trispinosus after

316

the exclusion of the environmental variables was consistent with the probability maps (Fig. 4A).

317

This effect showed (through the posterior mean) a large hot spot for this species on the northern

318

coast of Brazil and at specific locations (namely latitude 3°51'S and longitude 33°49'W, and latitude

319

20o29-32'S and longitude 29o17-21'W). These locations correspond with current biological reserves:

320

Atol das Rocas and the Trindade islands (Fig. 4A). Habitats associated with deeper and warmer

321

waters and at lower latitudes showed a greater probability for the presence of both species of

322

Sparisoma.

AC C

EP

TE D

315

323

In particular, the presence of Sparisoma frondosum was positively affected by bathymetry

324

(posterior mean = 6.55; 95% CI = [5.84, 8.56]) and sea surface temperature (posterior mean = 4.65;

325

95% CI = [3.67, 7.59]) but negatively influenced by a decrease in distance, indicating that the

326

further away from the coast, the higher the probability of occurrence of this species (posterior mean

13

ACCEPTED MANUSCRIPT

= -1.64; 95% CI = [-2.20, -0.03]) (Table 1). For this species, the median posterior showed a higher

328

probability of occurrence in deeper waters and oceanic islands along the Brazilian coast (Fig. 3B).

329

This distribution was confirmed by the spatial effect, which identified hot spots for their occurrence

330

in Parcel Manuel Luís, Fernando de Noronha, Atol das Rocas, Abrolhos Archipelago and Trindade,

331

and all oceanic islands or islets located at lower latitudes (Fig. 4B).

RI PT

327

Similarly, the prediction of occurrence of Sparisoma axillare indicated positive effects of

333

bathymetry (posterior mean = 11.34; 95% CI = [9.83, 15.81]) and sea surface temperature (posterior

334

mean = 3.63; 95% CI = [2.02, 4.9]) and a negative effect of distance from the coast (posterior mean

335

= -1.54; 95% CI = [-1.12, -0.02]; Table 1). Figure 3(C) and Figure 4(C) show that habitats

336

associated with deeper waters that are further away from the coast have a higher probability of

337

occurrence of Sparisoma axillare. Lower probabilities for the occurrence of this species were found

338

at higher latitudes.

M AN U

SC

332

The analyses of the model prediction performances found AUC values greater than 0.70 for all

340

models, indicating a good degree of discrimination between locations in which the species was

341

present and those in which it was absent (Table 2). All TSS values ranged between 0.60 and 0.80,

342

which also represents a good ability of the model to predict true negative and true positive

343

predictions. The values of AUC and TSS found using the AQUAMAPS datasets (macro-scale)

344

achieved lower values than the ones obtained using the Bayesian approach (Table 2); however, both

345

AUC and TSS showed values higher than 0.65, indicating predictions of good performance on a

346

macro-scale as well.

EP

AC C

347

TE D

339

348

Discussion

349

Environmental predictors of species distribution

350

Fishery-dependent data were used to improve the understanding of the distribution of three

351

different species of parrotfish of conservation concern on the entire Brazilian coast using Bayesian

14

ACCEPTED MANUSCRIPT

spatial-temporal hierarchical models. It was shown that depth was an important determinant for the

353

occurrence of the studied parrotfish. Scarus trispinosus was likely to be distributed in shallower

354

waters closer to the shore, while the other two species of Sparisoma were more likely to be

355

distributed in deeper waters but with increasing probabilities at increasing distances from the coast.

356

The distribution of all three species was negatively related to the latitude, showing that the spatial

357

effect is a relevant factor in their distribution.

RI PT

352

Despite the limited information available, Scarus trispinosus has been described as a species

359

with a more restricted distribution, establishing populations mostly near the coast. The positive

360

relationship between Scarus trispinosus and primary production found in this study may explain the

361

more restricted distribution of this species due to its specialized feeding habits (Bernardi et al.

362

2000, Streelman et al. 2002), which results in intense grazing that enhances the productivity of

363

benthic algal assemblages by removing large algae with low specific productivity (Hatcher 1988).

364

In addition, the distribution prediction shown in this study supports the patterns of a decreasing

365

occurrence at higher latitudes, which is explained by the fact that the tropics sustain higher primary

366

productivity (Carpenter 1986, Hatcher 1988).

TE D

M AN U

SC

358

Conversely, Sparisoma axillare and S. frondosum have been identified as species that occur

368

over a deeper geographical range, mainly in the Brazilian oceanic islands (Moura et al. 2001). These

369

descriptions are confirmed by the results found in this study regarding depth.

EP

367

Similarly, discussions on the relevance of the spatial factor on herbivore fish species

371

distribution have already shown that at high latitudes, richness and abundance decrease, mainly due

372

to the evolutionary histories of different species and habitat features, such as primary production

373

and temperature (Ferreira et al. 2004, Floeter et al. 2005). The positive relationship between

374

parrotfish species and temperature result from their temperature tolerance limits (Moyle and Cech

375

2003). This illustrates the dependence of their digestion process on a thermally sensitive, diverse

376

assemblage of intestinal microflora, indicating that the distribution patterns of herbivorous fish are

AC C

370

15

377

ACCEPTED MANUSCRIPT

driven by temperature-related feeding and digestive processes (Floeter et al. 2005).

378 379

Spatial predictors for species distribution Identifying the spatial distribution of threatened marine fish populations is essential for

381

developing appropriate marine spatial planning (Douvere 2008) and management measures, such as

382

the establishment of marine protected areas (Jones 2004). The understanding of spatial distribution

383

is also useful in defining essential fish habitats (Rosenberg et al. 2000), mainly for reef fish, which

384

have patchy distributions (Sale 1998). For parrotfish, this is particularly important because despite

385

their growing relevance for fisheries, little is known about their spatial ecology.

SC

RI PT

380

The spatial random effects of the three species showed different spatial patterns observed in

387

the maps of the spatial effect. This could be due to the fact that the spatial random component is

388

often used to capture the effect of important missing predictors (Dormann 2007, Dormann et al.

389

2007) or to account for ecological processes (e.g., dispersal or aggregative behavior) (Merow et al.

390

2014). In our case, the spatial effect probably reflects the type of seabed because it identified areas

391

in which there are known reef complexes that could be the potential habitats of these species.

TE D

M AN U

386

The Brazilian northeast and the northern portion of the southeastern coast were shown to be

393

areas with a positive spatial effect for the three species of parrotfish, mainly in regions with multiple

394

reef complexes. Hence, it would be advisable to establish spatial planning that ensures the

395

protection of such sensitive areas and perhaps to increase the protection in surrounding areas, such

396

as the major coral reef hot spot of Archipelago of Abrolhos.

AC C

EP

392

397

A species habitat analysis should be able to identify the areas within the distribution of a

398

species that contribute most to sustaining the long-term viability of a population. Although it may

399

be complicated to define the boundaries of sensitive habitats, the identification of these areas

400

combined with efficient fishery management recognizing the importance of such areas represents

401

the first step toward the conservation of these species; however, accuracy is not always easy to

16

ACCEPTED MANUSCRIPT

achieve because there is often a large amount of variability in the measurements of the response and

403

environmental variables (Latimer et al. 2006). This variability leads to uncertain predictions and,

404

consequently, to uniformed decision making. It is therefore important to develop and apply tools,

405

such as the statistical approach used in this study, that account for measurements with significant

406

variability.

RI PT

402

407 408

Fishers’ behavior as a predictor of species distribution

The results show that the random effect of fishers’ behavior was significant to explain part of the

410

variability in the distribution of the three species that is not explained by the environment. As

411

expected, this shows that some fishers have more success than others when exploiting parrotfish.

412

This could be due to their experience, the fishing gear used, the fishing ground, or some joint effect

413

of all of these factors.

M AN U

SC

409

More importantly, it is clear that fishers’ personal experience and choices will have an effect on

415

how fish are harvested. Therefore, in addition to establishing protected areas, fishery management

416

should focus on measures that regulate fishing operations, such as temporary closures to assure fish

417

reproduction and restrictions on non-selective fishing gear in unprotected places (McClanahan and

418

Mangi 2004). This is especially urgent given the already threatened conservation status of the three

419

studied species.

421

EP

AC C

420

TE D

414

Dataset limitations, statistical reliance, and ecological findings

422

Large-scale predictions, such as the entire Brazilian coast, allow for broader and more

423

complete views of ecosystem status, but their use often leads to some degree of compromise in the

424

analysis. This is because the quality and quantity of data available for large ecosystems and long

425

time-series are often insufficient, while mapping essential habitats requires the highest level of

426

accuracy. The Bayesian interpolation and extrapolation are sufficiently reliable for the purpose of

427

effective decision making and are supported by a range of evaluation criteria that demonstrate a 17

ACCEPTED MANUSCRIPT

good predictive performance of this approach as well as its advantages in terms of ecologically

429

interpretability. The data extracted from AQUAMAPS confirmed our large-scale extrapolations for

430

the entire Brazilian coast because the observed presences in the AQUAMAPS dataset matched the

431

predicted presences of the Bayesian models with a high level of accuracy, as demonstrated by the

432

AUC and TSS indexes.

RI PT

428

The analyses performed in this study also confirmed the suitability of the existing marine

434

protected areas, such as Parcel Manuel Luís, Atol das Rocas, Fernando de Noronha, Abrolhos

435

Archipelago and Trindade, reinforcing the ecological accuracy of the method. The Bayesian spatial-

436

temporal approach identified essential habitats along with a full specification of associated

437

uncertainty, therefore improving knowledge about the habitat of the parrotfish species along the

438

Brazilian coast and providing practical tools for reef conservation management. However, because

439

both species and environmental data were sampled over a limited period of time and space, the

440

models fitted here can only reflect a snapshot of the expected relationship. Future studies should

441

compare the spatial distribution of these species with more detailed data from fishery-independent

442

surveys (i.e., visual census abundance data), which is often considered to be a more reliable

443

abundance index because of its scientifically rigorous design.

TE D

M AN U

SC

433

Despite these issues, this first approximation for an area in which in-depth information is not

445

available at the time could direct preliminary choices of areas to be managed in addition to helping

446

direct efforts in data collection and future research.

448

AC C

447

EP

444

Threatened species and management suggestions

449

The restricted coastal distribution of Scarus trispinosus is worrisome due to the fact that small-

450

scale fisheries operate mostly in shallow reef areas where this species is present, making it highly

451

susceptible to fisheries. In fact, the decline of Scarus trispinosus is currently observed along the

452

Brazilian coast, mainly in areas with heavy fishing pressure (Floeter et al. 2006, Francini-Filho and

18

ACCEPTED MANUSCRIPT

Moura 2008). In some parts of the southeastern Brazilian coast, this species is already considered

454

ecologically extinct (Floeter et al. 2007), and the same fate has the potential to occur in the

455

northeastern coast as well. Similarly, Sparisoma axillare and S. frondosum have also declined in

456

abundance in some places on the Brazilian southeast coast (Bender et al. 2014) and have been

457

intensively caught on the northeast coast by trap fishery, including trap fishery for exportation

458

(Ribeiro 2003).

RI PT

453

For such a large coastline, protection measures must be fishery-specific and use a spatial

460

approach. In the northeastern region, for example, parrotfish fisheries have some sort of natural

461

closures from June to November due to the low underwater visibility (it affects spearfishing, but not

462

trap fishing) and the open season for the more profitable lobster, which attracts most fishers. On the

463

other hand, when lobster fishing is officially closed, parrotfish can suffer a much higher fishing

464

pressure. This exemplifies how the protection of one species (lobster) may increase the pressure

465

upon another. Moreover, it illustrates that local regulations on gears could be more suitable and

466

likely to promote compliance than a fishing ban period since fishers already experience four months

467

of natural interruption in parrotfish exploitation.

TE D

M AN U

SC

459

The recent inclusion of S. trispinosus as an endangered species and of Sparisoma axillare and S.

469

frondosum as vulnerable species in the Brazilian Red List of Threatened Species implies an

470

automatic banning of the fishery for the first species and the requirement of management measures

471

for the latter two over the entire coast. However, Brazil lacks the financial and human resources to

472

implement a de facto moratorium and/or management. The Brazilian coast is simply too large an

473

area with too many scattered fishing communities that would probably not even be properly

474

informed of such measures. Also, some of these communities may partially rely on the exploitation

475

of these species. Therefore, an effective permanent or temporary moratorium and management

476

actions on parrotfish harvesting would have a better potential of being effective if coupled with

477

programs that investigate and address the livelihood needs of the affected communities. Otherwise,

AC C

EP

468

19

ACCEPTED MANUSCRIPT

478

parrotfish harvesting could become another unreported, unregulated, and illegal fishery. Different

479

places around the world have faced similar dilemmas and have developed solutions, such as the

480

diversification of economic alternatives that support biodiversity conservation (e.g., parrotfish

481

ecotourism) (Tallis et al. 2010). Based on the results presented, we recommend the establishment of additional protected areas in

483

which the larger populations of the three studied parrotfish species are mostly expected to occur.

484

For some areas, it appears to be reasonable to allow tourism activities, such as parrotfish viewing.

485

This may be a feasible and liable opportunity primarily for S. trispinosus, which is predicted to

486

occur in shallow waters around the coral reef area. However, we also emphasize the need to address

487

the impacts such measures may have on local communities’ livelihoods and the need to mitigate

488

possible negative consequences. The identification of the occurrence areas of target species in

489

general could be an essential tool to direct the spatial management of the fleet as well as to direct

490

additional studies to detect nurseries and high-abundance areas.

M AN U

SC

RI PT

482

TE D

491

Acknowledgements

494

Authors would like to thank all the fishermen from the two localities for their kind support and

495

patience. NR and MGP received financial support from CNPq and CAPES, as a Master’s and Post-

496

Doctoral Grant (88881.030502. 2013-01), respectively.

AC C

497

EP

492 493

498

References

499 500 501

Allouche, O., Tsoar, A. & Kadmon, R. (2006) Assessing the accuracy of species distribution models: prevalence, kappa and the true skill statistic (TSS). Journal of Applied Ecology 43:1223–1232.

502 503

Amante, C. & Eakins, B.W. (2009) ETOPO1 1 Arc-Minute global relief model: procedures, data sources and analysis. 25. https://www.ngdc.noaa.gov.

504

Amaral, R.F., Feitoza, B.M., Attayde, J.L., Arantes, S.V., Batista, D.S., Costa Neto, L., Mendes, L., 20

505 506

ACCEPTED MANUSCRIPT

Martins, F.S., Silva, I.B., Soriano, E.M. & Ziggmond, M. (2005) Diagnóstico Ambiental da Área de Uso Turístico Intensivo (AUTI) no Parracho de Maracajaú. 128. Banerjee, S., Carlin, B.P. & Gelfand, A.E. (2004) Hierarchical Modeling and Analysis for Spatial Data. Chapman and Hall/CRC, Boca Raton.

509 510

Bellwood, D.R., Hoey, A.S. & Choat, J.H. (2003) Limited functional redundancy in high diversity systems: resilience and ecosystem function on coral reefs. Ecology Letters 6:281–285

511 512 513

Bellwood, D.R., Hoey, A.S. & Hughes, T.P. (2012) Human activity selectively impacts the ecosystem roles of parrotfishes on coral reefs. Proceedings of the Royal Society B: Biological Sciences 279:1621–1629.

514 515 516 517

Bender, M.G., Machado, G.R., Silva, P.J. de A., Floeter, S.R., Monteiro-Netto, C., Luiz O.J. & Ferreira, C.E.L. (2014) Local Ecological Knowledge and Scientific Data Reveal Overexploitation by Multigear Artisanal Fisheries in the Southwestern Atlantic. PLoS ONE 9:e110332.

518 519 520

Bernardi, G., Robertson, D.R., Clifton, K.E. & Azzurro, E. (2000) Molecular Systematics, Zoogeography, and Evolutionary Ecology of the Atlantic Parrotfish Genus Sparisoma. Molecular Phylogenetics and Evolution 15:292–300.

521 522 523 524 525 526

Bonaldo, R. M., Krajewski, J. P., Sazima, C., & Sazima, I. (2006). Foraging activity and resource use by three parrotfish species at Fernando de Noronha Archipelago, tropical West Atlantic. Marine Biology, 149(3), 423-433.

527 528

Carpenter, R.C. (1986) Partitioning Herbivory and Its Effects on Coral Reef Algal Communities. Ecological Monographs 56:345–363.

529 530 531

Cheal, A.J., MacNeil, M.A., Cripps, E., Emslie, M.J., Jonker, M., Schaffelke, B. & Sweatman, H.(2010) Coral–macroalgal phase shifts or reef resilience: links with diversity and functional roles of herbivorous fishes on the Great Barrier Reef. Coral Reefs 29:1005–1015 .

532 533 534

Choat, J.H. & Randall, J.E. (1986) A Review of the Parrotfishes (Family Scaridae) of the Great Barrier Reef of Australia with Description of a New Species. Records of the Australian Museum, 38:175–228.

535 536 537

Colloca, F., Bartolino, V., Lasinio, G.J., Maiorano, L., Sartor, P. & Ardizzone, G. (2009) Identifying fish nurseries using density and persistence measures. Marine Ecology Progress Series 381:287–296.

538 539 540

Di Dario, F., Alves, C.B.M., Boos H., Frédou F.L., Lessa, R.P.T., Mincarone, M.M., Pinheiro, M.A.A., Polaz, CNM, Reis, RE, Rocha, LA, Santana, FM, Santos, RA, Santos, SB, Vianna, M. & Vieira, F. (2015) A better way forward for Brazil’s fisheries. Science 347:1079.

541 542

Dormann, C.F. (2007) Effects of incorporating spatial autocorrelation into the analysis of species distribution data. Global Ecology and Biogeography Letters 16:129–138.

543

Dormann, C.F., McPherson, J.M., Araújo, M.B., Bivand, R., Bolliger, J., Carl, G., Davies, R.G.,

M AN U

SC

RI PT

507 508

AC C

EP

TE D

Bond, W.J. (1994) Keystone Species. In: Schulze E.-D., Mooney H.A. (eds) Biodiversity and Ecosystem Function. Springer Berlin Heidelberg, pp 237–253.

21

544 545 546 547

ACCEPTED MANUSCRIPT

Hirzel, A., Jetz, W., Kissling, W.D., Kühn, I., Ohlemüller, R., Peres-Neto, P.R., Reineking, B., Schröder, B., Schurr, F.M. & Wilson, R. (2007) Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography 30:609– 628. Douvere, F. (2008) The importance of marine spatial planning in advancing ecosystem-based sea use management. Marine Policy 32:762–771.

550 551 552 553

Edwards, C.B., Friedlander, A.M., Green, A.G., Hardt, M.J., Sala, E., Sweatman, H.P., Williams, I.D., Zgliczynski, B., Sandin, S.A. & Smith, J.E. (2014) Global assessment of the status of coral reef herbivorous fishes: evidence for fishing effects. Proceedings of the Royal Society B: Biological Sciences, 281(1774), 20131835.

554 555 556 557 558 559

Elith, J., Graham, C.H., Anderson, R.P., Dudík, M., Ferrier, S., Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M.M., Peterson, A.T., Phillips, S.J., Richardson, K.W., Scachetti-Pereira, R., Schapire, R., Soberón, J., Williams, S., Wisz & M., Zimmermann, N.E. (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography 29:129–151.

560 561

Ferreira, C.E.L., Floeter, S.R., Gasparini, J.L., Ferreira, B.P. & Joyeux, J. (2004) Structure in Brazilian Reef Fishes: A Latitudinal Comparison. 10th International Coral Reef Symposium.

562 563

Ferreira, C. & Gonçalves, J. (1999) The Unique Abrolhos Reef Formation (Brazil): need for specific management strategies. Coral Reefs 18:352–352.

564 565

Ferreira, C. & Gonçalves, J. (2006) Community structure and diet of roving herbivorous reef fishes in the Abrolhos Archipelago, south-western Atlantic. J. Fish Biol. 69:1533–1551.

566 567

Fielding, A.H. & Bell, J.F. (1997) A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation 24:38–49 .

568 569 570

Floeter, S.R., Behrens, M.D., Ferreira, C.E.L., Paddack, M.J. & Horn, M.H. (2005) Geographical gradients of marine herbivorous fishes: patterns and processes. Marine Biology 147:1435– 1447 .

571 572 573 574

Floeter, S.R., Ferreira, C.E.L. &Gasparini, J.L. (2007) Os Efeitos da Pesca e da Proteção através de UC’s Marinhas: Três Estudos de Caso e Implicações para os Grupos Funcionais de Peixes Recifais no Brasil. Áreas Aquáticas Protegidas Como Instrumento de Gestão Pesqueira. Ministério do Meio Ambiente, Brasília, pp 183–194.

575 576

Floeter, S.R., Halpern, B.S. & Ferreira, C.E.L. (2006) Effects of fishing and protection on Brazilian reef fishes. Biology Conservation 128:391–402.

577 578 579

Francini-Filho, R.B. & Moura, R.L. (2008) Evidence for spillover of reef fishes from a no-take marine reserve: An evaluation using the before-after control-impact (BACI) approach. Fishery Research 93:346–356.

580 581 582

Francini-Filho, R.B., Moura, R.L., Ferreira, C.M. & Coni, E.O.C. (2008) Live coral predation by parrotfishes (Perciformes: Scaridae) in the Abrolhos Bank, eastern Brazil, with comments on the classification of species into functional groups. Neotropical Ichthyology 6(2):191–200.

AC C

EP

TE D

M AN U

SC

RI PT

548 549

22

ACCEPTED MANUSCRIPT

583 584 585 586 587 588 589 590 591 592 593 594

Freeman, E. A., & Moisen, G. (2008). PresenceAbsence: An R package for presence absence analysis. Journal of Statistical Software 23(11):1–31.

595 596 597 598 599 600

Held, L., Schrödle, B., & Rue, H. (2010). Posterior and cross-validatory predictive checks: a comparison of MCMC and INLA. In Statistical modelling and regression structures (pp. 91-110). Physica-Verlag HD.

601 602 603 604

Hughes, T.P., Rodrigues, M.J., Bellwood, D.R., Ceccarelli, D., Hoegh-Guldberg, O., McCook, L., Moltschaniwskyj, N., Pratchett, M.S., Steneck, R.S. & Willis, B. (2007) Phase Shifts, Herbivory, and the Resilience of Coral Reefs to Climate Change. Current Biology 17:360– 365

605 606

Jones, P.J.S. (2004) Marine protected area strategies: issues, divergences and the search for middle ground. Review in Fish Biology and Fisheries 11:197–216

607 608

Kaschner, K.J., Rius-Barile, J., Kesner, Reyes K., Garilao, C., Kullander, S.O., Rees, T., Froese, R. (2013) AquaMaps: Predicted range maps for aquatic species. www.aquamaps.org

609 610 611 612 613 614 615

Katsanevakis, S., Stelzenmüller, V., South, A., Sørensen, T.K., Jones, P.J.S., Kerr, S., Badalamenti, F., Anagnostou, C., Breen, P., Chust, G., D’Anna, G., Duijn, M., Filatova, T., Fiorentino, F., Hulsman H, Johnson, K, Karageorgis, AP, Kröncke I, Mirto S, Pipitone C, Portelli S, Qiu W., Reiss, H., Sakellariou, D., Salomidi, M., van Hoof, L., Vassilopoulou, V., Vega Fernández, T., Vöge, S., Weber, A., Zenetos, A. & Hofstede, R. (2011) Ecosystem-based marine spatial management: Review of concepts, policies, tools, and critical issues. Ocean Coastal Management 54:807–820

616 617

Latimer, A.M., Wu, S., Gelfand, A.E. & Silander, J.A. (2006) Building statistical models to analyze species distributions. Ecological application 16:33–50

618 619 620

Lindgren, F., Rue, H. & Lindström, J. (2011) An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach.ournal of the Royal Statistical Society: Series B (Statistical Methodology). 73:423–498

621 622 623

Lokrantz, J., Nyström, M., Norström, A.V., Folke, C. & Cinner, J.E. (2009) Impacts of artisanal fishing on key functional groups and the potential vulnerability of coral reefs. Environmental Conservation 36:327–337

624 625

McClanahan, T.R. & Mangi, S.C .(2004) Gear-based management of a tropical artisanal fishery based on species selectivity and capture size. Fishery Management and Ecology 11:51–60

Gelfand, A. E., Silander, J. A., Wu, S., Latimer, A., Lewis, P. O., Rebelo, A. G., & Holder, M. (2006). Explaining species distribution patterns through hierarchical modeling. Bayesian Analysis 1(1): 41–92.

RI PT

Hastie, T., & Fithian, W. (2013). Inference from presence‐only data; the ongoing controversy. Ecography, 36(8): 864-867.

SC

Hatcher, B.G. (1988) Coral reef primary productivity: A beggar’s banquet. Trends in Ecology & Evolution 3:106–111.

AC C

EP

TE D

M AN U

Hughes, T.P. (1994) Catastrophes, Phase Shifts, and Large-Scale Degradation of a Caribbean Coral Reef. Science 265:1547–1551.

23

ACCEPTED MANUSCRIPT

Merow, C., Smith, M.J., Edwards, T.C., Guisan, A., McMahon, S.M., Normand, S., Thuiller, W., Wüest, R.O., Zimmermann, N.E. & Elith, J. (2014) What do we gain from simplicity versus complexity in species distribution models? Ecography 37:1267–1281

629 630

Mills, L.S., Soulé, M.E. & Doak D.F. (1993) The Keystone-Species Concept in Ecology and Conservation. BioScience 43:219–224

631 632 633 634

Moura, R.L., de Figueiredo, J.L. & Sazima, I. (2001) A new parrotfish (Scaridae) from Brazil, and revalidation of Sparisoma amplum (Ranzani, 1842), Sparisoma frondosum (Agassiz, 1831), Sparisoma axillare (Steindachner, 1878) and Scarus trispinosus Valenciennes, 1840. Bulletin of Marine Science 68:505–524

635 636

Moyle, PB & Cech JJ (2003) Fishes: An Introduction to Ichthyology. Benjamin Cummings, Upper Saddle River, NJ

637 638 639 640

Mumby, P.J., Dahlgren, C.P., Harborne, A.R., Kappel, C.V., Micheli, F., Brumbaugh, D.R., Holmes, K.E., Mendes, J.M., Broad, K., Sanchirico, J.N., Buch, K., Box, S., Stoffle, R.W. & Gill, A.B. (2006) Fishing, Trophic Cascades, and the Process of Grazing on Coral Reefs. Science 311:98–101

641 642

Mumby, P.J., Hastings, A. & Edwards, H.J. (2007) Thresholds and the resilience of Caribbean coral reefs. Nature 450:98–101

643 644 645

Muñoz, F., Pennino, M.G., Conesa, D., López-Quílez, A. & Bellido, J.M. (2012) Estimation and prediction of the spatial occurrence of fish species using Bayesian latent Gaussian models. Stochastic Environmental Research and Risk Assessment 27:1171–1180

646 647

Myers, R.A., & Worm, B. (2003) Rapid worldwide depletion of predatory fish communities. Nature 423:280–283

648 649 650

Padovani-Ferreira, B., Floeter, S.R., Rocha, L.A., Ferreira, C.E., Francini-Filho, R., Moura, R., Gaspar, A.L. & Feitosa, C. (2012) Scarus trispinosus. http://www.iucnredlist.org/details/190748/0

651 652 653 654

Parravicini, V., Villéger, S., McClanahan, T.R., Arias-González, J.E., Bellwood, D.R., Belmaker, J., Chabanet, P., Floeter, S,R., Friedlander, A.M., Guilhaumon, F., Vigliola, L., Kulbicki, M. & Mouillot, D. (2014) Global mismatch between species richness and vulnerability of reef fish assemblages. Ecological Letters 17:1101–1110

655 656

Pauly, D., Christensen, V., Dalsgaard, J., Froese, R., Torres, F. (1998) Fishing Down Marine Food Webs. Science 279:860–863

657 658

Pennino, M.G., Muñoz, F,. Conesa, D., López-Quίlez, A. & Bellido, J.M. (2014) Bayesian spatiotemporal discard model in a demersal trawl fishery. Journal of Sea Research 90:44–53

659 660

Pennino, M.G., Muñoz, F,. Conesa, D., López-Quίlez, A. & Bellido, J.M. (2013) Modeling sensitive elasmobranch habitats. Journal of Sea Research 83:209–218

661 662

R Development Team (20014) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria

663

Ribeiro, F.P. (2003) Composição da biocenose e abundância relativa de peixes capturados com

AC C

EP

TE D

M AN U

SC

RI PT

626 627 628

24

664 665

ACCEPTED MANUSCRIPT

covos nos estados do Rio Grande do Norte e Pernambuco (Brasil). Boletim Técnico Científico CEPENE 12:113–118 Roos, M. & Held, L. (2011) Sensitivity analysis in Bayesian generalized linear mixed models for binary data. Bayesian Analysis 6:259–278

668 669 670

Rosenberg, A., Bigford, T.E., Leathery, S., Hill, R.L., Bickers, K. (2000) Ecosystem approaches to fishery management through essential fish habitat. Bulletin Marine Sciences 66:535–542

671 672 673

Rue, H., Martino & S., Chopin, N. (2009) Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. Journal of the Royal Statistical Society: Series B (Statistical Methodology) 71:319–392

674 675

Sale, P.F. (1998) Appropriate spatial scales for studies of reef-fish ecology. Australian Journal of Ecology 23:202–208

676 677

Soberón, J.M. (2010) Niche and area of distribution modeling: a population ecology perspective. Ecography 33:159–167

678 679

Soberon, J. & Peterson, A.T. (2005) Interpretation of Models of Fundamental Ecological Niches and Species’ Distributional Areas. Biodiversity Informatics 2:1-10

680 681 682

Spiegelhalter, D.J., Best, N.G., Carlin, B.P. & Van Der Linde, A. (2002) Bayesian measures of model complexity and fit.Journal of the Royal Statistical Society: Series B (Statistical Methodology) 64:583–639

683 684 685

Streelman, J.T., Alfaro, M., Westneat, M.W., Bellwood, D.R. & Karl, S.A. (2002) Evolutionary History of the Parrotfishes: Biogeography, Ecomorphology, and Comparative Diversity. Evolution 56:961–971

686 687 688

Tallis, H., Levin, P.S., Ruckelshaus, M., Lester, S.E., McLeod, K.L., Fluharty, D.L. & Halpern, B.S. (2010) The many faces of ecosystem-based management: making the process work today in real places. Marine Policy 34(2): 340-348

689 690

Tewfik, A. & Béné, C. (2004). “The Big Grab”: non-compliance with regulations, skewed fishing effort allocation and implications for a spiny lobster fishery. Fisheries Research 69(1): 21-33.

691 692

Thrush, S.F. & Dayton, P.K. (2010) What can ecology contribute to ecosystem-based management? Annual Review in Marine Sciences 2:419–41

694 695

SC

M AN U

TE D

EP

AC C

693

RI PT

666 667

696 697 698

25

699

ACCEPTED MANUSCRIPT

Figure Legends

700 701

Figure 1: Map of the study area and sampling locations.

702

Figure 2: Sampling locations of our personal dataset (black circle) and AQUAMAPS (red circle)

704

dataset of the the studied species: Scarus trispinosus (A); Sparisoma frondosum (B); Sparisoma

705

axillare (C).

RI PT

703

706

Figure 3: Median of the posterior probability of the presence of the studied species: Scarus

708

trispinosus (A); Sparisoma frondosum (B); Sparisoma axillare (C) and MPA’s locations: Parcel

709

Manuel Luís (1), Atol das Rocas (2), Fernando de Noronha (3), Abrolhos Archipelago (4) and

710

Trindade Island (5).

SC

707

M AN U

711

Figure 4: Median of the posterior spatial effect of the studied species: Scarus trispinosus (A);

713

Sparisoma frondosum (B); Sparisoma axillare (C) and MPA’s locations: Parcel Manuel Luís (1),

714

Atol das Rocas (2), Fernando de Noronha (3), Abrolhos Archipelago (4) and Trindade Island (5).

AC C

EP

TE D

712

26

ACCEPTED MANUSCRIPT

Table 1: Numerical summary of the posterior distribution of the fixed effects for the best model of the three species studied. This summary contains mean, the standard deviation, the median and a 95% credible interval, which is a central interval containing 95% of the probability under the posterior distribution (SST = Sea Surface Temperature; NPP = Net Primary Production).

Predictor

Mean

SD

Q0.025

Q0.5

Q0.975

S. trispinosus

Intercept

-5.69

1.48

-6.86

-5.85

-3.37

Bathymetry

4.66

1.06

2.75

4.59

6.99

Distance to the coast

-1.22

0.54

-1.20

-0.92

-0.05

SST

2.74

1.12

1.92

2.66

5.13

NPP

0.53

1.42

0.78

0.51

2.95

Intercept

-7.46

1.92

-8.79

-7.33

-4.03

Bathymetry

11.34

2.16

9.83

12.77

15.81

Distance to the coast

-1.64

0.81

-2.20

-1.10

-0.03

SST

4.65

1.43

3.67

4.63

7.59

-3.95

0.78

-4.45

-3.92

-2.53

6.55

0.98

5.84

6.50

8.56

-1.54

0.45

-1.12

-0.98

-0.02

3.63

0.92

2.02

3.56

4.97

S. axillare

Intercept Bathymetry Distance to the coast

AC C

EP

TE D

SST

SC

M AN U

S. frondosum

RI PT

Species

ACCEPTED MANUSCRIPT

Table 2: Model prediction performance statistics for three species studied. AUC (Area Under the receiver-operated characteristic Curve) and TSS (True Skill Statistic). AUCaquamaps and TSSaquamaps indicate the measures obtained using the AQUAMAPS dataset.

AUC

TSS

AUCaquamaps

TSSaquamaps

Scarus trispinosus

0.78

0.68

0.69

0.66

Sparisoma frondosum

0.84

0.72

0.72

0.70

Sparisoma axillare

0.87

0.79

0.71

RI PT

Species

AC C

EP

TE D

M AN U

SC

0.73

AC C

EP

TE

D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT We map the distribution of parrotfish species using a hierarchical Bayesian approach

•

Our results identify sensitive habitats for parrotfish along the Brazilian coast

•

Bathymetry, SST and distance coast were the main predictors for species occurrence

•

The analyses performed confirmed the suitability of the existing Brazilian MPA’s

AC C

EP

TE D

M AN U

SC

RI PT

•