Geoscience Reference
In-Depth Information
Locally, however, it seems likely that annual totals will be
fairly consistent. It is also clear that, in the short term,
quite marked differences in rainfall may occur within
short distances, depending upon the route taken by a
particular storm or cyclone; indeed, it is possible for it to
be raining on one side of the street and dry on the other.
In order to study spatial variation on a local scale, we
need a dense network of recording gauges, for otherwise
individual storms may be missed as they pass between the
rain gauges. One such investigation was carried out in
Illinois, where fifty recording rain gauges were set up in
an area of 1,400 km
2
of flat rural land. The experiment
was maintained for five years, measuring individual
storms, and for thirteen years for monthly and seasonal
analyses. Comparisons were made by correlating rainfall
at a gauge at the centre of the area with all other gauges.
Correlation is a statistical measure which provides an
index of the strength of the linear relationship between
two variables; a value of +1·0 indicates a perfect positive
linear relationship, a value of -1·0 shows a perfect nega-
tive linear relationship, and a value of 0·0 shows no
relationship (
Figure 5.11
).
For the shortest time period studied (one minute) the
degree of correlation fell rapidly with distance from the
central gauge (
Figure 5.14
).
Thus at a distance of only 8
km from the central gauge the rainfall pattern is different
minute by minute; in many cases it may have been raining
at the central gauge but not 8 km away. This is what we
would expect if rainfall was produced by local summer
convection storms, each affecting an area of only a few
square kilometres. Even with cyclonic rainfall variability
at this time scale is normal.
N
•
0
5
km
Figure 5.12
Correlation patterns associated with one-minute
rainfall rates in warm season storms in Goose Creek, central
Illinois.
Source: After Huff and Shipp (1969)
At a longer time scale the degree of correlation is better.
Taking rainfall totals for whole storms
(
Figure 5.13a
),
it
is apparent that the gauges close to the central station are
quite strongly correlated. Nevertheless, by the distance
of about 20 km the degree of correlation is low. Again, this
is probably due to the effect of local variability caused
by the passage of small summer convection storms. If,
instead, we look at frontal storms or storms associated
with low-pressure systems, we get a different picture
(
Figure 5.13b
)
. Now most of the area shows a close
correlation with the central gauge - indeed, almost a
perfect correlation - indicating that these more wide-
spread systems affected the whole area equally, despite
occasions of variability referred to earlier.
These results indicate some of the atmospheric factors
controlling rainfall variability. Convectional storms give
high levels of spatial variation, while cyclonic rainfall is
spatially much more uniform. In the tropics, where a
great proportion of the rainfall comes from convectional
storms, the spatial variation is particularly marked. The
storms often build up without any significant movement,
so areas just beyond the limits of the cloud may receive
no rainfall at all. Sometimes the storms develop over a
wider area, perhaps 500 km
2
, but even so they do not give
rain everywhere. Using the correlation method, it has
been found that in some areas the relationship between
rain-gauge totals falls to zero within 100 km and is
negative beyond. In other words, if rainfall were high for
r = +1.0
r = -1.0
y
y
x
x
r = 0.0
y
x
tion coefficients.
