From AMCGMedia

Introduction

Accurate bathymetry is an important factor in any ocean model. However it is worth keeping in mind:

"The world’s ocean floors are vast regions covered only sparsely by ship surveys of depth. Only a few percent of the deep ocean floor has been mapped, and the density of ship tracks leaves areas as large as 10⁵ km² untraversed."

Marks and Smith (2006)

There are several publicly available global bathymetry datasets which are reviewed in Marks and Smith (2006). There are really two 'end-member' grids:

Smith and Sandwell (1997) (v9.1) a 1-min Mercator-projected grid based on bathymetry derived from satellite gravity data, combined with ship measurements.
GEBCO (The General Bathymetric Charts of the Oceans) a 1-min grid constructed from a series of paper and digital contour maps at 1:10 million scale. These maps were contoured at 500 m depth intervals, by hand, from both digital and analog ship soundings.

Other notable global bathymetry maps (e.g. DBDB2, DBDBv, ETOPO2, GINA) take the original Smith and Sandwell data and merge them with other high resolution datasets in localised areas. However, there are issues of smoothing and grid mislocation (Marks and Smith, 2006). Therefore we choose to compare 3 datasets: GEBCO, Smith and Sandwell (version 9.1) and a blend of GEBCO and Smith and Sandwell (S2004) produced by Walter Smith.

All these maps are available at /home/data/maps/

Bathymetry and Shoreline datasets available

GEBCO

The GEBCO data is taken from the DVD and the NetCDF file can be found at /home/data/maps/gridone.grd

Smith and Sandwell

The latest published Smith and Sandwell global 2 minute bathymetry/topography data (version 8.2) is available for download from ftp://topex.ucsd.edu/pub

However, it is constantly being updated and the most up to date version can be found on David Sandwell's ftp site: ftp://topex.ucsd.edu/pub/sandwell/ the current version there is 9.1

Both version 8.2 and 9.1 are downloaded here in .img format. This is easily converted to .grd format using the GMT commands:

img2grd topo_8.2.img -GSandS_8.2.grd -R-180/180/-72/72 -T1 -V
img2grd topo_9.1.img -GSandS_9.1.grd -R-180/180/-80/80 -T1 -m1 -D-80.738/80.738 -V

These are pixel registered grids whereas GEBCO and the GEBCO SandS blend are gridline registered. To be consistent (see figure below) we have decided to convert these to gridline registered using:

grdsample SandS_8.2.grd -T -GSandS_8.2.grd
grdsample SandS_9.1.grd -T -GSandS_9.1.grd

SandS_8.2.grd is 2 min and between 72^oN and 72^oS

SandS_9.1.grd is 1 min and between 80^oN and 80^oS.

In a grid-registered grid (A), data is stored at grid intersections, whilst in a pixel-registered grid (B), data is stored in cell centres

Contour plot (contours at 100 m intervals) of a small area of the ocean bottom illustrating the importance of pixel vs. gridline registration in comparing the datasets. In (A) the black contours are from the gridline registered S2004 map and the red contours from the pixel registered Smith and Sandwell v.8.2. Note the shift between the two, if we were plotting difference in depth between the two grids the values would be quite large. In (B) the black contours are once again from the gridline registered S2004 map, but the blue contours are from the gridline registered Smith and Sandwell v.8.2, note how the two contours now match up and there is little difference between the datasets. S2004 is derived from the Smith and Sandwell grid in deep regions such as these (see below).

S2004

Walter Smith has blended the GEBCO and Smith and Sandwell datasets in order to try and capture the best of both products and extend the Smith and Sandwell data to the poles. The blend uses Smith and Sandwell equatorward of 70^o and below 1000 m water depth and GEBCO poleward of 72^o and above 200 m water depth (including land). The two datasets are blended inbetween with a cosine taper.

The dataset can be downloaded from Walter Smith's ftp site: ftp://facon.grdl.noaa.gov/pub/walter/Gebco_SandS_blend.bi2

This is converted to a .grd using the GMT command:

xyz2grd /home/data/maps/Gebco_SandS_blend.bi2 -GGebco_SandS_blend_1m.grd -R-180/179:59/-90/90 -fig -I1m -ZTLhw -V

since this only extends to 179:59E, it is not periodic and can't be used for a global simulation (i.e. won't be stitched at 180^o). This is overcome by copying the -180^o values to 180^o:

grd2xyz Gebco_SandS_blend_1m.grd > Gebco_SandS_blend_1m.xyz
grd2xyz -R-180/-179:59/-90/90 Gebco_SandS_blend_1m.grd > min180s
grep -v 179.983 min180s | sed -e "s/-180/180/" > 180s
cat Gebco_SandS_blend_1m.xyz 180s > Gebco_SandS_blend_1m_global.xyz
xyz2grd -I1m -R-180/180/-90/90 Gebco_SandS_blend_1m_global.xyz -GGebco_SandS_blend_1m_global.grd

The global, 1 min converted .grd is called Gebco_SandS_blend_1m.grd.

Vector shoreline

The GSHHS (Global Self-consistent, Hierarchical, High-resolution Shoreline Database) is a high-resolution shoreline data set amalgamated from two data bases in the public domain. The data have undergone extensive processing and are free of internal inconsistencies such as erratic points and crossing segments. The shorelines are constructed entirely from hierarchically arranged closed polygons.

It was created and is maintained by Paul Wessel and Walter Smith and can be downloaded from Paul Wessel's ftp site: ftp://ftp.soest.hawaii.edu/pwessel/gshhs

gshhs_f.b Full resolution data

gshhs_h.b High resolution data

gshhs_i.b Intermediate resolution data

gshhs_l.b Low resolution data

gshhs_c.b Crude resolution data

Comparison of datasets

Histograms of bathymetry from the 3 datasets above reveal the 'terracing' problem common to grids derived from contour data (i.e. GEBCO). Note the 'spikes' in the number of counts at 500m intervals, the same spacing as the contour data from which the GEBCO grid was derived.

Comparison study area 1: The Mediterranean Sea

Bathymetry plots

Data for the Mediterranean region was extracted and analysed using this script, and then plotted using this script.

The graphics below illustrate the bathymetry for the Mediterranean Sea region taken from 5 different bathymetry datasets: 1) GEBCO, 2) Smith & Sandwell v8.2, 3) Smith and Sandwell v9.1, 4) S2004, and 5) DBDBv. All are plotted from 1 minute data, with the same colour bar and contours at 10, 20, 40, 60, 80, 100, 200, 500, 1000, 2000, 3000, and 4000 m water depth.

Maximum difference between bathymetries

The graphics below plot the maximum difference between the 5 bathymetry datasets used in turn, the last plot compares all the datasets together. All are drawn to the same colour scale.

'Relative error' between bathymetries

Here a measure of 'relative error' between the datasets is calculated as 'maximum difference / mean bathymetry'. This helps highlight significant differences between the different datasets in shallow regions where the absolute difference may only be small, but relative to the water depth it may be large. The Results are presented in the same order as the maximum difference plots above and are all drawn to the same colour scale.

Comparison study area 2: The North Sea

Bathymetry plots

Data for the North Sea was extracted and analysed using this script, and then plotted using this script.

The graphics below illustrate the bathymetry for the North Sea taken from 5 different bathymetry datasets: 1) GEBCO, 2) Smith & Sandwell v8.2, 3) Smith and Sandwell v9.1, 4) S2004, and 5) DBDBv. All are plotted from 1 minute data, with the same colour bar and contours at 10, 20, 40, 60, 80, 100, 200, 500, 600, and 700 m water depth.

Maximum difference between bathymetries

The graphics below plot the maximum difference between the 5 bathymetry datasets used in turn, the last plot compares all the datasets together. All are drawn to the same colour scale.

'Relative error' between bathymetries

Here a measure of 'relative error' between the datasets is calculated as 'maximum difference / mean bathymetry'. This helps highlight significant differences between the different datasets in shallow regions where the absolute difference may only be small, but relative to the water depth it may be large. The Results are presented in the same order as the maximum difference plots above and are all drawn to the same colour scale.