• Abstract
• Surveys
• Data Processing
• Data Tables
• Figures
• References
• Appendix A
• Metadata
• USGS Related Links
• WCMG Home
• Search WCMG

Geophysical Surveys of the San Andreas and Crystal Springs Reservoir System Including Seismic-Reflection Profiles and Swath Bathymetry, San Mateo County, California

Data Processing

GPS Data Processing

During the San Andreas Reservoir survey, the vessel was equipped with two positioning systems. The first was a CodaOctopus F180 attitude and position system. This system combines twin WAAS-enabled GPS signals with high-precision inertial moment measurements to provide real-time heave, roll, pitch, heading, and positioning information to the sonar-acquisition system. The second positioning system was an Ashtech Z-Xtreme, dual-frequency GPS receiver used to resolve 3-D boat positions during post-processing. The continuously operating GPS receiver (CGPS) station P178 was used as a base station. P178 had a 15-second epoch rate during the survey. These data were low-pass filtered using a 5-minute running median filter and used as a synthetic tide fixed in the WGS84 (G1150) / ITRF 2000 reference frame during post-processing (Figure 2). The WAAS-enabled position recorded by the F180 was used for horizontal positioning of the vessel.

For the Upper and Lower Crystal Springs survey, the CodaOctopus F180 firmware was upgraded to an F190 model in order to receive real-time kinematic (RTK) corrections directly, removing the need for the pseudo-tide procedure described above. The RTK GPS data (2-cm error ellipse) are combined with the inertial motion measurements directly within the F190 hardware so that high-precision position and attitude corrections are fed in real-time to the sonar-acquisition equipment. As before, the WGS84 (G1150) reference frame was used for both horizontal and vertical measurements.

Sound-Velocity Measurements

Sound-velocity profile measurements were collected daily throughout the survey with an Applied Micro Systems, SvPlus 3472. This instrument provides time-of-flight sound-velocity measurements using invar rods with a sound-velocity accuracy of +/- 0.06 m/s, pressure measured by a semiconductor bridge strain gauge to an accuracy to 0.15% (full scale), and temperature measured by thermistor to an accuracy of 0.05 C (Applied Microsystems Ltd., 2005). In addition, an Applied Micro Systems Micro SV, accurate to +/- 0.03 m/s, was deployed on the transducer frame for real-time sound-velocity adjustments at the transducer-water interface.

On all three days of the San Andreas Reservoir survey, the lake was isothermal to the depth measured and within 0.5 degrees C between locations (Figure 3). Similarly, the sound-velocity profiles were nearly uniform except for the very near-surface (<1 m depth).

In contrast to San Andreas Reservoir, the temperature-profile measurements of the Crystal Springs Reservoirs (Figure 4) show evidence of spatially variable, shallow mixing to about 5 to 7 m depth, below which a moderately strong thermocline reduces temperatures quickly to isothermal conditions by about 15 m depth. The temperature gradients in the lake are mirrored in the observed sound-velocity profiles, with strong gradients in sound velocity within the thermocline.

Sonar-Sounding Processing

The general processing workflow procedures for converting bathymetric soundings to a digital elevation model (DEM) are shown in Figure 5.

GPS data and measurements of vessel motion are combined in the F180 hardware to produce a high-precision vessel attitude packet. This packet is transmitted to the Swath Processor acquisition software in real-time and combined with instantaneous sound-velocity measurements at the transducer head before each ping. Up to 20 pings per second are transmitted, with each ping consisting of 2,048 samples per side (port and starboard). The returned samples are projected to the lake bottom using a ray-tracing algorithm working with the previously measured sound-velocity profiles. A series of statistical filters are applied to the raw samples that isolate the lake-bottom returns from other uninteresting targets in the water column. Finally, the processed data is stored, line-by-line in processed trackline files.

Digital Elevation Model Production

The digital elevation model (DEM) produced in this work is a combination of the bathymetric data collected by the USGS and airborne LiDAR collected as part of the National Science Foundation GeoEarthScope Northern California LiDAR Project (GESNCLP).

The individual soundings stored in the processed trackline files are combined into a 1-m resolution raster in the Grid Processor software. Each bin stores statistics on the samples that fell within the cell, including the mean elevation value, the sample count, and measurements of sample precision, such as standard deviation. The Grid Processor software allows for 3-D manual editing of the soundings that proved particularly useful for cleaning the data around the vertical wall of the Lower Crystal Springs Dam and in a few other locations where the statistical filtering process identifying the bottom returns broke down (see the Bathymetric Uncertainty Estimation section for more information). The 1-m binned elevation values and the associated bin standard-deviation values were exported from Grid Processor as an XYZ point cloud. The mean standard deviation of all cells in the San Andreas data set was 0.28 m (1 sigma); for Lower Crystal Springs it was 0.24 m (1 sigma); and for Upper Crystal Springs it was 0.16 m (1 sigma).

The XYZ point cloud was gridded by using Surfer 8 Kriging algorithm with the following parameters:

• Gridding Algorithm: Kriging
• Variogram Model: Linear
• Nugget Variance: (used standard deviation values listed above)
• MicroVariance: 0.00 m
• Number of Seach Sectors: 4
• Maximum Data Per Sector: 16
• Maximum Empty Sectors: 1
• Minimum Data: 8
• Maximum Data: 64
• Search Radius: 15 m

Each grid was exported to GRASS GIS. In GRASS, the bathymetry shoal of 1.5 m depth was trimmed to remove shoreline artifacts associated with nearshore vegetation. Similarly, GESNCLP LiDAR data was clipped below about 1 m above the lake level to remove the water surface from the LiDAR data and buffered to include only data within 250 m of the lake shoreline. The two data sets, both in WGS84 G1150, were then merged together to form a combined bathymetric/topographic DEM. For both San Andreas and Crystal Springs lakes, there was about 3 to 4 m of vertical separation between the lowest LiDAR elevations and the shoalest bathymetric soundings. This vertical gap manifested itself as a horizontal gap that occasionally exceeded 200-m width in certain sections of the lakes. The gap was interpolated by using a regularized spline with tension (Mitasova and Mitas, 1993). In most cases the spline smoothly transitions from the LiDAR data to the bathymetry, but in cases where the horizontal separation between the LiDAR and sonar data is large, linear segmentation artifacts are apparent. Also, in areas where the shoreline elevations are not smooth (such as where overlying vegetation confused either the laser or the sonar), the interpolation looks flawed.

To convert the data from the WGS84 (G1150) ellipsoid to NAD83 (CORS96), the combined topo/bathy data set was exported from GRASS to a point file. Next, a 14-point Helmert transformation was applied to the data set with time-dependent transformation parameters figured for January 1, 2008, according to methods outlined in Soler and Snay (2004). The transformed data was then regridded in GRASS using a nearest-neighbor algorithm using the original 1-m grid spacing. Table 2 shows the specific parameters in the format required by the cs2cs program found in the Proj4 library used to transform the points.

The NAD83 (CORS96) ellipsoid elevations in the combined bathy/topo data sets were converted to orthometric heights based on NAVD88 by using the National Geodetic Survey Geoid03 Model (NGS, 2003). Tile g2003u05 of the Geoid03 model was downloaded from the NGS Web site and converted to GRASS format. The grid was then projected into UTM Zone 10 North coordinates (the original data is 1 arc minute spacing in geographic coordinates) using a 10-m cell spacing with a cubic convolution interpolation. The resulting grid was subtracted from the bathy/topo data set resulting in a new bathy/topo data set in the NAVD88 vertical datum. The portion of the Geoid03 model which overlies the San Andreas and Crystal Springs Reservoirs is shown in Figure 6. On average, there is about 32.55 m difference between the NAD83 (CORS96) ellipsoid and NAVD88 over the area of interest. However, as Figure 6 shows, the difference varies by about 12 cm north-to-south in a nonlinear gradient.

Figure 7 and Figure 8 show the resulting merged bathymetery and lidar rasters for San Andreas Lake and Upper and Lower Crystal Springs Reservoirs, respectively.

Bathymetric Uncertainty Estimation

The mean standard deviation of bathymetric soundings in the 1-m cells of the San Andreas data set was 0.28 m (1 sigma), for Lower Crystal Springs it was 0.24 m (1 sigma), and for Upper Crystal Springs it was 0.16 m (1 sigma). These statistics are 2 to 5 times the 1-sigma elevation accuracy published for the GESNCLP lidar (0.05 to 0.10 m).

Areas of high standard deviation in Figure 9 and Figure 10 are associated with areas of strong along-track artifacts in the final DEMs. The artifacts are particularly problematic in rough terrain, such as along the steep valley walls of the reservoir and throughout most of Lower Crystal Springs. The errors seem to be associated with soundings directly beneath the vessel—a known nadir artifact that is characteristic of sidescan sonars—but also with the overlapping edges of the swath on steep slopes where the bottom detection algorithms of opposing track lines appear to be deriving different values. Some of this error may be associated with instability of the small vessel used during this survey; other errors appear to be intrinsic to the sonar system itself.