Scientific Investigations Report 2007–5224
U.S. GEOLOGICAL SURVEY
Scientific Investigations Report 2007–5224
Back to Table of Contents
Streamflow, optical-backscatter, and suspended-sediment data were collected and analyzed using USGS methods for streamflow (Rantz and others, 1982), in situ water-quality monitors (Wagner and others, 2006), and fluvial sediment (Edwards and Glysson, 1999). Daily mean streamflow and suspended-sediment discharge data, as well as annual minimum and maximum streamflow for the study period, are published separately in USGS Annual Water-Data Reports (Webster and others, 2005; U.S. Geological Survey, 2006; 2007). The published streamflow and suspended-sediment data also are available at http://waterdata.usgs.gov/nwis/ . Streamflow, gage heights, edited optical-backscatter data, and estimated suspended-sediment concentrations (calculated from calibrated optical-backscatter data) are presented at 15-minute intervals in Appendix A. Suspended-sediment concentrations, grain-size, and turbidity data for individual samples are presented in Appendix B.
Depth-integrated, single-vertical, and multi-vertical suspended-sediment samples were collected during water years 2004–06 following standard USGS procedures (Edwards and Glysson, 1999). Samples at two gaging stations (Lagunitas Creek SPT and Walker Creek) were collected 2–7 days per week, depending on hydrologic conditions, with increased frequency of sampling during periods of higher streamflow. A total of 14 periodic samples were collected throughout the 3-year project period at Lagunitas Creek PRS. Typically, two sequential suspended-sediment concentration (SSC) samples were collected during each visit, and additional samples were collected periodically throughout the water year for grain-size and turbidity analysis. Concentrations (SSC), grain size, and turbidity were measured at the USGS sediment laboratory in Marina, California, using methods described by Guy (1969) and Anderson (2005).
Optical backscatter has been used successfully as a surrogate for suspended-sediment concentration in low-gradient rivers such as the Lower Sacramento River (Schoellhamer and Wright, 2003) and steep mountain rivers such as the Yuba River (Curtis and others, 2006). Developed and tested by Downing and others (1981), optical backscatter sensors (OBS) emit infrared light into the water column, which is reflected on contact with suspended particles. A series of photodiodes positioned around the emitter detects the light reflectance (backscatter), and an empirical calibration is used to convert sensor output voltage into suspended-sediment concentration. One can think of the calibration as the relation between the mass of sediment per unit volume of water (suspended-sediment concentration) and the number of particles per unit volume of water (backscatter).
Backscatter is related inversely to particle size such that the sensitivity of the OBS changes with particle size by more than an order of magnitude; therefore, large changes in the particle size distribution of suspended material can be problematic when developing calibrations. For example, during a storm event, increasing streamflow results in the suspension of larger sized particles. As progressively more sand travels in suspension, the mass of sediment (per unit volume of water in suspension) increases at a different rate than the number of particles (per unit volume of water), resulting in a nonlinear calibration relation. Also, there is an upper limit to the concentrations that optical sensors can measure that may be exceeded in smaller, steeper creeks with flashier hydrographs. This complication can be addressed by decreasing the sensitivity of the OBS, which is accomplished by adjusting the sensor gain, whereby the highest output voltage (2,500 mV) corresponds to the maximum expected sediment concentration.
Continuously recording optical backscatter sensors (OBS-3) were installed at three gaging stations [Lagunitas Creek at Samuel P. Taylor (SPT) State Park, Lagunitas Creek at Point Reyes Station (PRS), and Walker Creek]. Data loggers stored output in millivolts from the optical sensors at 15-minute intervals and recorded data were telemetered from the sites or downloaded during site visits. Raw sensor output data (in millivolts) were archived in the USGS automated data-processing system (ADAPS) and retrieved and edited to remove invalid data. The criteria for identifying and removing invalid data included: (1) small groups (one to three) of anomalously high data points (assumed to represent debris or fish passage); (2) anomalous periods of moderate-to-high sensor output with no associated changes in the streamflow record or concentrations of sediment samples measured in the laboratory (assumed to represent bed aggradation); (3) progressive increases in sensor output prior to sensor cleaning followed by drastic reductions after cleaning (assumed to represent fouling); and (4) data that exceeded the upper limit of concentrations that the optical sensors were capable of measuring.
Sediment samples and optical data were collected seasonally (October to May), and measurement errors were assessed periodically throughout the water year. Measurement errors can lead to data loss and may occur due to instrumentation or environmental effects including: electronic sensor drift, fouling, and changes to the morphology of the channel bed. Sensors were field checked for electronic drift at the beginning and end of each water year using a 4,000-nephelometric turbidity unit (NTU) formazin standard, which was diluted using deionized water into a series of calibration solutions ranging from 40 to 1,200 NTU. Sensors were cleaned and immersed in the calibration solution and the voltage output was recorded. Sensor drift ranged from 1 to 4 percent, which fell within the recommended 5 percent calibration criteria (Wagner and others, 2006). Sensor fouling and aggradation of the channel bed occurred periodically throughout the project period, resulting in data loss. Fouling, due to biological or chemical particles accumulating on the sensors, occurred at all three sites and was of concern primarily during low-flow periods. Aggradation of the channel bed was problematic at Lagunitas Creek SPT and Walker Creek, resulting in extended periods of invalid data. Data loss was minimized by cleaning the sensor one to two times per week, as flow conditions allowed, and by periodically adjusting the distance of the sensors above the channel bed.
Calibration of sensor output voltage to suspended-sediment concentration can vary significantly with particle size and color (Conner and de Visser, 1992; Levesque and Schoelhammer, 1995; Sutherland and others, 2000). In lieu of using a laboratory-based calibration, the optical sensors were calibrated using field data on a site-specific basis. Therefore, the raw data, which represents sensor output voltages, is reported in millivolts (mV), as opposed to turbidity units such as backscatter units (BU) or formazin backscatter units (FBU).
Although the data collected at Lagunitas Creek PRS were not sufficient to develop a reasonable calibration, adequate data were available to develop empirical calibrations for Lagunitas Creek SPT and Walker Creek. OBS at these two sites were calibrated using SSC data from width/depth-integrated, suspended-sediment samples. The SSC and OBS data were nonlinear and displayed nonconstant variance of the residuals (heteroscedasticity). In cases such as this, Hersel and Hirsch (1992) recommend transforming the data (for example, log x, 1/x, x 1/2 ) to achieve constant variance of the residuals (homoscedasticity) required for ordinary least-squares regression. Analysis of the residuals indicated that only a log transformation resulted in a satisfactory decrease in the variance at higher concentrations and sensor output voltages such that homoscedasticity could be achieved.
The SSC and associated OBS data (collected contemporaneously) were transformed to base-10 logarithmic values, and the log values then were used to calculate a linear least-squares regression:
log SSC = log a + b log OBS (1)
The regression converted to power form is:
SSC = a OBS b (2)
The first significant runoff event at the beginning of each water year (first flush) and all subsequent storm events (greater than 200 percent of the mean daily flow) were analyzed if valid data were available. Mean daily streamflows at Lagunitas Creek SPT, Lagunitas Creek PRS, and Walker Creek were 44, 93, and 35 ft/s, respectively. Peak streamflow and storm durations were estimated using 15-minute streamflow data (Appendix A). Suspended-sediment yield and periods of time during which suspended-sediment concentrations were greater than 100 mg/L were estimated using calibrated optical backscatter data (Appendix A).
Back to Table of Contents