![[Link to USGS home page]](images/wrcmgbnr.gif)
| Predicted Seafloor Facies of Central Santa Monica Bay, California |
|
Introduction Study Area Data Classification Process Results Accuracy Assessments Misclassifications Conclusions References Mapping surficial seafloor facies (sand, silt, muddy sand, rock, etc.) should be the first step in marine geological studies and is crucial when modeling sediment processes, pollution transport, deciphering tectonics, and defining benthic habitats. Traditionally, the surficial facies of the seafloor is determined by collecting a suite of bottom samples, analyzing the samples for grain size and/or determining rock type, and then mapping the results by interpolating or extrapolating gaps between samples. However, a detailed surficial facies map is rarely achieved by this process because an insufficient number of samples are collected to adequately describe the variability of the surficial facies. Within the last few decades, acoustic systems have been developed that collect wide swaths of seafloor data and can map 100% of a study area. The advent of high-resolution multibeam echosounders (MBES) has provided a new technique to efficiently map large areas of the seafloor at meter-scale horizontal, and centimeter-scale vertical resolutions (Hughes Clarke et al., 1996; Hughes Clarke, 2000a; Hughes Clarke, 2000b). The most advanced MBES systems collect a swath of geo-referenced soundings that provide geodetic-quality bathymetry as well as calibrated acoustic backscatter values co-registered with each of the bathymetric soundings. However, as with all remote sensing data, these data require adequate ground truth to derive geologically meaningful maps. This study outlines an empirical technique that uses high-resolution multibeam bathymetry and co--registered calibrated backscatter data correlated to ground-truth sediment samples to predict the distribution of seafloor facies for a large area offshore Los Angeles, CA. The technique uses a series of procedures that involve supervised classification and a hierarchical decision tree classification that are now available in advanced image-analysis software packages. Derivative variance images of both bathymetry and acoustic backscatter are calculated from the MBES data and then used in a hierarchical decision-tree framework to classify the MBES data into areas of rock, gravelly muddy sand, muddy sand, and mud. A quantitative accuracy assessment on the classification results is performed using ground-truth sediment samples. The predicted facies map is also ground-truthed using seafloor photographs and high-resolution sub-bottom seismic-reflection profiles. This Open-File Report contains the predicted seafloor facies map as a georeferenced TIFF image as well as the multibeam bathymetry and acoustic backscatter data. The focus of the study is a marginal plateau, informally called Short Bank, located in central Santa Monica Bay, offshore Los Angeles, CA. (Figure 1, 67kb). Short Bank was chosen for this study because of the large amount of seafloor data available, including high-resolution multibeam bathymetry and acoustic backscatter, sediment samples, seafloor photographs, and seismic-reflection profiles. Short Bank is a relatively shallow continental-shelf region (40- to 110-m deep) that projects more than 18 km from the coastline and is bounded by two large submarine canyons; Santa Monica Canyon on the north and Redondo Canyon on the south (Gardner et al., 2003). The total study area covers 167.2 km2 in water depths that range from 40 m on the east to 450 m on the northwest. Previous work on and around Short Bank using seismic-reflection profiles and sediment samples identified rocky outcrops with as much as 12 m of relief, as well as areas of coarse gravelly sand, muddy sand, and sandy mud (Shepard and MacDonald, 1938; Terry and Stevenson, 1957; Junger and Wagner, 1977, Haner and Gorsline, 1978, Vedder et al., 1986, and Kolpack, 1987). Multibeam bathymetry (Figure 2, 55kb) and calibrated acoustic-backscatter data (Figure 3, 64kb) were collected using a Kongsberg Simrad EM1000 MBES (Gardner et al., 2003). The 95-kHz MBES transmitter/receiver used for this study was designed for operation in water depths from 5 to 800 m. The navigation subsystems provided horizontal position accuracies of better than ±1 m. A vehicle-motion sensor was used to compensate for the ship's roll and pitch during transmit and receive cycles with an accuracy of ±0.02o and yaw (heading) to ±0.05o. A CTD was used to determine the sound speed within the water column several times each day so that each acoustic path could be ray traced to the seafloor thereby allowing correction for refraction in the water column. In addition, sound speed at the transducer was continuously monitored. Shipboard processing corrected backscatter levels for source level, angular response, spherical spreading, attenuation in the water column, beam pattern, and ensonified area. Bathymetry data were corrected for measured tides that referenced all depth measurements to a mean lower low water datum. Ground-truth data used in this study included: 1) 60 sediment samples collected from the surface layer of box corers, 2) 35-mm color seafloor photographs collected from a towed camera sled, and 3) 53-km of high-resolution Huntec DTS seismic-reflection profiles. For a more in-depth discussion of the classification methodology, see Dartnell and Gardner, in press. The prediction of the distribution of seafloor facies from multibeam data requires a two-step empirical process that involves two classification procedures. The first step uses a supervised classification of the acoustic-backscatter with grain-size data and rock locations and the second step uses those results as rules for a hierarchical decision-tree classification that includes both bathymetry and acoustic backscatter. Although the hierarchical decision tree is a classification process, developing rules for the classification without some prior knowledge of the composition of the seafloor is difficult. Grain-size analysis were conducted on 60 surface samples collected from the Short Bank study area. Four sediment classes (facies) were identified using Folk's (1954) classification scheme based on percentages of gravel, sand, and mud. The four sediment classes are: 1) gravelly muddy sand (greater than 5 percent gravel, greater than 50 percent sand, and less than 50 percent mud), 2) muddy sand (less than 5 percent gravel, greater than 50 percent sand, and less than 50 percent mud), 3) sandy mud (less than 5 percent gravel, less than 50 percent sand, and greater than 50 percent mud), and 4) mud (less than 5 percent gravel, less than 10 percent sand and greater than 90 percent mud). Student's T-test comparing the sediment facies to multibeam backscatter showed that the backscatter intensity values of sandy mud and mud were not statistically different. Therefore, the sandy-mud and mud classes were grouped together into a single class called mud. The initial supervised classification of the acoustic backscatter data was made using the three sediment facies plus a rock facies. Rock was defined as a very high-backscatter surface with high bathymetric relief as identified from the multibeam bathymetry data. The predicted seafloor facies map generated from the supervised classification was used to test and refine the hierarchical decision-tree classification. This process used both the multibeam bathymetry and backscatter data to classify the data into the same four seafloor facies (rock, gravelly muddy sand, muddy sand, and mud). In addition, noise was added as a fifth class. Noise is defined as the abnormally high backscatter signal recorded at the MBES nadir. At near-vertical incidence, almost all of the transmitted acoustic pulse is returned to the transducer, resulting in a saturation of the signal in the inner beams. The decision-tree classification process used four raster images, the original backscatter-intensity image, and three derivative raster images calculated from the original bathymetry and backscatter images; a 3x3-filtered bathymetry-variance image, an 11x11-filtered bathymetry-variance image, and a 3x3-filtered backscatter-variance image. The four images were then analyzed using a hierarchical decision-tree classification that is part of the ERDAS Imagine 8.4 software package (ERDAS, 1999). The classification is a rules-based approach that uses a hierarchy of conditions to parse the input data into a set of classes. The decision-tree framework was developed from empirically determined textural rules, variables, and hypotheses. An hypothesis is an output-facies class, such as muddy sand, a variable is a raster image of derived values (i.e. 3x3 bathymetry variance), and a rule is a conditional statement about the variable's pixel (data) values that describes the hypothesis. The rules are based on the ground-truth areas from the previous supervised classification. The 60 ground-truth sediment samples were used to test and refine the classification. After each classifying iteration, a quantitative accuracy assessment was run that compared the predicted pixel of the new classified image to the ground-truth sediment type at each sample location. If required, the decision-tree rules were refined and the database reclassified until the highest accuracy assessment, 72%, was obtained for the ground-truth samples. The classification process resulted in a new predicted seafloor-facies thematic map composed of rock, gravelly muddy sand, muddy sand, mud, and noise. Noise was separated into its own class so it could be removed from the interpretations. The data were filtered to replace each noise pixel with the majority of non-noise pixels that surround it. For example, a noise pixel located in a field of mud was replaced with a mud classification pixel value. This process removed most of the noise and resulted in a predicted seafloor facies map that better represented the seafloor geology (Figure 4, 136kb). The predicted facies was combined with shaded relief bathymetry to visualize the facies with respect to the seafloor morphology (Figure 5, 101kb). The predicted facies classification of the Short Bank area shows that Short Bank has a more complex distribution of sediment composition than does the surrounding inner shelf and deeper Santa Monica Canyon (Figure 5, 101kb). Rock exposures were predicted throughout Short Bank and on the steeper flanks of the adjacent slope environments. The most prominent area of outcrops on the eastern edge of Short Bank protrudes above the surrounding seafloor by as much as 12 m (Figure 6, 98kb). Predicted gravelly muddy sand and muddy sand are found on the top flanks of the outcrops. Smaller ridges composed of individual boulders (seen in seafloor photographs) trend north to south on the western side and east to west on the northern side of Short Bank. These ridges have relatively low bathymetric relief (<2 m) compared to the larger rocky outcrops but trend for many kilometers. Rock is also exposed in places on the steep, western flanks of Short Bank. Rock covers 18.0 km2 or 10.8% of the study area. Gravelly muddy sand was predicted to occur in small patches throughout Short Bank and only covers 13.0 km2 or 7.8% of the study area. This facies is generally found close to the rocky outcrops where eroded rock material has been identified (Shepard and MacDonald, 1938). Muddy sand is the most prevalent sediment type predicted, covering 100.7 km2 or 60.2% of the area. This facies is found on the flat part of the plateau, in pockets within the elevated rocky outcrops, on the steep flanks, and on the smooth inner shelf (Figure 5, 101kb). Predicted mud facies dominate the inner shelf and floor of Santa Monica Canyon. Mud was also predicted on Short Bank in small pockets within the rocky outcrops as well as in larger bathymetric lows in the center and southern portion of the plateau. Predicted mud facies cover 35.4 km2 or 21.2% of the study area. The predicted seafloor facies map was tested for validity in three ways; 1) a quantitative accuracy assessment using sediment samples, 2) comparisons with seafloor photographs, and 3) comparisons with high-resolution seismic-reflection profiles. The first accuracy assessment compared sediment facies of the 60 samples to the predicted seafloor facies map. Forty-three of the 60 samples (72%) were correctly predicted. All of the 17 mis-classified samples were off by only one adjacent class (i.e., mud instead of muddy sand). In addition, there were a number of misclassifications where, if the ground-truth pixel location were moved over by one pixel, the classified pixel would be the correct facies. An accuracy assessment using seafloor photography shows good correlation between predicted seafloor facies and observed facies. Small rocky ridges with widths of only a few pixels on the predicted facies map are seen as areas of exposed rock (Figure 7A, 127kb). The predicted facies map shows a small region of muddy sediment surrounded by coarser sand and rocks in the center of Short Bank. A seafloor photograph of this area shows a field of muddy sediment with burrows and worm trails (Figure 7B, 127kb). Another example is an area where the predicted facies map shows a sharp boundary between a rocky ridge and a region of muddy sand. Seafloor photographs show that the ridge (Figure 7C, 127kb) in this area consists of rocks from 10 to 30 cm in length with pockets of coarse sand. Photograph "D" (Figure 7D, 127kb) taken approximately 3-m down track from Figure7C over the predicted muddy sand region documents sandy sediment but no rocks. The third accuracy test used 9 Huntec high-resolution seismic-reflection profiles to ground truth the rocky part of the predicted facies map. Whereas the seismic-reflection data cannot distinguish between sediment type, they do show zones of rock outcrops and intervening sediment sections. The predicted facies map correlates with outcropping rock that stand above the sedimented regions of the plateau (Figure 8, 119kb). This technique classified the high-resolution multibeam data on a pixel by pixel basis as contrasted with methods that interpolate between known ground-truth points spaced widely apart or by drawing polygons around similar ground-truth points. The technique also generated more accurate results than other automated classification techniques that classify backscatter alone (50% accuracy for an unsupervised classification of Short Bank and 58% for a supervised classification). The seafloor-classification scheme predicted some misclassifications. Most of the misclassified pixels occurred near the MBES nadir, a region dominated by noise. The most obvious errors are predicted rock pixels on the steep western flank of Short Bank near the MBES nadir. Also, the pixels at the edge of MBES nadir on top of Short Bank were misclassified as gravelly muddy sand instead of muddy sand. The misclassification occurred because of the abnormally high backscatter from the vertical- and near-vertical incidence at the MBES nadir. This high-backscatter zone rapidly decreases with across-track distance from nadir. Some of the pixels at the outer edge of the MBES nadir zone, above the noise level, have high bathymetric variance and high backscatter similar to rock, but based on the surrounding facies, these pixels should probably be classified as muddy sand. The predicted facies map generated by analyzing the high-resolution multibeam echosounder data with remote sensing analytical techniques is a complete and accurate map of the surficial geology of the seafloor of Short Bank. A hierarchical decision tree using bathymetry and backscatter data predicted the distribution of rock, gravelly muddy sand, muddy sand, and mud in this region. The predicted facies were ground-truthed with sediment samples, seafloor photography, and seismic-reflection profiles. Ground-truth with sediment samples shows that the predicted facies map is 72% accurate, better than other automated classification methods, and good correlations are demonstrated between seafloor photography and seismic-reflection profiles. This seafloor-classification technique provides a method to transform high-resolution multibeam bathymetry and calibrated acoustic backscatter data into meaningful geological information. Although this classification technique can be used on other areas of the seafloor, the exact values within the hierarchical decision tree cannot be transferred to a different area. Ground-truth samples and other geological information need to be gathered for each new study area to build that area's knowledge database. The predicted-seafloor-facies map from this study can be used not only to study the spatial distribution of geologic material on Short Bank, but also to model sedimentation and pollution transport processes, as well as for defining benthic habitats within central Santa Monica Bay. Dartnell, P and J.V. Gardner, in press. Predicting seafloor facies from multibeam bathymetry and backscatter data. Photogrammetric Engineering and Remote Sensing. ERDAS Field Guide, 1999, ERDAS Inc, Atlanta Georgia. 672p. Folk, R. L. 1954. The distinction between grain size and mineral composition in sedimentary rock nomenclature, Journal of Geology, 62: 344-359. Gardner, J.V., P. Dartnell, L.A. Mayer, and J.E. Hughes Clarke, 2003. Geomorphology, acoustic backscatter, and processes in Santa Monica Bay from multibeam mapping, Marine Environmental Research, 56: 15-46. Haner, B.E., and D.S. Gorsline, 1978. Processes and morphology of continental slope between Santa Monica and Dume Submarine Canyons, Southern California, Marine Geology. 28: 7787. Hughes Clarke, J.E., L.A. Mayer, and D.E Wells, 1996. Shallow-water imaging multibeam sonars: A new tool for investigating seafloor processes in the coastal zone and on the continental shelf, Marine Geophysical Researches, 18: 607-629. Hughes Clarke, J.E., 2000a. Acoustic Seabed Surveying: Meeting the new demands for Accuracy, Coverage, and Spatial Resolution. Geomatica, 54: 473-513. Hughes Clarke, J.E., 2000b. Present-day methods of depth measurements, Continental shelf limits: The scientific and legal interface (Peter J. Cook and Chris M. Carleton editors). Oxford University Press. p139-159. Junger, A. and H.C. Wagner, 1977. Geology of the Santa Monica and San Pedro Basins, California Continental Borderland. U.S. Geological Survey Miscellaneous Field Studies, Map, MF-820, 5 sheets, 1 pamphlet, scale 1:250,000. Kolpack, R.,L., 1987. Sedimentology of shelf and slope in Santa Monica Bay, California, American Association of Petroleum Geology, 71: 578. Shepard, F.P., and G.A. MacDonald, 1938. Sediments of Santa Monica Bay, California, Bulletin of the American Association of Petroleum Geologists, 22: 201-216. Terry, R.D., and R.E. Stevenson, 1957. Microrelief of the Santa Monica Shelf, California, Bulletin of the Geological Society of America, 68: 125-128. Vedder, J.G., H.G. Green, S.H. Clarke, M.P. Kennedy, 1986. Geologic Map of the Mid-Southern California Continental Margin. California Continental Margin Geologic Map Series, Map 2A. Division of Mines and Geology. Scale 1 : 250,000 |
 |
![[Reports]](images/nav_rprt.gif) |
 |
![[Home]](images/nav_home.gif) |
|
![[Search]](images/nav_srch.gif) |
 |