Geologic characteristics of benthic habitats in Glacier Bay, southeast Alaska

METHODS

Sea Floor Video

Video imagery of the sea floor in Glacier Bay at depths of 15-370 m was obtained in April 2004 on a USGS research cruise aboard the R/V Alaskan Gyre. (The cruise report can be viewed online at http://walrus.wr.usgs.gov/infobank/g/g104gb/html/g-1-04-gb.meta.html). The principal objectives during seafloor video data collection were to: (1) record geologic and biologic characteristics of the seafloor real-time, (2) ground-truth geophysical data (bathymetry and backscatter) by resolving both common and unique features of the sea floor, and (3) examine regions of transition between different substrate types suggested in acoustic backscatter data. Video observations would eventually be used to construct maps of seafloor morphology, substrate type, and habitat distribution. Therefore transect locations were selected on the basis of the existence, quality, and complexity of geophysical data and on regions of geologic transition and/or biologic significance.

A towed video sled was used to collect sea floor data and imagery . The sled was constructed of welded aluminum with the following dimensions: length = 1.36 m (53.5"), width = 0.44 m (17.5"), height = 0.52 m (20.6"). The sled was equipped with forward- and downward-looking video cameras, lights, altimeter, pressure sensor, and pitch and roll sensors. Two down-pointing lasers spaced 20 cm apart provided scale on the seafloor. As equipped (including ballast for deep marine operations), the sled weighed ~57 kg (125 lbs) in air. The rigging consisted of a four-point bridle assembly attached to a swivel. A 14-mm (0.56") electromechanical cable with a strain-relief grip was threaded through the rigging and connected to the electronic components mounted on the sled frame. A 1000-kg (~2200 lb) winch was mounted on the deck of the vessel and used to deploy the camera sled with ~600 m (~2000') of spooled electromechanical cable.

The sled was towed behind the vessel at speeds of 0.5-1.5 knots, and the winch operator maintained its altitude above the seafloor at 1-2 m as much as possible. Video footage was recorded to digital mini-DV tape and then copied to DVD. Ship position was determined by a CSI Wireless differential geographic positioning system (dGPS). All instrument data were multiplexed through a sub-sea housing and transmitted by the 12-conductor cable to a topside console. Latitude, longitude, height above the seafloor, pitch, roll, water depth, ship speed, ship heading, and Greenwich Mean Time (GMT) were continuously imprinted on the digital video tape while recording. These data were also automatically recorded once per second in a navigational text file.

Positional accuracy of the sled relative to the ship's dGPS position varied with water depth, current speed and direction, and environmental conditions. Cable layback was not measured directly but was estimated to be approximately equal to the water depth during most deployments. For example, along the deepest transects (370 m water depth), video observations may have positional uncertainties of ~200 m relative to the ship's location.

Visually observed sea floor characteristics (geomorphology, sediment texture, and biota) were digitally recorded in real time at 30 second intervals by a geologist and a biologist watching the towed video (after Anderson et al. in press). Observation codes were entered as "events" in G-Nav navigation software (see acknowledgments) using an "X-Keys" programmable keypad and a Dell Inspiron 8100 laptop computer. Time (GMT), dGPS position, and other ship data for were automatically recorded in the text file each time an observation event was entered. Observations at each event included:

• primary and secondary substrate type (e.g. boulder/cobble, rock/sand, mud/mud)
• substrate complexity (rugosity)
• seafloor slope
• benthic biomass (low, medium, or high)
• the presence of benthic organisms and demersal fish
• small-scale sea-floor features (e.g., ripples, tracks, and burrows).

Nearly 42 hours of underwater video were collected and logged real-time in this manner on 52 transects in the lower and central bay, in the Beardslee and Marble Islands, offshore of Tlingit Point, and in parts of the east and west arms (view a location map of video observations).

Processing and Analytical Methods

Seafloor observations and geophysical data were co-registered, integrated, and analyzed using ArcGIS, ArcGrid, and ERDAS Imagine software to formulate predictions of benthic habitat distribution in the central and lower bay (Dartnell and Gardner 2004; Chavez 1984).

All grid calculations were performed using ArcGrid. (Commands and code are available on request.) Multibeam bathymetry and acoustic reflectance data grids (each composed of more than three million pixels) were first de-sampled from 5 m to 20 m pixel resolution. ArcGrid calculations were then performed on both 20-m data grids to quantify the variance of bathymetry and backscatter values within each "kernel" (in this case, 3x3 group of pixels).

"Maximum" acoustic images were calculated for each grid by running a filter that returned the maximum value within a kernel to the center cell of that kernel. "Minimum" acoustic images were similarly calculated such that the center cell of each kernel was the minimum value observed in that kernel. The "variance" was then calculated as the difference between the maximum and minimum images. Results were binned and assigned relative index values of 1-5 using an unsupervised classification in ERDAS. The central pixel of each kernel was assigned the index value which corresponded to the level of variance observed in the surrounding 8 pixels (Table 1). These calculations and steps were performed on four raster images: bathymetric variance, backscatter variance, backscatter intensity, and seafloor slope.

Table 1. Index of values assigned to the central pixel of each kernel in the raster images to express the relative variability in acoustic data observed within each 3x3 group of pixels.

Bathymetric variance (depth variability)
A measure of the variance in water depth observed in the kernel. When depths were homogeneous within a kernel, the bathymetric variance of the central pixel was low (1); when depths were more variable within a kernel, the index of the central pixel was high (5). For example, high bathymetry variance values resulted for areas of seafloor having a wide range of depths over relatively small area, such as rocky outcrops. Low bathymetry variance values resulted for smooth, flat, or homogeneous areas of seafloor.

Backscatter variance (textural variability)
A measure of the variance in acoustic reflectance (backscatter) observed in the kernel. When the range of backscatter values within a kernel was low, the textural variability index of the central pixel was low (1); when the backscatter values were more variable within a kernel, the index of the central pixel was high (5). For example, high backscatter variance values resulted for areas of seafloor having a wide range of backscatter values, such as a rocky outcrop with pockets of fine sediment (mixed patches of high and low backscatter). Low backscatter variance values resulted for homogeneous areas of seafloor, such as in muddy basins.

Top of Page

Geomorphic Classification

We defined four principal geomorphic classes on the basis of our analysis of acoustic data combined with video observations (Table 2). We constructed a hierarchical decision tree using ERDAS Imagine 8.4 software. The decision-tree framework consists of hypotheses, variables, and rules that use a conditional hierarchy to parse and classify the input data (ERDAS 1999):

• hypothesis = the geomorphic class into which the pixel will be classified
• variable = input data; raster image of derived values (e.g. bathymetry variance)
• rule = a conditional statement about the variable’s pixel (data) values that describes the hypothesis.

Because the variables (raster images) were co-registered, rules are established that compared and tested pixel values within and between images. Multiple rules and hypotheses could be placed in a hierarchy that best described the hypothesis.

Hypotheses were formulated that correlated general substrate type observed in video with bathymetric variance, backscatter variance, backscatter intensity, and slope. Rules were first made to correctly classify the (known) substrate type along video transect lines. These rules were then applied to acoustic data in other areas for which no video observations existed.

We tested each hypothesis by running its set of rules through the supervised decision-tree classification and comparing the geomorphic class output to known substrate type at ~30,000 data points. The final classification scheme that produced the best fit to observational data is summarized in Table 2. These hierarchical rules were eventually used to produce maps of substrate type and sediment texture in Glacier Bay. Further details on how this was accomplished in ArcGIS are provided in the following sections.

Table 2. Characteristics of principal geomorphic classes used to define "rules" in a supervised ERDAS decision-tree classification of acoustic data in Glacier Bay.

Bathymetric Classification

Depth is an important factor in the distribution and life history of benthic organisms. Gridded multibeam bathymetry data were sorted into three classes, converted to isobaths, and used to subdivide habitat polygons (Table 3). Bathymetric classes were selected on the basis of biological and geological observations, as well as with an understanding of oceanographic patterns in the bay (Etherington et al. 2004). Depths less than 75 m represent environments with the highest energy and a well-mixed surface layer, particularly those areas south of Sitakaday Narrows and toward the mouth of Glacier Bay. The 200 m boundary was selected on the basis of observed transitions in seafloor geology, benthic community structure, and oceanographic properties that occur bay-wide at this depth and represent a shift to conditions typical of deep-sea environments. Bathymetry values in the ArcGIS data table are in the form of negative elevations relative to sea level.

Table 3. Depth range of bathymetric class isobaths used to subdivide habitat polygons.

Seafloor Slope and Complexity

Slope and complexity refer to the bathymetric characteristics of a polygon as observed in sonar data. The degree of slope (0-90°) was calculated from the multibeam bathymetry grid. Complexity is a relative description of seafloor rugosity (roughness), where low values are characteristic of flat, homogeneous sea beds and high values are characteristic rocky, rough, and variable sea beds. Relative values of complexity were assigned to GIS polygons on the basis of video observations, original ERDAS classification, and geomorphic variability within the polygon. For polygons in which bathymetry data did not exist or were inconclusive, slope and complexity fields in the ArcGIS data table were intentionally left blank. Table 4 defines the range of values applied, which are consistent with the Greene et al. (1999) classification system.

Table 4. Index of seafloor slope and complexity values. Degree of slope was calculated from multibeam bathymetry and binned into three classes (consistent with Greene et al. (1999)). Seafloor complexity was recorded in video observations and is an attribute of those point features. Seafloor complexity is also an attribute of habitat polygon features that was assigned on the basis of bathymetry, backscatter, and video ground-truthing.

Top of Page