Suspended Sediment and Turbidity Patterns in the Middle Truckee River, California for the Period 2002 - 2003

URL:

Abstract:

EXECUTIVE SUMMARY
The principal goal of the Truckee River suspended sediment study was to estimate the
sediment loads for the Truckee River in California, and to characterize the existing range of
sediment loads and variability according to total amount, maximum, duration, timing and
frequency of sediment transport events. Our primary objective in support of this goal was to
develop a sediment surrogate that could be measured continuously in the Truckee River. This
work supports development of numeric targets and load allocations for the Truckee River
sediment Total Maximum Daily Load (TMDL) in California. The report is divided into three
chapters: 1) Develop a model to predict suspended sediment concentrations from multiple
variables, including turbidity; 2) Estimate suspended sediment loads; and 3) Assess spatial
and temporal properties of turbidity.
Transport changes in suspended sediment concentration (SSC) due to natural or
human-induced causes are difficult to characterize because SSC varies rapidly and
unpredictably during storm events. Capturing the extreme variation in SSC during storms
requires sampling at high temporal frequency, which is usually impractical and expensive. As
such, easier-to-measure surrogate variables are monitored continuously with in-situ
instrumentation. Thus, a continuous turbidity record, supplemented with selected
measurements of SSC to derive the turbidity-SSC relationship, can provide an efficient and
cost effective method for estimating transported suspended sediment loads.
Our project builds on and extends previous work by the U.S. Geological Survey
(USGS) in the choice of the general class of statistical modeling techniques-regression
methods. First, we use multivariate regression techniques, which allows for inclusion of
multiple variables characterizing the river processes to the model. Second, since the Truckee
River is expected to have high variability in sediment and turbidity, we explored parametric
as well as nonparametric (robust) statistical techniques. Third, since we have evidence from
previous work that the relationship between turbidity and sediment may be nonstationary, we
used local regression models. These allow for different functional relations to be built for
different parts of the data set that increases the overall fit of the model to the data.
Additionally, since extreme sediment discharge events may be of major interest, we made
every effort to capture them with our models, and not discard them (before modeling) as
outliers. Fourth, we included error estimates (confidence intervals) for the predictions made
using our models.
The primary data sources for building the model and calculating sediment loads were
flow, measured by the USGS at three sites, and turbidity measured by the California
Department of Water Resources (CalDWR) at four sites. In addition, the turbidity
instrumentation measured water temperature and specific conductivity. In order to develop a
statistical relationship between suspended sediment concentrations and the explanatory
variables (here discharge, turbidity, specific conductivity, and water temperature), SSC
samples needed to be collected throughout the year as the explanatory variables varied.
Therefore SSC was collected monthly at all four turbidity collection sites and weekly during
snowmelt. Additionally, SSC was collected at Farad during thunderstorm events.
Model Development to Predict Suspended Sediment Concentrations
We developed a multiple linear regression model (MLR) for predicting SSC from
various combinations of turbidity, water temperature, stream flow and specific conductivity:
iv
a) Full model: turbidity, flow, water temperature and specific conductivity;
b) Turbidity, water temperature and specific conductivity;
c) Flow only; and
d) Water temperature only
Models with fewer predictive variables were explored because it is often the case that
only one or two (e.g., flow, temperature) of these variables are measured continuously in the
field. A statistically “good” model with few predictive variables could be very useful for
future studies. As part of the process of developing the predictive models, the statistical
relationships between individual variables were analyzed.
The “best” model developed was one in which SSC was predicted from all four
explanatory variables (turbidity, flow, water temperature, and specific conductivity). The R2
was 0.73, that is, the explanatory variables explained about 73% of the variability of SSC.
From a statistical standpoint, this was found to be a “good” model, in that the mathematical
assumptions under which the model was constructed satisfied statistical diagnostic tests. In
this model, site was not found to be significant, so one equation can be used for all four sites
(rather than a separate equation having to be used for each site).
The model developed to predict SSC from three explanatory variables (turbidity,
water temperature and specific conductivity; flow removed from the model) explained about
58% of the variation in SSC. In this model, site was significant and it was necessary to
develop a separate predictive equation for each site. The model with flow as the only
explanatory variable resulted in a multiple R2 value of 0.39, but the statistical properties of
this model were not good. There was no statistical relationship between SSC and water
temperature, so a model could not be built with temperature as the only explanatory variable.
Executive Summary continued on website or attachment below.