A data acquisition framework for runoff prediction in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 48

Figure 3.17. Cross-section of the
Chicken Creek catchment. From
Bormann et al.( 2011b ).

Watershed

Sandy cover

Clay base

Clay barrier

Depression

Outlet

Figure 3.18. Differences in
frequency distributions of the
different models during a-priori
simulation, the first prediction for the
Chicken Creek. From Holländer
et al.( 2009 ).

100

75

50

Catflow
SIMULAT

CMF CoupModel

SWAT Topmodel

25

Hill-Vi
WaSiM-ETH

Hydrus-2D

NetThales

0

1 10 100 1000

Runoff (m 3 /d)

variables (e.g., soil moisture, groundwater tables) were
withheld as well in the first modelling steps. Therefore,
model predictions were done a priori, without guidance
from gauged data, although the catchment is intensely
monitored with respect to water and matter dynamics as
well as system characteristics (Gerwin et al., 2009 ).
The comparison study is broken up into four different
modelling stages, reflecting the hierarchy of data acquisi-
tion discussed in the rest of this chapter: (i) A-priori mod-
elling based only on data about soil texture, soil thickness,
clay layer, topography, vegetation cover, hourly climate
data, air photography and initial groundwater levels. Mod-
ellers were not allowed to visit the catchment. (ii) A walk
through the catchment, after which the modellers discussed
and compared their a-priori model simulations as a group.
(iii) Additional observations including soil hydraulics, soil
physical data, soil water content and infiltration capacity.
(iv) Runoff observations (for calibration) from a subcatch-
ment (1.8 ha). Three of these modelling steps have so far

been executed and are summarised briefly below. The
models applied encompass different modelling philoso-
phies, ranging from one-dimensional to three-dimensional
models regarding their spatial representation. Most of the
models describe the hydrological processes in a physical
way, whereas only a few models are based on a lumped,
conceptual concept. Eight of the twelve models describe
the unsaturated soil water flow by the Richards equation,
and ten models use the Penman

-

Monteith approach to

calculate potential evaporation.
The results of the study showed a very large spread of
model results (runoff at the outlet) across the models after
stage one ( Figure 3.18 ). Models simulated from 10% to
330% of observed mean annual runoff. According to
Holländer et al.( 2009 ), those differences could mainly be
attributed to differences in model parameterisation and
conceptualisation. Unknown initial conditions in terms of
soil moisture content were another important issue to be
tackled by the modellers. ' Runoff was mainly predicted as