Geoscience Reference
In-Depth Information
failed completely to predict a later “validation” storm. In fact, the volume of rainfall recorded by the
raingauges was far less than the volume of discharge recorded in the stream (there was a line of rain-
gauges at different elevations on one side of the catchment, but the intense storm was centred on the
other side). It is difficult in such circumstances for any model to predict the response accurately. In-
adequate estimation of the rainfall inputs to a catchment must therefore increase the uncertainty of
runoff predictions.
A number of models reported in the literature have included a rainfall multiplier as a parameter to
be calibrated as one way of trying to allow for the fact that the raingauge data available might not be
a good characterisation of the rainfall inputs to a catchment area. It is not clear that this is generally a
good strategy in that it may be only some events for which the rainfall inputs are not well estimated. For
extreme events, as in the White Oak Run example above, it may be quite obvious that there is a problem.
For more moderate events, it may be suspected that some events are not well estimated, but it will not be
clear which events might be problematic. Thus, a constant rainfall multiplier would not be an appropriate
way of adjusting the catchment inputs. It would, then, be better to implement some quality controls (see
Section 3.2) and, if necessary, exclude some periods of data from the modelling exercise as unreliable or
disinformative
(see Beven and Westerberg, 2011).
However, there may be cases when an adjustment might be justified. It is a fairly common situation, for
example in mountainous terrain, that one or more raingauges might be available in the valley bottoms, but
none in the higher elevations where it is expected that precipitation inputs might be greater. The average
catchment rainfall input might then be consistently higher than that recorded by the valley bottom gauges.
Some adjustment would be necessary to achieve a reasonable water balance. Even in such a case, however,
calibration of a rainfall multiplier might not be the best solution, since there would be a distinct possibility
that the calibration process would result in interaction with other parameters being calibrated at the same
time. This is definitely the case when event by event rainfall multipliers are used to try to correct for
input errors (see Section 7.8). Thus, it might be better to make a prior adjustment on the basis of physical
reasoning rather than allowing the multiplier to vary during calibration. This would also allow for the
possibility of making different adjustments in different periods, although it will be rare that there would
be sufficient information on which to base such a variable adjustment. However, calibration of any
other parameter values would still be conditional on the adjusted inputs. The general case, as noted in
Section 1.8, is that any calibrated parameter values must be conditional on the sequence of inputs used,
even if no such adjustments are made.
The estimation of precipitation inputs in the form of snow raises a whole range of additional problems.
The hydrologist is interested in the water equivalent of the snow, which depends on both its depth and
profile of density, both of which change over time as the snowpack structure evolves and ripens. Snow
water equivalent may be measured directly at a point or on transects known as
snow courses
by field
measurements of snow depths and density profiles but this can be arduous and expensive to maintain
at frequent intervals. The best continuous measurements method available is to measure the weight of
snow above a point using a pressure measurement device, such as a snow pillow. An increase in the
pressure indicates a new fall of snow; a decrease, loss by
sublimation
or, more importantly, melting. The
continuous measurement of pressure can then give a good indication of the rates of melt that are required
for hydrograph modelling.
Unfortunately, such installations are expensive and remain relatively rare. They also, like a raingauge,
only give an indication of conditions at a single point and snow packs are renowned for their variability
both in terms of water equivalent and rates of melt, particularly in mountainous terrain and where
vegetation extends above the pack. The redistribution of snow by wind; the effects of topography and
vegetation on snow collection, temperature and insolation conditions; freeze-thaw cycles; and changing
pack albedos over time are all factors that affect this variability and make modelling snowmelt very
difficult indeed (see, for example, the study by Bathurst and Cooley, 1996). This is one area of hydrology
where remote sensing has proven especially useful (see Section 3.7).
