This dataset contains processed and raw backscatter data in matlab format produced by the CMST-GA MB Toolbox from various swath surveys in and around Australian waters.

Geoscience Australia carried out a marine survey on Lord Howe Island shelf (NSW) in 2008 (SS06_2008) to map seabed bathymetry and characterise benthic environments through colocated sampling of surface sediments and infauna, rock coring, observation of benthic habitats using underwater towed video, and measurement of ocean tides and wave generated currents. Subbottom profile data was also collected to map sediment thickness and shelf stratigraphy. Data and samples were acquired using the National Facility Research Vessel Southern Surveyor. Bathymetric data from this survey was merged with other preexisting bathymetric data (including LADS) to generate a grid covering 1034 sq km. As part of a separate Geoscience Australia survey in 2007 (TAN0713), an oceanographic mooring was deployed on the northern edge of Lord Howe Island shelf. The mooring was recovered during the 2008 survey following a 6 month deployment. The "2461_ss062008" folder contains raw multibeam backscatter data of the Lord Howe Rise. The raw multibeam backscatter data were collected along survey lines using SIMRAD EM300 from aboard RV Southern Surveyor

Multibeam sonars provide co-located high-resolution bathymetry and acoustic backscatter data over a swath of the seafloor. Not only does backscatter response vary with incidence angles but it also changes with different seabed habitat types as well. The resulting imagery depicts spatial changes in the morphological and physical characteristics of the seabed that many use to relate to other dataset such as biology and sediment data for seabed habitat classification purposes. As a co-custodian of national bathymetry data, Geoscience Australia holds massive volumes of multibeam data from various systems including comprehensive collection from its own SIMRAD EM3002D multibeam sonar system. Consequently, Geoscience Australia is researching the application of acoustic backscatter data for seabed habitat mapping to assist with deriving an inventory of seabed habitats for Australia's marine jurisdiction. We present a procedure and a technique developed for our SIMRAD EM3002D multibeam sonar system to derive meaningful angular backscatter response curves. The ultimate goal of this excersie is to try to make use of the angular backscatter response curve that many believe is unique and is an intrinsic property of the seafloor for seabed habitat classification purposes. Adopting the technique intially developed by the Centre for Marine Science and Technology at Curtin University of Technology, Geoscience Australia has further improved these techniques to suits its own sonar system. Issues surrounding the production of the angular backscatter response curves and their solutions will be discussed. We also present results derived from multibeam data acquired in the Joseph Bonaparte Gulf, NT and from the Carnarvorn Shelf (Point Cloates), WA from aboard AIMS Research Vessel Solander. This includes potential use of the angular backscatter response curves for seabed classification and results from a simple analysis using the Kolmogrov-Smirnov goodness of fit.

Geoscience Australia provides spatial information of seabed environment to support Australian marine zone management. Central to this approach is the prediction of Australia's seabed biodiversity from spatially continuous data of seabed biophysical properties. Seabed hardness is an important environmental property for predicting marine biodiversity and is often inferred from multibeam backscatter data. Although seabed hardness can be measured based on video images, they are only available at a limited number of sampled locations. In this study, we attempt to predict the spatial distribution of seabed hardness using random forest based on video classification and available marine environmental properties. We illustrate the effects of cross-validation methods including a new cross-validation function on the selection of optimal predictive models. We also test the effects of various predictor sets on the predictive accuracy. This study provides an example for predicting the spatial distribution of environmental properties using random forest in R.

Deep sea environments occupy much of the sea floor, yet little is known about diversity patterns of biological assemblages from these environments. Physical mapping technologies and their availability are increasing rapidly. Sampling deep-sea biota over vast areas of the deep sea, however, is time consuming, difficult, and costly. Consequently, the growing need to manage and conserve marine resources, particularly deep sea areas that are sensitive to anthropogenic disturbance and change, is leading the promotion of physical data as surrogates to predict biological assemblages. However, few studies have directly examined the predictive ability of these surrogates. The physical environment and biological assemblages were surveyed for two adjacent areas - the western flank of Lord Howe Rise (LHR) and the Gifford Guyot - spanning combined water depths of 250 to 2,200 m depth on the northern part of the LHR, in the southern Pacific Ocean. Multibeam acoustic surveys were used to generate large-scale geomorphic classification maps that were superimposed over the study area. Forty two towed-video stations were deployed across the area capturing 32 hours of seabed video, 6,229 still photographs, that generated 3,413 seabed characterisations of physical and biological variables. In addition, sediment and biological samples were collected from 36 stations across the area. The northern Lord Howe Rise was characterised by diverse but sparsely distributed faunas for both the vast soft-sediment environments as well as the discrete rock outcrops. Substratum type and depth were the main variables correlated with benthic assemblage composition. Soft-sediments were characterised by low to moderate levels of bioturbation, while rocky outcrops supported diverse but sparse assemblages of suspension feeding invertebrates, such as cold water corals and sponges which in turn supported epifauna, dominated by ophiuroids and crinoids. While deep environments of the LHR flank .

Geoscience Australia carried out marine surveys in Jervis Bay (NSW) in 2007, 2008 and 2009 (GA303, GA305, GA309, GA312) to map seabed bathymetry and characterise benthic environments through colocated sampling of surface sediments (for textural and biogeochemical analysis) and infauna, observation of benthic habitats using underwater towed video and stills photography, and measurement of ocean tides and wave generated currents. Data and samples were acquired using the Defence Science and Technology Organisation (DSTO) Research Vessel Kimbla. Bathymetric mapping, sampling and tide/wave measurement were concentrated in a 3x5 km survey grid (named Darling Road Grid, DRG) within the southern part of the Jervis Bay, incorporating the bay entrance. Additional sampling and stills photography plus bathymetric mapping along transits was undertaken at representative habitat types outside the DRG. The "kimbla" folder contains raw multibeam backscatter data from four surveys archived seperately in 0303_jervis_trials, 0305_jervisbay2, 0311_jervisbay3 and 0313_jervis_trials4. The raw multibeam backscatter data were collected along survey lines using GAs Kongsberg SIMRAD EM3002 in single head and dual head configuration from aboard Work Boat Kimbla.

Acoustic backscatter from the seafloor is a complex function of signal frequency, seabed roughness, grain size distribution, benthos, bioturbation, volume reverberation and other factors. Angular response is the variation in acoustic backscatter with incident angle and it is considered be an intrinsic property of the seabed. The objective of the study was to illustrate how the combination of a self-organising map (SOM) and hierarchical clustering can be used to develop an angular response facies map for Point Cloates, northwest Australia; demonstrate the cluster visualisation properties of the technique; and highlight how the technique can be used to investigate environmental variables that influence angular response.

This study used angular response curves of multibeam backscatter data to predict the distributions of seven seabed cover types in an acoustically-complex area. Several feature analysis approaches on the angular response curves were examined. A Probability Neural Network model was chosen for the predictive mapping. The prediction results have demonstrated the value of angular response curves for seabed mapping with a Kappa coefficient of 0.59. Importantly, this study demonstrated the potential of various feature analysis approaches to improve the seabed mapping. For example, the approach to derive meaningful statistical parameters from the curves achieved significant feature reduction and some performance gain (e.g., Kappa = 0.62). The first derivative analysis approach achieved the best overall statistical performance (e.g., Kappa = 0.84); while the approach to remove the global slope produced the best overall prediction map (Kappa = 0.74). We thus recommend these three feature analysis approaches, along with the original angular response curves, for future similar studies.

Seabed sediment textural parameters such as mud, sand and gravel content can be useful surrogates for predicting patterns of benthic biodiversity. Multibeam swath mapping can provide near-complete spatial coverage of high-resolution bathymetry and backscatter data that are useful in predicting sediment parameters. The multibeam acoustic data at a ~1000 km2 area of the Carnarvon Shelf, Western Australia was used in a predictive modeling approach to map eight seabed sediment parameters. The modeling results indicates overall satisfactory statistical performance, especially for %Mud, %Sand, Sorting, Skewness, and Mean Grain Size. The study demonstrated that predictive modelling using the combination of machine learning models has several advantages over the interpolation of Cokriging. Combing multiple machine learning models can not only improve the prediction performance but also provides the ability to generate useful prediction uncertainty maps. Another important finding is that choosing an appropriate set of explanatory variables, through a manual feature selection process, is a critical step for optimizing model performance. In addition, machine learning models are able to identify important explanatory variables, which is useful in explaining underlying environmental process and checking prediction against existing knowledge of the study area. The sediment prediction maps obtained in this study provide reliable coverage of key physical variables that will be incorporated into the analysis of co-variance of physical and biological data for this area. International Journal of Geographical Information Science

Acoustic remote sensing is the only effective technique to investigate deep sea bottom. Modern high-frequency multibeam echosounders transmit and receive backscatter signals from hundreds of narrow-angle beams which enlighten small footprints on the seabed. They can produce bathymetry and backscatter data with a spatial resolution around 2% of water depth, which enables us to map the seabed with great detail and accuracy. After calibration, the backscatter intensity is largely controlled by three seabed physical properties: the acoustic impedance contrast (often called hardness), apparent interface roughness (relative to acoustic frequency) and volume inhomogeneity [3, 4, 7]. These seabed physical properties are directly related to sediment grain size characteristics at the sedimentary areas. Studies showed that backscatter intensity had a moderate and positive correlation with sediment mean grain size [1, 3, 6]. Also, backscatter intensity was found to be positively correlated with coarse fractions and inversely correlated with finer fractions [2, 5, 6]. Other sediment grain size properties, especially sorting may also play important roles in the backscatter-sediment relationship [3, 5, 6]. The backscatter-sediment relationship, however, is complex in nature. Research is needed to better understand how acoustic sound interacts with sediment. This study aims to explore this relationship using a set of high quality sediment and multibeam backscatter data, and a robust spatial modelling technique. The co-located sediment and multibeam data were collected from four different areas of Australian margin which represent different sedimentary environments. Five hundred sixty-four sediment grab samples were taken from these survey areas. They were analysed in laboratory using the same procedure to generate grain size properties of %gravel, %sand, %mud, mean grain size, sorting, skewness and kurtosis. The multibeam data were collected using Kongsberg's 300 kHz EM3002 system. The raw multibeam backscatter was processed using the CMST-GA MB Process v8.11.02.1 software developed by Geoscience Australia and the Centre for Marine Science and Technology at Curtin University of Technology. As a result, the backscatter mosaics from incidence angles of 1o to 60o, at an interval of 1o, were generated. The backscatter intensity values from these 60 incidence angles were extracted for all of the sediment samples. The machine learning model Random Forest Decision Tree (RFDT) was used to investigate the backscatter-sediment relationship. The seven sediment grain size properties were the explanatory variables. The response variable was the backscatter intensity from each incidence angle. The model performance was evaluated using 10-fold cross-validation. For incidence angles between 1o and 42o, the RFDT models achieved fairly good performance, with a percentage of variance explained around 70% (Figure 1). The model performance gradually decreased for the outer beam range (incidence angle > 42o). Mud content was consistently identified as the most important explanatory variable to the backscatter strength. The second most important explanatory was usually sediment mean grain size. The RFDT models were also able to generate predicted response curves to quantitatively investigate the relationships between the important explanatory variables and individual response variables. The predicted relationship between %mud and the acoustic backscatter intensity is shown in Figure 2. This indicates a negative but non-linear relationship, with the increase of mud content in the sediment, the backscatter intensity decreases. This finding is consistent with that of previous studies [2, 5, 6]. Fine sediment with high mud content not only is soft (e.g., low impedence contrast) but also has high acoustic penetration (e.g., high attenudation in sediment), which naturally incurs low backcatter return