Geography Reference
In-Depth Information
Figure 12.6. (a) Dependence of cross-validation performance of runoff prediction methods in ungauged basins on catchment area (Level 1 and
Level 2 assessments from Chapters 5 - 10). (b) Schematic of how performance decreases with decreasing catchment scale (relative to large
catchments).
how well we can predict runoff in ungauged basins that go
beyond individual case studies.
Results from the assessment show that prediction per-
formance increases with size of catchment for almost all
signatures ( Figure 12.6a ). Two possible reasons can be
attributed for this. First, larger catchments tend to produce
smoother responses because, with increased area, the stor-
age tends to be larger and causes an attenuation of small-
scale variability, thereby increasing predictability. Second,
larger catchments tend to have more observations that help
to condition the predictions.
The degree to which performance depends on area
differs between signatures. Floods, low flows and hydro-
graphs show the strongest dependence. Area is important
for floods because of the relatively short spatial scales of
flood generating processes (depending on the event type),
which affects the magnitude and timing of floods. The
more extreme a flood, the stronger the attenuation with
catchment scale because of the averaging of extremes
(Sivapalan and Blöschl, 1998 ). The dependence is some-
what less for low flows because these tend to be large-scale
processes of longer duration and have longer flow paths
( Chapter 4 ). Because of the space
Spacing of data and size of region with respect to natural
variability
As mentioned above, the size of a catchment may impact
predictive performance through the availability of data. In
the context of regionalising runoff signatures, a key issue
is how many stream gauges are available for estimating
the runoff signatures in ungauged basins. The dependence
of cross-validation performance on number of stream
gauges per study from the L1 analyses is shown in Figure
12.7 . For most signatures there is a positive dependence,
i.e., performance increases with increasing number of
stream gauges. This is not surprising, as more robust
regionalisation estimates can be obtained with larger
numbers of stations. An exception is the runoff hydro-
graph signature. Here, it should be noted that (in the
assessment) hydrographs were estimated by process-based
methods (using rain gauge data), while the other signa-
tures were computed mainly using statistical methods.
This may explain why the performance of runoff hydro-
graph estimates in ungauged basins is more dependent on
number and quality of rain gauges (conditioned, of
course, on the suitability of the model structure) than on
the number of stream gauges in the region. Also, in the
case of predicting hydrographs, the number of stream
gauges was more closely related to the overall size of
the study region than to the stream gauge density. One
would expect that the performance of predicting hydro-
graphs in ungauged basins will actually increase with
increasing stream gauge density. This is illustrated in
Figure 10.25 for a French case study. The relationship
in Figure 12.7 encompasses all climates and catchment
characteristics. The data of the assessment were not
detailed enough to be stratified by climate/catchment
time connections of
runoff processes, both spatial and temporal scales are
reflected in the areal dependence. Area is less important
for annual runoff, seasonal runoff and FDCs because
they relate to larger time scales of aggregation. The
areal dependence of runoff prediction performance is
therefore strongly dependent on the runoff signature exam-
ined. The shorter the time scale of a signature, the smaller
is the attenuating role of catchment storage and the
stronger the dependence. This is illustrated schematically
in Figure 12.6b .
-
Search WWH ::




Custom Search