Development of a baited video technique and spatial models to explain patterns of fish biodiversity in inter-reef waters
31
0
0
Full text
(2) Development of a baited video technique and spatial models to explain patterns of fish biodiversity in inter-reef waters. Thesis submitted by Mike CAPPO BSc (Hons) MSc (University of Adelaide) in February 2010. for the degree of Doctor of Philosophy in the School of Marine and Tropical Biology James Cook University, Townsville.
(3)
(4) Statement of Access I, the undersigned author of this work, understand that James Cook University will make this thesis available for use within the University Library and via the Australian Digital Theses network, and elsewhere as appropriate. I understand that as an unpublished work, a thesis has significant protection under the Copyright Act and I do not wish to place any further restriction on access to this work.. Signature:. ............................................................................ Date:. ............................................................................ i.
(5) Statement of Sources I declare that this thesis is my own work and has not been submitted in any form for another degree or diploma at any university or other institution of tertiary education. Information derived from the published or unpublished work of others has been acknowledged in the text and a list of references is provided.. Signature:. ........................................................................... Date:. ........................................................................... ii.
(6) Contributions I declare that this thesis is an output from the ‘Great Barrier Reef Seabed Biodiversity Project’, a collaboration between the Australian Institute of Marine Science (AIMS), the Commonwealth Scientific and Industrial Research Organisation (CSIRO), Queensland Primary Industries & Fisheries (Department of Employment, Economic Development and Innovation (DEEDI), formerly QDPIF) and the Queensland Museum (QM). The project was funded by the CRC Reef Research Centre, the Fisheries Research and Development Corporation (FRDC) and the National Oceans Office, and led by Drs R. Pitcher (Principal Investigator, CSIRO), P. Doherty (AIMS), J. Hooper (QM) and N. Gribble (QDPIF). Chapter 3 was supported with earlier funding from the FRDC to the CSIRO and AIMS (FRDC 97/205). I declare that the environmental covariates involving sediments, epibenthic cover and properties of the water column were supplied for my use in a basic form by Dr Roland Pitcher (CSIRO), and that I was responsible for the inception of the baited video database, and the training and data verification for its use by others under my direction and supervision. To this end, there were 1,438 baited video tapes analysed for this thesis, and the primary data gathering from these tapes was split by the author (620), Mr Peter Speare (649) and Ms Helen Sturmey (169). I checked the data and images from all tapes read by others, and re-analysed tapes where I deemed that identifications were unreliable. Mr Gavin Ericson (AIMS) developed the ‘BRUVS2.1.mdb’ tape reading interface in collaboration with myself and the other users, and it proceeded through numerous versions. Dr Glenn De’ath (AIMS) provided me with numerous private libraries and functions for univariate and multivariate analyses in the statistical environment ‘R’. As a result, the analyses in this thesis closely follow the approaches and techniques published by Dr De’ath and his collaborators. Dr Jari Oksanen (University of Oulu) and Dr De’ath also wrote custom functions for my use. These functions are specified in the thesis. The candidature within the School of Marine and Tropical Biology, James Cook University, was supported with the provision of AIMS resources and one day of study leave per fortnight by Professor S. Hall (CEO, AIMS). A small student stipend was made available and some was spent on stereo-video software and the professional formatting of this thesis by Ms Shannon Hogan (Adelpha Publishing and Design).. iii.
(7) My primary supervisors were Professor Michael Kingsford (JCU) and Dr Glenn De’ath. Secondary supervisors at the beginning of the candidature were Dr Garry Russ (JCU) and Dr Julian Caley (AIMS).. Signature:. ........................................................................... Date:. ........................................................................... iv.
(8) Electronic Copy I, the undersigned author of this work, declare that the electronic copy of this thesis provided to the James Cook University Library is an accurate copy of the printed thesis already submitted, within the limits of the technology available.. Signature:. ............................................................................ Date:. ............................................................................ v.
(9) Acknowledgements This study would not have been possible without the loving support of my wife, Fairlie Sandilands. I dedicate this thesis to her efforts in both encouraging me and maintaining the fabric of our family during my long physical and mental absences from my roles as parent and husband. Isaac Newton once said, “If I have seen further than others, it is because I have stood on the shoulders of giants”. It is with this quotation I recognise the patient contributions of Dr Glenn De’ath in thrusting me to the forefront of ‘R’ multivariate analyses in community ecology, and of Gavin Ericson who developed the BRUVS1.5.mdb© image analysis ‘environment’ at the AIMS. I could not have accomplished this thesis without their tutelage. I wish to thank the multi-agency field teams and the crews of the RV Lady Basten, FRV Gwendoline May and RV James Kirby for assistance with the many months of fieldwork, and Peter Speare and Helen Sturmey for assistance with tape reading. Dr Euan Harvey and Theodore Wassenberg provided fundamental advice and assistance in developing my enthusiasm for the baited video approach, and the AIMS engineering workshops fabricated the frames and camera housings. I am grateful to my fellow workers Charlotte Johansson and Marcus Stowar for their help in allowing me to keep my focus when at the AIMS, and Belinda Curley and Professor Howard Choat for their input during my days on campus at JCU.. vi.
(10) Abstract A baited, remote, underwater video sampling technique (BRUVS) was developed to survey the patterns of diversity and abundance of fishes, sharks, rays and sea snakes in all shelf depths and ‘inter-reef’ habitats throughout the Great Barrier Reef Marine Park (GBRMP). The use of bait greatly enhanced the ability to distinguish fish assemblages, including those containing functional groups thought to be either unresponsive to bait, or shy of the carnivores present in the field of view (e.g. scarids and siganids). A field comparison showed the BRUVS recorded more, larger, mobile species (e.g. carangids, scombrids, carcharhinids) than demersal trawls and performed best in daylight hours. Although the BRUVS did not record many sedentary and cryptic families collected as trawl ‘bycatch. fauna’ (e.g.. apogonids,. priacanthids,. pleuronectiformes), they did discriminate the same site groups – and with less classification error than the trawls. BRUVS could be used on any seafloor topography and any zone of the GBRMP, but their effectiveness was restricted by high turbidity and low irradiance at the seabed. As a result, BRUVS sampling was included in the largest exploration yet undertaken of seafloor biodiversity on a tropical shelf. Replicate BRUVS were deployed at 366 sites throughout the length (‘along’) and breadth (‘across’) of the GBRMP, enabling a comprehensive examination of the spatial patterns in vertebrate richness, assemblage structure and species occurrences in terms of the major environmental covariates presumed to govern fish distributions. These analyses were conducted with gradient boosting models and multivariate classification and regression trees. These approaches are robust and flexible and allow visualisation of complex interactions. The latitudinal gradient in richness of the 347 species was relatively weak, but there were strong cross-shelf gradients, with a ‘hump shaped’ peak in richness about a position ~0.8 ‘across’ the shelf. This was shown not to be the result of a random, mid-domain effect in species ranges but rather a response to the topographic complexity, epibenthic marine plants, low currents and mixed carbonate/mud sediments found (on average) at this position. Hierarchical, multivariate regression tree analyses (MRT) constrained by 28 selected environmental covariates, showed ten assemblages characterised by Dufrêne-Legendre species indicators. There were strong faunal boundaries, or ecotones, about Bowen in the south and Cooktown to Cape Flattery in the north. A position about ~0.5-0.8 across the continental shelf, where carbonate content of the sediments rose to ~84%, separated inshore ‘lagoonal’ assemblages from offshore ‘reefal’ assemblages. On either side of this demarcation there were fish assemblages distinguished by their association with finer or coarse sediments, beds of seagrass (Halophila) and banks of marine algae (Halimeda and others). There were more lethrinids, labrids, serranids and scarids offshore, but there was not strong replacement or zonation of vertebrate families along environmental gradients – unlike the generalisations from low latitude shelves in the Atlantic. Instead, there were changes amongst. vii.
(11) species within genera that followed sedimentary facies and other gradients. In the central section such gradients varied simply with position across the shelf, for example, from Nemipterus hexodon and N. peronii inshore, to N. furcosus and N. nematopus on the mid-shelf, to N. balinensoides on the outer shelf. Ubiquitous families of the highly evolved tetraodontiformes had advanced dentition and anti-predator defenses that no doubt enabled them to take advantage of the abundant, but often poor quality, sources of food in the vast plains of muddy and carbonate sediments. Boosted regression trees (BRT) were used to predict species responses to the environmental covariates, to identify important gradients and understand surrogacy and competition amongst the spatial and environmental predictors in models. The position of sites across and along the shelf gave the most parsimonious and easily interpretable models of species richness and assemblage structure, but the underlying gradients in properties of the sediments, epibenthos and water column were not so linear in these two dimensions. There were clearly three sections of the GBRMP, separated by ecotones, which differed in their flushing regime, tidal energy, oceanic influences, epibenthic habitats and sedimentary facies. The species responses and assemblage structure are discussed in terms of these influences on benthic communities and productivity of the water column, the functional morphology of the inter-reef vertebrates and the prevailing paradigms for tropical shelf faunas. In this study I conclude that spatial position and depth on linear tropical shelves are fundamental surrogates to provide insight on the major environmental drivers acting together to shape spatial gradients in vertebrate distributions, as well as identifying boundaries for management interventions.. viii.
(12) Contents Statement of Access ....................................................................................................................... i Statement of Sources ..................................................................................................................... ii Contributions................................................................................................................................ iii Electronic Copy..............................................................................................................................v Acknowledgements ...................................................................................................................... vi Abstract ....................................................................................................................................... vii List of Tables and Appendices ................................................................................................... xiii List of Figures ........................................................................................................................... xvii Acronyms and Abbreviations ................................................................................................... xxvi 1.. GENERAL INTRODUCTION .........................................................................................1. 2.. GENERAL METHODS ....................................................................................................7 2.1. A review of baited video techniques to estimate relative abundance of fish .............7 2.1.1 General approaches and applications.............................................................8 2.1.2 Estimation of abundance from fish sightings ................................................12. 2.2. Baited remote underwater video stations (BRUVS) ................................................15. 2.3. Procedures for tape interrogation ............................................................................23. 2.4. Statistical analyses of indices of richness and abundance .......................................25 2.4.1 Classification and regression trees (CART) ..................................................25 2.4.2 Boosting ........................................................................................................27 2.4.2.1 Reporting prediction errors ..............................................................28 2.4.3 Multivariate regression trees (MRT) .............................................................29 2.4.4 Species indicators for site groups ..................................................................30 2.4.5 Smoothing splines .........................................................................................30 2.4.6 Statistical software ........................................................................................31. 3.. HOW DOES THE USE OF BAIT AFFECT ABILITY TO DISTINGUISH DEMERSAL FISH ASSEMBLAGES? ..........................................................................32 3.1. Introduction .............................................................................................................32. 3.2. Methods ...................................................................................................................33 3.2.1 Univariate analyses .......................................................................................34 3.2.2 Multivariate analyses.....................................................................................35. 4.. 3.3. Results .....................................................................................................................36. 3.4. Discussion................................................................................................................54. HOW DOES TIME OF DAY AND FUNCTIONAL MORPHOLOGY OF FISHES AFFECT THE ABILITY OF BRUVS TO DISTINGUISH INTERREEF FISH ASSEMBLAGES? ......................................................................................59. ix.
(13) 4.1. Introduction............................................................................................................. 59. 4.2. Methods .................................................................................................................. 60 4.2.1 Statistical analysis......................................................................................... 63. 4.3. Results..................................................................................................................... 64 4.3.1 Species richness ............................................................................................ 64 4.3.2 Description of patterns in fish assemblages ................................................. 73 4.3.3 Indicator species for the ten fish assemblages .............................................. 73 4.3.4 Predicting group membership....................................................................... 74 4.3.5 Logistical consideration................................................................................ 79. 4.4. Discussion ............................................................................................................... 79 4.4.1 ‘Trawl ground’ species show over-dispersion in relative abundance ........... 80 4.4.2 BRUVS discriminated better amongst site groups ....................................... 81 4.4.3 Both techniques show selectivity ................................................................. 81 4.4.4 The role of BRUVS in assessments of seafloor biodiversity ....................... 83. 5.. ENVIRONMENTAL GRADIENTS AND SHELF-SCALE PATTERNS OF SPECIES RICHNESS IN INTER-REEFAL WATERS ............................................... 91 5.1. Introduction............................................................................................................. 91. 5.2. Methods .................................................................................................................. 93 5.2.1 Sampling area and species identifications .................................................... 93 5.2.2 Environmental covariates ............................................................................. 96 5.2.3 Data analysis............................................................................................... 100 5.2.3.1 Correlations amongst explanatory variables................................. 100 5.2.3.2 Mapping of dependencies of environmental variables and species richness ............................................................................. 100 5.2.3.3 Testing for a ‘mid-domain effect’ along environmental gradients ........................................................................................ 101. 5.3. Results................................................................................................................... 103 5.3.1 Environmental covariates ........................................................................... 103 5.3.2 Broad-scale spatial patterns in hydrological and sedimentary explanatory variables .................................................................................. 107 5.3.3 Broad-scale patterns of epibenthic habitats ................................................ 116 5.3.4 Species occurrence and site richness .......................................................... 119 5.3.5 The influence of shelf position and depth on species richness ................... 122 5.3.6 The comparative influence of spatial and environmental covariates on species richness .......................................................................................... 125 5.3.7 Did a random ‘mid-domain effect’ account for the ‘hotspots’ in species richness? ..................................................................................................... 130. x.
(14) 5.4. Discussion..............................................................................................................135 5.4.1 Broad regional differences in the three sections of the GBR ......................135 5.4.2 The nature of species records in BRUVS data ............................................137 5.4.3 Spatial and environmental influences on species richness ..........................138 5.4.4 Relevance to biogeographic models ............................................................139. 6.. SHELF-SCALE PATTERNS OF VERTEBRATE ASSEMBLAGES IN THE INTER-REEFAL WATERS OF THE GREAT BARRIER REEF MARINE PARK ..............................................................................................................................143 6.1. Introduction ...........................................................................................................143. 6.2. Methods .................................................................................................................146 6.2.1 Statistical analysis .......................................................................................146. 6.3. Results ................................................................................................................... 147 6.3.1 Patterns in vertebrate communities .............................................................147 6.3.2 Diversity and abundance of assemblages ....................................................153 6.3.3 Indicator species for assemblages ...............................................................157. 6.4. Discussion..............................................................................................................162 6.4.1 Cross-shelf patterns .....................................................................................162 6.4.2 Along-shelf patterns ....................................................................................163 6.4.3 Comparison with other tropical shelves ......................................................166. 7.. SPATIAL MODELS EXPLAINING AND PREDICTING THE OCCURRENCE OF COMMON INTER-REEF VERTEBRATES OF THE GREAT BARRIER REEF MARINE PARK ...............................................................169 7.1. Introduction ...........................................................................................................169. 7.2. Methods .................................................................................................................171. 7.3. Results ...................................................................................................................173 7.3.1 Prediction of species occurrence using environmental explanatory variables ......................................................................................................173 7.3.2 ‘Across-shelf’ influences on species-environment relationships ................186 7.3.3 ‘Along-shelf’ influences on species-environment relationships .................195 7.3.3.1 Spatial predictions of species occurrence ......................................204. 7.4. Discussion..............................................................................................................216 7.4.1 The influence of depth and position on the shelf ........................................216 7.4.2 The influence of sediment composition ......................................................218 7.4.3 Positive and negative responses to seafloor complexity .............................219 7.4.4 Are influences of the water column surrogates for nutrient inputs? ...........221 7.4.5 Cryptic species and sampling gaps may explain high predictions at extremes in parameter space .......................................................................223 xi.
(15) 8.. 9.. xii. GENERAL DISCUSSION ............................................................................................ 225 8.1. BRUVS as a powerful sampling tool .................................................................... 225. 8.2. Biophysical maps show ‘hotspots’ in species richness ......................................... 228. 8.3. Ecotones separate ‘lagoonal’ and ‘reefal’ vertebrate assemblages........................ 230. 8.4. Species replacements across ecotones and along gradients .................................. 234. 8.5. The most prevalent species were ‘habitat generalists’ .......................................... 235. 8.6. Higher primary productivity is reflected in regional species responses ............... 237. 8.7. Cross-chapter comparisons of heuristic and mechanistic predictors .................... 240. 8.8. Conclusion ............................................................................................................ 244. BIBLIOGRAPHY ......................................................................................................... 246.
(16) List of Tables and Appendices Table 2.1.. Examples of baited video studies. Abbreviations are HBRUVS/VBRUVS (Horizontal/Vertical baited remote underwater video stations), VBUV /HBUV (Vertical (V) or Horizontal (H) baited underwater closed circuit television), SBRUVS (Stereo horizontal baited remote underwater video stations) and MPA (Marine Protected Areas) .......................................................9. Table 3.1.. Fitted, mean values of the effects of bait X location +(1|location) on species richness (S) and abundance (MaxN), using a quasipoisson link function to account for overdispersion in the raw data. The number of pairs of sets of baited (BRUVS) and unbaited (RUVS) video units are shown for each location (n). On average across locations, there were about 2.3 times as many species and 3.1 times as many individual fish recorded on baited BRUVS when compared to unbaited units………… ..........38. Table 3.2.. Species sighted only on baited BRUVS and only on unbaited RUVS ...............39. Table 3.3.. Node names and species indicators (DLIs) for the terminal nodes (leaves) of the MRT examining the effects of bait (Figure 3.3). The number of sites in the groups are shown, and the percentage of moderate DLIs (>19) are outlined in brackets for each node. The ranges and means ± standard deviation (in brackets) are given for abundance (abund= ∑in MaxNi) and richness ...............................................................................................................45. Table 3.4.. Distance-based redundancy analysis of full dissimilarity matrices for transformed baited (BRUVS) and unbaited (RUVS) samples at 63 sites in six locations. MSEd = the Mean Square of Euclidean distance .........................49. Table 3.5.. Prediction of membership of video sets by location using single drop-out linear discriminant analysis for varying numbers of discriminants from principal coordinates analysis (PCoA) ...............................................................53. Table 3.6.. Misclassification table for the use of baited (BRUVS) video and the first two discriminants (dimensions 1 and 2). Diagonal numbers in bold font show successful predictions by the model..........................................................53. Table 3.7.. Misclassification table for the use of unbaited (RUVS) video and the first two discriminants (dimensions 1 and 2). Diagonal numbers in bold font show successful predictions by the model..........................................................54. Table 4.1.. Total number of families, species, and individuals (n fish) recorded by 19 trawls and 95 BRUVS sets along 19 transects. Average number of species and n fish for these transects are shown with standard errors and ranges ..........65. xiii.
(17) Table 4.2.. Number of species and families (brackets) recorded by day and night and in total by 19 trawls and 95 BRUVS sets along 19 transects ............................ 65. Table 4.3.. Families recorded by only BRUVS or by only trawls in descending order of abundance. The number of species (n spp) is shown for each family in brackets, with total number of fish (n fish), the number of transects along which the family was recorded (n transects) and the average number of that family recorded on those transects, with standard errors............................ 68. Table 4.4.. Abundance (n fish) and occurrence (n transects) of species common to both techniques, ranked in descending order of abundance in the BRUVS records for the 19 transects. Average abundance and standard errors are shown for each species for those transects on which they were recorded ......... 69. Table 4.5.. Distance-based redundancy analysis of full dissimilarity matrices for all transformed BRUVS and trawl data .................................................................. 75. Table 4.6.. Distance-based redundancy analysis of full dissimilarity matrices for transformed BRUVS and trawl data restricted to 38 species recorded by both techniques .................................................................................................. 75. Table 4.7.. Prediction of treatment groups using single drop-out linear discriminant analysis (LDA) for varying numbers of variables from principal coordinates analysis (PCoA).............................................................................. 76. Table 4.8.. The functional morphology, habits and approximate, reported size range of adults (or juveniles, as indicated with an asterisk) (after Gloerfelt-Tarp and Kailola 1984; Sainsbury et al. 1985) of the genera and families mentioned in the text and tables, grouped by their form and trophic level ....... 86. Table 5.1.. Definition of 28 explanatory variables supplied, or derived, for use in multivariate analyses. ........................................................................................ 98. Table 5.2.. Spearman rank correlation matrix for the entire set (n=28) of explanatory variables where the modulus of correlations >0.29. Field names are defined in Table 5.1. The largest correlations (>0.69) are highlighted in bold. In this case, average water temperature (Temp.av) was positively correlated, and salinity (Salin.av) was negatively correlated, with increasing position northward along the shelf (‘along’) .................................. 104. Table 5.3.. Variation in predictor variables explained by general additive models with smoothed spline terms, and three hundred degrees of freedom. The models were based on information supplied by the CSIRO for 1,531 sites. The high values of the adjusted r2 show that these environmental variables, supplied by the CSIRO, were spatially interpolated by latitude and longitude to a high degree ......................................................................... 106. xiv.
(18) Table 5.4.. The diagnostics for the predictors of species richness, showing the relative importance of each predictor (% var. rel. influence), the percentage change in prediction error (% change) after its omission from the model, and the significance of the omission of predictors based on permutation tests (Pr>|z|; n = 5000 permutations). A decline is denoted by the – symbol. The full model, with monotonic main effects, had a relative prediction error (%PE) ~65.2% ........................................................................126. Table 5.5.. Percentage of the variation in interpolated (bold) covariates and other environmental variables, at the 366 BRUVS sites, predicted by spatial position and depth on the shelf of the GBR .....................................................127. Table 5.6.. The numbers of 347 species with locations and ranges across and along the GBR shelf between (within), and outside (below and above) the bounds of boot-strapped random distributions based on their probability of occurrence in the dataset ..............................................................................131. Table 6.1.. Values for environmental covariates producing primary and surrogate splits (nodes) on the left (1, 2, 4, 5) ‘reefal’ and right (3, 6, 7, 12, 14) ‘lagoonal’ sides (branches) of the tree in Figure 6.1. The ‘improvements’ in the model at each split are represented by the decrease in relative error from the first to the last split. The model has explained (1-0.737=26.3%) of the variation amongst the occurrence of 172 species amongst 366 BRUVS sites. The percentage improvement in each split by the primary and surrogate splitting variables shows that there were numerous correlations amongst the spatial and environmental covariates – especially the interpolated values for salinity, temperature and sediment composition ......................................................................................................151. Table 6.2.. Summaries of richness and abundance in the ten terminal fish assemblages. See Table 6.1 and Figure 6.1 for an explanation and summary of MRT node numbers and abbreviated node names. Overall richness and (raw) abundance are shown for the membership of each node, together with ranges, means and standard deviations in these parameters for n sites within nodes. The MRT was based on 172 species of vertebrates, with prevalence >3, and 19 spatial and environmental covariates ..........................................................................................................154. Table 6.3.. Summaries of the Dufrêne- Legendre Index (DLI) for all 172 species in all nine higher nodes (branches – B) and ten terminal nodes (leaves – L). For a given species and a given group of sites, the DLI was defined as the product of the mean species abundance occurring in the group divided by xv.
(19) the sum of the mean abundances in all other assemblages (specificity), times the proportion of sites within the assemblage where the species occurred (fidelity), multiplied by 100 .............................................................. 160 Table 7.1.. Results of the univariate and multivariate BRT for the 36 most prevalent species (y) at n=366 sites using all 28 explanatory variables (x). The species were present at ‘occ’ sites, and ‘sdt’ was the number of correct predictions, on average, over 1,000 univariate BRT with five-fold crossvalidation. Relative prediction error was ‘rel.PE’ = (1 –(sdt/n))%, and the percentage of the variation in occurrence of each species explained by the best gbm model was ‘%Var’. The species were ranked in decreasing order of their average ‘predictability’, represented by the summed, average influence of all 28 predictor variables (‘%var.infl‘) ........................... 174. Table 7.2.. The 28 explanatory variables (x) sorted by descending order of the average influence of each explanatory variable (%var.infl) across all the 36 responses (species occurrences). In combination, these 28 covariates accounted for 83.1% of the prediction success and 86.8% of the variation for the 36 species in Table 7.1. The column rel.var.infl% expressed this average influence as a percentage of the grand average %var.infl = 13.2 in Table 7.1. Variable names were defined in Table 5.1 ...................................... 178. Table 7.3.. The influence (%var.infl) of 28 explanatory variables (x) on 36 species (y) in the BRT. Cell values show the relative amount of influence by each variable in explaining occurrence of species, so row totals equal %Var in Table 7.1. Variable names were defined in Table 5.1 ...................................... 181. Table 8.1.. Summary of the importance of the families of heuristic, spatial variables, and the mechanistic, environmental covariates used in data exploration and predictive models of richness, assemblage structure and species occurrence in the final three data chapters. Importance is expressed as increasing rank of influence (in models), primacy or surrogacy as splitting variables in regression trees, and qualitative notes on the strength of responses. Comments are also made on the sources of influences of the variables and some key species and assemblages associated with each variable……………………………………………………………................. 241. Appendix 5.1. Definition of species groups identified with the suffix ‘_grp’ in this thesis. The use of ‘_grp’ indicated that there was uncertainty in identification, due to the limitations of video footage, but also that the name preceding the suffix was the most plausible level of identification.................................. 141. xvi.
(20) List of Figures Figure 2.1.. BRUVS prototypes used in Chapters 3 and 4 (A) and subsequent chapters (B). For night use, prototype (A) had lights powered by a 12 Volt gel-cell battery enclosed in a housing .............................................................................17. Figure 2.2.. Rubber bonnet tie-downs held housings on camera arm clamps (A, C). Bolting camera arms through slots allowed 10 degrees of tilt (B). A locator lug mated with a socket in the camera housing faceplate (D) ................18. Figure 2.3.. The lug and socket in each leg of the frame (A) used #18 gauge (1.25mm) galvanised wire and a B10 ‘R’ Clip to enable (B,C) or bypass the weaklink (D) ...............................................................................................................19. Figure 2.4.. Rear (A) and front view (B) of a housing showing the fixed, female dovetail plate for camera, and the lug in the faceplate (B) that fitted a locking pin on the camera arm (Figure 2.2). The locator pin on the dovetail plate (B) locked onto a female lug on the camera baseplate (C). The camera was screwed onto the male plate (C) and slid into and out of the female dove-tail joint with the aid of a wire handle (D) ................................................20. Figure 2.5.. Loading of camera (A,B) and ballast weights (C,D) to a BRUVS .....................21. Figure 2.6.. Once the camera was loaded (A), the floats and rope were streamed astern of the vessel first (B) , and tied off (C) to await the final position of the drop (D) ....................................................................................................22. Figure 2.7.. A grapple was cast to snag the buoy line and bring it through a snatch block for hauling with an hydraulic pot-hauler wheel (A) .................................22. Figure 2.8.. Tape interrogation interface from BRUVS2.1.mdb© (A). Reference image for Pristipomoides multidens, with Lutjanus sebae, L. adetii and Epinephelus undulatostriatus and E. areolatus in the background (B) ..............24. Figure 3.1.. Location of video sampling sites in the central GBRMP. Triangle symbols represent baited BRUVS (filled symbols point upwards) and unbaited RUVS (open symbols point downwards). ..........................................................34. Figure 3.2.. Box and whisker plots of the raw species richness (S) and transformed abundance (log10 MaxN). The boxplots show the median and 95% Confidence Intervals. The notches represent 1.5 x (interquartile range of MaxN/SQRT(n)). If the notches do not overlap this is ‘strong evidence’ that the two medians differ, independent of any assumptions about normality of data distributions or equivalence of variances (see Chambers et al. 1983, p. 62). ...............................................................................................37. xvii.
(21) Figure 3.3.. Multivariate regression tree analysis (MRT) of the transformed abundance of all 210 species at 126 sites. The top six species indicators are shown with DLI values at each node ........................................................... 44. Figure 3.4.. Multivariate regression tree analysis (MRT) of the transformed abundance of all 40 families at 126 sites. The top six indicator families are shown with DLI values at each node ........................................................... 47. Figure 3.5.. Species accumulation curves for the nine terminal nodes of the MRT in Figure 3.4 ........................................................................................................... 48. Figure 3.6.. Principal Coordinates Analysis (PCoA) of baited (BRUVS, A) and unbaited (RUVS, B) video units. Group means of the six locations are shown ................................................................................................................. 50. Figure 3.7.. Linear Discriminant Analysis (LDA) of baited (BRUVS, A) and unbaited (RUVS, B) video units. Centroid means ± 1 Standard Error of the six locations are shown ........................................................................................... 52. Figure 4.1.. Location of 19 trawl and BRUVS transects either side of Cape Grafton. The arrows are scaled precisely to the trawl path and point in the trawl direction. Adjacent coordinates of each numbered BRUVS replicate are shown. Bolded arrows represent night samples ................................................. 62. Figure 4.2.. Species accumulation curves (method = ‘random’) for the data pooled by sampling technique ............................................................................................ 65. Figure 4.3.. Scatter plots of the average total length of each species, measured or estimated, from trawl catches and BRUVS sightings. Abundance has been scaled using the fourth root transformation. Each point represents the number of individuals of a species. Filled symbols represent species recorded by both techniques and open symbols represent species unique to each technique ............................................................................................... 71. Figure 4.4.. Multivariate regression tree (MRT) for BRUVS and trawl transects, showing top six DLI species scores for nodes. The bar plots show the distribution of species abundance at each of the terminal nodes, ranked from left to right in decreasing order of prevalence in the entire data set, with each vertical bar representing the mean abundance of a species in that group ........................................................................................................... 72. Figure 4.5.. Principal Coordinates Analysis (PCoA) for species recorded by BRUVS (A) and trawls (B). The PCoA was based on extended dissimilarities calculated from species abundances which were transformed and row standardised. The 19 transects within each of the six treatment groups (three locations by day-night) are outlined by polygons. The locations are. xviii.
(22) Double Island (DI), Double Island Wide (DIW) and Scott Reef (SR), the filled and open symbols are night and day sets ..................................................77 Figure 4.6.. Linear discriminant analysis (LDA) plots based on the first two principal coordinates and six fish assemblages, defined by location and day-night, for the transformed BRUVS data (A) and trawls (B). The circles denote one Standard Error about the group means, and the symbols denote the transect means and assemblage membership. All conventions are the same as those for Figure 4.5 ...............................................................................78. Figure 5.1.. Rotated maps of the Great Barrier Reef Marine Park (GBRMP) showing (a) locations of the reef matrix (olive). All 1,531 sampling sites (blue), including the 381 BRUVS sampling sites (orange), are shown without the reef matrix (b).....................................................................................................95. Figure 5.2.. Patterns of variation of location (a) across and (b) along the shelf for the study area (rotated) smoothed using thin plate splines with three hundred degrees of freedom (see Chapter 2.4.5). Distance along was set to range from 0 at the southern end to 1 at the far northern end. Distance across was 0 on the coast and 1 on the 80m isobath ...................................................102. Figure 5.3.. Plots of (a) depth and (b) the seabed current shear stress (Current; Newtons per square metre) interpolated for the entire GBRMP by general additive models with smoothed spline terms. The models were based on information supplied by the CSIRO for 1,531 sites ........................................109. Figure 5.4.. Plots of (a) the average (Salin.av) and (b) standard deviation (Salin.sd) in salinity at the seabed. Heat colour contours represent the relationship between each covariate and the position of 1,531 locations, interpolated to the entire GBRMP by general additive models with smoothed spline terms ................................................................................................................. 110. Figure 5.5.. Plots of (a) the average (Temp.av) and (b) standard deviation (Temp.sd) in water temperature at the seabed interpolated by CSIRO in the GBRMP. All other conventions as per Figure 5.4............................................................ 111. Figure 5.6.. Plots of (a) the percentage composition of ‘carbonate’ (carbnte.pc) and (b) ‘mud’ (mud.pc) in sediment fractions interpolated by the CSIRO in the GBRMP. All other conventions as per Figure 5.4 ...................................... 112. Figure 5.7.. Plots of (a) the percentage of the towed video track on each site where ‘rocky’ substratum was present, and (b) measures of ‘location’ (mean: ‘rugosity.vid.av’) and (c) ‘spread’ of seabed rugosity (standard deviation: ‘rugosity.vid.sd’) on tracks. Symbols portray site measurements scaled to. xix.
(23) the maximum value at all sites for that covariate. Coloured symbols show sites with values less than (green) or greater than (red) the overall mean. ...... 113 Figure 5.8.. Plots of (a) the average percent cover within still video frames of ‘large boulders’ at sites, and measures of (b) ‘location’ (mean: ‘rugosity.pho.av’) and (c) ‘spread’ (standard deviation: ‘rugosity.pho.sprd’) of an index of seabed rugosity in still video frames. All other conventions as per Figure 5.7. ................................................................................................................... 114. Figure 5.9.. Plots of the percentage composition of (a) ‘sand’ (sand.pc) and (b) ‘gravel’ (gravl.pc) in sediment fractions interpolated by the CSIRO in the GBRMP. All other conventions as per Figure 5.4............................................ 115. Figure 5.10.. Plots of (a) the average percent of video tracks where marine ‘plants’ were recorded (plant.pc.vid), and average percentage cover of still video frames occupied by (b) ‘seagrasses’ (seagr.pc.pho) and (c) ‘algae’ (including Halimeda spp; algae.pc.pho). Symbols portray site measurements scaled to the maximum value at all sites for that covariate. Symbol colours show sites with values less than (green) or greater than (red) the overall mean. ..................................................................................... 117. Figure 5.11.. Plots of (a) the average percentage of video tracks where ‘megabenthos’ was present (mgbnths.pc.vid), and (b) the average number of still video frames on which ‘megabenthos’ occurred. All other conventions follow Figure 5.10. ...................................................................................................... 118. Figure 5.12.. Summaries of (A) species richness by cumulative number of sites, and (B) prevalence of 347 species at 366 BRUVS sites ranked in descending order of occurrence .......................................................................................... 120. Figure 5.13.. A smoothed spline fit of species richness (k=150 degrees of freedom) by latitude and longitude (a). The right panel shows the observed richness on which the fit was based (b), scaled to the maximum value and colourcoded according to the mean value amongst all sites ...................................... 121. Figure 5.14.. Relative influence of location and depth as predictors of species richness in a gradient boosting model (a), and dependency of site species richness on location (b, d) and depth (c). Richness was constrained to increase monotonically in response to distance along the shelf. Distance along the shelf ranges from 0 at the southern end of the GBRMP to 1 at the far northern end (corresponding degrees in latitude are given in brackets). Distance across takes the value 0 on the coast and 1 on the 80m isobath. The short dashed lines (rugs) along the x-axes indicate the ten percentiles in location of the BRUVS sites. Values were predicted for each variable,. xx.
(24) holding values for both other variables at their mean for the BRUVS dataset. Grey lines indicate two standard errors for the predicted values, estimated from predictions made from five hundred trees fitted in fivefold cross validation of the BRUVS dataset .....................................................123 Figure 5.15.. Partial dependency of site species richness at five distances across the shelf on depth (a), and the distance along the shelf (b). Other conventions described in Figure 5.14 ...................................................................................124. Figure 5.16.. Partial dependency plots of richness and six environmental predictors (untransformed) as a function of spatial location and depth. The responses are centered on their mean values at y=0. The dotted line indicates values of ‘Across’ ~0.8. All other conventions and definitions as per Figure 5.15 and Table 5.1 ....................................................................................................129. Figure 5.17.. Species ranges across the GBRMP. The 347 species were ranked by prevalence from y=1 (Nemipterus furcosus at 192 sites) to the numerous singletons at y=250:347. Symbols representing the median value of ‘across’ for each species were scaled and coloured by Log10 of the number of BRUVS sites on which the species was found. Species found on more than 100, 40-100, and less than 40 sites were represented by light blue, orange and black symbols respectively. Horizontal lines show species ranges. Vertical lines show the median value of all the 366 sites, and across=0.8 ..................................................................................................132. Figure 5.18.. Spatial ranges of 347 species along the GBRMP. All conventions follow Figure 5.17 .......................................................................................................133. Figure 5.19.. Plots of locations and ranges of all 347 species across and along the GBRMP shelf. The vertical bars indicate 95% confidence intervals under the assumption that the taxa are randomly distributed conditional on their observed probability of occurrence. Species beyond the bounds expected under this assumption (open circles), and species within these bounds (grey circles) are shown with mean observed values (dashed) and maximum values expected under random distributions (dotted)......................134. Figure 6.1.. A MRT based on 172 species of vertebrates and the nineteen explanatory variables. Terminal node numbers were summarised in Table 6.1, and mapped with abbreviated names in Figure 6.2 .................................................150. Figure 6.2.. The location of BRUVS sites within the ten vertebrate assemblages divided into ‘lagoonal’ (A) and ‘reefal’ (B) groups within regions of the GBRMP (rotated). The node numbers in the legend link to Figure 6.1 and Table 6.1, where full descriptions were provided .............................................152 xxi.
(25) Figure 6.3.. Species-accumulation curves for the ‘reefal’ (a) and ‘lagoonal’ (b) assemblages of vertebrates. For definitions of the assemblage names see Table 6.1 .......................................................................................................... 156. Figure 6.4.. Node names and top ten species indicators (DLIs) for the terminal nodes (leaves) of the MRT defined by the 19 explanatory variables. Terminal node numbers were summarised in Table 6.1 and mapped with abbreviated names in Figures 6.1 and 6.2........................................................ 159. Figure 7.1.. The performance of three scenarios of covariate selection in predicting the occurrence of vertebrate species in the GBRMP. The heuristic variables “across”, “along” and depth (aad), and 25 mechanistic, environmental variables (env) were analysed together and separately in the three scenarios. The top 36 species present on at least nine percent of sites were selected for plotting, and these were not necessarily ranked in the same order by prediction success for each scenario. The Y axes show the sum of average influences of explanatory variables (%var.infl in Table 7.1) from the cross-validated gbm models for each species in each scenario. A mixture of spatial and environmental covariates produced the optimal predictions (grey line), but use of position and depth alone produced reasonable prediction success (green line). ...................................... 176. Figure 7.2.. A ‘Heatmap’ plot showing the 36 most prevalent species clustered by their responses to the influence of 28 covariate predictors, which are also clustered on the horizontal axis by their level of influence on the species. The ‘hotness’ of cell colours reflects the strength of the correlations. Each cell is a partial regression plot, where the curves are logit functions (log2[p/(1-p)]) of the occurrence of each species explained by levels of each predictor. The curve colour ranges from blue for strong responses to grey for little influence. The y-axes of these plots were adjusted to the maximum y value for each species; so many minor relationships appear flat. Covariates were clustered in the upper dendrogram by their similarity with each other in terms of relative influence on the species, and species were clustered on the left dendrogram by the similarity in response to the influence of the covariates. For example, average temperatures and salinities were similar in their levels of influence on species (upper x clusters) , and the presence of Upeneus tragula_grp, Pentapodus paradiseus and Lethrinus genivittatus were highly influenced by high records of ‘marine plants’ on towed video footage (y clusters). The direction of influence was not always the same amongst species in. xxii.
(26) these clusters. For example, four species clustered by their response to position ‘across’ the shelf, but Pentapodus nagasakiensis increased with distance offshore, unlike the other three species most common inshore ..........184 Figure 7.3.. Species occurrence as a function of location across the shelf. Plots are ranked in descending order of relative influence of the predictor variable for the 25 most predictable species. The ‘rugs’ on the X axes are 10 percentiles in the distribution of the predictor variable. The Y axes (logodds) are logit log2[p/(1-p)]) and centered (dashed line; y=0) on a probability of occurrence of p=0.5 ...................................................................188. Figure 7.4.. Species occurrence as a function of a measure of the percentage of carbonate in sediments at BRUVS sampling sites (carbnte.pc). All conventions follow Figure 7.3 ..........................................................................189. Figure 7.5.. Species occurrence as a function of a measure of the percentage of mud in sediments at BRUVS sampling sites (mud.pc). All conventions follow Figure 7.3 .........................................................................................................190. Figure 7.6.. Species occurrence as a function of a measure of the percentage of the coarse fraction of sediments at BRUVS sampling sites (coarsns.pc). All conventions follow Figure 7.3 ..........................................................................191. Figure 7.7.. Species occurrence as a function of a measure of location (rugosity.vid.av) in towed video classification of substratum rugosity at BRUVS sampling sites. All conventions follow Figure 7.3 .............................192. Figure 7.8.. Species occurrence as a function of a measure of depth at BRUVS sampling sites. All conventions follow Figure 7.3 ...........................................193. Figure 7.9.. Species occurrence as a function of distance to the nearest coral reef centroid for BRUVS sampling sites (dist.reef). All conventions follow Figure 7.3 .........................................................................................................194. Figure 7.10.. Species occurrence as a function of position along the shelf. All conventions follow Figure 7.3 ..........................................................................197. Figure 7.11.. Species occurrence as a function of average annual salinity at the seabed (Salin.av; ppt). All conventions follow Figure 7.3 ...........................................198. Figure 7.12.. Species occurrence as a function of average annual temperature at the seabed (Temp.av; degrees Celsius). All conventions follow Figure 7.3 ...........199. Figure 7.13.. Species occurrence as a function of the seabed current shear stress (Current; Pascals; Newtons per square metre) at BRUVS sampling sites. All conventions follow Figure 7.3 ....................................................................200. xxiii.
(27) Figure 7.14.. Species occurrence as a function of the percentage of towed video footage at BRUVS sampling sites comprised of ‘marine plants’ (plant.pc.vid). All conventions follow Figure 7.3 ............................................ 202. Figure 7.15.. Species occurrence as a function of the percentage of towed video footage at BRUVS sampling sites comprised of ‘megabenthos’ (mgbnths.pc.vid). All conventions follow Figure 7.3 ...................................... 203. Figure 7.16.. Smoothed spline predictions (± 3 std errors) of the probability of presence of ‘ubiquitous’ species across and along the GBRMP. Adjusted r2 values are shown for each fit. Symbols show presence (blue) or absence (open) at BRUVS sites, scaled by abundance to the maximum measurement for each species. The contribution to the total dataset is shown in terms of occurrence (occ%) and abundance (abun%). The top three environmental influences are shown from Table 7.3 ............................................................... 206. Figure 7.17.. Smoothed spline predictions (± 3 std errors) of the percentage of mud in sediments (a) and the probability of presence of some common Nemipterus influenced by mud (b,c). Other conventions follow Figure 7.16 .................................................................................................................. 207. Figure 7.18.. Smoothed spline predictions (± 3 std errors) of the percentage of mud in sediments (a) and the probability of presence of common demersal and semi-pelagic predators influenced by mud (b,c). Other conventions follow Figure 7.16 ........................................................................................... 208. Figure 7.19.. Smoothed spline predictions (± 3 std errors) of the percentage of mud in sediments (a) and the probability of presence of some regionally common demersal predators influenced by mud (b,c). Other conventions follow Figure 7.16 ....................................................................................................... 209. Figure 7.20.. Smoothed spline predictions (± 3 std errors) of the percentage of gravel in sediments (a) and the probability of presence of some common benthic microcarnivores influenced by gravel (b,c). Other conventions follow Figure 7.16 ....................................................................................................... 211. Figure 7.21.. Smoothed spline predictions (± 3 std errors) of the percentage of carbonate in sediments (a) and the probability of presence of a common benthic microcarnivore (b) and a macrocarnivore (c) influenced by carbonate levels. Other conventions follow Figure 7.16 ................................. 212. Figure 7.22.. Smoothed spline predictions (± 3 std errors) of the average annual salinity (a) and the probability of presence of some common Nemipterus influenced by salinity (b,c). Other conventions follow Figure 7.16 ................ 213. xxiv.
(28) Figure 7.23.. Smoothed spline predictions (± 3 std errors) of the average annual water temperature (a) and the probability of presence of some (b) semi-pelagic and (c) demersal microcarnivores influenced by temperature. Other conventions follow Figure 7.16 ........................................................................214. Figure 7.24.. Smoothed spline predictions (± 3 std errors) of the probability of presence of a small lethrinid (b) and a pomacentrid (c) influenced by covariates representing marine plants. The left panel (a) shows sampling stations scaled by the percentage cover of plants on towed video footage where red symbols indicate measurements higher than the mean values. Other conventions follow Figure 7.16. .......................................................................215. xxv.
(29) Acronyms and Abbreviations ABT ...................... Aggregated boosted regression trees AIMS .................... Australian Institute of Marine Science ANOVA ................ Analysis of Variance AUDOS ................ University of Aberdeen deep ocean submersible camera system BRD...................... Bycatch reduction devices BRT ...................... Boosted regression trees BRUVS................. Baited remote underwater video stations BUV ..................... Baited underwater video CAP ...................... Canonical analysis of principal coordinates CART ................... Classification and regression trees CCD...................... Charge coupling device CEO ...................... Chief Executive Officer CPUE .................... Catch-per-unit-effort CRC ...................... Cooperative Research Centre CSIRO .................. Commonwealth Scientific and Industrial Research Organisation DLI ....................... Dufrêne-Legendre Index EAC ...................... East Australian Current EFH ...................... Essential fish habitat FRV ...................... Fisheries Research Vessel GBRMP ................ Great Barrier Reef Marine Park H ........................... Horizontal HBUV................... Horizontal baited underwater closed circuit television HBRUVS .............. Horizontal baited remote underwater video stations LDA...................... Linear discriminant analysis LME ..................... Large marine ecosystem MaxN…. ............... Maximum number (of individuals for each species) seen at one time on tape MAXNO ................ Maximum number (of individuals for each species) seen per specified time interval MPA...................... Marine Protected Area MRT ..................... Multivariate regression tree analysis PE ......................... Prediction error PVC ...................... Polyvinyl chloride QDPIF .................. Queensland Department of Primary Industries and Fisheries QM ....................... Queensland Museum PCoA .................... Principal coordinates analysis. xxvi.
(30) rda ........................ Redundancy analysis ROBIO ................. RObust BIOdiversity lander (University of Aberdeen autonomous underwater camera system) RV ........................ Research Vessel S…........................ Species richness SAC ...................... Species accumulation curve(s) SBRUVS .............. Stereo baited remote underwater video station(s) SCUBA ................ Self contained underwater breathing apparatus SEC ...................... South Equatorial Current SSD ...................... Sum of squared deviances SST ....................... Sea-surface temperature TED ...................... Turtle exclusion device TFAP..................... Time of first arrival TITO..................... Time in/ time out TL ......................... Total length TOTTM ................. Total duration of visit during a sequence UVC ..................... Underwater visual census UVS...................... Underwater visual surveys V ........................... Vertical VBUV .................. Vertical baited underwater closed circuit television VBRUVS ............. Vertical baited remote underwater video stations. xxvii.
(31) “The true spirit of delight, the exaltation, the sense of being more than Man, which is the touchstone of the highest excellence, is to be found in mathematics as surely as poetry.” Bertrand Russell (1872-1970), philosopher.
(32)
Related documents