Figures
Abstract
Nature conservation and fisheries management often focus on particular seabed features that are considered vulnerable or important to commercial species. As a result, individual seabed types are protected in isolation, without any understanding of what effect the mixture of seabed types within the landscape has on ecosystem functions. Here we undertook predictive seabed modelling within a coastal marine protected area using observations from underwater stereo-video camera deployments and environmental information (depth, wave fetch, maximum tidal speeds, distance from coast and underlying geology). The effect of the predicted substratum type, extent and heterogeneity or the diversity of substrata, within a radius of 1500 m around each camera deployment of juvenile gadoid relative abundance was analysed. The predicted substratum model performed well with wave fetch and depth being the most influential predictor variables. Gadus morhua (Atlantic cod) were associated with relatively more rugose substrata (Algal-gravel-pebble and seagrass) and heterogeneous landscapes, than Melanogrammus aeglefinus (haddock) or Merlangius merlangus (whiting) (sand and mud). An increase in M. merlangus relative abundance was observed with increasing substratum extent. These results reveal that landscape effects should be considered when protecting the seabed for fish and not just individual seabed types. The landscape approach used in this study therefore has important implications for marine protected area, fisheries management and monitoring advice concerning demersal fish populations.
Citation: Elliott SAM, Sabatino AD, Heath MR, Turrell WR, Bailey DM (2017) Landscape effects on demersal fish revealed by field observations and predictive seabed modelling. PLoS ONE 12(12): e0189011. https://doi.org/10.1371/journal.pone.0189011
Editor: Stephan C.F. Neuhauss, University Zürich, SWITZERLAND
Received: April 5, 2017; Accepted: November 16, 2017; Published: December 11, 2017
Copyright: © 2017 Elliott et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All files have been uploaded to the PANGAEA database. The link to the data is publicly available on: https://doi.org/10.1594/PANGAEA.879306.
Funding: This work was funded by Marine Scotland (Clyde 2020) fieldwork funding to SE, Scottish Natural Heritage grant fieldwork funding to SE, The ClimateXChange PhD scholarship to SE, and Natural Environment Research Council National Facility for Scientific Diving grant (Grant NFSD/13/01) to SE.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Protecting species requires good knowledge of species distribution and the role of their habitat including the wider landscape effects. Unfortunately, there is often little information on landscape effects on fish abundance and survival [1–3]. This is particularly important given the mobility of many demersal and commercial fish species. It is well recognised that particular seabed types and the biodiversity they provide is of significance to demersal fish (e.g. [4,5]). The roles of different seabed types are especially important during demersal fish juvenile stages, where they are more vulnerable to predation (e.g.[3,6,7]). Understanding landscape ecology provides a more complete understanding of the ecological function and broader spatial pattern of fish [8]. Here, we define landscape as the composition, distribution, and topography of substratum types within a given area [7].
Fisheries affect fish populations through direct mortality as catch or bycatch and also through indirect effects on the success of individual fish [2,9,10]. Direct mortality can be decreased by a reduction in fishing effort, modifications to gear, or by avoiding times and places where there is high density of the species in question [11,12]. Avoidance of indirect effects on the resources needed by fish is most commonly achieved through spatial measures such as Marine Protected Areas (MPAs) or fisheries closures [13,14]. Understanding the role of juvenile commercial fish habitat can be particularly important since juvenile gadoid survival has been observed to have the greatest effect on population recruitment (e.g. [15–17]).
Protecting areas important to fish remains a less common approach to conserving fish than reducing fishing effort or modifying gear impact [2,18]. This is mainly because of the difficulties in understanding which habitat components are important to fish and how the extent (area of each substratum type) and heterogeneity (diversity and pattern of substratum types and patches within a landscape) of substrata affect fish populations [3,7,19]. Nonetheless, with the rise in spatial protection measures, knowledge of the distribution of species and their habitat is increasingly important [20,21].
Within coastal areas, mapping has been derived from aerial and satellite images. Use of optical imaging techniques in high visibility waters can provide useful information. Unfortunately such techniques are less useful in more turbid waters [3]. Acoustic methods can also provide detailed maps of the seabed [22], but can be resource intensive and prohibitively expensive [23,24]. Predictive methods can therefore be an important tool to overcome such issues [25]. Predictive outcomes will, however, vary according to the spatial scales used [26–28]. For individual MPAs and other spatial management plans to be effective, adequate spatial scales need to be used [26,28,29].
Given the need for higher resolution seabed maps to implement adequate spatial management measures, and the lack of knowledge of landscape effects on fish abundance and distribution, the aims of this study were two-fold. Firstly a range of environmental variables were used to undertake fine-scale predictive mapping of substrata within South Arran Nature Conservation MPA (NCMPA) [5]. The predicted seabed map was then used to understand how substratum type, extent and heterogeneity affected juvenile G. morhua, M. aeglefinus and M. merlangus relative abundance.
Material and methods
Study location
Research was conducted from June to September 2013 and 2014 within the recently designated (2014) South Arran NCMPA, located within the Firth of Clyde, southwest coast of Scotland (Fig 1). The MPA encompasses an area of 250 km2 and contains within its boundaries a 2.67 km2 No Take Zone (NTZ) [30].
Squares represent substratum types from stereo-video deployments.
Data collection
Gadoid sampling was undertaken using three Stereo Baited Remote Underwater Video (SBRUV) frames as described in [5]. The maximum number of individuals of each gadoid in the field of view at the same time (MaxN) [31,32] was used to assess the relative abundance of gadoids within each deployment. The methods used were based on those of [33,34], with modifications for lower light and visibility (< 6 m). Still images were extracted from the SBRUV and stereo-video SCUBA transect recordings to classify substrata and undertake substratum prediction modelling. The substratum types comprised of seagrass (Zostera marina), sand, mud, algal-gravel-pebble (AGP) which contained maerl, and algal-boulder-cobble (ABC). Two divisions of the Wentworth grain scale [35] were used to classify sediment type [36]. Refer to [5,37] for a detailed description of substratum categorisation. A total of 289 ground-truthed data points were used for substratum prediction analysis consisting of 258 SBRUV deployments and 31 SCUBA transects (Fig 1).
Substratum model environmental descriptors
Five abiotic environmental predictor variables were used for predictive substratum modelling (Table 1). Depth data were obtained from vessel echosounders. Distance from shore was calculated using the GPS fix made at the time of deployment and ArcGIS v. 10.1 (EDINA digimap, British National Grid 1984). Maximum tidal speed was obtained from a three-dimensional Finite Volume Community Ocean Model (FVCOM) implementation for the Firth of Clyde [38,39]. The maximum tidal speed, evaluated over a whole year, was modelled at the location of each data collection point for the deeper layer of the model, as described by [38]. Wave fetch values were used as described in [40]. Information on the underlying geology of the MPA was obtained from the British Geological Survey (BGS) and downloaded from EDINA.
Substratum predictive modelling
A multinomial generalized linear model was applied in order to undertake substratum distribution modelling. A multinomial distribution was selected since five possible discrete outcomes restricted between zero and one were plausible [41]. nnet R package in R software (v. 3.03, R Core Team, 2015) was used for the multinomial model. Prior to statistical analysis, continuous variables were standardised by dividing the mean by the standard deviation. One SBRUV deployment was removed as a result of an erroneous depth reading. Automated model selection was undertaken to find the model of best fit using the difference between Akaike’s Information Criterion (AIC) scores. Eq 1, provides the model of best fit for substratum prediction.
Where Yi is the response variable and β are the modelled coefficients for sample i.
The multinomial model performance was tested on 25% of the collected SCUBA and SBRUV substratum dataset, by randomly splitting the data into 217 samples (75%) to fit the data and 72 samples (25%) to validate the data. To evaluate the accuracy of the model of best fit, an area under the curve (AUC) was performed [42] to give the percentage of correct predictions. Correct classification of the individual substratum categories was calculated to understand sensitivity of the model’s predicted substratum types. ROCR R package was used to discern how well each of the variables explained the presence of the modelled substrata.
To undertake substratum prediction modelling for the wider area, the environmental variables were extracted at each node point of the hydrodynamic model [38]. Depth data for the wider area was sourced from the General Bathymetric Chart of the Oceans (GEBCO) overlapped with SeaZone (v. 1.1). The predictor variables were then standardised, and the multinomial model of best fit was used to predict substratum type. The resulting data frame containing substratum predictions was imported into ArcGIS, converted into a point shapefile and validated with ground-truthed substratum types. Polygons were created joining the ground-truthed and predicted substratum data points to create a smooth continuous surface for each substratum type across the predicted area (Fig 1).
Gadoid landscape calculations
Very little information exists on in situ juvenile gadoid movement patterns due to the difficulties in using acoustic tags on such small fish (below 10 cm) [43] and difficulties in tracking marked, recaptured individuals [44–46]. The horizontal distance and time which the juveniles moved within the field of view of the cameras was measured to calculate the speed and approximate distance they can move within one hour. To undertake these distance calculations, the cruising speeds (when the gadoids were moving in a straight line away from the bait) of ten individuals of each gadoid species of less than 20 cm were used.
Existing literature on gadoid swimming speeds and home ranges was also gathered (i.e. [44,46–49]). Both measures (distance speed calculations and information from existing literature) were used since, fish behaviour around baited cameras is not usually classed as normal, whereas existing literature on juvenile gadoid movement is sparse and varied. From the combined distance speed calculations and existing literature on gadoid speed and movement behaviour, a radius of 1500 m (covering an area of 7.07 km2) was created in ArcGIS around the gadoid point data collection.
Hill number N∞ (inverse of Berger-Parker dominance index) [50,51], was used to understand how landscape heterogeneity affected gadoid relative abundance, using the predicted substratum point data. N∞ gives the reciprocal of the proportional abundance from the most common specie [51,52], or in this case substratum type. Substratum richness was therefore the number of types of substrata within the 1500 m radius, and evenness the frequencies of the substrata within the radius. Hill numbers are inclusive of well-known indices and are expressed as the effective numbers [53,54]. N∞ (substratum dominance) was used since N1 (exponential of Shannon) and N2 (the inverse of Simpson’s index) are more sensitive to varying sample size [55,56]. The extents of each substratum polygon within the radii were calculated in ArcGIS.
A negative binomial error distribution using R package glmmADMD was used to model SBRUV fish counts. Random effects to account for varying location and grouped days of data collection were incorporated into the models where significant. Year was included into the model due to slight differences in sampling between years. Stepwise backwards selection was used for model selection by AIC minimisation. Tukey tests were performed to test for differences between substratum categories using the R package multcomp [57].
Results
Substratum distribution model
Evaluation of the multinomial model using the validation dataset indicated ‘excellent’ predictive power [41] (AUC score of 0.88) (l = -131.21, d.f. = 32, p < 0.001). Correct classification for seagrass was 100%, followed by mud with 89%. AGP, sand and ABC accurate classification received scores of 70%, 69% and 68% respectively. Predictor variable individual AUC scores were particularly strong for wave fetch and depth (0.71 and 0.69 respectively). Distance from coast and maximum tidal speed had AUC scores of 0.66 and 0.62.
Landscape effects on gadoids
G. morhua highest MaxN was observed over AGP substratum type with the lowest average MaxN observed over sand (Table 2), no G. morhua were observed over mud. A decrease in MaxN was observed with increasing substratum dominance (N∞) (l = -311.87, d.f. = 9, θ = 0.98, p < 0.001; Table 3; Fig 2; Table A in S1 File).
Arrows indicative whether the predictor variable significantly increased or decreased gadoid MaxN. N∞ refers to the dominance of the most common substratum type.
M. aeglefinus and M. merlangus highest MaxN was observed over sand, with lowest MaxN observed over ABC (Table 2). An increase in M. aeglefinus and M. merlangus MaxN was observed with increasing substratum dominance (N∞) (l = -286.45, d.f. = 9, θ = 1.36, p < 0.001; Table 3; Fig 2; Table B in S1 File). A slight increase in M. merlangus MaxN was also observed with increasing substratum extent (l = -272.45, d.f. = 11, θ = 0.88, p < 0.001; Fig 2; Table C in S1 File).
Discussion
This research demonstrates the importance of considering landscape effects on demersal fish for conservation and fisheries spatial management purposes. Using a range of predictor variables the distribution and configuration of seabed was able to be predicted within an MPA, saving on resource intensive costs of acoustic monitoring methods. The substratum prediction model enabled the heterogeneity and the extent of different substratum types, at ranges of up to 1500 m around gadoid data collection locations to be linked to gadoid relative abundance.
The multinomial model used for substratum prediction performed well. Depth and wave fetch had the greatest influence on substratum prediction. Depth affects many species and is commonly used as a surrogate for light, temperature and benthic shear stress from ocean swell [23,58]. Wave fetch and maximum tidal speed will also have an effect on benthic shear stress and light attenuation through varying levels of exposure and turbidity [23,24,59]. Maximum tidal speed additionally provided point-specific information of hydrodynamic regimes within the area.
Individual substratum correct classification showed that seagrass and mud substrata had the highest classification accuracy. Seagrass only grows in shallow areas where sufficient light can penetrate [24], and mud was only observed within deeper or more sheltered areas. ABC substrata also occurred at shallower depths (< 20 m) and more exposed areas composed of larger sediment grain sizes, where the macro-algae can anchor itself [58–60]. AGP substratum type largely consisted of varying percentages of maerl (coralline red algae, Phymatolithon calcareum), which together with red algae requires a lesser degree of light penetration (≤ 30 m) [61]. Sand can be found within a broad range of environmental conditions which may have led to sand having a lower prediction accuracy [62].
Using the predicted full coverage map, seabed landscape effects on gadoid relative abundance were modelled at a range relevant to the movement behaviour of the gadoids studied. In situ landscape effects on commercially important gadoid fish have not previously been explored [8]. Research on landscape effects of marine organisms has demonstrated the importance of such larger scale processes on the distribution of fish. For example [8] explored a range of landscape metrics on demersal reef fish relative abundance within a 200 m radius of SBRUV deployments in south-eastern Australia. A combination of different landscape measures, including distance from the reef and the length of the edges of the reef were found to influence their distribution.
G. morhua were observed over a variety of substrata occurring at shallower depths, with higher relative abundance observed over AGP containing maerl. G. morhua have a checkerboard brown and white colour which lend them to being more difficult to distinguish in gravel-pebble substratum types [37,63]. This substratum may therefore be favoured as an anti-predator mechanism. G. morhua are also known to inhabit shallow coastal waters where algal-gravel-pebble and algal-boulder-cobble occur [5,37,64]. Conversely M. aeglefinus and M. merlangus were observed in higher relative abundance over sand and mud. M. aeglefinus and M. merlangus have been known to occur in deeper waters than G. morhua, where sand and mud are more prevalent within the Firth of Clyde [5]. The species specific segregation observed between these gadoids may result in reduced interspecific predation and competition [65,66].
A decrease in G. morhua relative abundance was observed with increasing N∞ (substratum dominance). Indicating that juvenile G. morhua are associated with more heterogeneous landscapes rather than landscapes dominated by one specific substratum type. M. aeglefinus and M. merlangus were observed in higher relative abundance with increasing substratum dominance. M. merlangus were also observed in higher relative abundance with increasing substratum extent. The increase in G. morhua observed with increasing landscape heterogeneity may enable G. morhua to access areas with possibly increased food availability and areas with sufficient refuge [8,67,68]. An experimental study undertaken by [69] demonstrated that juvenile G. morhua seem to differentiate between substratum types, selecting areas where growth and survival were highest. The increase in M. aeglefinus and M. merlangus observed with increasing substratum dominance, and extent for M. merlangus, may be an indication of how these gadoids are better adapted to prey found within sand and mud substrata.
Few demersal species are associated with a single seabed type, but instead use a combination of substrata according to foraging, shelter and tidal behaviours [68,70]. Substratum boundaries are also thought to be important foraging and refuge areas for fish depending on the extent of the patches [69,70]. Attempts to identify fish nursery areas are frequently static processes identifying individual homogeneous seabed types [68]. However, seascapes are often dynamic and varied and the focal species may also undergo ontogenetic shifts as a result of changing resource needs [5,68,71]. Such behaviour (foraging, shelter and ontogenetic shifts in resource), may have explained why all three gadoids were observed over the range of substratum types and G. morhua were observed in higher relative abundance in more heterogeneous landscapes.
Conclusion
This study demonstrates the importance of considering landscape measures in demersal fish distribution given that landscape heterogeneity had differing effects on the gadoids studied here. This is in stark contrast with the typical approaches of either disregarding habitat requirements, or seeking to ensure the availability of a single seabed type. The use of a range of environmental variables enabled an accurate prediction of substrata and the creation of a fine scale seabed map. The approaches used in this study could be applied on a larger scale for the selection of areas for stock improvement or trialled on other species. For G. morhua, the selection and prioritisation of more heterogeneous landscapes for protection as nurseries should be prioritised.
Supporting information
S1 File. Gadus morhua, Melanogrammus aeglefinus and Merlangius merlangus models of best fit.
https://doi.org/10.1371/journal.pone.0189011.s001
(DOCX)
Acknowledgments
The first author would like to thank to E. Argaña Pérez and Glasgow University Spatial group for statistical and GIS advice. Thanks to B. Allan, C. Willmott and J. Clarke for support in data collection and H. Wood, R. Cheshire and Millport Field Studies Council for the boat support for data collection.
References
- 1. Mangel M, Levin P, Patil A. Using Life History And Persistence Criteria To Prioritize Habitats For Management And Conservation. Ecological Applications 2006;16:797–806. pmid:16711063
- 2. Armstrong CW, Falk-Petersen J. Habitat-fisheries interactions: a missing link? ICES Journal of Marine Science 2008;65:817–21.
- 3. Sundblad G, Bergstrom U, Sandstrom A, Eklov P. Nursery habitat availability limits adult stock sizes of predatory coastal fish. ICES Journal of Marine Science 2014;71:672–80.
- 4. Seitz RD, Wennhage H, Bergström U, Lipcius RN, Ysebaert T. Ecological value of coastal habitats for commercially and ecologically important species. ICES Journal of Marine Science 2014;71:648–65.
- 5. Elliott SAM, Turrell WR, Heath MR, Bailey DM. Juvenile gadoid habitat and ontogenetic shift observations using stereo-video baited cameras. Marine Ecology Progress Series 2017;568:123–35.
- 6. Ralph GM, Seitz RD, Orth RJ, Knick KE, Lipcius RN. Broad-scale association between seagrass cover and juvenile blue crab density in Chesapeake Bay. Marine Ecology Progress Series 2013;488:51–63.
- 7. Elliott SAM, Milligan RJ, Turrell WR, Heath MR, Bailey DM. Disentangling habitat concepts for demersal marine fish management. Oceanography and Marine Biology Annual Review 2016;54:173–91.
- 8. Moore CH, Van Niel K, Harvey ES. The effect of landscape composition and configuration on the spatial distribution of temperate demersal fish. Ecography 2011;34:425–35.
- 9. Auster PJ, Malatesta RJ, Langton RW, Watting L, Valentine PC, Donaldson CLS, et al. The impacts of mobile fishing gear on seafloor habitats in the gulf of Maine (Northwest Atlantic): Implications for conservation of fish populations. Reviews in Fisheries Science 1996;4:185–202.
- 10. Jennings S, Kaiser MJ. The Effects of Fishing on Marine Ecosystems. Advances in Marine Biology, vol. 34, 1998, p. 201–352.
- 11. Fernandes PG, Cook RM. Reversal of Fish Stock Decline in the Northeast Atlantic. Current Biology 2013;23:1432–7. pmid:23871238
- 12. Clarke J, Bailey DM, Wright PJ. Evaluating the effectiveness of a seasonal spawning area closure. ICES Journal of Marine Science: Journal Du Conseil 2015;72:2627–37.
- 13. Worm B, Hilborn R, Baum JK, Branch TA, Collie JS, Costello C, et al. Rebuilding global fisheries. Science 2009;325:578–85. pmid:19644114
- 14. Halpern BS, Lester SE, McLeod KL. Placing marine protected areas onto the ecosystem-based management seascape. Proceedings of the National Academy of Sciences of the United States of America 2010;107:18312–7. pmid:20176945
- 15. Olsen EM, Moland E. Fitness landscape of Atlantic cod shaped by harvest selection and natural selection. Evolutionary Ecology 2011;25:695–710.
- 16. Laurel BJ, Knoth BA, Ryer CH. Growth, mortality, and recruitment signals in age-0 gadids settling in coastal Gulf of Alaska. ICES Journal of Marine Science: Journal Du Conseil 2016;73:2227–37.
- 17. Myers RA, Cadigan NG. Density-Dependent Juvenile Mortality in Marine Demersal Fish. Canadian Journal of Fisheries and Aquatic Sciences 1993;50:1576–90.
- 18. Kritzer JP, DeLucia M-B, Greene E, Shumway C, Topolski MF, Thomas-Blate J, et al. The Importance of Benthic Habitats for Coastal Fisheries. BioScience 2016;66:274–84.
- 19. Moore CH, Harvey ES, Van Niel K. The application of predicted habitat models to investigate the spatial ecology of demersal fish assemblages. Marine Biology 2010;157:2717–29.
- 20. Howell KL, Holt R, Endrino IP, Stewart H. When the species is also a habitat: Comparing the predictively modelled distributions of Lophelia pertusa and the reef habitat it forms. Biological Conservation 2011;144:2656–65.
- 21. Moore CH, Drazen JC, Radford BT, Kelley C, Newman SJ. Improving essential fish habitat designation to support sustainable ecosystem-based fisheries management. Marine Policy 2016;69:32–41.
- 22. Brown CJ, Smith SJ, Lawton P, Anderson JT. Benthic habitat mapping: A review of progress towards improved understanding of the spatial ecology of the seafloor using acoustic techniques. Estuarine, Coastal and Shelf Science 2011;92:502–20.
- 23. Reiss H, Birchenough S, Borja A, Buhl-Mortensen L, Craeymeersch J, Dannheim J, et al. Benthos distribution modelling and its relevance for marine ecosystem management. ICES Journal of Marine Science 2015;72:297–315.
- 24. Schubert PR, Hukriede W, Karez R, Reusch TBH. Mapping and modeling eelgrass Zostera marina distribution in the western Baltic Sea. Marine Ecology Progress Series 2015;522:79–95.
- 25. Guisan A, Zimmermann NE. Predictive habitat distribution models in ecology. Ecological Modelling 2000;135:147–86.
- 26. Levin S. The problems of pattern and spatial scale in ecology. Ecology 1992;73:1943–67.
- 27. Elith J, Leathwick JR. Species Distribution Models: Ecological Explanation and Prediction Across Space and Time. Annual Review of Ecology, Evolution, and Systematics 2009;40:677–97.
- 28. Chave J. The problem of pattern and scale in ecology: what have we learned in 20 years? Ecology Letters 2013;16:4–16. pmid:23351093
- 29. Crain CM, Halpern BS, Beck MW, Kappel C V. Understanding and Managing Human Threats to the Coastal Marine Environment. Annals of the New York Academy of Sciences 2009;1162:39–62. pmid:19432644
- 30. Thurstan RH, Roberts CM. Ecological Meltdown in the Firth of Clyde, Scotland: Two Centuries of Change in a Coastal Marine Ecosystem. PLoS ONE 2010;5:e11767. pmid:20686614
- 31. Priede IG, Bagley PM, Smith A, Creasey S, Merrett NR. Scavenging deep demersal fishes of the Porcupine Seabight, north-east Atlantic: observations by baited camera, trap and trawl. Journal of the Marine Biological Association of the United Kingdom 1994;74:481–98.
- 32. Watson DL, Harvey ES, Anderson MJ, Kendrick GA. A comparison of temperate reef fish assemblages recorded by three underwater stereo-video techniques. Marine Biology 2005;148:415–25.
- 33. Harvey ES, Shortis MR. A System for Stereo-Video Measurement of Sub-Tidal Organisms. Marine Technology Society Journal 1995;29:10–22.
- 34. Harvey ES, Shortis MR. Calibration Stability of an Underwater Stereo Video System: Implications for Measurement Accuracy and Precision. Marine Technology Society Journal 1998;32:3–17.
- 35. Wentworth CK. A Scale of Grade and Class Terms for Clastic Sediments. The Journal of Geology 1922;30:377–92.
- 36. Connor DW, Allen JH, Golding N, Kerry L, Lieberknecht LM, Northen KO, et al. The Marine Habitat Classification for Britain And Ireland. Peterborough: 2004. Available from: http://jncc.defra.gov.uk/marinehabitatclassification.
- 37. Elliott SAM, Ahti PA, Heath MR, Turrell WR, Bailey DM. An assessment of juvenile Atlantic cod Gadus morhua distribution and growth using diver operated stereo-video surveys. Journal of Fish Biology 2016;89:1190–207. pmid:27221152
- 38. Sabatino AD, O’Hara Murray RB, Hills A, Speirs DC, Heath MR. Modelling sea level surges in the Firth of Clyde, a fjordic embayment in south-west Scotland. Natural Hazards 2016;84:1601–23.
- 39. Chen C, Liu H, Beardsley RC. An Unstructured Grid, Finite-Volume, Three-Dimensional, Primitive Equations Ocean Model: Application to Coastal Ocean and Estuaries. Journal of Atmospheric and Oceanic Technology 2003;20:159–86.
- 40. Burrows MT, Harvey R, Robb L. Wave exposure indices from digital coastlines and the prediction of rocky shore community structure. Marine Ecology Progress Series 2008;353:1–12.
- 41.
Hosmer DW, Lemeshow S, Sturdivant RX. Applied Logistic Regression. 3rd ed. John Wiley & Sons, Inc; 2013.
- 42. Hand DJ, Till RJ. A Simple Generalisation of the Area Under the ROC Curve for Multiple Class Classification Problems. Machine Learning 2001;45:171–86.
- 43. Campana SE. Accuracy, precision and quality control in age determination, including a review of the use and abuse of age validation methods. Journal of Fish Biology 2001;59:197–242.
- 44. Laurel BJ, Gregory RS, Brown JA, Hancock JK, Schneider DC. Behavioural consequences of density-dependent habitat use in juvenile cod Gadus morhua and G. ogac: the role of movement and aggregation. Marine Ecology Progress Series 2004;272:257–70.
- 45. Olsen EM, Gjøsæter J, Stenseth NC. Evaluation of the use of visible implant tags in age-0 Atlantic cod. North American Journal of Fisheries Management 2004;24:282–6.
- 46. Wright PJ, Galley E, Gibb IM, Neat FC. Fidelity of adult cod to spawning grounds in Scottish waters. Fisheries Research 2006;77:148–58.
- 47. Grant SM, Brown JA. Nearshore settlement and localized populations of age 0 Atlantic cod (Gadus morhua) in shallow coastal waters of Newfoundland. Canadian Journal of Fisheries and Aquatic Sciences 1998;55:1317–27.
- 48. Løkkeborg S, Fernö A, Jørgensen T. Effect of position-fixing interval on estimated swimming speed and movement pattern of fish tracked with a stationary positioning system. Hydrobiologia 2002;483:259–64.
- 49. Cote D, Moulton S, Frampton PCB, Scruton DA, McKinley RS. Habitat use and early winter movements by juvenile Atlantic cod in a coastal area of Newfoundland. Journal of Fish Biology 2004;64:665–79.
- 50. Berger WH, Parker FL. Diversity of planktonic Foraminifera in deep-sea sediments. Science 1970;168:1345–7. pmid:17731043
- 51. Hill MO. Diversity and evenness: A unifying notation and its consequences. Ecology 1973;54:427–32.
- 52. Buckland ST, Yuan Y, Marcon E. Measuring temporal trends in biodiversity. AStA Advances in Statistical Analysis 2017;101:461–74.
- 53. Magurran AE. Measuring Biological Diversity. African Journal of Aquatic Science 2004;29:285–6.
- 54. Jost L. The Relation between Evenness and Diversity. Diversity 2010;2:207–32.
- 55. Buckland ST, Magurran AE, Green RE, Fewster RM. Monitoring change in biodiversity through composite indices. Philosophical Transactions of the Royal Society B: Biological Sciences 2005;360:243–54. pmid:15814343
- 56. Buckland ST, Studeny AC, Magurran AE, Illian JB, Newson SE. The geometric mean of relative abundance indices: a biodiversity measure with a difference. Ecosphere 2011;2:1–15.
- 57. Hothorn T, Bretz F, Westfall P. Simultaneous Inference in General Parametric Models. Biometrical Journal 2008;50:346–63. pmid:18481363
- 58. Holmes KW, Van Niel KP, Radford B, Kendrick GA, Grove SL. Modelling distribution of marine benthos from hydroacoustics and underwater video. Continental Shelf Research 2008;28:1800–10.
- 59. Burrows MT. Influences of wave fetch, tidal flow and ocean colour on subtidal rocky communities. Marine Ecology Progress Series 2012;445:193–207.
- 60. Christie H, Norderhaug KM, Fredriksen S. Macrophytes as habitat for fauna. Marine Ecology Progress Series 2009;396:221–33.
- 61. Hall-Spencer JM, Grall J, Moore PG, Atkinson RJA. Bivalve fishing and maerl-bed conservation in France and the UK?retrospect and prospect. Aquatic Conservation: Marine and Freshwater Ecosystems 2003;13:S33–41.
- 62. Reiss H, Cunze S, König K, Neumann H, Kröncke I. Species distribution modelling of marine benthos: a North Sea case study. Marine Ecology Progress Series 2011;442:71–86.
- 63. Gregory RS, Anderson JT. Substrate selection and use of protective cover by juvenile Atlantic cod Gadus morhua in inshore waters of Newfoundland. Marine Ecology Progress Series 1997;146:9–20.
- 64. Gibb FM, Gibb IM, Wright PJ. Isolation of Atlantic cod (Gadus morhua) nursery areas. Marine Biology 2007;151:1185–94.
- 65. Fromentin JM, Stenseth NC, Gjøsæter J, Bjørnstad ON, Falck W, Johannessen T. Spatial patterns of the temporal dynamics of three gadoid species along the Norwegian Skagerrak coast. Marine Ecology Progress Series 1997;155:209–22.
- 66. Bastrikin DK, Gallego A, Millar CP, Priede IG, Jones EG. Settlement length and temporal settlement patterns of juvenile cod (Gadus morhua), haddock (Melanogrammus aeglefinus), and whiting (Merlangius merlangus) in a northern North Sea coastal nursery area. ICES Journal of Marine Science 2014;71:2101–13.
- 67. Kamenos NA, Moore PG, Hall-Spencer JM. Substratum heterogeneity of dredged vs un-dredged maerl grounds. Journal of the Marine Biological Association of the UK 2003;83:411–3.
- 68. Nagelkerken I, Sheaves M, Baker R, Connolly RM. The seascape nursery: a novel spatial approach to identify and manage nurseries for coastal marine fauna. Fish and Fisheries 2013;16:1–10.
- 69. Laurel BJ, Gregory RS, Brown JA. Settlement and distribution of Age-0 juvenile cod, Gadus morhua and G. ogac, following a large-scale habitat manipulation. Marine Ecology Progress Series 2003;262:241–52.
- 70. Gorman AM, Gregory RS, Schneider DC. Eelgrass patch size and proximity to the patch edge affect predation risk of recently settled age 0 cod (Gadus). Journal of Experimental Marine Biology and Ecology 2009;371:1–9.
- 71. Laurel BJ, Ryer CH, Knoth B, Stoner AW. Temporal and ontogenetic shifts in habitat use of juvenile Pacific cod (Gadus macrocephalus). Journal of Experimental Marine Biology and Ecology 2009;377:28–35.