-
School of Civil Engineering and GeoSciences
Catchment and River Modelling CEG 8506 (Hydrosystems Modelling)
Coursework
Chinedum C. Eluwa (140538430) 7th January, 2015
This report compares the functionality and comparative
performance of event-based and
continuous models in the prediction of floods, especially as
regards model sources of uncertainty
Merits and demerits are investigated and expounded and model
structure is found to be the
highest source of uncertainty in 1-dimensional river models.
Continuous based models on the
other hand, are found to be data intensive and are susceptible
to parameter uncertainties, and in
cases of sparse data sets, may not function efficiently.
These various merits and demerits, widen risks during non-expert
model application and this
report demonstrates the importance of in-depth understanding of
model limitations, and the
competent application of such models within these limits.
EXECUTIVE SUMMARY
-
i
Contents 1. Introduction
....................................................................................................................................
1
1.1. Aims and Objectives
................................................................................................................
1
1.2. Study Area
...............................................................................................................................
1
1.2.1. Location
...........................................................................................................................
1
1.2.2. Weather
..........................................................................................................................
2
1.3. Data
.........................................................................................................................................
2
2. Transfer Function Modelling: Generating a 100 - Year Flood
Event ............................................... 2
3. NOAH 1-D: Routing a Flood
.............................................................................................................
4
3.1. Model Calibration
...................................................................................................................
5
3.2. Model Validation
.....................................................................................................................
5
3.3. Model Results
.........................................................................................................................
6
4. SHETRAN: Simulating Catchment Response to a Flood Event
........................................................ 7
4.1. Model Calibration
...................................................................................................................
7
4.2. Calibration Results
..................................................................................................................
7
5. Results Comparison and Critical Assessment of Methods
..............................................................
9
6. References
....................................................................................................................................
10
List of Tables Table 1: Important Physical Catchment Descriptors
of the Wansbeck Catchment ................................ 2
Table 2: Table of Peak Discharge and Loss Factor
..................................................................................
4
Table 3: Calibration Process (Manning's Coefficient trial
values) ...........................................................
5
Table 4: Calibration Results (Wansbeck stage at Oldgate Bridge)
and Errors ........................................ 5
Table 5: Validation Results from 2nd Alteration of Mannings
Coefficient on Validation period .......... 6
Table 6: Nash Sutcliffe and Logarithmic Nash Sutcliffe
Efficiency Values ..............................................
8
List of Figures Figure 1: Map of Morpeth, (Source: Google Maps)
................................................................................
1
Figure 2: Storm Profile for 1% AEP for Wansbeck Catchment
................................................................
3
Figure 3: Transfer Function Model for Wansbeck Catchment
................................................................
3
Figure 4: Various Flood Hydrographs on the 1% AEP rainfall event
....................................................... 4
Figure 5: Sensitivity plot of Discharge to Loss Factor
.............................................................................
4
Figure 6: Sensitivity plot of low and high river level
simulations
........................................................... 6
Figure 7: Cross Section at selected portion of Wansbeck River.
..... 6
Figure 8: Response Surface of Nash-Sutcliffe Efficiency value
...............................................................
8
Figure 9: Time series of modelled and observed flow at Mitford
.......................................................... 8
-
1
1. Introduction Relevant information is necessary to support
decision making at all levels. Therefore, hydrologic
models, through predictions and estimations, are designed to
provide valuable spatial and temporal
information on catchment or regional scale responses to specific
hydrologic events. Hydrologic models
are built to extrapolate measurements from available data
(Pechlivanidis, et al., 2011) and also
incorporate the impact of future alterations within the
hydrologic cycle at various levels. Due to their
important role in disaster mitigation, flood models must
represent (as accurately as possible) expected
discharges and consequent water levels.
1.1. Aims and Objectives This paper is aimed at understanding
and critically examining the various means to successfully
calibrate hydrologic models and realize best estimates of
desired simulations. To achieve this, data
from a recent hydrologic and hydraulic event which occurred in
Morpeth (a city in the Wansbeck
catchment) would be estimated by calibrating two different
hydrologic models A 1-dimensional river
flow model (Noah-1D) and, a physically based distributed
catchment model (SHETRAN).
Both models would be calibrated using data from the 2012 flood
event which occurred in Morpeth.
However, only the 1-D model would be used to forecast responses
to an artificial flood event. This
artificial flood event would be generated using a transfer
function model.
1.2. Study Area
1.2.1. Location Morpeth (55.1630 N, 1.6780 W) is a market town
which lies 13 miles north of Newcastle upon Tyne,
a similar distance to the south of Alnwick and 12 miles from the
North Sea, extending over both north
and south banks of the River Wansbeck where it takes a broad
southward loop (Northumberland
County Council; English Heritage, 2009). The river Wansbeck is a
major river which flows through
Morpeth. It begins at the Sweethope Loughs lakes in the southern
part of Northumberland, about
18 miles (29 km) west of Morpeth (Wikipedia, 2013). It is joined
by Swilder Burn, Hart Burn and the
Font before reaching Morpeth and flowing to the coast south of
Newbiggin-by-the-sea (Environment
Agency, 2009).
Oldgate Bridge (River level gauge)
River flow gauging station
Morpeth Mitford Wansbeck River North
0m 200m Map Scale
Figure 1: Map of Morpeth, showing the River Wansbeck and
important data source points (Source: Google Maps)
-
140538430 Hydrosystems Modelling CEG 8506
2
1.2.2. Weather The general climate in Morpeth is typical of
usual maritime climate in the United Kingdom with no
defined dry season and precipitation quite evenly dispersed
throughout the year. However, its
location in the North Eastern part of England leaves it prone to
east or NE winds on the northern flank
of depressions passing to the south of the area (Met Office,
2013). Such frontal interactions (like the
occluded fronts) in the September months of 2008 and 2012 caused
periods of intense and extended
rainfall duration. Catchment features responded to these events
and riparian areas of the Wansbeck
in Morpeth were inundated.
1.3. Data In this study, observed hydrological, geological,
geographical, hydraulic and meteorological data from
various sources (including Centre for Ecology and Hydrology;
Environment Agency; Met Office;
Newcastle University Water Resources Group) was used. Simulated
data from deterministic and
stochastic models was also used. The observed data was primarily
from records taken during years
2008 and 2012 for the Morpeth town and Wansbeck River. Observed
hydrological and hydraulic data
came from flow station at Mitford, and stage gauge at Oldgate
Bridge, both logged in 15-minute time
series. Observed meteorological data included rainfall data
(from 3 stations: Wallington, Harwood and
Font Res at 15-minute resolutions), temperature data (aggregated
into two hourly resolved sets of
maximum and minimum), and potential evaporation data (aggregated
into one dataset of hourly
resolutions). Other observed data such as cross sections of the
Wansbeck River, vegetation, aquifer
characteristics, surface elevation, and soil cover were from GIS
and geological surveys of the
catchment. The data used to generate a transfer function for the
Wansbeck catchment came from the
Centre for Ecology and Hydrologys Flood Estimation Handbook. The
transfer function model then
simulated hourly flood time series flow data which was routed
through the 1-D model.
2. Transfer Function Modelling: Generating a 100 - Year Flood
Event The objectives of this task are to design a rainfall event of
1% exceedance probability and generate
the runoff hydrograph resulting from this storm. The transfer
function model (unit hydrograph) to be
used is based on the simple triangular version in the Centre for
Ecology and Hydrologys Flood
Estimation Handbook. This simple triangular method used to
generate the Unit Hydrograph is defined
by three parameters Time to Peak (Tp), Base time (TB) and Peak
Discharge (Up), to be estimated for
the particular catchment under consideration. These parameters
were estimated using standard
formulae and physical catchment descriptors from the Flood
Estimation Handbook. Of these three
parameters, the most important parameter is the catchment time
to peak. This is because, the other
parameters are based on various manipulations of the time to
peak.
Table 1: Important Physical Catchment Descriptors of the
Wansbeck Catchment River Wansbeck AREA 286.88 Area of catchments
(square kilometres)
PROPWET 0.45 Proportion of Time with Soil Moisture above field
capacity
DPLBAR 21.85 Average drainage path length (kilometres)
DPSBAR 50.8 Average drainage path slope (metre per
kilometre)
SAAR 793 Standard Annual Average Rainfall (1961-1990)
(millimetres)
BFIHOST 0.347 Base Flow Index derived from Hydrology of Soil
Types
URBEXT1990 0.002 Extent of urban and suburban land cover (year
1990)
Important catchment descriptors which help determine the
catchment time to peak, using the
formulae below, are given in Table 1.
-
140538430 Hydrosystems Modelling CEG 8506
3
() = 1.56 1.09 0.60 (1 + 1990)
3.34 0.28;
=0.65 ()
3.6 () ; = 2.52
The unit hydrograph is based on the rainfall-runoff relationship
and therefore its convolution requires
a design storm as input. In order to ensure that the entire
catchment is contributing to runoff, the
storm duration is calculated using the formula below.
= (1 +
1000)
Using this storm duration, a combination
of Areal Reduction Factors, Seasonal
Correction Factors and Depth-Duration-
Frequency curves (Kjeldsen, 2007), the
total rainfall with a 100-year return period
was estimated. This rainfall depth was
then transformed to a dimensionless curve
(storm profile) also known as mass curve,
with cumulative fraction of time (storm
duration) and total precipitation on the
horizontal and vertical axes, respectively
(Ellouze, et al., 2009). For the Wansbeck
catchment, the storm duration was 15
hours. This yielded a storm profile shown
in Figure 2.
Usually, the peak discharge from the 100
year hyetograph may be assumed to be
the 100 year flood (Viglione & Bloschl,
2009). However, there is great uncertainty
associated with this assumption because
of the difficulty in correctly assessing the
current soil moisture conditions at the
onset and during the design storm. To
incorporate soil moisture conditions, a loss
factor is applied to the storm profile
before it is combined with the unit
hydrograph. Kjeldsen (2007) describes a
method of properly estimating and
applying this loss factor to derive effective
rainfall based partly on the likelihood that the catchment soil
moisture would be above or below field
capacity (PROPWET) at the onset of a rainfall event and
infiltration capacity during the event;
however, this is beyond the scope of this section.
For the given catchment characteristics, the time to peak for
the Wansbeck catchment was calculated
to be 8 hours. The peak discharge was calculated to be 6.52
m3/s/mm-of-effective-rain; base time was
calculated to be 19 hours. These parameters defined the transfer
function model and produced the
profile in Figure 3.
0
2
4
6
8
10
12
14
0 5 10 15
Pre
cip
itat
ion
(St
orm
) D
ep
th
(mm
)
Time (hours)
Figure 2: Storm Profile for 1% AEP for Wansbeck Catchment
0
1
2
3
4
5
6
7
0 5 10 15 20
Dis
char
ge p
er
un
it e
ffe
ctiv
e
rain
fall
(m3/s
/mm
)
Time (hours)
Figure 3: Transfer Function Model for Wansbeck Catchment
-
140538430 Hydrosystems Modelling CEG 8506
4
The combination of the design 100-year
storm, unit hydrograph convolutions,
baseflow and other catchment
characteristics, such as the antecedent soil
moisture conditions, produced the
required flood hydrograph. The variability
of effective rainfall due to soil moisture
conditions represented by various loss
factors gave a range of peak discharges,
shown in Figure 4 which all correspond to
the 1% AEP rainfall event. This range
introduces high uncertainty and makes it
difficult to determine the 100-year flood
based on the 100-year storm alone.
Table 2: Table of Peak Discharge and Loss Factor
Loss Factor
Change in Loss Factor
Peak Discharge
(m3/s)
Change in Peak
Discharge
0.3 0 107.3 0
0.4 0.1 141 33.7
0.5 0.2 176 68.7
0.6 0.3 210 102.7
0.7 0.4 244 136.7
0.8 0.5 278 170.7
0.9 0.6 312 204.7
1 0.7 346 238.7
The range of possible peak discharges
from application of loss factors from 0.3 to
1 is 238.7 (m3/s), as can be seen from
Figure 4. This is a highly significant value
that reflects the level of uncertainty
involved in selecting a 100 year design flood based on
rainfall.
It is also evident from Figure 5 that the value of peak
discharge is highly sensitive to changes in the
loss factor with a regression slope of 341. This means that a
change in loss factor of 0.1 accounts for
a change in discharge by about 34 (m3/s). This sensitivity
represents the magnitude of the likely error
made in discharge estimation if the wrong loss factor is
used.
3. NOAH 1-D: Routing a Flood In the event of a storm, generated
runoff from all areas of the catchment would travel towards the
main drainage channel. If the generated flood supersedes the
volumetric capacity (design discharge)
of the channel, one of two things occur depending on whether the
channel is open or closed. In an
open channel, water levels rise above channel edges (banks) and
surrounding areas are inundated.
Otherwise, excess storm water backs up through closed channels
(pipes) and out into drains. This risk
of fluvial or pluvial flooding depends on the intensity of the
storm generating runoff. Fluvial flooding
is the focus of this paper.
The previous chapter generated a storm over the Wansbeck
catchment and consequent hydrograph.
This chapter will present the modelled response of the Wansbeck
River to the design flood. The
0
50
100
150
200
250
300
0 4 8 12 16 20 24 28 32 36
Dis
char
ge (
m3/s
)
Time (Hours)
Loss Factor = 0.7
Loss Factor = 0.3
Loss Factor = 0.5
Loss Factor = 0.58
Figure 4: Various Flood Hydrographs corresponding various loss
factors on the 1% AEP rainfall event
y = 341.18x + 0.075
0
50
100
150
200
250
300
0 0.2 0.4 0.6 0.8
Ch
ange
of
Dis
char
ge r
elat
ive
to a
lo
ss f
acto
r o
f 0
.3
Change of Loss Factor relative to 0.3
Peak DischargeVariation
Linear (PeakDischargeVariation)
Figure 5: Sensitivity plot of Discharge to Loss Factor
-
140538430 Hydrosystems Modelling CEG 8506
5
accuracy of the modelled response is dependent on the ability of
the model to reflect actual behaviour
of the river and channel. Model parameters were altered until a
suitable fit was found.
3.1. Model Calibration The model used in this section was the
NOAH 1-D river routing model. Being a 1-D hydraulic model, it
assumes flow as predominantly one dimensional (in the
stream-wise direction) and follows the St.
Venant equations. In the model, estimation of Mannings roughness
coefficient is very important to
the simulation of open channel flows because the coefficient
includes the components of surface
friction resistance, form resistance, wave resistance and
resistance due to flow unsteadiness (Ding &
Wang, 2004). Because the 1-D model aggregates flow velocity over
the cross-sectional area, there is
the underlying assumption that stream flow is normal to the
cross section. This means that the
geometry of individual cross sectional areas are quite important
to the model.
After model configuration using data from GIS surveys of cross
sectional area, initial conditions were
included. Initial conditions require constant inflow over the
inflow cross section, this is due to the
model representation of the St Venant momentum equation
(University of Technology Hamburg,
2010). This means that the model starts up with an empty channel
and constantly fills until it matches
the set initial conditions at the particular time step. An
eleven-day period in the 2012 flow time series
(20/09/2012 30/09/2012) was selected, to calibrate the model.
The flow hydrograph of this period
was used as initial conditions of the model. The model
calibration took place through a repeated
procedure of trial and error involving visual comparisons
between field measurements and
simulations of water level (at Oldgate Bridge) whilst changing
the roughness coefficient (Table 3). Few
number of trials are due to familiarity with the study area
through online maps and tutorials on the
model.
Table 3: Calibration Process (Manning's Coefficient trial
values)
Mannings Coefficient
Calibration Steps River Bed Riparian Areas Special Features
(Weirs, stepping stones etc.)
Model (as configured) 0.03 0.03 0.03
1st Alteration 0.03 0.05 0.05
2nd Alteration 0.03 0.05 0.015
Table 4: Calibration Results (Wansbeck stage at Oldgate Bridge)
and Errors
The adopted calibration (2nd Alteration) was the trial which
resulted in the least combined error (Table
4) averaged over high and low flows.
3.2. Model Validation When the proper Manning coefficients
resulting in satisfactory error margins (considered acceptable
due to other model constraints) were established, the model was
validated using a different period
(also of high and low flows) of the same 2012 flow dataset
(28/06/2012 07/07/2012). Similar results
were observed for the validation period (Table 5). This verified
the acceptability of the model to be
used for design purposes.
Observed
River Stage Model (as
configured) Error (%)
1st Alteration
Error (%)
2nd Alteration
Error (%)
Average of High Flows
25.72 25.35 -1.45% 25.27 -1.74% 25.34 -1.49%
Peak Flow 27.12 26.65 -1.75% 26.54 -2.15% 26.64 -1.76%
Average of Low Flows
24.47 24.08 -1.60% 24.10 -1.50% 24.11 -1.47%
Lowest Recorded Flow
24.33 24.04 -1.21% 24.07 -1.08% 24.08 -1.04%
-
140538430 Hydrosystems Modelling CEG 8506
6
Table 5: Validation Results from 2nd Alteration of Mannings
Coefficient on Validation period
Observed River Stage 2nd Alteration Error (%)
Average of High Flows 25.34 24.98 -1.44%
Peak Flow 25.80 25.48 -1.24%
Average of Low Flows 24.46 24.11 -1.43%
Lowest Flow 24.37 24.08 -1.17%
3.3. Model Results The hydrograph generated from the
100-year
storm event (with antecedent conditions of
0.7) was then set up as model initial conditions,
and routed through the calibrated river
channel. Model watches were set up at the
Oldgate Bridge to log the time at exceedance
of average low flow levels and maximum levels.
Map views and photographs of sections of the
river (from Google Maps) corresponding to
survey chainages (as shown in Figure 7) were
used to determine river banks for the model
watch.
The design flood from the 100-year storm caused an overflow in
the river banks after about 5 hours,
53 minutes. The maximum flood level recorded at Oldgate Bridge
(26.67m about 2.29m above
average low flows) occurred 15 hours 23 minutes into the event.
This peak flood with river levels
increased 2.29 metres above average low flow was less than the
recorded levels from the peak flood
in 2012 as observed from dataset (2.65m above low flows). This
may be used to roughly validate the
assumed approximation for catchment antecedent conditions,
because the inundation from the 100
Figure 7: Cross Section at selected portion of Wansbeck River.
Reach 5 WANS05_1147 Top: Google Map image of the section
immediately downstream of Road bridge A192; red line showing
approximated riparian inundation line during design flood event
(loss factor 0.7). Bottom: NOAH 1-D Model section diagram of same
section with water levels (blue line) at low flow; purple line
showing maximum water levels during design flood event.
0.25
0.30
0.35
0.40
0.45
0.50
0 0.0025 0.005 0.0075 0.01 0.0125
Dif
fere
nce
be
twe
en
ob
serv
ed
an
d s
imu
late
d le
vels
(m
)
Change in Manning's Coefficient (relative to default value
0.3)
High Flows
Low Flows
Figure 6: Sensitivity plot of low and high river level
simulations against changes in Manning's coefficient during
calibration
-
140538430 Hydrosystems Modelling CEG 8506
7
year event supersedes that caused by 2012 flood event (estimated
as a 1 in 16 year event) (Carlyon,
et al., 2013).
Sensitivity tests of the model show (Figure 6) steeper curve
gradients on visual inspection for the high
flows. This means that the model sensitivity to the Mannings
roughness coefficient varies for low and
high flows. This discrepancy shows that the prediction of the
low flows may depend on some other
parameter which either is not included in the model, or is not
adequately represented or
compensated. This highlights one area of uncertainty in the
model as well as a general limitation of
the application of 1-D models on complex flows. Both of these
would will be covered more in chapter
5.
4. SHETRAN: Simulating Catchment Response to a Flood Event In
the previous chapter, a specific event was generated and used to
predict river levels during the
event. Most of the uncertainty of that method arose from the
initial soil moisture due to antecedent
conditions at the start time of the model. Continuous catchment
simulation seeks to minimize this
uncertainty by initializing an entire time series and constantly
updating antecedent conditions during
simulation.
This chapter will focus on calibrating a physically based
distributed model for continuous simulation
of the catchment in focus. The model used here is the SHETRAN
model, set up on a 1km square grid
for use with the yearlong 2012 flow data series from the Mitford
gauge. For this calibration only one
aspect of the catchment response (discharge within the major
drainage channel at the Mitford flow
gauge) would be optimized. Usually, in physically distributed
models, the number of parameters
contained in the model and potentially subjected to calibration
is huge and increases with model
complexity. According to the principle of parsimony (Hill,
1998), the calibration problem is better
posed if its dimensionality is reduced whilst retaining
satisfactory results (Blasone, et al., 2007).
4.1. Model Calibration Prior to calibration the current model
was set up with initial conditions to simplify computations.
Two
parameters (hydraulic conductivity and Stricklers surface
roughness coefficient) were identified as
particularly influential on model simulations. These were
singled out for the calibration process. Other
parameters in the model are soil type, soil depth, and residual
and saturated water content values.
Data required by the model was mainly from GIS survey of
elevations, land cover, rainfall over the
catchment, potential evaporation and temperature. A systematic
method of alteration of parameter
values was adopted. The realistic ranges of parameter values
were first established and combined in
the model to determine the practical extremities of responses.
After these extreme value sets, other
combinations were taken. For saturated hydraulic conductivity
(K), the values for sandstone, which
make up majority of the subsurface formation in the catchment
were used. In the case of surface
roughness, the entire range of Stricklers roughness coefficient
(C) was used.
Model calibration was assessed using the Nash-Sutcliffe
efficiency value because it is sensitive to
timing errors, is suitable for continuous simulation modelling
and it can be easily transformed (by the
logarithmic functions) to give more emphasis to low flows. Also
other objective functions a multi-
objective calibration was used to assess the optimum parameter
sets for better understanding.
4.2. Calibration Results In this exercise, the highest
Nash-Sutcliffe value attained was 0.75. This showed generally
satisfactory calibration, at least at an operational level (Zhang
& Savenije, 2005). This highest efficiency occurred for only
one of the parameter sets (K = 0.000026 (wet soil); C = 20 (slow
flow)) from the selected range
-
140538430 Hydrosystems Modelling CEG 8506
8
(Table 6). This case of single-valued finality rarely occurs in
reality and may have happened due to the coarse resolution of
manually selected and modelled conductivity values. Certain
parameter sets produced poor Nash-Sutcliffe values but better
values for logarithmic Nash Sutcliffe values, showing overland flow
sensitivity to aquifer conditions. Table 6: Nash Sutcliffe and
Logarithmic Nash Sutcliffe Efficiency Values
Mass balance (Inflow-Outflow-Storage) for the highest model
efficiency was calculated to be 349mm. This difference from zero
may be responsible for antecedent moisture conditions in the
catchment quite sensible for a wet soil of depth 20.4 metres (from
model input file) and model accuracy with low flows. These little
verification details available in the model demonstrate one of the
merits of continuous catchment models. It would be worthwhile to
investigate the reaction of the model to deeper soils and to verify
storage changes.
Multi-criteria calibration was used to generate a response
surface (Figure 8) for the Nash-Sutcliffe efficiency value. This
surface (within the uncertainty due to resolution of selected
values) shows that the model simulates peak catchment contribution
to river flows better at low values of hydraulic conductivity. The
model predictive efficiency is relatively less sensitive to
Stricklers surface roughness coefficient. This can be seen in the
steeper
Saturated Hydraulic Conductivity (K) m/day
Strickler's Roughness Coefficient (C)
20 40 50 60 80
NSE Ln (NSE) NSE Ln (NSE) NSE Ln (NSE) NSE Ln (NSE) NSE Ln
(NSE)
0.52 0.07 0.473 0.08 0.48 0.08 0.48 0.08 0.48 0.08 0.473
0.26 0.02 0.35 0.02 0.354 0.03 0.353 0.03 0.353 0.03 0.35
0.026 -0.03 -0.055 -0.03 -0.066 -0.03 -0.066 -0.02 -0.067 -0.03
-0.059
0.0026 0.01 -0.05 0.01 -0.049 0.01 -0.050 0.01 -0.051 0.01
-0.05
0.00026 0.60 0.299 0.63 0.339 0.62 0.335 0.62 0.332 0.60
0.299
0.000026 0.75 0.598 0.73 0.588 0.68 0.573 0.69 0.57 0.68
0.559
0
20
40
60
80
100
120
140
160
180
1
11
21
31
41
51
61
71
81
91
10
1
11
1
12
1
13
1
14
1
15
1
16
1
17
1
18
1
19
1
20
1
21
1
22
1
23
1
24
1
25
1
26
1
27
1
28
1
29
1
30
1
31
1
32
1
33
1
34
1
35
1
36
1
Dis
char
ge (
m3/s
)
Time (days)
Observed Data
Modelled Data (N&SE = 0.75)
Figure 9: Time series of modelled and observed flow at
Mitford
20
50
80
-0.100.000.100.200.300.400.500.600.700.80
NA
SH-S
UTC
LIFF
E EF
FIC
IEN
CY
VA
LUE
-0.10-0.00 0.00-0.10 0.10-0.20 0.20-0.30 0.30-0.400.40-0.50
0.50-0.60 0.60-0.70 0.70-0.80
Figure 8: Response Surface of Nash-Sutcliffe Efficiency value to
calibration parameters (Hydraulic Conductivity and Surface
roughness)
K (m/day)
(C) Stricklers Roughness
-
140538430 Hydrosystems Modelling CEG 8506
9
gradient along the hydraulic conductivity axis on the response
surface plot.
5. Results Analysis and Critical Assessment of Methods In the
generation of storm profiles, errors accumulated from frequency
statistics of extreme rainfall
contained in rainfall depth-duration-frequency curves (Overeem,
et al., 2008). The effects of these
errors on model are however reduced on appropriate application
to the specific catchment. Because
this sort of modelling is based on a particular event, not an
entire time series, its greatest uncertainty
is due to initial conditions. The loss factor applied in this
method was a simplistic aggregation which
tried to account for antecedent soil moisture conditions. It
would be worth investigating the
differences when hydrograph results are compared to the more
robust initial conditions method as
stated in the revitalized flood handbook (Kjeldsen, 2007).
Another drawback of this method is that it
transfers its uncertainty to any other model which depends on
its output.
Predicting flood levels from NOAH 1-D was associated with
uncertainty from input hydrograph, model
structure and parameters. The model was configured using
physical survey data which may also
contain errors. More importantly, the model was limited in the
simulation of certain natural hydraulic
scenarios as are typical of the constantly meandering reaches
used in this study. Curved reaches were
assumed straight, therefore water levels over riparian regions
enclosed within meanders are not as
realistic as expected. The model was also unable to accurately
simulate hydraulic behaviour around
and above obstacles (such as weirs) during low flows (which do
not drown such obstacles), especially
when these obstacles are not exactly perpendicular to the flow,
thus reducing the reliability of
predictions at low flows. Also, without physical site
inspection, or additional data, modellers cannot
define where river banks begin from surveyed data alone. This
creates uncertainties in the
reconstruction of flood level progress in time and in the
application of results. Also, it is quite difficult
(without adequate knowledge of the varied features and surfaces
within and around the river banks)
to apply correct parameter values for Mannings coefficient.
These limitations impose uncertainties
(especially underestimations) in the model prediction of flood
water levels. Nevertheless, the model
performance, with a few corrections for the systematic errors
and proper understanding of its
limitations, was simple enough and sufficient to understand
catchment behaviour during high flood
conditions. Results were sufficient for the level of detail and
uncertainty involved. For low flows, other
higher level models are more appropriate.
As a distributed model, the large data requirements of SHETRAN
expose it to wide uncertainty margins
through error propagation. A major advantage of the physically
distributed model is the continuous
simulation which constantly updates initial conditions to
current model status. The most uncertainty
was associated with the choice of parameters to collectively and
concurrently enhance optimization.
The problem of equifinalty did not occur in this study probably
because the model was significantly
simplified and only a few combinations were tried. The model
assumed parameter calibration values
for the entire catchment. This is not exactly so in reality, as
such values vary spatially. Another
drawback was that the continuous model required about six months
of data (half the dataset) to
stabilize during the spin up period after initialization. For
this sort of data intensity, a split sample
calibration would yield abysmal results. It would be worth
investigating methods and algorithms to
incorporate spatial variation in parameterization (to more
realistically represent catchment variables)
and also to reduce spin up periods in models for application to
sites with short data series. This lack
of data also affected the Nash-Sutchliffe Efficiency as it was
calculated using the entire hydrograph
series including spin up times. The model timings were
appropriate, but observed discharges were
underestimated, by about 22% (linear regression () = . () + . ;
R2 = 0.83)
even at a Nash-Sutcliffe Value of 0.75 and index of agreement
(as defined by Krause, et al. (2005)) of
0.95.
-
140538430 Hydrosystems Modelling CEG 8506
10
In conclusion, both models, performed satisfactorily within
their limitations. However, the continuous
catchment simulation model performed better during low flows.
Large data requirements are the
major drawback of physically distributed models, however, their
ability to simulate entire catchment
responses over long periods complements the simplicity of river
routing models. After ensuring data
quality, reducing uncertainty in model predictions depends
finally on the competence of the modeller
and familiarity with the limitations of the modelling tools.
6. References Blasone, R. S., Madsen, H. & Rosbjerg, D.,
2007. Parameter estimation in distributed hydrological
modelling:
comparison of global and local optimisation techniques. Nordic
Hydrology, 38(4), pp. 451-476.
Carlyon, H., Hitching, J. & McNeill, A., 2013. Flood
Investigation Report: Investigation of the summer 2012 Floods,
Morpeth: Northumberland County Council.
Ding, Y. & Wang, S. S., 2004. Identification of Mannings
Roughness Coefficients in Channel Network Using Adjoint Analysis.
International Journal of Computational Fluid Dynamics, 00(0), pp.
1-11.
Ellouze, M., Abida, H. & Safi, R., 2009. A triangular model
for the generation of synthetic hyetographs. Hydrological Sciences
Journal, 2(54), pp. 287-299.
Environment Agency, 2009. Wansbeck and Blyth Catchment Flood
Management Plan, Leeds: Environment Agency.
Hill, M. C., 1998. METHODS AND GUIDELINES FOR EFFECTIVE MODEL
CALIBRATION, Denver: U.S. Geological Survey .
Kjeldsen, T. R., 2007. Flood Estimation Handbook: Supplementary
Report No. 1 (The revitalised FSR/FEH rainfall-runoff method),
Wallingford: Centre for Ecology & Hydrology.
Krause, P., Boyle, D. P. & Base, F., 2005. Comparison of
different efficiency criteria for hydrological model assessment.
Advances in Geosciences, pp. 89-97.
Met Office, 2013. North East England: climate. [Online]
Available at:
http://www.metoffice.gov.uk/climate/uk/regional-climates/ne
[Accessed 29 December 2014].
Northumberland County Council; English Heritage, 2009. Morpeth:
Northumberland Extensive Urban Survey, Morpeth: Northumberland City
Council.
Overeem, A., Buishand, A. & Holleman, I., 2008. Rainfall
depth-duration-frequency curves and their uncertainties. Journal of
Hydrology, Volume 348, pp. 124-134.
Pechlivanidis, I., Jackson, B., McIntyre, N. & Wheather, H.,
2011. CATCHMENT SCALE HYDROLOGICAL MODELLING: A REVIEW OF MODEL
TYPES, CALIBRATION APPROACHES AND UNCERTAINTY ANALYSIS METHODS IN
THE CONTEXT OF RECENT DEVELOPMENTS IN TECHNOLOGY AND APPLICATIONS.
Global NEST Journal, 13(3), pp. 193-214.
University of Technology Hamburg, 2010. 1D Hydrodynamic Models.
[Online] Available at:
http://daad.wb.tu-harburg.de/tutorial/flood-probability-assessment/hydrodynamics-of-floods/1d-hydrodynamic-models/theory/fundamentals-of-mathematical-river-flow-modelling-1d-water-level-calculation/derivation-of-the-basic-equation/
[Accessed 31 December 2014].
Viglione, A. & Bloschl, G., 2009. On the role of storm
duration in the mapping of rainfall to flood return periods.
Hydrology and Earth System Science, Issue 13, pp. 205-216.
Warmink, J. J., Klis, H. V. d., Booij, M. J. & Hulscher, S.
J. M. H., 2011. Identification and Quantification of Uncertainties
in a Hydrodynamic River Model Using Expert Opinions. Journal of
Water Resource Management, Volume 25, pp. 601-622.
Wikipedia, 2013. Sweethope Loughs. [Online] Available at:
http://en.wikipedia.org/wiki/Sweethope_Loughs [Accessed 29 December
2014].
Zhang, G. P. & Savenije, H. H. G., 2005. Rainfall-runoff
modelling in a catchment with a complex groundwater flow system:
application of the Representative Elementary Watershed (REW)
approach. Hydrology and Earth System Sciences Discussions, pp.
345-359.
1. Introduction1.1. Aims and Objectives1.2. Study Area1.2.1.
Location1.2.2. Weather
1.3. Data
2. Transfer Function Modelling: Generating a 100 - Year Flood
Event3. NOAH 1-D: Routing a Flood3.1. Model Calibration3.2. Model
Validation3.3. Model Results
4. SHETRAN: Simulating Catchment Response to a Flood Event4.1.
Model Calibration4.2. Calibration Results
5. Results Analysis and Critical Assessment of Methods6.
References