Geoscience Reference
In-Depth Information
perspective. The polls used to gauge public opinion, for example, typically sample
a few thousand out of several tens of millions of people, and so are a sample of
order 0.1%. In comparison, the rain gauge network used in the United Kingdom
has about two gauges with funnel area 0.0127 m
2
to sample an area of 100 km
2
(a 5 × 10
−12
% sample), and the network used in the United States of America has a
sampling density approximately one tenth of this.
The global challenge of providing estimates of area-average precipitation is
further exacerbated by the fact that gauge densities are very substantially less
than those just mentioned for the UK and USA. The location of available gauges
is also heavily biased, with most gauges located in wealthier, developed coun-
tries. Even in countries where there are many gauges, the gauge sample is heavily
biased toward centers of population and, because of this, to lowland sites. This is
an issue because much of the surface water used to provide water resources for
human use falls as precipitation in sparsely populated regions with significant
topography.
When seeking to evaluate the representativeness of available gauge data, typical
questions asked are, 'How representative are a set of gauges in giving average rain-
fall when there is systematic spatial variability which may be related to topography
and/or mean rainfall gradients?, or, 'What is the minimum gauge density needed
to sample spatial rainfall pattern and total water volume of a rain event?' In gen-
eral, there is a need for a denser network when seeking to determine short-term
rainfall totals. For example, in a temperate climate a study in which a 9-gauge
network was used to measure rainfall across a small, 20 m grid gave gauge-to-
gauge variations of ±5% in the measured monthly total, but ±8% in single storm
totals. Studies in the semi-arid climate of southern Arizona, where much of the
rain falls in thunderstorms, suggest that one gauge every 2.4 km is needed to pro-
vide an adequate estimate of the annual water balance for a catchment of area
25 km
2
(Sumner 1988).
When installing a gauge network to document precipitation pattern and
area-average precipitation, the optimum sampling pattern should recognize
known systematic variations. In flat regions and in urban environments, it is
generally considered best to install gauges on a rectangular grid, with a gauge
at each intersection of the grid. In regions with steep topography, which is often
the case for experimental catchments in which hydrological studies are made,
the recommended procedure is to define 'hill slope elements' on the basis
of these having similar slope and aspect, and to locate a rain gauge in the mid-
dle of each. If there is evidence of a rainfall gradient, or such a gradient might
realistically be expected (because the ground is sloping, or sampling is away
from a coast where weather systems move onshore), the preferred sampling
will be along the gradient.
One strategy that can be used is first to make an educated guess at the most
appropriate sampling strategy for the situation under study, then collect data for a
trial period and analyze the data, and then optimize the gauge arrangement, if
necessary. The correlation coefficient between all the gauges in the network for the
duration over which rainfall estimates are required may be used as a numerical
