Artigo

1. Título: Sampling Designs for Bottom Trawl Surveys: The Portuguese Autumn Survey Field Experience
2. Novo título: Novo Título: Geostatistical Tools for Assessing Sampling Designs Applied to a Portuguese Bottom Trawl Survey Field Experience
3. Autores:
   • Ernesto Jardim
   • Paulo Justiniano RIBEIRO JUNIOR
4. Periódicos:
   • Fisheries Research
     • submetido em 06/Dez/2006, Texto Submetido: versão submetida
     • Status: rejeitado, revisado, resubmetido e novamente rejeitado
   • Scientia Marina
     • submetido em 23/10/2007. Texto submetido

• Texto submetido
• arquivos fonte
• Referees reports: Referee 1 e Referee 2 (the texts are also copied below)
• Artigo resubmetido após sugestões dos revisores
  1. arquivo PDF
  2. arquivos FONTE
• Resposta dos editores (aceitação com “minor changes”):

Date: Mon, 11 Feb 2008 11:03:34 +0100 From: Francesc maynou maynouf@icm.csic.es To: ernesto@ipimar.pt Subject: Scientia Marina - SM 2593

Dear Ernesto, I write as Scientific Editor of your manuscript “Geostatistical tools for assessing samplimg designs …” submitted to Scientia Marina. First, I apologize for the delay in handling this manuscript, which was due to the difficulty in finding qualified and available reviewers. Now I have received 2 opinions and they're both positive to your manuscript, suggesting minor changes (attached). I will appreciate receiving a revised version of the manuscript, with detailed responses to the referees' opinions on separate sheets, at your earliest convenience. Please note also, from an editorial point of view, that units should be separated from the magnitudes throughout the manuscript (eg. line 29, page 2: 20-100 m, 101-200 m, 201-500 m). Avoid also colloquial spellings such as “we'd” in line 190, page 7. All species names should be italicized: lines 301-303, page 11.

I look forward to receiving a revised version of your manuscript.

With best regards,

Francesc Maynou, Scientific Editor, Scientia Marina. http://www.icm.csic.es/scimar

This study aims to assess and compare two experimental designs modifying a standard bottom trawling survey with the help of advanced geostatistical tools. This applied research is important to improve abundance estimates and to derive continuous resource maps. Before publication minor changes are required to improve the organization of the manuscript and to insert adequate references. The introduction is not very well organized and it lacks of both general references for geostatistical applications in fisheries and references discussing the geostatistical assessment of sampling designs in fisheries (see comments below). The analytical standard is high and results are in general clearly presented. However, the discussion of the results seems incomplete as important issues such as the impact of sample size or species studied are not discussed.

Introduction

1. Move description of data collection p2 line 41 – p3 line 52. to methods section.
2. Move last sentence of the introduction (p.3 line 69-72) to discussion.

Methods

1. In methods section 2.1 p. 4 line 91 the authors discuss the proposal of sampling designs mixing a set of locations with additional sampling stations at short distances. Here the authors are lacking to discuss this concept, its necessity and its application in a fisheries context see work of Simard et al. (1992); Petitgas, P. (2001) or Doonan et al. (2003). Further, the author’s statement p.4 line 93-94: “Such designs were not considered for bottom trawl surveys until now,….” is not correct. In Stelzenmüller et al. (2005) a bottom trawl sampling design is described which aimed to improve geostatistical estimates by adding sampling stations at shorter distances. The whole methods section 2.1 (p.4 lines 79-99) on the description of sampling designs in theory and praxis should be moved to the introduction.
2. A table should be inserted stating the number of sampling stations within each design. For example only by counting visually the number of sample positions in Fig. 2 it became clear that the hybrid design contained 17 regular grid stations, while the systematic design contained 19 regular grid stations. Further, this table should also contain a summary statistics with mean, variance, and coefficient of variation for both sampling designs.

Results

1. Figure 1: Replace axis labels with latitude and longitude; give location reference and scale bar, and label the 500 m isobaths.
2. Figure 2: Either label axis with lon/lat or X (km)/Y (km) but not with lon (km)/lat (km).

Discussion

1. p.8 lines 212-214. In order to underpin the use of the geostatistical estimator and its variance the authors should discuss their results in relation to the measures of sample average, variance etc. for both designs (see comment 2-methods).
2. p.8 line 222-Results showed that the hybrid design performed better.

The number of samples is often a crucial point in geostatistical analyses especially for applications in fisheries where expensive ship times often limit the number of hauls. Although both sampling designs comprised 36 sampling stations, the composition between numbers of stations of the regular grid and number of stations of the additional stations differed between the hybrid (17/19) and the systematic (19/17) designs. Though, I would expect a sensitivity of the analysis to sample size in design composition (see also Rufino et al. 2006). This issue should be included in the discussion of the results.

1. The performance of both sampling designs was assessed on the base of hake abundance data. Do the authors expect different results when testing the designs for different species or even biological groups such as different size classes (see Stelzenmüller et al. 2005)? I would assume before recommending a new survey design that this design performs well for most species.

Specific comments

1. Remove citations from abstract
2. Replace “yield” with “abundance”, p1, line 7; p6, line 166; p6, line 168; p14 Figure 2 caption
3. Replace “tools” with “measures”, p1 line 8
4. Check units of abundance throughout the text, kg / km2 not kg / km
5. Check space between numbers and their units throughout the text for e.g. p2 lines29, 43, 47, 48; etc.
6. Correct spelling of reference from Muller 2001 to Müller 2001, p4, lines 83,86, 92; p10 line 287

Literature cited

• Doonan, I.J., B. Bull and R.F. Coombs. (2003). Star acoustic surveys of localized fish aggregations. ICES J. Mar. Sci. , 60:132-146.
• Petitgas, P. (2001). Geostatistics in fisheries survey design and stock assement: models, variances and applications. Fish Fish., 2:231-249.
• Rufino, M.M., V. Stelzenmüller, F. Maynou and G.P. Zauke (2006). Assessing the performance of linear geostatistical tools applied to artificial fisheries data. Fisheries Research 82 (1-3):263-279
• Simard, Y., P. Legendre, G. Lavoie and D. Marcotte. (1992). Mapping, estimating biomass, and optimizing sampling programs for spatially autocorrelated data: case study of the northern shrimp (Pandalus borealis). Can. J. Fish. Aquat. Sci., 49:32-45.
• Stelzenmüller, V., S. Ehrich and G.P. Zauke. (2005). Impact of additional small-scale catch data (star survey design) on the geostatistical analyses of demersal fish species in the North Sea. Scientia Marina, 69 (4): 587-602.

This paper describes a comparison study between a hybrid design and a systematic design using a bottom trawl survey data. The paper is well written and provides useful information for revising sampling designs. Some minor revision is recommended.

1. Page 4, line 99-102, the authors may include a more detailed description of the two designs and why they are chosen for this comparison study instead of other designs.
2. Page 5, have the authors considered the possibility of non-stationarity in S(x)?
3. Page 5-6, the discussion of estimating the Box-Cox parameter may be moved to Section 2.2, as in 2.3 only \lambda=0.5 is discussed.
4. Page 6, line 154, the symbol \epsilon has already been used no page 4 for the geostatistical model. The authors may use a different symbol here for the coverage probability. On line 161-162, \xi as is defined should be \epsilon, while \epsilon should be \xi.
5. Page 7, line 178-179, do you mean “hybrid design had lower sill and range”?
6. Page 8, although the results presented in the paper are useful and informative, a better comparison could be made by estimating the covariance parameters using the pooled observations, and do conditional simulation with the pooled parameter estimators to compare the two designs.

Ref.: Ms. No. FISH1094 Sampling Designs for Bottom Trawl Surveys: The Portuguese Autumn Survey Field Experience Fisheries Research

Dear Dr. Jardim,

I can now inform you that the Editorial Board has evaluated the manuscript FISH1094: Sampling Designs for Bottom Trawl Surveys: The Portuguese Autumn Survey Field Experience.

I am sorry to have to inform you that your manuscript has been found to be unacceptable for publication in our journal. An explanation for this decision is given in the review report(s) appended below and I hope that the comments contained therein will be of use to you. Thank you for considering our journal as a possible publication medium.

With kind regards,

Antoinette van den Brakel Journal Manager Fisheries Research fish@elsevier.com

General comments:

While this paper had a number of interesting ideas in it, I am not sure how many of these ideas are unique to this paper and whether or not the primary objective of this paper represented anything more than an extremely limited case study. The introduction (line 104) states that three of the four designs were already proposed in another submitted manuscript. The estimation methods and performance statistics have been published before. The title of the other submitted paper by these authors suggests that there may be some redundancy with the current manuscript. The main conclusions from the study recommend a particular design (R36) of only the four evaluated here but also they seem to be implying some generality to these conclusions beyond this study.

We need more details on the different designs used here. How many strata were used in the ACTUAL design and how were the samples allocated? It would be helpful to include the strata boundaries on Figure 2 to see how the samples from the other designs were placed relative to these boundaries.

The authors make a very strong statement about the appropriateness of design-based estimates in the presence of spatial autocorrelation but they need to be explicit about what they are basing this on. See for example, Smith and Robert (1998) where it is shown that even in the presence of strong spatial autocorrelation, the standard design-based estimates are appropriate for the variance that they are supposed to be estimating.

It would be informative to be able to compare estimates between the design-based methods and the spatial methods. In a number of cases my experience has been that despite the structure and assumptions built into the spatial estimates there results are not that different. I gather the authors found the same thing here with the hake data (lines 359 to 360).

While the design-based estimates seem to appear in Table 1, it is not clear what they are. How were the mean, variance and confidence intervals calculated for each of the survey designs in Table 1? Were stratified random formula used for the ACTUAL design? How was the variance calculated for the systematic and hybrid designs? Was this variance of the sample or variance of the mean? Did you use the normal approximation (with Satterthwaite's approximation for the stratified design) for the confidence intervals as per Cochran (1977)? How about bootstrap confidence intervals for the stratified case (e.g., Smith 1997, Schnute and Haigh 2003)?

Overall, these results seem to be specific for a few designs suggested for hake and horse mackerel off of Portugal for this one particular survey. It is not clear how these results could be generalized or used by any one else designing a similar type of survey.

Specific comments:

The term “spots” for trawl locations should be changed to “sites”. The term “haul's tray” seems to refer to the track of the survey haul.

Line 32: Sampling statistics are routinely referred to as design-based estimates in the literature now days. Update your reference to Cochran (see below).

Line 38: Changes considered by these authors also include station allocation schemes.

Line 65: “clear” tow positions: Do you mean trawlable grounds or bottom (see below)?

Line 116: “… Sines' Canyon, with an area of 4300 km2 and …”

Line 132: “clear” grounds: Do you mean trawlable grounds?

Line 151: “geostatistical jargon”: The word jargon tends to have a pejorative sense that is not intended here. Use the word notation instead.

Line 221: Give justification for grid size (1070 locations) for guidance for the reader.

Line 390-398: I am trying to understand what the point of the opening paragraph of the discussion is. Here is my interpretation of what you may be trying to say. “There are several kinds of information collected from a population, such as length, sex, maturity and age composition plus stomach contents, etc. Spatial patterns for these measurements may be important to know. The bottom trawl survey is used to collect information on several species to study their relationships with each other in an ecological context. It is very difficult to define one optimal survey design for all of the species being sampled in a design-based framework.”

Lines 398-404: I do not understand the rest of the paragraph (What exactly is a scholastic species?). My interpretation of your “… complexity makes it difficult…” sentence suggests that your approach provides a way around this issue. It seems to me that either your approach does not rely on any optimality with respect to survey design because it models the spatial distribution or design R36 solves all problems. Your approach does not “capture … the large [sic] complexity of bottom trawl surveys”. What about different catchabilities between species, size selectivity of the gear, availability issues due diurnal behaviours, etc? These are also complexities associated with bottom trawl surveys.

Lines 405-412: This may be an additional source of variability but what can impact does it have on the analysis and what do you do about it?

Lines 413-422: If trawling is so destructive then how does using different designs and sharing observations help address reducing this destruction? What exactly is the point of this paragraph in the context of the methods used in this paper?

Lines 423-429: Design-based methods use the design-effect approach to address differences between designs with different samples sizes. Apparently, all you have done is avoid the problem.

References:

• Cochran, W.G. 1977. Sampling Techniques. John Wiley & Sons. New York, NY.
• Schnute, J. and R. Haigh. 2003. A simulation model for designing groundfish trawl surveys. Canadian Journal of Fisheries and Aquatic Science. 60: 640-656
• Smith, S. J. 1997. Bootstrap confidence limits for groundfish trawl survey estimates of mean abundance. Canadian Journal of Fisheries and Aquatic Science. 54: 616-630.
• Smith, S. J. and G. Robert. 1998. Getting more out of your survey information: an application to Georges Bank scallops (Placopecten magellanicus). In North Pacific Symposium on invertebrate stock assessment and management. Edited by G. S. Jamieson and A. Campbell. Special Publication of Fisheries and Aquatic Science. 125: 3-13.

Reviewer #2: The manuscript considers a topic (i.e. bottom trawl survey design) that is clearly within the scope of the journal. The manuscript also is, I believe, a new and original contribution because I have never seen a comparison of the relative efficiency of survey designs using geostatistical modeling. However, I also believe that the statistics used are much too sophisticated considering the paucity of data (only 19 sampling sites). As a consequence they have little statistical power to support claims of the superiority of one sampling design relative to another, as is implied in the title. Although I suspect that the authors can do little to improve their data shortage in any revision, I feel uncomfortable about recommending that the manuscript be rejected, because they did a very thorough job in describing their methodology and I learned a lot from my review about this approach to comparing sampling designs when spatial correlation is important. As a consequence, I recommend a revision that stresses the shortcomings of the data and perhaps has a new title that emphasizes the methodology rather than results for a specific survey.

Substantive comments:

1. Line 141. The Box- Cox transformation will not create a near-normal distribution of the catch per unit effort data when there is a substantial proportion of zero catches, as there is for Horse mackerel. Since the authors are using Likelihood methods anyway, a simpler and more appropriate approach would be to use a statistical distribution that allows zeros (for example, gamma) rather than trying to force the data into a normal distribution. However, for hake their approach looks valid because its distribution has few zeros.
2. Line 152. After describing geometric anisotropy and introducing the <PSI> parameters, they go on to present their model (equation 2) which lacks these parameters. Although they later show that these parameters are not important, I feel that equation 2 should be complete, then later describe the simplification. This would help users of the methodology in cases where these parameters are important.
3. Line 235. How was m (1070 locations) determined?
4. Line 283. Although the authors admit here that the small size of their data sets limits their ability to distinguish between sampling designs, this point needs to be repeated in the discussion and conclusions.
5. The two species examined differ in the degree of their patchiness and as a consequence may differ in the suitability of the two contrasting survey designs. However, one survey is used to assess both species simultaneously. Although the authors deal with the practical aspects of conducting trawl surveys, I feel they did not clearly say that “optimum” survey designs must be globally optimum over all surveyed species.
6. Jon Simmons, I believe in an ICES CM document, described a simulation study he did that showed that systematic designs out performed random designs where there was strong spatial correlation. I think that if the authors could find and include a citation for this document it would add to their manuscript.
7. There are some problems with the author's use of English, especially tenses and the use of prepositions, that should be cleaned up. Some things are very unclear: (line 402) “scholastic” should be changed to “schooling” and (line 411) “haul's tray” should be changed to “trawl path”. In addition, I don't understand the meaning of the sentence starting on line 359.
8. Line 434. Non-stationarity, I believe, is often present in fisheries survey data and, unfortunately, is usually ignored in the spatial models I am aware of.

From the Editor-in-Chief :

In the light of the many comments made by the reviewers I conclude with regret that this manuscript is not acceptabe as it stands.Questions on the paucity of the data, the lack of statistical power, difficulty of generalising the results, and the possibility of redundancy with other papers are only some of the matters raised. However, there is good material in this manuscript, and if the authors are able to take full account of the reviewers concerns, a resubmission is possible.

Idéia básica ao redor da sugestao do segundo referee de enfatizar a metodologia, eventualmente em maiores detalhes e aplicando aos dados com ressalvas às limitações impostas pelos dados em questão

I recommend a revision that stresses the shortcomings of the data and perhaps has a new title that emphasizes the methodology rather than results for a specific survey.

1. Intro (enviar 12/02)
   • bts generico
   • bts pt
   • justificação geostats
   • justificação designs alternativos
   • objectivos, ressaltar a apresentação da metodologia sob o enfoque MB, e a contribuição na proposição das medidas usadas na avaliação
2. Material
3. Meth (temos que ser didáticos e claros aqui)
   • Intro aos meth ?? + didático + detalhado ??
   1. geostats framework
   2. Inference and prediction
   3. performance stats
4. Resultados (apresentamos somente para Hake, eventualmente só comentando que foi feito para HM tb)
   • genéricos
   1. Hake
   2. H.Mac
5. Discussion
   • justifica nosso trabalho
   • justifica importância das questões operacionais
   • confounding effects
   • Consistência de theta hat
   • justifica performance stats
   • justifica resultados
   • conclusão

Intro 120207

lyx 160207 - revisto intro, material, methods, results.

1. comentários PJ (19/02/2007)
   
   1. acho que o novo encaminhamento está ficando bom, certamente temos que arrumar o texto aqui e lá
   2. na motivação talvez valha e pena referir a algo como: need of objective geostatistical methodology for the problem addressed here (within the context of this work), this is not a common practice and therefore the presentation of the geostatistical framework is a main issue here.
   3. no penúltimo paragrafo da introdução algo como: the extra burden … is compensated by …
   4. acho que vale a pena chamar mais atenção das apresentação das “performance statistics”
   5. Abstract: acredito q ainda não foi revisto mas temos que remover o “secondary”

Draft final 020307 - Artigo revisto, tabelas e figuras.

Revisão de PJ (20/03/2007, 09:50 BRT)