Geoscience Reference
In-Depth Information
point measurements. A number of techniques are available for such spatial integration: simple averaging,
Thiessen polygons and inverse distance weighting and a variety of others (see, for example, Shaw
et al.
,
2010). None of these techniques can be more than an approximation to the actual volume of rainfall over
the catchment and accuracy of a particular technique is likely to change from storm to storm.
The development of radar rainfall measurement has led to a much greater appreciation of the temporal
and spatial variation of rainfall intensities than was previously available from raingauge measurements
alone. Much of Europe and large areas of the USA are now routinely monitored by ground-based radar
rainfalls. The radar has a revolving antenna that sends regular electromagnetic pulses at a low upward angle
into the atmosphere. A detector measures the strength (and, in some cases, the frequency attenuation) of
the return signal. The principle is that the return signal to the radar is strongly dependent on the intensity
of the precipitation in the path of the radar beam at different distances from the measurement site. A
calibration function then allows the intensity of rainfall at each distance to be estimated; the estimated
intensities are then normally interpolated onto a square grid, commonly with a resolution of 2 or 4 km
for operational radars.
This would appear to be a very important development in the data available for rainfall-runoff mod-
elling, and indeed it is, but there are some important limitations that must be recognised. The first is
that the radar does not measure the rainfall at ground level but at some distance above the ground (often
hundreds of metres and increasing away from the radar station). There is thus potential for changes in
the patterns of intensities at ground level, particularly where winds are strong and where there is a strong
orographic effect. Secondly, the calibration of the radar depends not only on rainfall intensity but also
on the type of precipitation, particularly the drop size distribution and whether the rain is all liquid water
or a mixture of water and ice (which produce very different return signals). Thus it may be necessary to
“correct” the basic calibration of the radar in different ways. This is most often done by continuously
adjusting the estimates of intensity produced by the radar using online data from recording raingauges at
the ground surface. In some sense then, the radar becomes an expensive (but very effective) spatial inter-
polation technique. It is effective because it can give an indication of cells of high rainfall intensity that
might be completely missed by the network of ground-level gauges. This may be particularly important
in high magnitude, localised rainfall events (such as that of Figure 3.1; from Smith
et al.
(1996) for an
example of a high-magnitude event in Virginia).
Direct measurement of drop size distributions using disdrometers has also been used to try to improve
calibration (but is subject to the limitation that the size distributions measured at ground level might not
be the same as those at the height of the radar beam, even allowing for wind drift effects). It might also
be possible to implement local radar systems with overlapping coverage of a catchment to resolve some
of the problems with the existing network of weather radars making use of cheaper X-band radars (see,
for example, Anagnostou
et al.
2004; Matrosov
et al.
2010; van de Beek
et al.
, 2010). X-band radar has
a much more limited areal coverage but can give more local detail that might be particularly useful, for
example, in forecasting runoff in urbanised catchments (Maki
et al.
, 2005).
An alternative method of obtaining spatial rainfall information is to make use of the attentuation of
existing microwave signals, such as the extensive network of mobile phone transmitter-receiver links.
This requires no additional equipment, but can only estimate rainfalls averaged over the length of a linear
link. It also requires negotiations with mobile phone companies to obtain the data. This work is still in
its early stages but might prove useful in future (see, for example, Leijnse
et al.
, 2007, 2008).
The estimation of rainfall is very important in rainfall-runoff modelling, since no model, however
well-founded in physical theory or empirically justified by past performance, can produce accurate
hydrograph predictions if the inputs to the model do not adequately characterise the rainfall inputs (the
well-established GIGO principle of Garbage In Garbage Out applies). A good example is reported by
Hornberger
et al.
(1985). In their study, a calibrated rainfall-runoff model for the 5 km
2
White Oak
Run catchment in Virginia, which had previously performed well in reproducing observed discharges,
