Marine benthic biodiversity can be measured using a range of sampling methods, including benthic sleds or trawls, grabs, and imaging systems, each of which targets a particular community or habitat. Due to the high cost and logistics of benthic sampling, particularly in the deep sea, studies are often limited to only one or two biological sampling methods. Results of biodiversity studies are used for a range of purposes, including species inventories, environmental impact assessments, and predictive modelling, all of which underpin appropriate marine resource management. However, the generality of marine biodiversity patterns identified among different sampling methods is unknown, as are the associated impacts on management decisions. This report reviews studies that have used two or more sampling methods in order to determine the consistency of their results among gear types, as well as the optimum combination of gear types. In addition, we directly analyse data that were acquired using multiple gear types to examine the consistency of biodiversity patterns among different gear types. These data represent two regions: 1) Joseph Bonaparte Gulf (JBG) in northern Australia, and 2) Icelandic waters as part of the Benthic Invertebrates of Icelandic Waters (BIOICE) program. For each dataset, we investigate potential patterns of biodiversity (measured by species richness, diversity indices, abundance, and community structure) in relation to environmental variables such as depth, geomorphology, and substrate. The availability of worldwide data from benthic marine biodiversity surveys reporting the results of two or more gear types is generally poor. Surveys were concentrated in the coastal regions of UK, Norway and Australia, with limited or no studies elsewhere and only 13% including the slope or deep sea. Between different gear groups, our review and analysis of datasets from two regions (northern Australia and Iceland) demonstrates there is little consistency in marine biodiversity trends, with only one study yielding consistent ecological patterns between sampling gear groups (imagery and epifaunal). This indicates that ideal gear combinations are not easily able to be generalised among studies and regions. In addition, the lack of consistency between sampling gear groups highlights the need to analyse gear-specific data and avoid amalgamation. Even among gear that yielded relatively consistent ecological relationships, results varied across biological or environmental factors. Within a gear group, there are more consistencies in ecological relationships, with only two out of the eight studies compiled showing inconsistent ecological relationships A lack of gear-specific studies precluded the determination of the optimal combination of gear types for a particular regions or environments. Nevertheless, based on our findings, we provide preliminary recommendations and inform further research: 1) If general biodiversity patterns are to be investigated, sampling for marine benthic surveys should be carried out using multiple gear types that are concurrently deployed; 2) Target measures of biodiversity need to be decided a priori and appropriate gear used; 3) Preliminary data will help determine the optimal combination of gear types used to sample that region and address a given hypothesis; and 4) If only two gear types are able to be deployed, a grab or corer should be one of them, as this sampling gear type samples a different habitat than other gear groups.

We undertook a biological data acquisition program as part of the transit of the R.V. Southern Surveyor between Darwin and Cairns 15-24 October 2012. The overarching aim of this program was to use an ROV and benthic sled to collect benthic marine information and specimens for biodiversity and biodiscovery research in areas previously mapped by Geoscience Australia during survey GA-276, including a bank (Area I) and terrace/hole feature within the proposed Wessel Islands CMR (Area II). This study focuses on sessile invertebrates such as sponges and octocorals due to their ecological importance as habitat providers and their chemical importance as sources of marine natural products and medicines. In less than 24 hours of sampling effort, survey SS2012/t07 resulted in 261 voucher specimens which will be used for biodiversity and natural products research. A total of 49 samples are to be lodged at the ABL, and samples with weights larger than 300 g will be sent to the NCI for screening of active compounds against cancer and HIV. Sponges were the most abundant group collected based on both biomass (~ 139 kg) and number of voucher specimens (93), followed by cnidarians (30 kg, 73 vouchers), particularly hard corals (23 kg, 11 vouchers). As expected the top of the bank in Area I had a seemingly diverse and abundant sessile invertebrate community, with consistent patchy occurrence of sponges, octocorals, and hard corals. The terrace at in Area II supports moderate densities of sponges and octocorals, while the adjacent deep hole at ~ 100 m seems to be covered with muddy gravel and supports scattered mobile and sedentary invertebrates, of which crinoids dominate, as well as skates and numerous small demersal fish.

Northern Australia has been the focus of recent marine biodiversity research to support resource management for both industry and conservation. Much of this research has targeted habitat-forming sessile invertebrates and charismatic megafauna, but smaller macrofauna and infauna must also be considered due to their important roles in ecosystem functions. In this study, a Smith-McIntyre grab was used during two surveys in 2009 and 2010 to the Joseph Bonaparte Gulf to collect sediment samples which were then elutriated over a 500µm sieve. The associated polychaetes were identified to species-level. A total of 2224 individual polychaetes were collected from 133 grabs and represent 43 families, including several new species, at least one new genus (Pilargidae) and many new distribution records. Biodiversity patterns were also analysed according to environmental and spatial factors (grain-size, carbonate, total organic content, depth, distance offshore) in order to inform predictive models and further our understanding of ecosystem processes in the region. These patterns differ from those of larger epifauna collected on the same surveys, highlighting the need to consider small macrofauna in biodiversity research and associated marine management.

Geoscience Australia's GEOMACS model was utilised to produce hindcast hourly time series of continental shelf (~20 - 300 m depth) bed shear stress (unit of measure: Pascal, Pa) on a 0.1 degree grid covering the period March 1997 to February 2008 (inclusive). The hindcast data represents the combined contribution to the bed shear stress by waves, tides, wind and densitydriven circulation. Ecological data collected from Torres Strait suggests that bed shear stresses exceeding 0.4 Pa are important in determining the species present (Long, Bode, & Pitcher 1997). Although this data may not be representative of other regions or benthic communities, it has been utilised to calculate two parameters for determining the relationship between shear bed stress and the benthic community. One of the parameters, which is denoted by , and is calculated using; represents the stress in excess of 0.4 Pa integrated over time as a proportion of the total stress integrated over time, and is intended to represent the proportion of the total integrated stress that has some control on the benthic community (Hughes & Harris 2008).

The accuracy of spatially continuous environmental data, usually generated from point samples using spatial prediction methods (SPMs), is crucial for evidence-informed environmental management and conservation. Improving the accuracy by identifying the most accurate methods is essential, but also challenging since the accuracy is often data specific and affected by multiple factors. Because of the high predictive accuracy of machine learning methods, especially random forest (RF), they were introduced into spatial statistics by combining them with existing SPMs, which resulted in new hybrid methods with improved accuracy. This development opened an alternative source of methods for spatial prediction. In this study, we introduced these hybrid methods, along with the modelling procedure adopted to develop the final predictive models. These methods were compared with the commonly used SPMs in R using cross-validation techniques based on both marine and terrestrial environmental data. We also addressed the following questions: 1) whether they are data-specific for marine environmental data, 2) whether input predictors affect their performance, and 3) whether they are equally applicable to terrestrial environmental data? This study provides suggestions and guidelines for the application of these hybrid methods to spatial predictive modelling not only in environmental sciences, but also in other relevant disciplines.

Although marine reserves are becoming increasingly important as anthropogenic impacts on the marine environment continue to increase, we have little baseline information for most marine environments. In this study, we focus on the Oceanic Shoals Commonwealth Marine Reserve (CMR) in northern Australia, particularly the carbonate banks and terraces of the Sahul Shelf and Van Diemen Rise which have been designated a Key Ecological Feature (KEF). We use a species-level inventory compiled from three marine surveys to the CMR to address several questions relevant to marine management: 1) Are carbonate banks and other raised geomorphic features associated with biodiversity hotspots? 2) Are there environmental or biogeographic variables that can help explain local and regional differences in community structure? 3) How do sponge communities differ between individual raised geomorphic features? Approximately 750 sponge specimens were collected in the Oceanic Shoals CMR and assigned to 348 species, of which only 18% included taxonomically described species. Between the eastern and western CMR, there was no difference between sponge species richness or assemblages on raised geomorphic features. Within individual raised geomorphic features, sponge assemblages were significantly different (ANOSIM: Global R = 0.328, p < 0.001), but species richness was not. There were no environmental factors related to sponge species richness, although sponge assemblages were weakly but significantly related to several environmental variables (mean depth, mean backscatter, mean slope). These patterns of sponge diversity are considered in the context of marine reserve management in order to explore how such information may help support the future management of this region.

Geoscience Australia's GEOMACS model was utilised to produce hindcast hourly time series of continental shelf (~20 - 300 m depth) bed shear stress (unit of measure: Pascal, Pa) on a 0.1 degree grid covering the period March 1997 to February 2008 (inclusive). The hindcast data represents the combined contribution to the bed shear stress by waves, tides, wind and densitydriven circulation. Ecological data collected from Torres Strait suggests that bed shear stresses exceeding 0.4 Pa are important in determining the species present (Long, Bode, & Pitcher 1997). Although this data may not be representative of other regions or benthic communities, it has been utilised to calculate two parameters for determining the relationship between shear bed stress and the benthic community. One of the parameters is the total percentage of time the bed shear stress exceeds 0.4 Pa, and this is denoted (Hughes & Harris 2008).

This dataset provides the spatially continuous data of predicted seabed gravel content (sediment fraction greater than 2000 µm) expressed as a weight percentage ranging from 0 to 100%, presented in 0.0025 decimal degree (dd) resolution raster grids format and ascii text file. The dataset covers the north-northwest region of the Australian continental EEZ. This dataset supersedes previous predictions of seabed gravel content for the region with demonstrated improvements in accuracy. Accuracy of predictions varies based on density of underlying data and level of seabed complexity. Artefacts occur in this dataset as a result of insufficient samples in relevant areas. This dataset is intended for use at regional scale. The dataset may not be appropriate for use at local scales in areas where sample density is insufficient to detect local variation in sediment properties. To obtain the most accurate interpretation of sediment distribution in these areas, it is recommended that additional samples be collected and interpolations updated.

Spatially continuous predictions of seabed hardness are important baseline environmental information for sustainable management of Australia's marine jurisdiction. Seabed hardness is often inferred from multibeam backscatter data with unknown accuracy, can be inferred based on underwater video footage at limited locations. It can also be predicted to two classes. In this study, we classified the seabed into four classes based on two new seabed hardness classification schemes (i.e. hard90 and hard70) for seabed video footage by. We developed optimal predictive models to predict the spatial distribution of seabed hardness using random forest (RF) based on point data of hardness classes and spatially continuous multibeam backscatter data. Five feature selection (FS) methods that are variable importance (VI), averaged variable importance (AVI), the combined, Boruta, and RRF were tested. Effects of highly correlated, important and unimportant predictors on the accuracy of RF predictive models were also examined. Finally, the most accurate models were used to predict the spatial distribution of the hardness classes and the predictions were visually examined and compared with the predictions based on two-class hardness classification. This study confirms that: 1) hard90 and hard70 are effective seabed hardness classification schemes; 2) seabed hardness can be predicted into a spatially continuous layer with a high degree of accuracy; 3) the typical approach used to pre-select predictors by excluding highly correlated predictors needs to be re-examined when using machine learning methods, at least, for RF, in the environmental sciences; 4) the identification of the important and unimportant predictors provides useful guidelines for further improving the predictive models; 5) FS is essential for identifying an optimal RF predictive model and the RF methods select the most accurate predictive model(s) instead of the most parsimonious ones, and AVI and Boruta are recommended for future studies; and 6) RF is an effective modelling method with high predictive accuracy for multi-level categorical data, can be applied to `small p and large n problems in environmental sciences, and is recommended for future studies. In addition, automated computational programs for AVI need be developed to improve its computational efficiency and caution should be taken when applying filter FS method in selecting predictive models in future studies.

Multibeam sonar data incorporates a wide range of metrics of physical seabed properties that can be utilised to generate substrate maps for marine habitat mapping. In particular, statistical descriptors of seabed form and texture can be derived to maximise the information provided by multibeam data. This study investigates the full potential of multibeam data for mapping seabed properties for an area of geomorphically complex seabed on the continental shelf offshore from Point Cloates, Western Australia. In 2008, as part of a collaborative survey within the Commonwealth Environmental Research Facilities (CERF) Marine Biodiversity Hub, Geoscience Australia acquired high resolution multibeam data and sediment samples across a 280 km2 area of the shelf, using a Kongsberg EM 3002 (300 kHz) system. Using this data, a two stage analysis was developed to: (i) separate 'hard seabed (e.g., reefs, ridges and mounds) from 'soft' sediments, and; (ii) predict textural properties for seabed sediments, including %Gravel, %Sand, %Mud, mean grain size and sorting. For a mapping tool, we chose the Random Forest Decision Tree technique. This entailed using ten combinations of input datasets as explanatory variables, including morphometric variables derived from bathymetry, and angular response curves and related statistics derived from backscatter mosaics. The training dataset was derived by combining sediment data from grab samples with locations of hard substrate inferred from bathymetry data. The predictive mapping of 'hard' and 'soft' seabed types resulted in predictions with very strong confidence levels, especially when bathymetry information was combined with backscatter data (i.e., cross-validated Area Under Curve = 0.99). The five sediment properties were predicted with moderate to good cross-validation accuracies (Figure 1). The highest accuracies were achieved for %Mud and Sorting, (R2s equal 0.73 and 0.68, respectively).