Multibeam Sonar Mapping and Modeling of a Submerged Bryophyte Mat in Crater Lake, Oregon
Summary
The remarkably clear water of Crater Lake, Oregon (fig. 1, 172 Kb) helps support a deep-water benthic bryophyte community of aquatic moss (fig. 2, 500Kb), which has been observed on submersible dives to a depth of approximately 140 m (McIntire and others, 1994; Buktenica, 1996). Thick mats of moss were observed on the Wizard Island platform as well as hanging from rock outcrops around the caldera wall. While the moss has been observed in a number of locations, its overall distribution is poorly understood. Mapping the distribution and calculating the aerial coverage of the moss would be extremely useful for understanding the overall ecology of the lake, especially the lake's nitrogen balance.
In the summer of 2000, the U.S. Geological Survey, in cooperation with the National Park Service, mapped Crater Lake using a multibeam echosounder (Gardner and others, 2000; Gardner and Dartnell, 2001). The mapping mission collected bathymetry and co-registered acoustic backscatter. Traditionally, multibeam data have been used to map seafloor or lake-floor morphology as well as the distribution of surficial facies or the geological component of benthic habitats. Rarely has multibeam data been used to map the distribution of benthic organisms directly.
This report discusses methods used to map the distribution of the deep-water moss based on the multibeam data and lake floor video collected using both a towed camera sled and a remotely operated vehicle (ROV). This report also presents the moss distribution map, base layers, and ground-truthing data in an ESRI ArcMap project.
The 2000 survey used a Kongsberg Simrad EM1002 high-resolution multibeam mapping system owned and operated by C&C Technologies, Inc. of Lafayette, LA. The survey took place from July 28 to August 2, 2000, and collected over 16 million soundings. The soundings were gridded into 2-m resolution geo-referenced and co-registered maps of bathymetry (fig. 3, 340 Kb) and acoustic backscatter (fig. 4, 528 Kb).
Although bathymetry can be interpreted using relatively straightforward geomorphological principles, the interpretation of acoustic backscatter is more complicated because it represents a complex interaction between the acoustic pulse, the lake floor, and the subsurface. Backscatter strength is dependent on the acoustic source level, the frequency used to image the lake floor (95 kHz for this study), the grazing angle, the composition of the lake floor (namely, grain size, water content, bulk density), seafloor roughness, volume reverberations to a few meters depth (Urick, 1983; Gardner and others, 1991; Augustin and others, 1996; Blondel and Murton, 1997), and possibly biomass.
The Crater Lake backscatter has an unusual pattern. Generally in the marine environment, harder grounds, such as rock outcrops, reefs, and coarse gravels and sands, have relatively high backscatter intensity, whereas loosely consolidated, finer grained sediments have low backscatter intensity (Hughes-Clarke and others, 1996; Dartnell and Gardner, 2004). In Crater Lake however, submerged outcrops along the caldera wall (Bacon and others, 2002) as well as the platform surrounding Wizard Island have relatively low backscatter intensity (-35 to -40 dB) compared to the steep talus slopes (-15 to -20 dB), and the deep sediment filled basins (-25 to -35 dB)(fig. 4, 528 Kb). A number of reasons for this reversal include differing sediment grain-size and/or lake-floor roughness. However, this lower backscatter may be attributed to the thick layer of moss diffusing the acoustic signal.
To obtain visual observations of the moss that we could relate to the remotely collected multibeam data, the U.S. Geological Survey, National Park Service (NPS), Oregon State University, and Southern Oregon University collected lake-floor video from a towed camera sled and a remotely operated vehicle (ROV) in August 2006. The camera sled towed behind the NPS RV Neuston collected over 23 line kilometers of regular and high-definition video. The sled is approximately 2 m by 1 m and houses a downward-looking camera, a forward-looking camera, and a downward-looking high-definition camera (fig. 5, 136 Kb). Paired lasers associated with each video camera are spaced 16 cm apart and provide scale.
Both the downward- and forward-looking regular video cameras had a real-time feed up to the R/V Neuston where the video could be observed as well as recorded onto miniDV tapes (fig. 6, 136 Kb). From the streaming video, observations, such as presence or absence of the bryophyte mat and the lake floor composition (such as, rock or sediment), were recorded every minute into a navigational system (YoNav). Each observation was stamped with a time and geographic coordinates based on differential GPS. While a few video transects were collected along the shoreline, the majority of the video was collected over the Wizard Island platform.
The second August 2006 research cruise collected lake-floor video using an ROV to traverse the steeper caldera walls where it was unsafe to tow the camera sled. Observations from both the camera sled and ROV cruises were then compared with the multibeam bathymetry and backscatter data to develop a moss-distribution model.
The moss-distribution model was developed by analyzing the bathymetry, backscatter, and video observations using a series of GIS and remote sensing techniques. Three variables used in a classification process include backscatter intensity, depth range, and a derivative bathymetric index called Bathymetric Position Index (BPI), all at 2-meter spatial resolution. Because these grids were co-registered, not only could relationships be made between pixel values within a grid, but comparisons could also be made between pixel values between grids.
BPI is a measure of relative depths that indicate the position of a given point to the overall surrounding landscape (Kvitek and other, 2003; Murphy, 2007). This index maps the overall shape of the lake floor, identifying concave shapes such as rock outcrops, pinnacles, and reefs, and convex shapes such as canyons, gullies, and pits. The index values are generated using the following map algebra formula (Weiss, written comm., 2001):
BPI = int((bathymetry - focalmean(bathymetry, annulus, irad, orad)) + 0.5)
where:
int = GIS integer function
focalmean = GIS area function
annulus = processing neighborhood comprised of one smaller circle within a larger circle (donut shape)
irad = inner radius of annulus in cells
orad = outer radius or annulus in cells
Positive values represent concave surfaces, while negative values represent convex surfaces. It was determined that lake-floor slope was not a reliable indicator of moss distribution because moss has been observed on the gentler slopes of Wizard Island platform as well as on the steeper rock outcrops around the caldera wall. BPI may be a better indicator because it distinguishes between the outcrops around the caldera wall where moss attach and the less stable talus slopes and gullies where moss is absent.
The three variables were analyzed using a hierarchical decision-tree classification process (ERDAS, 1999; Dartnell and Gardner, 2004) to produce a new classified map showing the distribution of moss. The classification is a rules-based approach that uses a hierarchy of conditions to parse the input data into one or more classes. The decision-tree framework was developed from empirically determined "hypotheses", "rules", and "variables" (fig. 7, 164 Kb). In this case, a "hypothesis" is an output moss class, a "variable" is a raster image of derived values (such as, backscatter intensity), and a "rule" is a conditional statement about the variable's pixel (data) values that describes the hypothesis. The rules were established based on the observations recorded during the camera sled and ROV operations. This was an iterative process where rules were tested, and the resulting map was compared with the video observations. The rules were then refined to achieve the highest possible agreement. The resulting classification not only maps the distribution of the moss but also distinguishes between thick, continuous mats of moss and thinner, patchy moss.
The combination of "hypotheses", "rules", and "variables" in the hierarchical decision tree produced a map showing the distribution of "continuous" and "patchy" moss (fig. 8, 256 Kb). In this map, a 2-meter pixel is tagged as continuous moss (green) if the backscatter intensity is within the range of -30 to -45 dB, has a BPI value greater than -5, and is within the depth range of 29 to 96 meters (fig. 7). A pixel is tagged as patchy moss (yellow) if the backscatter intensity is within the range of -22 to -30 dB, has a BPI value greater than -5, and is within the depth range of 35 to 96 meters.
In addition to the empirical classification process described above, an attempt was made to manually outline the boundary of the moss based on backscatter alone (fig. 4). In some areas of the moss's shallower limit, there was a clear boundary between high and lower backscatter. However, in other areas including the deeper limit, the boundary was much less distinctive. Whereas the empirical approach and the manual approach show similar moss coverage, the empirical approach was able to distinguish between the continuous and patchy mats of moss.
Results show that moss covers most of the Wizard Island platform (fig. 8, 256 Kb) as well as numerous spots around the caldera wall. On the Wizard Island platform, the continuous moss mat is most abundant in the northeast and southeast quadrants, whereas thinner, patchy moss is found in the northwest and southwest quadrants. One possible reason for this pattern is the eastern quadrants receive more sunlight throughout the day, while the western quadrants become shadowed in the early afternoon as the sun falls below the caldera wall. Around the caldera wall, patches of moss are attached to many of the protruding outcrops but not to the less stable talus slopes. Based on this map, the thicker, continuous layers of moss covers about 2 sq km or 0.77 sq mi whereas the thinner patchy moss covers a much smaller area of 0.2 sq km or 0.08 sq mi.
The accuracy of the moss-distribution map was tested by overlaying the camera sled observations onto the map in a GIS. Points where we observed moss were highlighted one color while points where we did not observe moss (detritus) were colored a different color (fig. 8, 256 Kb). At each point where we observed moss, the underlying pixel in the classified map that was tagged as either "continuous moss" or "patchy moss" was counted as correct. An underlying pixel that was tagged with no moss was counted as incorrect. Out of a total of 374 observations around the Wizard Island platform, 343 of the underlying pixels were correct (tagged continuous or patchy) whereas 31 were incorrect (tagged as no moss) for an overall accuracy of 91.7 percent. On the other hand, 27 of a total of 186 detritus observation points (14%) were recorded over a "continuous" or "patchy" classified pixel.
There is also good agreement between the moss distribution map and video stills (images) captured from the video. Figure 9 (256 Kb) shows images captured from video north of Wizard Island. The image at point A located outside the area of moss on the map shows little to no live moss. The image at point B located in an area of patchy moss on the map shows thin moss with open patches. Point C on the map shows the sharp, shallow boundary of the moss and this is confirmed by the image captured at this location. Similarly, figure 10 (252 Kb) shows images captured from video collected south of Wizard Island. The image captured at point A located in a small area of patchy moss on the map shows thin patches of moss while the image captured at point B located in an area of thick, continuous moss shows full coverage of live green moss.
The focus of the accuracy assessment was over the Wizard Island platform. Visual comparisons between the observations and the classified map around the caldera wall show that the accuracy of these regions is less. The moss coverage around the caldera wall is more thin and patchy than on the Wizard Island platform. The decision tree classification process, mainly due to backscatter intensity, does not seem to accurately map these regions.
Multibeam bathymetry and backscatter data effectively contribute to map the distribution of the deep-water bryophyte moss community, especially over the Wizard Island platform. Empirical techniques, analyze the multibeam bathymetry, backscatter, and lake floor video observations to reliably map the distribution of both the thicker, continuous mats as well as the thinner, patchy mats. The manual outline of the moss boundary around the Wizard Island platform matches well with the empirical approach. However, this manual process does not seem possible without first completing the empirical approach.
Disclaimer
Any use of trade, firm, or product names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. government.
Augustin, J.M., Le Suave, R., Lurton, X., Voisset, M., Dugelay, S., and Satra, C., 1996, Contribution of the multibeam acoustic imagery to the exploration of the sea-bottom: Marine Geophysical Research, v. 18, p. 459-486.
Bacon, C.R., Gardner, J.V., Mayer, L.A., Buktenica, M., Dartnell, P., Ramsey, D.W., and Robinson, J.E., 2002, Morpholgy, volcanism, and mass wasting in Crater Lake, Oregon: GSA Bulletin, v. 114, no. 6, p. 675-692.
Blondel, P., and Murton, B.J., 1997, Handbook of seafloor sonar mapping, New York, John Wiley and Sons, 314 p.
Buktenica, M., 1996, Why enter a sleeping volcano in a submarine?. Nature Notes: Crater Lake National Park, v. 27, Available at http://www.nps.gov/archive/crla/notes/vol27c.htm#2.
Dartnell, P., and Gardner, J.V., 2004, Predicting seafloor facies from multibeam bathymetry and backscatter data: Photogrammetric Engineering and Remote Sensing. v. 70, no. 9, p. 1081-1091.
ERDAS Field Guide, 1999, ERDAS Inc, Atlanta, Georgia, 672 p.
Gardner, J.V., Field, M.E., Lee, H., and Edwards, B.E., 1991, Ground-truthing 6.5-kHz side scan sonographs: what are we really imaging? Journal of Geophysical Research, v. 96, p. 5955-5974.
Gardner, J.V., Mayer, L.A., and Buktenica, M., 2000, Cruise Report RV Surf Surveyor Cruise S1-00-CL, Mapping the Bathymetry of Crater Lake, Oregon: U.S. Geological Survey Open-File Report 00-405.
Gardner J.V., and Dartnell, P., 2001, 2000 Multibeam Sonar Survey of Crater Lake, Oregon—Data, GIS, Images, and Movies: U.S. Geological Survey Digital Data Series, DDS-72.
Hughes Clarke, J.E., Mayer, L.A., and Wells, D., 1996, Shallow-water imaging multibeam sonars: A new tool for investigating seafloor process in the coastal zone and on the continental shelf: Marine Geophysical Research, v. 18, p. 607-629.
Kvitek, R., Iampietro, P., Summers-Morris, E., 2003, Integrated spatial data modeling tools for auto-classification and delineation of species-specific habitat maps from high-resolution, digital hydrographic data: NOAA Technical Report, 74 p.
McIntire, D.C., Phinney, H.K., Larson, G.L., and Buktenica, M., 1994, Vertical distribution of a deep-water moss and associated epiphytes in Crater Lake, Oregon: Northwest Science, v. 68, no. 1, p. 11-21.
Murphy, J., 2007, Benthic Terrain Modeler, Available at http://www.csc.noaa.gov/products/btm/.
Urick, R.J., 1983, Principles of underwater sound, 3rd edition, New York, McGraw-Hill Book Company, 423 p.
For more information contact: Peter Dartnell
