A new bathymetric grid for the Gulf of Papua and northern Australia was produced for the area 140°–150°E, 6°–14°S, with a 3.6″ (∼110 m) cell size. New multibeam sonar surveys have added much needed detail to a region of the seabed where previously little was known. In shallow Australian waters, bathymetry derived from Landsat satellite imagery was used to supplement traditionally acquired bathymetric data. For onshore areas, Shuttle Radar Topography Mission data were used for topographic control. The final grid revealed numerous features not observed in previous compilations of bathymetric data for the region. Bathymetric surveys on the continental shelf revealed an incised shelf with highly variable valley morphology. Prograding clinoforms are infilling valleys on the continental shelf and mark the seaward extension of the Fly River delta. A linear, relict shelf-edge barrier marks the shelf break in the northern Ashmore Trough region, elsewhere the shelf break is scalloped and incised by canyons. Large mass transport deposits are widespread on the continental slope and indicate regions where mass wasting is common.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 This study describes the production of a new bathymetric grid for the Gulf of Papua region (Figure 1). The grid covers 140.0°–150.0°E, 6.0°–14.0°S with a cell size of 3.6″ (∼110 m). The regional scale of the grid and the 3.6″ cell size is intended to provide (1) the most “up-to-date” compilation of bathymetric data for the Gulf of Papua and (2) a tool for understanding sediment transport and accumulation for the region. Previous bathymetric compilations that cover the Gulf of Papua region include the General Bathymetric Chart of the Oceans (GEBCO) series of maps and grids [Intergovernmental Oceanographic Commission, International Hydrographic Organisation and British Oceanographic Data Centre, 2003] and the ETOPO2 [National Geophysical Data Center, 2001] bathymetric grid. The GEBCO bathymetric grid is based on bathymetric contours of world's oceans derived from digital and analog ships soundings. By comparison, the ETOPO2 grid uses bathymetry derived from ship track bathymetry and satellite altimetry using the method of Smith and Sandwell . Considerable differences are to be expected between GEBCO and grids derived from satellite altimetry (such as ETOPO2) because the ship track coverage is different and because the interpolation between the tracks is different. Marks and Smith  demonstrated that at long wavelengths both grids showed similar features where control data existed, where control data did not exist the grids could differ by >250 m. While providing global coverage, both GEBCO and ETOPO2 lack the resolution (60″ and 120″ cell size, respectively) and detail needed to gain an understanding of the morphology and sedimentary processes within the Gulf of Papua. Compared to GEBCO and ETOPO2, the bathymetric grid presented here has a significantly higher density of data points available to control the interpolation routine (Figure 2).
 The increase in data available for the grid is primarily a result of recent multibeam sonar surveys and the use of hydrographic data sets. In shallow waters around northern Australia bathymetric information derived from satellite imagery was used to assist in the definition of seabed features. Multibeam sonar surveys have resolved many seabed features that are indicative of past and present sedimentary environments. Harris et al.  and Crockett et al.  have documented the presence of several incised shelf valleys. These valleys were not extensively modified during the Holocene sea level transgression and thus represent an insight into interglacial fluvial processes operating in the region. The modern Fly River delta has been shown to be prograding seaward, gradually infilling the valleys on the Fly River shelf. The morphology of the continental slope is inferred to be dominated by sediment gravity flows. The morphology and sedimentary processes in this region are described by Francis et al. .
2. Data Types and Preprocessing
2.1. Multibeam Data
 All multibeam sonar data, in appropriate data formats, were processed using the Caris HIPS/SIPS data processing software. Bathymetric data were corrected to mean sea level (MSL) using tide information acquired during surveying or a tide model derived from known harmonic constituents using the XTide software (http://www.flaterco.com/xtide). A despiking filter was applied to all multibeam data sets to remove the majority of “erroneous” data. The filter calculates the slopes between every ping and its neighbor's port and starboard. If both slopes exceed a defined value (a value of 5° was typically used) and are of opposite sign then the ping is rejected. Most data sets were also visually inspected to further remove spikes and to check the results of the despiking filter. Overlapping multibeam data sets were then used to correct roll biases and refraction errors. A number of the deepwater multibeam data sets had refraction errors (typically in depths >2000 m) because of inaccurate sound velocity profiles applied during data acquisition. Refraction artifacts were minimized using the refraction editor in the Caris HIPS/SIPS software. Other multibeam data sets that were not compatible with the Caris software were edited using the MBSystem software suite [Caress and Chayes, 1995]. Presently, multibeam data cover 11.6% of the marine grid area (Figure 2a), the bulk of these data sets are located in the Gulf of Papua and were acquired as part of the MARGINS source-to-sink research program. The multibeam data sets used to construct the bathymetry grid are summarized in Table 1.
Table 1. Summary of Multibeam Bathymetric Data Sets Used to Create the Bathymetric Grida
Abbreviations are RAN, Royal Australian Navy; SIO, Scripps Institution of Oceanography; LDEO, Lamont-Doherty Earth Observatory; CSIRO, Commonwealth Scientific and Industrial Research Organisation; IFREMER, French Research Institute for Exploitation of the Sea; IPEV, Institut Polaire Français; JAMSTEC, Japanese Marine Science and Technology Centre; and JCU, James Cook University.
Gloria–seismic project, northeastern Australia
Atlas Hydrosweep DS2
Geoscience Australia Survey 234 East Torres Strait
MARGINS S2S Gulf of Papua
Simrad EM3000 & Seabeam 2000
Geoscience Australia Survey 266 Turnagain Island
R/V James Kirby
MARGINS S2S Gulf of Papua
Simrad EM3000 & Seabeam 2000
East Cay/Torres Strait James Cook University
R/V James Kirby
PECTIN Gulf of Papua
R/V Marion Dufresne
2.2. Australian Hydrographic Service Fair Sheet and Survey Data
 The Royal Australian Navy Australian Hydrographic Service (AHS) is responsible for publishing and maintaining Australian nautical charts. Data held by the AHS include echo sounder surveys, multibeam sonar surveys (not included in section 2.1), Laser Airborne Depth Sounder surveys [Penny et al., 1989], and digitized fair sheets (Figure 2b). AHS data sets are assumed to be correct to lowest astronomical tide (LAT). To convert from LAT to MSL corrections were applied to all data sets within the grid region using tide datum information from the Australian National Tide Tables [Australian Navy Office, 1998]. Data quality for AHS data sets is considered to be high because of rigorous quality control standards during acquisition. However, factors effecting data set quality include the following: (1) In regions that experience strong tidal currents and bed load transport (such as Torres Strait) data sets may become unreliable because of modification of the seabed over time. (2) As data sets are reduced to “shoal” soundings a shallow bias is introduced. (3) An error is introduced when converting data sets from LAT to MSL because of distances between survey areas and tide stations.
 The data coverage for AHS data sets is typically limited to Australian waters (Figure 2b). northern Australia has a patchy coverage and sizable gaps are present in the Torres Strait and northern Great Barrier Reef. This is due to region's remote location, complex topography (including numerous islands, reefs and sand banks which limit vessel access), and turbid water limiting the use of optical remote sensors such as Laser Airborne Depth Sounder. The patchy coverage, but relatively high data density, of AHS data sets provided adequate ground truthing to extract bathymetry from Landsat imagery (section 2.5).
2.3. Ship Track Data and Miscellaneous Sources
 Echo sounder data from research vessels were extracted from Geoscience Australia's marine survey database OZMAR. Some additional data sets were obtained from other research institutes (Table 2) and added to the OZMAR data set. The overall data coverage of the ship track data set (Figure 2c) is comparable to that of the ETOPO2 and GEBCO (Figures 2g and 2h) data sets. It is evident that all three data sets share common survey data, however, distinct differences do occur. ETOPO2 data are generally restricted to the Coral Sea and the Gulf of Papua with little data around northern Australia. GEBCO has fewer surveys than ETOPO2 in the grid area, but more surveys in Australian waters (specifically toward the west of the grid area). The final ship track data set (Figure 2c) benefited for the addition of numerous data sets from the northern Great Barrier Reef and Torres Strait that were not available to GEBCO or ETOPO2. To the west of the grid area there remains little ship track data; however, in Australian waters, this region is adequately covered by AHS data sets.
Table 2. Summary of Ship Track and Miscellaneous Data Sources Used in the Gulf of Papua Bathymetric Grida
Abbreviations are CSIRO, Commonwealth Scientific and Industrial Research Organisation; GBRMPA, Great Barrier Reef Marine Park Authority; AIMS, Australia Institute of Marine Science; and SMEC, Snowy Mountains Engineering Corporation.
ship track data sets within Geoscience Australia's marine surveys database
Other sources of ship track bathymetric data
digital ship track data held by GBRMPA, AIMS, and CSIRO
Geoscience Australia marine sediments database [Passlow et al., 2005]. Sample depths are held for many grab and core samples
SMEC Fly River Hydrographic Chart Series
bathymetry data points digitized from paper charts. Charts used were 05-215-(1-8) (SMEC, 1982)
Australian Nautical Chart series
bathymetry and selected contours digitized from paper charts. Charts used were AUS 299, 374,375, 376, 377, 378, 379, 380, 410, 839, and 840
 Some shallow-water echo sounder data sets were not provided with time stamps and therefore could not have a tide correction applied, potentially introducing a large error. The error is equal to the difference between LAT and MSL and variable over the region, but is typically on the order of ∼2 m. Once the ship track data had been visually inspected, they were reduced to a 36″ (∼1100 m) grid using the Generic Mapping Tools module block mean [Wessel and Smith, 1998]. This was done to reduce artifacts caused by crossover errors in the ship track data sets.
 Some of the poorest data regions around the Gulf of Papua were on the continental shelf (<100 m), so digital soundings were supplemented with points and contours digitized from nautical charts (Table 2 and Figure 2d). Additional bathymetric data extracted from Geoscience Australia's Marine Sediments database [Passlow et al., 2005] were also used, because water depths were recorded when sediment samples were collected.
2.4. Topography and Coastline
 The Shuttle Radar Topography Mission (SRTM) onboard the Space Shuttle Endeavour obtained the most complete near-global high-resolution data set of the Earth's topography during its 10 day operation [Farr and Kobrick, 2000; Rabus et al., 2003]. The topographic data were derived by radar interferometry and covers 90% of the Earth's land surface at 3″ (∼90 m) resolution. The cell size of the SRTM data was less than that of the new grid (3″ compared to 3.6″, respectively) therefore there were no gaps in the data coverage for all subaerial areas.
 High-resolution vector coastlines were obtained for both Papua New Guinea (1:50000 Coastline of Papua New Guinea, available at http://gis.mortonblacketer.com.au/upngis/downloads.htm) and Australia (GEODATA Coast 100k 2004, available at http://www.ga.gov.au/nmd/products/thematic/coast.jsp). The coastlines were used in two ways: (1) the vector files were used to clip the SRTM topographic data, and remove any data that occurred outside the areas defined as land by the coastline file; and (2) both vector coastlines were converted to a grid with 3.6″ resolution grid and converted to an XYZ ASCII text file, each latitude/longitude pair being given a depth value of 0. The coastline ASCII file ensured that a cell value of 0 m would mark a coastline around all topographic data, and that the topographic data did not “bleed out” into the bathymetry during the gridding process.
2.5. Bathymetry Derived From Landsat
 Bathymetry derived from Landsat imagery was used to supplement current holdings of traditionally acquired bathymetric data. Band ratio methods, typically the ratio of the Landsat bands 1 and 2 (blue and green wavelengths, respectively), have been suggested as a way to estimate depth and minimize the effects of changes in bottom albedo [Lyzenga, 1978; Dierssen et al., 2003; Stumpf et al., 2003]. Band ratio methods make an assumption that different bottom albedoes at the same depth have the same ratio and, when rescaled using known bathymetry values, provide an approximation of depth.
 For this study, the ratio of the blue and green pixel intensities (denoted as B1:B2) from 12 Landsat (Table 3) scenes covering the Torres Strait and the northern Great Barrier Reef were compared to AHS bathymetric data. For each Landsat scene, pixels that had colocated bathymetric data were identified. The average ratio of B1:B2, for each 1 m of depth, from 1 to 20 m, was then calculated. The resulting data sets provided an estimation of how B1:B2 changed with depth over each Landsat scene. Scene-specific algorithms, based on second-order polynomials, were then calculated to estimate depth in each scene (Table 3). Owing to the increasing attenuation of light with increasing depth, the relationship between B1:B2 and depth is asymptotic in optically deep water, (i.e., no signal from the seabed will be received by the satellite). The depth at which this asymptote occurs is variable between the Landsat scenes (Table 3) but generally occurs at ∼15 m. This marks the maximum depth down to which bathymetry can be estimated by this technique.
Table 3. Landsat Scenes Used in Bathymetry Processinga
Maximum Depth, m
Scene Center, Latitude/Longitude
The “x” in each scene-specific second-order polynomial is equal to the ratio B1:B2.
depth = −471.98x2 + 2666.5x – 3759.9
depth = −11.721x2 + 55.653x – 66.518
depth = −419.37x2 + 1351x –1089.8
depth = −277.48x2 + 918.11x – 760.82
depth = −61.95x2 + 163.55x – 107.73
depth = −30.43x2 + 70.56x – 38.176
depth = −67.712x2 + 190.75x – 133.87
depth = −88.122x2 + 236.12x – 157.21
depth = −31.563x2 + 62.345x – 21.306
depth = −49.07x2 + 130.46x – 84.9
depth = −292.61x2 + 826.21x – 584.1
depth = −2.233x2 +1.466x + 5.8912
 The application of passive remote-sensing techniques to bathymetric mapping requires shallow and clear water, minimal changes in bottom type, and no atmospheric contamination. Concentrations of turbidity and chlorophyll are assumed to be variable (though not directly measured) in all the scenes and would account for some of the variability in the algorithms used to convert B1:B2 to bathymetry. Regions within Landsat scenes that were observed to be highly turbid (particularly northern Torres Strait and costal regions around northern Australia) were masked out of the analysis to limit the influence of terrestrial runoff. As a result, a good correlation was observed between measured depths and B1:B2 (r2 > 0.97 typically). This technique was not extended to the Gulf of Papua because of the high turbidity in the region, the greater depths involved, and the lack of adequate bathymetric data in clear waters.
3. Grid Development
3.1. Grid Interpolation
 The ANUDEM software [Hutchinson, 1988, 1989] was used for creation of the grid. The interpolation algorithm implemented by ANUDEM uses an iterative finite difference technique that combines the surface continuity of global interpolation methods (i.e., splines or kriging) with the computational efficiency of local interpolation methods [Hutchinson, 1989]. ANUDEM obtains realistic drainage structures through two different processes: (1) the use of a roughness penalty and (2) the option of a drainage enforcement algorithm. The roughness penalty is introduced into the algorithm to provide a mechanism to assign relative weightings to first derivatives (slope) and second derivatives (curvature), and thus determine the nature of the interpolating function. A minimum-curvature surface is created when the roughness penalty approaches 0 (indicating a heavily weighted second derivative), and a minimum potential surface is created when a roughness penalty approaches 1 (indicating a heavily weighted first derivative). A roughness penalty of 0.5 was used for the interpolation of the Gulf of Papua grid. Hutchinson  recommends that a roughness penalty of 0.5 be used when gridding data sets containing randomly spaced points. The adjustable roughness penalty used in ANUDEM is similar to the use of a tension parameter [Smith and Wessel, 1990] in minimum curvature splines. Both techniques provide a compromise between minimum curvature and minimum potential surfaces to maintain artifact-free behavior of the interpolated surface in areas with few, widely spaced data points.
 The drainage enforcement algorithm of ANUDEM is typically used to create hydrologically sound digital elevation models in areas of sparse data coverage (specifically in the subaerial environment). The suitability of such an algorithm to the marine environment is not well documented though Harris et al.  and Crockett et al.  describe continental shelf valleys in the Gulf of Papua and eastern Torres Strait characterized by closed bathymetric contours. These features suggest that the Gulf of Papua shelf is not a hydrologically sound surface and hence drainage enforcement was not appropriate. The effect of gridding the SRTM data set with a drainage enforcement algorithm is beyond the scope of this paper. Other gridding techniques such as kriging and triangular irregular networks were used to interpolate the grid. However, both methods took an excessive amount of time to complete and were abandoned.
 The grid bounds were 6.0°–14.0°S 140.0°–150.0°E with a 3.6″ cell size. The 3.6″ cell size (∼110 m) was considered sufficiently fine to resolve structures of interest within the grid area. The final grid size was 10001 × 8001 cells.
3.2. Data Coverage Statistics
Figures 2f–2h show the relative data coverage for the Gulf of Papua, GEBCO and ETOPO2 grids, respectively. Minor differences occur in the different ship track data sets; however, the Gulf of Papua grid benefits from the addition of multibeam, hydrographic data sets, digitized charts and bathymetry derived from Landsat (Figures 2a, 2b, 2d, and 2e). The total number of data points used in the Gulf of Papua grid (when reduced to a 3.6″ grid) is ∼9.1 × 106. By comparison, ETOPO2 has ∼1.5 × 105 data points for the same area. No comparison is made with the GEBCO data set as ship tracks are supplied without bathymetry (i.e., navigation only).
 Key statistics for each data set used in the construction of the bathymetric grid are shown in Table 4. In the marine zone, 19.3% of grid cells had associated bathymetric soundings or Landsat-derived bathymetry, 59.8% of the final data set was multibeam bathymetry, 23.4% was Landsat-derived, and 10.9% was provided by the Australian Hydrographic Service.
Table 4. Data Coverage Statistics Generated for Each Data Type Used in Bathymetry Grid
Points After Reduction to 0.001° Grid
Coverage of Sea Surface, %
Total Bathymetry Data Points, %
3.3. Accuracy of Grid Data Sets
 A relative root-mean-square error (RRMSE) was used to compare the accuracy of overlapping data sets used in the grid (Table 5). When comparing two overlapping bathymetric (or topographic) grids a root-mean-square error (RMSE) represents the standard deviation between overlapping grid cells. The RRMSE expresses the difference between grid cells as a percentage rather than a magnitude [Kienzle, 2004].
Table 5. RRMS Error Analysis of Overlapping Bathymetric Data Setsa
RRMS Error, %
RRMS Error, %
RRMS Error, %
RRMS Error, %
Data set comparisons with RRMSE values >10% are bold.
 Comparisons between traditionally acquired bathymetry and derived bathymetry (i.e., from Landsat) typically had the highest RRMSE values (17.9–35.53%). The high RRMSE values are inferred to be a result of variable water column (turbidity, chlorophyll etc) and seabed properties (albedo and benthic habitats) within Landsat scenes on the scene specific depth algorithms. While the band ratio algorithm minimizes the influence of variable bottom types it is evident that the relationship between pixel radiances and bathymetry is complex and not addressed fully with this method. Comparisons between traditionally acquired bathymetric data had RRMSE values <10% (except for AHS v Misc) indicated a much higher overall level of agreement between data sets. The Landsat bathymetry is retained within the grid data, despite its low RRMSE values, because of its extensive data coverage (Figure 2) and its capacity to delineate seabed features in regions where little or no bathymetric data exist presently. Errors between traditionally acquired bathymetric data sets are assumed to be a result of (1) incorrect tide correction or not applied, (2) inaccurate sound velocity profile corrections used to convert ping travel times to depth, (3) inaccurate positioning in older ship track data set (i.e., pre GPS), (4) errors converting data sets to MSL, and (5) erroneous data sounding not identified during initial data processing.
 The final grid revealed an array of seabed features not seen in previous compilations of bathymetric data for the region. Both the continental shelf and slope benefited from the addition of new bathymetric data sets that allowed for the interpretation of past and present sedimentary environments.
4.1. Continental Shelf
 On the continental shelf two contrasting sedimentary environments were identified: (1) the modern Fly River delta and (2) a system of relict, incised valleys seaward of the delta. The transition between these two environments is shown regionally in Figure 2b and occurs at ∼40 m depth. The modern Fly River delta is prograding in response to sediment input from the Fly River and has been shown to be infilling valleys incised into the Gulf of Papua shelf [Harris et al., 1993; Crockett et al., 2008]. These valleys occur between ∼40–80 m depth and were not extensively modified during the Holocene sea level transgression and thus represent an insight into fluvial processes operating in the region during the last glacial maximum. The Kiwai and Umuda valleys (Figures 3c and 3d) represent two different, sedimentary systems. Crockett et al.  interprets the Kiwai Valley as a relict, incised valley. The valley is relatively long (28 km) but narrow (0.8–1.1 km) and consists of three, steep sided, meanders. The Umuda Valley by contrast, is interpreted as a modern, alluvially influenced, coastal plain distributary channel because of its broad nature (∼28 km) and low relief (<20 m).
4.2. Shelf Break and Slope
 The shelf break is characterized by a sharp change in gradient between the continental shelf and slope. In the Gulf of Papua the shelf break occurs between 135–140 m (Figure 3b). The Gulf of Papua shelf break is characterized by two distinct morphologies. Francis et al.  document a highly linear shelf-edge barrier feature along the northern Ashmore Trough (Figure 3e). Sampling of the shelf-edge barrier has revealed the presence of a relict barrier reef system that occupied the ridge during the transgression [Tcherepanov et al., 2008].
 Elsewhere the shelf break is scalloped and/or incised by northwest-southeast trending submarine canyons. This is most evident in the southern shelf of the Pandora Trough region (Figure 3f). Francis et al.  interpret the scalloped shelf break to be indicative of mass wasting with the channels acting as conduits for sediment gravity flows. Figure 3f indicates the location of a mass flow deposit off the Gulf of Papua shelf.
5. Discussion and Conclusions
 A new bathymetric grid was created for the region 140.0°–150.0°E, 6.0°–14.0°S with a cell size of 3.6″. Compared to GEBCO and ETOPO2, the region benefited from the addition of numerous new data sets including multibeam sonar surveys, hydrographic data sets, and bathymetry derived from Landsat. This significant increase in data availability added much needed detail to the Gulf of Papua and assisted with identifying past and present sedimentary environments both on the continental shelf and slope [Tcherepanov et al., 2008; Crockett et al., 2008; Francis et al., 2008].
 RRMSE analysis of overlapping data sets has suggested that errors between traditionally acquired bathymetric data sets were on the order of <10%. When traditionally acquired data sets were compared to bathymetry derived from Landsat the error increased to 17.9–35.53%. The significantly higher errors associated with Landsat-derived bathymetry were considered to be a result of variable water quality and seabed types within individual Landsat scenes. The Landsat bathymetry was retained within the grid, despite it relatively low accuracy, as it assisted with delineating seabed features and covered areas that were data poor.
 The data density maps (Figures 2a–2f) highlight many areas where very little bathymetric data presently exist. The continental shelf around Papua New Guinea and Indonesia and the deep portion of the southern Coral Sea remain data poor.
 The construction of this bathymetry grid would not have been possible without data provided by the MARGINS Source-to-Sink research group (specifically Chuck Nittrouer, Andre Droxler, and Brian Donahue), Geoscience Australia, Royal Australian Navy (RAN) Australian Hydrographic Service (AHS), Scripps Institution of Oceanography (SIO), Lamont-Doherty Earth Observatory (LDEO), Australian Institute of Marine Science (AIMS), Commonwealth Scientific and Industrial Research Organisation (CSIRO), French Research Institute for Exploitation of the Sea (IFREMER), French Polar Institute (IPEV), Great Barrier Reef Marine Park Authority (GBRMPA), Japanese Marine Science and Technology Centre (JAMSTEC), and National Aeronautics and Space Administration (NASA). Landsat imagery was provided by the Australian Centre for Remote Sensing (ACRES). The MARGINS Source-to-Sink research group was supported by National Science Foundation. I also wish to thank Peter Petkovic, Craig Smith, Michael Hughes, and Peter Harris for reviewing this paper.