Estimating the precision of regression-based load estimates is not easy. Bennett and Sabol [1973] considered several rating-curve-type estimators, and concluded that errors could be as low as 20%. Others have predicted errors of 100% or more [ Walling, 1977], and in some cases order-of-magnitude errors.
Long-term estimates of loads are usually approximated by sums of daily
(or higher-frequency) load estimates.
Gilroy et al. [1990] and
Gilroy [1991]
develop exact expressions for the bias and variance of such sums
for the rating-curve-based
estimators, including
, 
, 
and
.
Cohn et al. [1992a]
tested these estimators in boot-strap
[ Efron, 1982]
experiments using nutrient data from four tributaries to the Chesapeake
Bay.
Gilroy et al. [1990]
also provides simplified expressions that can be used for
sampling design.
The approximate variance of long-term load estimates is
given as a function of rating-curve parameters, the mean and variance of
for the long-term discharge record, and the mean and variance of
the discharge data used to calibrate the rating curve.