Environmental Engineering Reference
In-Depth Information
time series without the need for interpolation or
estimation. Guidelines based on this approach for
computing SSC values from continuous turbidity
data (or, when appropriate, continuous turbidity and
streamfl ow data) have been produced by Rasmussen
et al.
(2009) and endorsed for collecting and storing
SSC and SSL data by the USGS.
The turbidity-based computational scheme has
several benefi ts:
•
no subjective interpolation or estimation is
required, although the hydrologic judgment and sta-
tistical prowess of the analyst may be important in
the derivation of the equation used to convert turbid-
ity, or turbidity and water discharge, to SSCs;
•
the computational procedure is precisely
reproducible;
•
the scheme takes full advantage of the available
data and computational resources, hence, substan-
tially reduces the time and effort to compute SSL
records;
•
estimates of uncertainty can be computed for the
SSC time series.
An adequate model calibration dataset consists of
an appropriate number of instantaneous SSC samples
and concurrent turbidity and streamfl ow measure-
ments made over most of the observed range of
hydrologic conditions for the period of record.
Another factor that should be considered when
determining the adequacy of the number of samples
in a calibration dataset is the amount of variability
in the relation between turbidity and SSC. The larger
the variability in the relation between turbidity and
SSC at a site, the greater the need to collect more
calibration data.
The key factor for computing time series of SSC
data from periodic instantaneous SSC, time series of
turbidity, and streamfl ow data is the type and good-
ness-of-fi t of the regression model used in the com-
putation. A simple linear regression model relating
turbidity to SSC is often suffi cient for reliable com-
putations of SSC. A multiple linear regression model
relating both turbidity and streamfl ow to SSC may
signifi cantly improve the usefulness of the simple
turbidity linear regression model. Typically, addition
of a streamfl ow variable is more likely to improve
the turbidity-SSC regression if more than about 20%
of the suspended-sediment mass is sand-size material
(between 62 and 2000
inferred from research by Gray
et al
. (2000) on dif-
ferences between SSCs and total suspended solids
measurements.
Prediction intervals are determined to evaluate the
uncertainty of SSC regression-computed values
(Helsel & Hirsch 2002). Prediction intervals defi ne
a range of values for the regression estimate associ-
ated with a known level of uncertainty. For a given
turbidity value, the 90% prediction interval repre-
sents a range of values within which there is a 90%
certainty that the true SSC value lies.
Once an acceptable regression model is developed,
it can be used to compute SSC within and outside of
the period of record used in model development.
Maintaining a long-term SSC record requires ongoing
collection of turbidity and streamfl ow time-series
data and sample collection for reanalysis and verifi -
cation of the current SSC regression model. The
method for validating the regression model is affected
by the frequency of sample collection and the purpose
of the study. Regression models can be validated
annually (or at some other frequency as needed
based on the nature of the monitored hydrologic
system and its watershed), after new data have been
collected, or on the basis of other valid criteria.
Owing to variability in hydrology and other factors,
one such period may experience an extreme condi-
tion compared with another, such as in fl oods or
droughts, urbanization, wildfi re, or implementation
of best-management practices. Ergo, a regression
model to compute SSC should never be considered
static, but rather to represent a set period in a
dynamic system in which additional data will help
verify changes in the SSC regression relation.
1.2.1.2 Example fi eld evaluations
Continuous turbidity measurements have been
shown to provide reliable continuous SSC values
with a quantifi able uncertainty at the USGS stream-
gage on the Little Arkansas River at Sedgwick,
Kansas, USA. The adequacy of the calibration dataset
was evaluated using duration curves of turbidity and
streamfl ow (Fig. 1.7). The number of samples is
often cited as the primary criterion for determining
if a dataset is adequate. Although the sample total is
important, their broad distribution over the range of
μ
m median diameter), as
