Geography Reference
In-Depth Information
Figure 11.21. Google Earth images of the upper parts of the basin in Lesotho (P1), the middle parts of the basin on the border between Lesotho
and South Africa (P2), and a reservoir on one of the tributaries in the lower part of the basin (P3).
slopes (the north-western edge of the Drakensberg Moun-
tains) with grassland vegetation. Land use consists of
extensive rain-fed cultivation and cattle grazing (mostly
subsistence agriculture) on the valley sides and bottoms.
The topography in most of the South African parts of the
basin is undulating, while land use is based on intensive
cultivation with a mixture of rain-fed (mostly maize) and
irrigated crops together with some cattle grazing. The
majority of the basin is underlain by sandstones and shales,
while the south-western parts of the basin are underlain by
shales and mudstones. Soil characteristics are highly vari-
able both in terms of depth and texture. Mean annual
precipitation varies from over 1000 mm/yr in the Drakens-
berg Mountains to less than 600 mm/yr in the lower parts
of the basin. Potential evaporation ranges from less than
1300 mm/yr in the headwaters to 1600 mm/yr downstream.
The rainfall regime is highly seasonal, with approximately
70% of the rain falling between November and March.
uncertainty, and is based on the use of estimation equations
using physical sub-basin properties (topography, soils, geol-
ogy and vegetation) available from various sources (e.g.,
AGIS,
2007
). The parameter uncertainty is estimated from
the spatial variation in the physical sub-basin properties and
expressed as means and standard deviations of normal distri-
butions or maximum and minimum values of uniform distri-
combined approach has been used. The Kapangaziwiri
et al. (2009) approach has been applied to selected represen-
tative sub-basins (steep headwater areas of the north-east, less
steep headwaters in the north-west and flatter downstream
sub-basins) to establish the likely ranges of parameter values
and their uncertainty distributions across the whole basin.
The Midgley et al.(
1994
) parameter sets were used as a guide
to extrapolating from the sample sub-basins to establish
uncertain parameter sets for the whole basin.
The uncertainty version of the model (Hughes et al.,
2010
) generates output ensembles (typically 10 000) based
on Monte Carlo sampling of the parameter distributions
independently for all 31 sub-basins. The original approach
was based on unstructured (or unconstrained) sampling,
but this resulted in highly reduced uncertainty in the down-
stream reaches compared to the upstream sub-basins. This
reduced uncertainty is associated with the total possible
parameter space being far greater than the number of
ensembles and the fact that each of the downstream ensem-
ble results is generated from a mix of upstream parameter
effects. A revised approach involves structured sampling to
ensure that more of the extreme parameter uncertainty
effects are propagated downstream.
Method
The hydrological model used in this study is the Pitman
monthly model (see also Sections 6.4.2 and 7.4.2 for other
examples) with revised surface
groundwater interaction rou-
tines (Hughes,
2004
; Hughes et al.,
2006
). The model has
been widely used in the Southern Africa region for practical
water resource assessments, traditionally based on parameter
sets that were established through calibration at a limited
number of gauging sites, followed by regionalisation using
a relatively subjective approach based on perceived basin
similarity. Midgley et al.(
1994
) provide parameter values
for the 1946 sub-basins covering the whole of South Africa,
Lesotho and Swaziland. Of these, 31 sub-basins form the
Caledon River study area (see
Figure 11.20
). Kapangaziwiri
and Hughes (
2008
)and
Kapangaziwiri et al.(2009)
report on
an alternative approach to parameter estimation for the
Pitman model that does not rely on calibration, includes
-
Results
Figure 11.22
presents the results as 1-month flow duration
curves for the lower and upper bounds encompassing 90%
of all the simulated ensembles. The only part of the model
Search WWH ::
Custom Search