RAINFALL-RUNOFF

MODELLING TOOLBOX
USER MANUAL

Thorsten Wagener
Matthew J. Lees
Howard S. Wheater

Civil and Environmental Engineering Department

Imperial College of Science Technology and Medicine
London, SW7 2BU, UK

http://ewre.cv.ic.ac.uk/
 RRMT User Manual

CONTENTS
1 INTRODUCTION _____________________________________________________ 4
1.1 THE TARGET GROUP OF THIS MANUAL __________________________ 5
1.2 OBJECTIVES OF THE RRMT _____________________________________ 5
1.3 HARDWARE AND SOFTWARE REQUIREMENTS ___________________ 5
1.4 ASSUMPTIONS __________________________________________________ 5
1.5 BACKGROUND __________________________________________________ 6
1.6 HOW TO USE THIS MANUAL _____________________________________ 7
1.7 PUBLICATIONS _________________________________________________ 7
1.8 PROGRAM STRUCTURE _________________________________________ 8
2 USER INSTRUCTIONS ________________________________________________ 9
2.1 INSTALLATION PROCEDURE ___________________________________ 10
2.2 EXAMPLE FILES _______________________________________________ 10
2.3 STEP BY STEP MANUAL ________________________________________ 10
2.4 MAIN GRAPHICAL USER INTERFACE DESCRIPTION _____________ 12
2.5 VISUALISATION INTERFACE DESCRIPTION _____________________ 18
2.6 INPUT _________________________________________________________ 18
2.6.1 Data Array - Structure ___________________________________________ 18
2.6.2 Create a new structure input array __________________________________ 20
2.7 BATCH-FILES __________________________________________________ 20
2.8 HOW TO ADD NEW MODULES __________________________________ 21
3 AVAILABLE MODULES _____________________________________________ 24
3.1 VISUALISATION MODULES _____________________________________ 25
3.1.1 Input Data – PL_INPUT __________________________________________ 25
3.1.2 Model Output – PL_OUTPUT _____________________________________ 25
3.1.3 Actual Evapotranspiration – PL_AET _______________________________ 25
3.1.4 Confidence Limits – PL_CFL _____________________________________ 25
3.1.5 Residual Confirmation – PL_CONFIRMATION_______________________ 25
3.1.6 Double Mass – PL_DATACHECK _________________________________ 26
3.1.7 Flow Duration + Vol. Fit – PL_VFaFDC _____________________________ 26
3.1.8 Obs. Vs Calc. And Error – PL_MODELFIT __________________________ 26
3.1.9 Parameter Population – PL_MC ____________________________________ 26
3.1.10 Search Steps – PL_SCE ________________________________________ 27
3.1.11 Modelfit (Log Scale) – PL_LOGFIT ______________________________ 27
3.1.12 MCAT _____________________________________________________ 27
3.1.13 Threshold ___________________________________________________ 27
3.1.14 HMLE______________________________________________________ 27
3.2 OPTIMISATION MODULES ______________________________________ 27
3.2.1 Shuffled Complex Evolution Algorithm of the University of Arizona – SCE _ 27
3.2.2 Uniform Random Search – URS ___________________________________ 30
3.3 SOIL MOISTURE ACCOUNTING (SMA) MODULES ________________ 31
3.3.1 Catchment Wetness Index – SMA_CWI _____________________________ 31
3.3.2 Ye et al. Model Structure – SMA_YE1 ______________________________ 32
3.3.3 Penman – SMA_PEN ____________________________________________ 33

2
 RRMT User Manual

3.3.4 Catchment Moisture Deficit – SMA_CMD ___________________________ 35

3.3.5 Bucket Structure – SMA_BUC_____________________________________ 37
3.3.6 Penman Version – SMA_IC1 ______________________________________ 38
3.3.7 No Soil Moisture Accounting Module – SMA_NON ___________________ 38
3.3.8 Probability distribution of soil moisture stores – SMA_PD3 and SMA_PD4 _ 39
3.4 ROUTING (R) MODULES ________________________________________ 41
3.4.1 Conceptual Reservoir – R_CRES ___________________________________ 41
3.4.2 Two conceptual reservoirs in parallel - R_2PAR _______________________ 41
3.4.3 Three conceptual reservoirs in parallel - R_3PAR ______________________ 42
3.4.4 Leaky Aquifer Model Structure – R _LEAK __________________________ 42
3.4.5 No Routing Component – R_NORO ________________________________ 43
3.4.6 Macro-pore approach, single reservoir _______________________________ 43
3.4.7 Macro-pore approach, parallel structure ______________________________ 44
3.5 OBJECTIVE FUNCTION (OF) MODULES __________________________ 44
3.5.1 Nash Sutcliffe Efficiency – OF_1_NSE ______________________________ 45
3.5.2 Coefficient of Determination – OF_2_R2 ____________________________ 45
3.5.3 Root Mean Squared Error – OF_3_RMSE ____________________________ 45
3.5.4 Absolute Bias - OF_4_ABIAS _____________________________________ 46
3.5.5 Heteroscedastic Maximum Likelihood Estimator – OF_5_HMLE _________ 46
3.5.6 RMSE Segmentation A – OF_6_FDA, OF_7_FQA, OF_8_FSA __________ 47
3.5.7 RMSE Segmentation B – OF_9_FDB, OF_10_FQB, OF_11_FSB _________ 48
3.5.8 RMSE High Flows – OF_12_FH ___________________________________ 49
3.5.9 RMSE Low Flows – OF_13_FL____________________________________ 49
3.5.10 RMSE Medium Flows – OF_14_FM ______________________________ 49
3.5.11 RMSE Warm Up Period – OF_15_FWU___________________________ 49
3.5.12 Deviation of Runoff Volumes – OF_16_DRV_______________________ 50
3.5.13 RMSE Segmentation B – OF_17_FDBL and OF_18_FDBH ___________ 50
3.5.14 Box-Cox Transformation _______________________________________ 50
3.6 OFF-LINE (OL) FUNCTIONS _____________________________________ 50
3.6.1 OF_FLOWCONVERT ___________________________________________ 50
3.6.2 OL_SUBAVER ________________________________________________ 51
3.6.3 OL_SUBSUM__________________________________________________ 51
4 EXPERIMENTS _____________________________________________________ 52
4.1 PENMAN MODEL – SMA_PEN.M _________________________________ 53
4.2 PROBABILITY DISTRIBUTED MOODEL – SMA-PDx _______________ 55
4.3 CONCEPTUAL LINEAR RESERVOIR – R-CRES____________________ 56
4.4 TWO CONCEPTUAL LINEAR RESERVOIRS IN PARALLEL – R-2PAR 56
5 REFERENCES ______________________________________________________ 58

3
 RRMT User Manual

1 INTRODUCTION

1
INTRODUCTION

4
 RRMT User Manual

1.1 THE TARGET GROUP OF THIS MANUAL

This manual was written for users of the Rainfall-Runoff Modelling Toolbox
(RRMT). Its emphasis lies on application of the toolbox. Only that part of the
toolbox visible to the user is described. No detailed information about the pro-
gramming code is given. However, the user is informed about how to extend
the toolbox by adding new modules.

1.2 OBJECTIVES OF THE RRMT

The toolbox simulates rainfall-runoff relationships in natural catchments. It
also offers tools for data analysis and manipulation, parameter uncertainty
analysis, and investigations into the nature of the model residuals.
An interface to the Monte Carlo Analysis Toolbox (MCAT), which is a collec-
tion of Matlab functions for model evaluation described in a separate manual,
is also available.

1.3 HARDWARE AND SOFTWARE REQUIREMENTS

HARDWARE
• Any computer able to run Matlab version 5.2 or higher (Mathworks, 1996).
SOFTWARE
• Any platform supporting Matlab
• Matlab Version 5 or higher (RRMT is developed in Matlab 5.2 (Release
10), it might therefore not work properly in older versions).

1.4 ASSUMPTIONS
It is assumed that the user has a basic knowledge of the Matlab programming
environment. Explanations about how to use Matlab are kept to a minimum
within this manual.
The user is referred to the help documentation delivered with every Matlab
version or to the Mathworks website (http:\\www.mathworks.com) for more de-
tailed information. A large number of books about Matlab are also available.
Two examples are Etter, D.M. (1997): Engineering problem solving with Mat-
lab - 2nd Edition. Prentice-Hall, USA (for a first introduction) and Hanselman,
D. and Littlefield, B. (1998): Mastering Matlab 5 - A comprehensive tutorial
and reference. Prentice-Hall, USA (for the more advanced user).

5
 RRMT User Manual

GUI

HYBRID MODEL ARCHITECTURE

OPTIMIZATION
MODULE AET

VISUAL ER
ANALYSIS P MOISTURE
ROUTING
MODULE T ACCOUNTING Q
MODULE
PET MODULE

OFF-LINE DATA
PROCESSING
MODULE MOISTURE STATUS

Figure 1 General structure of the RRMT.

1.5 BACKGROUND
A Rainfall-Runoff Modelling Toolbox (RRMT) has been developed in
order to produce parsimonious, lumped model structures with a high level of
parameter identifiability. Such identifiability is crucial if relationships between
the model parameters representing the system and catchment characteristics
(e.g. dominant soil types, land use, etc.) are to be established. RRMT is a
modular framework that allows its user to implement different model structures
to find a suitable balance between model performance and parameter identifi-
ability. Model structures that can be implemented are lumped, relatively sim-
ple (in terms of number of parameters), and are of the conceptual or hybrid
metric-conceptual type (Wheater et al., 1993).
The architecture of the RRMT is shown in Figure 1. All structures con-
sist of a moisture accounting and a routing module. Available moisture ac-
counting modules are the Catchment Moisture Deficit (CMD, Evans and
Jakeman, 1998), and the Catchment Wetness Index (CWI) and its modifica-
tions which are all used in the IHACRES (Jakeman and Hornberger, 1993)
model structure, the Penman model structure (Penman, 1948) and its modifi-
cations, different modules using a probability distribution of moisture stores
(e.g. Moore, 1999) etc. The inputs into all modules are precipitation and tem-
perature or potential evapotranspiration time series. They produce an esti-
mate of the effective rainfall, i.e. that part of the rainfall that is contributing to
runoff. Some modules will also calculate estimates of actual evapotranspira-
tion and the moisture status of the catchment. Routing modules use the effec-
tive rainfall as input and translate it into streamflow. Available are different
combinations of conceptual reservoirs. The user also has to select an optimi-
sation module during calibration. Available modules are a uniform random
search (URS), a version of the shuffled complex evolution algorithm (SCE,
Duan et al., 1992). A variety of visual analysis modules can be selected to
view input data and results for further analysis. Off-line data processing mod-

6
 RRMT User Manual

RRMT

STRUCTURE
CONTROLLER
ARRAY

Figure 2 The user can also communicate with the program by changing the structure array
and calling the controller function. He doesn’t have to use the main window. All information
and data required are stored in the structure array.

ules are a range of files which can be used outside the RRMT to, for example,
aggregate data from daily to monthly time steps.

1.6 HOW TO USE THIS MANUAL

This manual only contains the basic information required to use the RRMT. A
detailed theoretical treatment of structural components and procedures can be
found in Wagener, Lees and Wheater (2001) as listed below.
The appropriate use of this manual depends on the level of expertise of
the reader. Chapter 2 contains the user instructions, i.e. the steps which need
to be taken to set up the RRMT, and then to implement any modelling proce-
dure. It is placed at the beginning because it is assumed that this is the part of
the manual which will be most frequently consulted. It does however not in-
clude any detailed descriptions of the individual components available.
These are given in chapter 3. The user might want to read this chapter in
the beginning to familiarise her- or himself with how the components work and
what their parameters represent in detail before any modelling is undertaken.
Chapter 4 is a short description of the structural implementation of the
software framework described here. It is not necessary for the user to know
about this. It was simply added for completeness.
Chapter 5 shows a number of experiments performed using some of the
available modules. It is meant to increase the understanding of the working of
these modules and the effects of changes in individual parameters.
Relevant references are listed at the end. Again, a much more compre-
hensive list of references is given in Wagener, Lees and Wheater (2001).

1.7 PUBLICATIONS
The following publications contain information and application examples of the
RRMT:

Wagener T., Boyle D.P., Lees M.J., Wheater H.S., Gupta H.V., Sorooshian S. 2001. A
framework for the development and application of hydrological models. Hydrology Earth
System Sciences, 5(1), 13-26.
Wagener T., Lees M.J., Wheater H.S. 2001. A framework for the development and application
of parsimonious hydrological models. To appear in Singh, Frevert, Meyer (Eds.) Mathe-
matical models of small watershed hydrology – Volume 2. Wat. Resour. Publ. LLC: USA.
Wagener T., McIntyre N., Lees M.J., Wheater H.S. and Gupta H.V. 2003. Towards reduced
uncertainty in conceptual rainfall-runoff modelling: Dynamic identifiability analysis. Hydro-
logical Processes , 17, 2, 455-476.

7
 RRMT User Manual

Wagener T., Lees M.J., Wheater H.S. 1999. A generic rainfall-runoff modelling toolbox. Eos.
Trans. AGU, 80(17), Fall Meet. Suppl., F203.

CONTROLLER

WORKFUN

SOIL MOISTURE
ACCOUNTING

ROUTING
OPTMET
(optional)
SRIV
(optional)

PERFORMANCE

Figure 3 The controller function calls workfun. Workfun combines SMA function, routing func-
tion and performance function. The returned output variables are the performance crite-
rion (objective function value), effective rainfall and internal state variables of the catch-
ment, i.e. catchment moisture status. The optimisation function placed in the optmet field
of the structure array is used as an interface between controller and workfun during the
calibration mode

1.8 PROGRAM STRUCTURE

The general architecture of the RRMT is built as follows. All data and all in-
formation are stored in a structure array. When starting a simulation using the
main GUI (Graphical User Interface) or through the workspace window, the
function 'controller' is called. This controller extracts the necessary information
from the structure array (Figure 2) and either calls a calibration function or di-
rectly the function 'workfun' (Figure 3). Workfun then combines the selected
SMA module, routing module, and optimisation module (if in calibration or
validation mode), and calculates all objective function values requested.

8
 RRMT User Manual

2 USER INSTRUCTIONS

2
USER
INSTRUCTIONS

To err is human. To really foul up, use a computer.

Author: unknown
Source: Goldberg (1989)

9
 RRMT User Manual

2.1 INSTALLATION PROCEDURE

Follow the instructions given below:

1. Download the RRMT zip-file from the project web page.

2. Place it in a directory (e.g. called RRMT) within the Matlab directory.
3. Unzip the file using win-zip.
4. Start Matlab.

[for Release 10 and 11]

5. Select Set Path… from the File menu.
6. Select Add to Path… from the Path menu.
7. Press the '…' button and locate the RRMT directory.
8. Press the OK buttons in 'Browse for Folder' and the 'Add to path' win-
dow.
9. Select Save Path from the File menu.
10. Select Exit Path Browser from the File menu.

[for Release 12]

5. Select Set Path… from the File menu.
6. Press the Add Folder… button.
7. Locate the RRMT directory.
8. Press the OK, Save and Close buttons.

You will now be able to access the RRMT from the Matlab command window
without having to be in the RRMT directory.

2.2 EXAMPLE FILES

An examples is available. It can be started by typing rrmt_eg1 into the work-
space window, followed by pressing return. Then just follow the instructions
on the screen. The example is of a multi-objective analysis (see chapters
3.5.6 and 3.5.7) of a rainfall-runoff model.

2.3 STEP BY STEP MANUAL

(1) START THE MAIN MENU WINDOW
Type gui_main in the Matlab command window. The startpage is shortly
shown. Then the main window appears.
You can also use the toolbox without using the main and visualisation
window. Every action can be performed in the Matlab command window. See
therefore the chapter about batch files later in this manual.
(2) LOAD THE INPUT DATA FILE
Select Load Structure Array from the Menu option in the menu bar. Find the
mat-file you want to use, select and open it. Use the file rrmt_eg1.mat as a
trial.

10
 RRMT User Manual

You have to press the Refresh button after you opened the file to see
the file contents in the edit fields.
(3) CHANGE THE SETTINGS OF THE MODELLING PROCEDURE
You can change the current settings by typing the new settings directly in the
edit fields. However, they will not be stored in the structure array.
(4) START THE CALCULATION
Press the Run button to start the calculation (calibration/validation/simulation).
(5) VISUALISE THE RESULTS
The numerical output of the calibration is plotted in the workspace window.
The best five calibration results are shown. The best result is at the top, then
decreasing downwards. Each column shows one parameter. The first columns
show the loss function parameters, followed by the routing parameters. The
last column shows the objective function (optimisation criterion) used. All ob-
jective functions are minimised during the calibration process!
You have to copy the parameter values from the workspace window
and paste them in the corresponding edit fields in order to validate the model.
WATCH OUT: Make sure that you copy and paste all parameter values
(loss function and routing) in one go! Otherwise RRMT can't recognise the pa-
rameters as one vector.
Select Visualisation Tools from the Menu option in the menu bar. The
visualisation tools window opens. Select the plots you want to see by select-
ing the checkboxes. Press the Show button to open the plots.
WATCH OUT: Only parameter values and objective function values are
stored during the calibration stage, not the calculated runoff time series. To
calculate runoff one has to select a certain parameter set and run the program
in validation ('val') mode.
(6) SAVE THE RESULTS
Select Save Structure Array from the Menu option in the menu bar. Type the
file name in the edit field and select the directory where you want to place the
file, followed by pressing the Save button.
(7) EXIT RRMT
Press the Exit button if you want to leave the RRMT. Type exit in the Matlab
command window to leave the program completely.

11
 RRMT User Manual

2.4 MAIN GRAPHICAL USER INTERFACE DESCRIPTION

GENERAL
Table 1 Description of ‘general’ field contents.

Field Name Content

Information Information about the modelling procedure.

Catchment The name of the catchment.
River The name of the river.
Last Changed The date is updated after every calculation
(press refresh).
Mode [cal/val/sim] The modelling mode [cal(ibration) /
val(idation) / sim(ulation)].
Time Step [minutes] The modelling time step. (e.g 1440 when us-
ing daily data)
Validation Result The objective function (OF) value after run-
ning a selected model in validation mode
(press refresh).

OPTIMISATION
Table 2 Description of ‘Optimisation’ field contents.

Field Name Content

% Warm Up Percentage time series neglected when cal-

culating the objective function value.
Search Method The search method used to optimise the
model parameters.
Objective Function One or more objective functions selected to
describe the model performance (input is the
number corresponding to the OF, e.g. select
1 for the Nash-Sutcliffe Efficiency measure)
SRIV Not used.
Population Size Size of the population sampled when using a
random search method.
SCE-UA Options of the SCE algorithm (see more de-
tailed description in chapter 3.2.1).
Box-Cox Transformation Enter a value to Box-Cox transform the data
as described in chapter 3.5.14.
Threshold [low/high] A low and a high threshold can be put here.
they will be used for the high, medium and
slow respectively low flow OF as described
later in chapters 3.5.8/9/10.

12
 RRMT User Manual

SOIL MOISTURE ACCOUNTING MODULE

Table 3 Description of ‘Soil Moisture Accounting Module’ field contents.

Field Name Content

Soil Moisture Accounting Name (i.e. filename) of the selected soil

moisture accounting ('SMA_NAME'). All loss
function names have a three letter (or letter +
number) code.
Parameters The parameter names are displayed here
after a calculation has been performed (press
refresh). Names have to be available in the
file 'INFO_SMA_NAME.M' corresponding to
the selected loss function.
Boundaries The boundary values for each parameter.
Lower and upper boundary for each parame-
ter have to be on one line.
Values The parameter values of a selected model
have to be put here. Each value has to be on
the correct line. These values are used when
the mode is 'val(idation)' or 'sim(ulation)'.

ROUTING MODULE
Table 4 Description of ‘Routing Module’ field contents.

Field Name Content

Routing Function Name (i.e. filename) of the selected routing

function ('R_NAME'). All routing function
codes have four letters (or letters + num-
bers).
TF Structure Not used
Parameters The parameter names are displayed here
after a calculation was performed (press re-
fresh). Names have to be available in the file
'INFO_R_NAME.M' corresponding to the se-
lected loss function.
Boundaries The boundary values for each parameter.
Lower and upper boundary for each parame-
ter should be on one line. Separate bounda-
ries for different parameters by pressing en-
ter.
Values The parameter values of a selected model
have to be put here. Each value has to be on
the correct line. These values are used when
the mode is 'validation' or 'simulation'.

13
 RRMT User Manual

Figure 5 Main GUI of the RRMT.

Figure 4 The 'open new structure array' window.

14
 RRMT User Manual

Figure 6 The ‘save structure array’ window.

15
 RRMT User Manual

SOIL MOISTURE ACCOUNTING AND ROUTING MODULE PARAMETERS

Table 5 Description of SMA and R module parameters.

Parameters Description [possible range]1

SMA_CWI
• Tau Time constant of catchment losses [1 : 40]
• refp Reference parameter [0 : 10]
• mf Modulation factor [0 : 5]
• s(1) Initial s value (initial moisture state of catchment) [0 : 1]
• volc Volumetric constant (set boundaries from 0 to 0 during calibration)
SMA_CMD
• c1 Evapotranspiration parameter (Transfers temperature to PE) [0 : 1]
• c2 Evapotranspiration parameter (Transfers PE to AE by scaling the influence
of the catchment moisture deficit (cmd)) [0 : 0.1]
• c3 Maximum drainage while a cmd exists [0 : 10]
• c4 Maximum cmd before water ceases draining to stream [1 : 100]
• in. cmd Initial catchment moisture deficit
SMA_YE1
• tau Time constant of catchment losses [1 : 40]
• refp Reference parameter [0 : 10]
• mf Modulation factor [0 : 5]
• l Threshold value [0 15]
• p Power value [0 : 2]
• s(1) Initial s value (initial moisture state of catchment) [0 : 1]
• volc Volumetric constant (set boundaries from 0 to 0 during calibration)
SMA_PEN
• root con- Root constant [10 : 200]
stant
• in. def. Initial deficit upper store (size upper store = rc + 25)
SMA_IC1
• root con- Root constant [10 : 200]
stant
• bypass Bypass (default is 0.15) [0 : 0.25]
• in. def. Initial deficit upper store (size upper store = rc + 25)
• Drying Drying curve (default is 0.08) [0 : 0.2]
curve
SMA_PD3/4
• cmax Maximum storage capacity [0 : 500]
• b Shape of Pareto distribution [0 : 2.5]
• c(1) Initial critical capacity
R_CRES
• k Time constant [1 : 15]
• n Non-linearity parameter (a value of 1 is linear, lower values are non-linear)
R_2PAR
• k(q) Time constant quick flow reservoir [1 : 15]
• n(q) Non-linearity parameter (a value of 1 is linear, other values (usually <1) are
non-linear)
• k(s) Time constant slow flow reservoir [15 : x00]
• n(s) Non-linearity parameter (a value of 1 is linear, other values (usually <1) are
non-linear)
• %(q) Fraction of flow through quick flow reservoir [0 : 1]
R_3PAR
• k1 Time constant first reservoir [1 : 15]
• n1 Non-linearity parameter (a value of 1 is linear, other values (~<1) are n-l)
• k2 Time constant second reservoir [15 : x00]
Non-linearity parameter (a value of 1 is linear, other values (~<1) are n-l)

16
 RRMT User Manual

• n2 Time constant third reservoir [15 : x00]

• k3 Non-linearity parameter (a value of 1 is linear, other values (~<1) are n-l)
• n3 Fraction of flow through first reservoir [0 : 1]
• a Fraction of flow through second reservoir [0 : 1], a + b <= 1
• b
R_LEAK
• k1 Residence time [example 214.05]
• k2 Residence time [example 97.64]
• k3 Residence time [example 21.84]
• h1 Lower threshold [example 150.0]
• h2 Upper threshold [example 273.48]
R_MAPO
• k(t) Time constant top outlet
• k(b) Time constant bottom outlet
• h Threshold
R_2PMP
• k(qt) Time constant top outlet quick reservoir
• k(qb) Time constant bottom outlet quick reservoir
• h Threshold quick reservoir
• k(s) Time constant slow reservoir
• n(s) Non-linearity parameter (a value of 1 is linear, other values (~<1) are n-l)
• %(q) Fraction of flow through quick flow reservoir [0 : 1]
1
these are just guidelines, wider ranges might be needed depending on catchment characteristics

17
 RRMT User Manual

Figure 7 Visualisation GUI.

2.5 VISUALISATION INTERFACE DESCRIPTION

The use of the visualisation tools window (Figure 5) is very straightforward.
Mark the checkboxes of the figures you want to be opened. Then press the
Show button, this will initialise all selected plotting functions and all corre-
sponding windows will appear. The following figures are currently available:

2.6 INPUT
2.6.1 Data Array - Structure
All necessary information and data are stored in a structure array. One can
imagine a structure array as a combination of post boxes. Each individual box
can be addressed by its name.
The structure array is basically split into two parts: the first part containing all
input data and information, and the second containing the output of the calcu-
lation(s). The first part can be viewed by typing structure.input in the Matlab
command window. The second by typing structure.output. Further information
about data handling is given in the Step by Step Manual.

18
 RRMT User Manual

STRUCTURE INPUT
Table 6 Input fields of the structure array.

FIELDNAME TYPE4 CONTENT

info1 ‘‘ File information

catchment1 ‘‘ Catchment name
river1 ‘‘ River name
date [] Automatically updated
mode ‘‘ ‘cal’ ‘val’ ‘sim’
dt [] Time step in minutes
cr [] Objective function
optmet [] Optimisation method
sriv ‘‘ Not used
samples [] Number of samples for urs
sma_boundaries2 [] When using URS:
Lower (lbp) and upper (upb) boundary values
[lbp1 ubp1;lbp2 ubp2]
r_boundaries2 [] When using URS:
Lower (lbp) and upper (upb) boundary values
WATCH OUT: Only neces- [lbp1 ubp1;lbp2 ubp2]
sary when using conceptual
reservoirs for the routing!
sma ‘‘ ‘sma_cmd’ ‘sma_cwi’ ‘sma_pen’
‘sma_ye1’…
sma_par3 [] Selected loss function parameter set
routfun ‘‘ ‘r_2par’…
tf_struct [] Not used
r_par3 [] Selected routing parameter set
flow [] Flow data vector {mm/dt}
temp [] Temperature data vector {degree C}
et [] Evapotranspiration vector {mm/dt}
rain [] Rainfall vector {mm/dt}
lambda [] Box-cox transformation coefficient
warmup [] Initial time-series period neglected {%/100}
options [] SCE options as described later in manual
epsilonL [] Lower flow threshold value
epsilonH [] Upper flow threshold value
1
Add one blank field at the end.
2
For urs and sce.
3
For validation and simulation.
4
{} is unit, [ ] is numerical input, ' ' is character input.

STRUCTURE OUTPUT
Table 7 Output fields of the structure array.

FIELDNAME TYPE1 CONTENT

population [] [lf_par r_par objective_function]

catchstatusinfo ‘‘ 'cwi' 'cmd'
catchstatus [] Variable describing catchment state
val_result [] Objective function value2
searchsteps [] [par1 par2...parn objective_function]
er [] Effective rainfall {mm/dt}
cal_f [] Calculated streamflow {mm/dt}
threshold [] Threshold for MCAToolbox
aet [] Actual evapotranspiration calculated by the

19
 RRMT User Manual

model {mm/dt}
ofs [] Remaining objective functions
variables [] Variables calculated {depending}
flowsep [] Flow separated in quick and slow component
{mm/dt}
1
{} is unit, [ ] is numerical input, ' ' is character input
2
Only the objective function used for calibration

2.6.2 Create a new structure input array

Creating a new structure array is very simple. All one has to do is to copy the
predefined structure array template from the file input_template.txt which can
be found in the RRMT directory.

• Open the file “input_template.txt” by double clicking on it in your Ex-

plorer.
• Select the complete text of the template.
• Copy the text.
• Paste the text into your Matlab workspace and press the return button.
• Now all that needs to be done is filling the empty fields of the structure
array. Use the table in chapter 6.1 to find out whether the field requires
a string or a numerical input.
• After filling the fields, the user can start the main window by simply typ-
ing gui_main in the workspace window and pressing the return button.
Within the main window one can do all the input except the time series.
That has to be done in the workspace window.
• Concerning the time series. The user has to prepare (n,1) vectors for
every time series. The vectors have to be of equal length. Imagine that
the vectors are called f, r, e, and t. The commands would then look like:

Structure.input.flow =f; [return]

Structure.input.rain =r; [return]
Structure.input.et =e; [return]
Structure.input.temp=t; [return]

The user can then save the structure array using the Save structure array
from the menu command.

2.7 BATCH-FILES
Batch-files are very useful when a large number of calculations with changing
parameter boundaries, input data files, etc have to be performed.
The user can write an m-file loading the initial structure array into the work-
space, changing elements of the array if required and starting the calculation.
The results can be saved in a file defined by the user after the calculation has
terminated. The structure array can then be changed and a new calculation
can be started.
This can all be done in a single m-file. Large modelling jobs can be performed
without interference of the user.

An example of a (general) batch-file structure is given below.

20
 RRMT User Manual

EXAMPLE

function batchjob;

load example1.mat;

structure.input.optmet =’urs’;
structure.input.samples=2000;
% etc.

[structure]=controller(structure);

save test1.mat structure;

structure.input.optmet =’sce’;
structure.input.complexes=10;

[structure]=controller(structure);

save test2.mat structure;

2.8 HOW TO ADD NEW MODULES

New modules can be added very simply. The user just has to use the struc-
ture, i.e. the input and output parameters, described below. The other con-
straint is that all sma names have to contain 3 letters, while all the routing
module names must be 3 letters long.
The user also has to supply an function related to every new module to pro-
vide the parameter names to be displayed.

SOIL MOISTURE ACCOUNTING MODULE

INPUT

pars sma parameter(s)

r vector with rainfall data
t vector with temperature data
et vector with ET data
f vector with observed streamflow

SMA MODULE NAME

must be 3 letters, e.g. “sma_buc.m”

OUTPUT

er effective rainfall
s catchment status (e.g. moisture deficit)
ae actual ET (set aet=0 if not calculated)
pars sma parameter(s)

21
 RRMT User Manual

EXAMPLE

[er,dd,ae,pars]=sma_cmd(pars,r,t,et,f);

You will have to replace catchstatus by a dummy (dummy = [ ];) if no moisture

content is produced.

ROUTING MODULE

INPUT

r_par routing parameter(s)

er effective rainfall
structure structure array

ROUTING MODULE NAME

must be 4 letters/numerals, e.g. “r_2par.m”

OUTPUT

cal_f calculated (stream)flow

structure structure array
r_par routing parameter(s)

EXAMPLE

[cal_f,structure,r_par]=r_2par(r_par,er,structure);

INFO MODULE
Every loss function and every routing module requires a related info module to
provide the parameter names.

INPUT

structure structure array

INFO MODULE NAME

must be: info_'name of the sma or r module', e.g.

“info_sma_buc.m”

OUTPUT

par_names a matrix containing the parameter names, in

created the order they are read in by the lf or r
module. The matrix must be created using the
str2mat function.

EXAMPLE

22
 RRMT User Manual

[par_names]=info_r_2par(structure);

23
 RRMT User Manual

3 AVAILABLE MODULES

3
AVAILABLE
MODULES

24
 RRMT User Manual

3.1 VISUALISATION MODULES

3.1.1 Input Data – PL_INPUT

The plot shows rainfall, streamflow, temperature and evapotranspiration (ET)
time series. The plotting function will also work when temperature or/and
evapotranspiration are not available.

3.1.2 Model Output – PL_OUTPUT

Rainfall, effective rainfall, observed and calculated streamflow time series are
visualised using this plotting function. The variable describing the catchment
status, will be either an indicator of the moisture deficit (e.g. in the case of the
cmd loss function) or the wetness of the catchment (e.g. in case of the cwi
loss function), if available.

3.1.3 Actual Evapotranspiration – PL_AET

The plot shows the Actual ET, calculated by the chosen sma module, versus
the Potential or Actual ET data stored in the input part of the structure array. It
can only be used in connection with the sma modules that calculates an esti-
mate of the actual ET (e.g. sma_cmd).

3.1.4 Confidence Limits – PL_CFL

This plot displays the time-series output with associated confidence intervals
calculated using the GLUE methodology (Beven, 1998; Beven and Binley,
1992). For each point in time a cumulative frequency distribution of calculated
streamflow values is generated using the selected objective function (con-
verted to likelihood as defined in the GLUE context) and the 90% confidence
intervals calculated using linear interpolation.

The bottom graph is added to make it easier to spot regions with large uncer-
tainties. It shows the difference between upper and lower confidence limits,
normalised by the maximum difference between the two over the investigated
period of time, i.e.
UCFLi − LCFLi
∆CFLi =
max(UCFL − LCFL )
where
∆ CFL normalised difference between upper and lower confidence limit
UCFL upper confidence limit
LCFL lower confidence limit
max( ) maximum over time series

This plot requires the specification of a threshold value.

3.1.5 Residual Confirmation – PL_CONFIRMATION

Certain assumptions are made when using an objective function based on a
simple least squares approach. These assumptions are that the residuals are
homoscedastic (i.e. constant variance across all values of the predictor vari-
able (flow)), independent of each other, and normally distributed with zero
mean (i.e. white noise) (see for example McCuen and Snyder, 1986, p. 81).

25
 RRMT User Manual

The residual confirmation plot visualises the residuals with respect to these
assumptions.

Residuals versus
Four different plots are available in order to spot trends. These are residuals
versus observed and calculated flows (an increasing spread of the residuals
with increasing flow values indicates heteroscedasticity), residuals versus
rainfall, and residuals versus time. The last plot shows whether the data show
a trend to over- or underestimate the observed hydrograph. The number of
over- and underpredictions is also shown since it is sometimes difficult to spot
a tendency when many data points are available.

Normality
The blue histogram shows the distribution of the residuals. Normally distrib-
uted residuals would have a mean of zero and show a Gauss bell shaped
form.

Autocorrelation
Residuals that do not show autocorrelation (also called serial correlation)
would completely lie within the shown boundaries in 95% of the cases.

3.1.6 Double Mass – PL_DATACHECK

Simple double mass plot showing cumulative rainfall versus cumulative
streamflow in order to identify inconsistencies in the relationship between the
two time series (Marsh, 1999). The double mass plot should be a relatively
straight line in case of consistency (Clarke, 1994, p.26).

3.1.7 Flow Duration + Vol. Fit – PL_VFaFDC

A flow duration curve 'is in effect a cumulative frequency diagram with specific
time scales' (Kottegoda and Rosso, 1997, p.9). 'The shape of the flow dura-
tion curve gives a good indication of a catchment's characteristic response to
its average rainfall history. An initially steep sloped curve results from a very
variable discharge, usually from small catchments with little storage where the
stream flow reflects directly the rainfall pattern. Flow-duration curves that have
a very flat slope indicate little variation in flow regime, the resultant of the
damping effects of large storages. Groundwater storages are provided natu-
rally by extensive chalk or limestone aquifers, and large surface lakes or res-
ervoirs may act as runoff regulators either naturally or controlled by man.'
(Shaw, 1994).

3.1.8 Obs. Vs Calc. And Error – PL_MODELFIT

Observed (q) and calculated (c) flow time series are plotted. Additionally the
residuals, i.e. e(i) = q(i) - c(i), are shown.

3.1.9 Parameter Population – PL_MC

This plot shows (individual) the result of a Monte-Carlo sampling procedure.
Each dot is one parameter value against its corresponding objective function
value. This plot requires the specification of a threshold value.

26
 RRMT User Manual

3.1.10 Search Steps – PL_SCE

Shows the best parameter set of each step of the calibration process using
the SCE-UA algorithm (see later in this manual).

3.1.11 Modelfit (Log Scale) – PL_LOGFIT

Observed (o) and calculated (c) flow are shown on a logarithmic scale. This
scale puts an emphasis on the model fit during low flow periods. Additionally
the residuals, i.e. e(i) = q(i) - c(i), are shown.

3.1.12 MCAT
Starting the Monte Carlo Analysis Toolbox (MCAT) for further analysis of pa-
rameter identifiability and uncertainty (see MCAT manual for further informa-
tion). This option requires the specification of a threshold value.

3.1.13 Threshold
A threshold value has to be selected for some plots so that only parameter
sets which produce an objective function value below the selected one, i.e.
those which are better, are chosen for visualisation.

3.1.14 HMLE
The Heteroscedastic Maximum Likelihood Estimator is an objective function
based on a weighted least square approach in case the modelling residuals
show heteroscedascity, i.e. changing variance with changing flow values (see
chapter on objective functions). It is often excluded from the standard objec-
tive functions calculated during calibration, because its calculation is very time
consuming. The reason for that is that each calculation of the HMLE requires
the iterative optimisation of the lambda parameter (using the fmin function in
Matlab).

3.2 OPTIMISATION MODULES

3.2.1 Shuffled Complex Evolution Algorithm of the University of Arizona –

SCE
% OPTIMISATION MODULE: SHUFFLED COMPLEX EVOLUTION ALGORITHM
%
% [result,structure]=sce(structure);
%
% INPUT
% structure structure array
% options for sce within structure array:
% (1) number of complexes, between 2 and 20,
% depending on no. parameters and complexity of problem
% (2) convergence threshold, use 10^-2 as default
% (3) convergence criterion, [1] parameter, [2] objective function
% (4) maximum number of iterations, recommended value is 100
%
% OUTPUT
% result matrix containing parameter values and corresponding function values
% structure updated structure array
%
% T. Wagener, Imperial College, February 2000
%
% Reference: Duan et al. 1992. WRR.

27
 RRMT User Manual

The Shuffled Complex Evolution algorithm (Duan et al, 1992; 1993; 1994) is a
population-based search algorithm. An initial population is sampled from the
feasible parameter space. This population is then split into a number of com-
plexes and each complex is evolved using a deterministic simplex-based ap-
proach. The complexes are combined and sorted (the shuffling step) after
each iteration and new complexes are formed, which are then again individu-
ally evolved. This communication between the complexes allows a better in-
vestigation of the whole feasible parameter space. The steps of the algorithm
are described in Figures 6 and 7.

28
 RRMT User Manual

A
SCE

Select q, α and
β, where 2≤q≤m,
α ≥1, β≥1
Set e=1

Assign triangular
probability distribution to A

Select q points from

A according to rho
Store points in B and their
relative positions in A in L
Set j=1

Sort B and L in order of

increasing function value
Compute centroid of
u(1)...u(q-1), with q(u) the
worst point in B

Compute rs=2g-u(q)
(reflection step)

NO Generate point ms at
rs within Ω random in H (mutation step)
Set rs=ms
YES

Compute fr

YES
Set u(q)=rs, fq=frs fr<fq

NO

Compute cs=[g+u(q))]/2, fcs

NO Generate point ms at
random in H (mutation step)
fcs<fq
Compute fz
Set u(q)=ms, fq=fms
YES

Set u(q)=cs and fq=fcs

NO
Set j=j+1 j>α

YES

Replace B into A according

to L and sort A in order of
increasing function value

NO
Set e=e+1 e≥β

YES

SCE

Figure 8 Flowcharts showing the SCE-UA algorithm (A) and the CCE algorithm (B) which is
part of the SCE-UA.

29
 RRMT User Manual

B
START

0 Input n=dimension. p=no.

complexes, m=points in 1
complex
Compute sample size s=p*m
and m=2n+1

1 Sample s points at random

(uniform) in feasible
parameter space omega
Compute function value at
each point

2
Sort s points in order of
decreasing performance
Store points in D

3
Partition D into p
complexes of m points,
i.e. D={A^k, k=1,2,...,p}

4
Evolve each complex CCE
A(k), k=1,2,...,p ALGORITHM

5
Shuffle complexes replace
A(k), k=1,2,...,m, into D
Sort D in order of
decreasing performance

6
Convergence
NO ?

YES

STOP

Figure 9 Flowcharts showing the SCE-UA algorithm (A) and the CCE algorithm (B) which is
part of the SCE-UA.

3.2.2 Uniform Random Search – URS

% OPTIMISATION MODULE: UNIFORM RANDOM SEARCH
%
% [result,structure]=urs(structure);
%
% INPUT
% structure structure array
%
% OUTPUT
% result matrix with all parameter sets + calculated o.f. values (ranked)
%
% T. Wagener, Imperial College, April 1999

In a simple Uniform Random Search, a parameter population, with user de-

fined size, is sampled from the feasible parameter space based on a uniform

30
 RRMT User Manual

s k −1 s k-1/c
rk rk cτ( t k )
s k/c

uk=0.5(s k+s k-1)*r k

Figure 10 Schematic description of the CWI function.

(also called rectangular) distribution. The user has to specify lower and upper
boundaries for each parameter.

3.3 SOIL MOISTURE ACCOUNTING (SMA) MODULES

3.3.1 Catchment Wetness Index – SMA_CWI

% SMA MODULE: CATCHMENT WETNESS INDEX (implemented in IHACRES)
%
% [er,s,dummy,pars]=sma_cwi(pars,r,t,et,f);
%
% INPUT
% pars contains
% tau time constant of catchment losses
% refc reference constant
% mf modulation factor
% s(1) initial value
% volc volumetric constant
%r rainfall (mm/dt)
%t temperature (deg. C)
% et potential evapotranspiration (mm/dt)
%f flow (mm/dt)
%
% OUPUT
% er effective rainfall (mm/dt)
%s catchment wetness index
% dummy empty array
% pars contains as above
%
% M.Lees and T.Wagener, Imperial College, March 1999
%
% Reference: Jakeman AJ and Hornberger GM 1993. WRR, 29(8), 2637-2649

The catchment wetness index loss function is used in the IHACRES model
structure introduced by Jakeman et al (1990). It has undergone various modi-
fications since then. The form used here is illustrated in Figure 10.
τ w (t k ) = τ ⋅ e mf ( refc −t k )

respectively

τ w (t k ) = τ ⋅ e mf ( refc − pet k )

where
τw the time constant (in days) of the catchment losses (i.e. the time it
takes for the catchment losses to decay to exp(-1) or 36,8% of their
peak value)

31
 RRMT User Manual

s k-1/c
rk s k −1
rk cτ(t k ) s k/c

uk=[0.5(s k+s k-1)] p*r k

Figure 11 Schematic description of the Ye et al. function.

τ τ w at a reference temperature of refc (or at a reference potential

evapotranspiration of refc)

t the temperature in °C
mf parameter describing the effect of a unit change in temperature on the
loss rate
k the time step
⎧ 1 ⎫
s k = Vc rk + ⎨1 − ⎬ ⋅ s k −1
⎩ τ w (t k ) ⎭
where
s the catchment storage index or catchment wetness [0 : 1]
r the rainfall
Vc parameter describing the impact that a unit input of rainfall has on the
catchment storage

The following form of Equation 2 helps to understand Figure 4,

sk s s
= rk + k −1 − k −1
Vc Vc Vc ⋅ τ k
[new state = rain + old state - depletion]

The effective rainfall can then be calculated using,

1
uk = ( s k + s k −1 ) ⋅ rk
2
where
u the effective rainfall

3.3.2 Ye et al. Model Structure – SMA_YE1

% SMA MODULE: MODIFIED CATCHMENT WETNESS INDEX
%
% [er,s,dummy,pars]=sma_ye1(pars,r,t,et,f);
%
% INPUT
% pars contains
% tau time constant of catchment losses
% refc reference constant
% mf modulation factor
% l threshold

32
 RRMT User Manual

% p peakiness factor
% s(1) initial value
% volc volumetric constant (set vc = 0 during calibration)
%r rainfall (mm/dt)
%t temperature (deg. C)
% et potential evapotranspiration (mm/dt)
%f flow (mm/dt)
%
% OUPUT
% er effective rainfall (mm/dt)
%s catchment wetness index
% dummy empty array
% pars contains as above
%
% M.Lees and T. Wagener, Imperial College, March 1999
%
% Reference: Ye W, Bates BC, Viney NR, Sivapalan M, Jakeman AJ 1997. WRR, 33, 153-166

Ye et al. (1997) introduced two additional parameters to the CWI loss function
described in an earlier chapter. This was necessary in order to model the be-
haviour of low-yielding ephemeral catchments which have different response
mechanisms than catchments in humid areas. The effective rainfall is now
calculated as follows,

uk =
1
(( s k + s k −1 ) − l ) p ⋅ rk if s k > l
2

uk = 0 otherwise

where

l the threshold below which no effective rainfall is produced

p a peakiness factor
The remaining equations are identical to the CWI loss function:
⎛ 1⎞
sk = Vc rk + ⎜⎜1 − ⎟⎟ sk −1
⎝ τk ⎠

τ k = τ ⋅ e f (R k −t k )

The equation for s k can also be written in the following form which makes it
easier to understand Figure 5:
sk s s
= rk + k −1 − k −1
Vc Vc V c ⋅ τ k

[new state = rain + old state - depletion]

3.3.3 Penman – SMA_PEN

% SMA MODULE: TWO STORE STRUCTURE BASED ON PENMAN DRYING CURVE
%
% [er,dd,ae,pars]=sma_pen(pars,r,t,et,f);
%
% INPUT
% pars contains:

33
 RRMT User Manual

% root constant (mm)

% initial deficit upper store (mm),
% [i.d. values larger than Smax1 are replaced by smaller random value]
%r rainfall (mm/dt)
%t temperature (deg. C)
% et potential evapotranspiration (mm/dt)
%f flow (mm/dt)
%
% OUTPUT
% er effective rainfall (mm/dt)
% dd=[d1 d2]
% d1 deficit upper store (mm)
% d2 deficit lower store (mm)
% ae actual evapotranspiration (mm/dt)
% pars contains as above
%
% T. Wagener, Imperial College, June 1999
%
% Reference: Jolley TJ 1995 PhD Dissertation Imperial College, Civil Eng. Dep.

The Penman sma module is based on an empirical drying curve developed by

Penman (1949). The version implemented here is taken from Jolley (1995,
p124). The general working of the approach is described in Rodda (et al.,
1976, p115) as follows:

Transpiration from a crop proceeds at the potential rate until the 'root reser-
voir' has been depleted. From that point about 25mm more water can be with-
drawn from the soil at close to the potential rate, before a change occurs and
transpiration proceeds at an actual rate that is about one twelfth of the poten-
tial.
The lower layer cannot be recharged until the surface layer has been re-
charged.

The conceptual structure of the loss function is a two-layer or two-store con-

cept (see Figure 6). The size of the upper store (Smax1) is equal to the root
constant value. The size of the lower store (or storage) is fixed to 1000 mm,
as was recommended by Moore (1992, in Jolley, 1995) to ensure that the
store never dries out.

The loss function also has a bypass mechanism to account for the quick re-
sponse of the catchment. The bypass parameter φ determines the amount of
rainfall directly converted to runoff is usually fixed to 15% (to be used as a
value of 0.15) following recommendations by Mander and Greenfield (1978, in
Jolley 1995). However, it can be freely calibrated here.

The second parameter (next to the root constant) defining the form of the dry-
ing curve is the rate of depletion after the upper store is emptied (see Figure
2). A value of 0.08 is chosen for the parameter g (following Penman, 1949, in
Jolley, 1995), i.e. the actual evapotranspiration reduces to 1/12 of the poten-
tial. This parameter is fixed.

34
 RRMT User Manual

ae k rk
uk2
Smax1
1
uk1 g
AE
Smax2 1
1

uk PE
Figure 12 A Schematic of the Penman loss function (left). Penman's soil drying curve is
shown on the right.

The model has only two free parameters, the root constant and the bypass
parameter, since all others are fixed to values which have found to be reason-
able for the UK.

Some values for the root constant proposed by Grindley (1969, in Sherratt
1985, p213 and *Rodda et al. 1976, p115) are shown in the table below

Table 8 Root constant values for Penman model structure.

CROP ROOT MAXIMUM

CONSTANT SMD
[mm] [mm]

Beans/Brassica 56 100
Cereals 140 150
Grass (Permanent) 75 125
Grass (Temporary) 56 / 50* 100
Orchards (Commercial) 160 225
Orchards (Ornamental) 140 200
Root Crops 95 150
Rough Grazing 12 50
Woodland 200 / 100* 250
Beet 115* -
Potatoes 65* -

3.3.4 Catchment Moisture Deficit – SMA_CMD

% SMA MODULE: CATCHMENT MOISTURE DEFICIT
%
% [er,cmd,ae,pars]=sma_cmd(pars,r,t,et,f)
%
% INPUT
% pars contains:
% c1 actual Evapotranspiration parameter
% c2 actual Evapotranspiration parameter
% c3 maximum drainage that can occur whilst a moisture deficit exists
% c4 maximum CMD that can occur before water ceases draining to the stream
% in. cmd initial deficit in bucket
%r rainfall (mm/dt)
%t temperature (deg. C)
% et potential evapotranspiration (mm/dt)
%f flow (mm/dt)
%
% !!! Delete temperature from the structure array if you want to use !!!

35
 RRMT User Manual

% !!! pe instead. !!!

%
% OUTPUT
% er effective rainfall (mm/day)
% cmd catchment moisture deficit (mm/day)
% ae actual evapotranspiration (mm/day)
% pars contains as above
%
% T. Wagener, Imperial College, April 1999
%
% Reference: Evans JP and Jakeman AJ 1998. Environmental Modelling & Software, 13, 385-393

The catchment moisture deficit approach used by Evans and Jakeman (1998)
is based on the following equation describing the water balance of the catch-
ment:
CMDk = CMDk −1 − Pk + Ek + Dk
where
CMD the Catchment Moisture Deficit
P the Precipitation
E the (actual) Evapotranspiration loss
D the drainage
k the time step

Dk is introduced due to the assumption that there is a certain amount of runoff

to the stream, even during a positive catchment moisture deficit (Evans and
Jakeman, 1998). This can for example take the part of rainfall into account
which directly falls on open water, or drainage from the soil near to the
stream, since it occurs even in periods without rainfall. This phenomenon was
already recognised by others as for example stated by Hough et. al. (1997):
Many experiments have shown that drainage continues into early summer in
some soils even when the soil moisture deficit has become substantial near
the surface.
The drainage can be calculated as follows:
⎧ − c3
⎪ CMDk + c3 for CMDk < c4
Dk = ⎨ c4
⎪⎩ 0 for CMDk ≥ c4
where
c3 a parameter equal to the maximum drainage that can occur whilst a
moisture deficit exists
c4 the maximum CMD that can occur before water ceases draining to the
stream

The (actual) evapotranspiration equation is based on the assumption that the

effects of vegetation can be represented by taking the ‘surface resistance’ into
account, which itself can be estimated by creating a relationship with the
available soil moisture.
Ek = c1Tk exp(−c2CMDk )
where
T Temperature (can be replaced by potential ET)
c1 ET parameter (can be set to 1 if potential ET is used)

36
 RRMT User Manual

ae k rk
c3

+
CMD c4 Dk

Figure 13 Schematic description of the catchment moisture deficit loss function.

c2 ET parameter

Parameter c2 was found to be equal to zero in the simulations done by Evans

and Jakeman (1998).

To make the calculation procedure easier, a simplification as used by Evans

and Jakeman (1998) was also used here. CMDk was replaced by [CMDk-1 –
Pk] in the equations to calculate Evapotranspiration and the drainage. The ex-
tra requirement that [CMDk-1 – Pk] is considered to be equal to zero in the
case that it has a negative value.

The effective rainfall can than be calculated as follows:

⎧ Dk for CMDk ≥ 0
Uk = ⎨
⎩Dk − CMDk for CMDk < 0

The second part of equation four ( U k = Dk − CMDk ) can also be written as the
following, if equation 1 is considered:
U k = Dk − CMDk −1 + Pk − Ek − Dk

U k = Pk − CMDk −1 − Ek
The rainfall (minus the evapotranspiration losses) is therefore used to fill the
soil moisture store. Surplus rainfall is assumed to be effective rainfall.

After a calculation for one time step was made, CMDk has to be set to zero for
the next time step, in case the second condition (CMDk < 0) is true. If the rain-
fall in a time step is larger than the existing catchment moisture deficit, then
the surplus rainfall is ‘going to the stream’ and the saturated soil remains. The
catchment moisture deficit will then start to increase again in the next time
step, caused by evapotranspiration and drainage (if there is no precipitation
surplus in this time step, too).

3.3.5 Bucket Structure – SMA_BUC

% SMA MODULE: SIMPLE OVERFLOW BUCKET
%
% [er,state,ae,pars]=sma_buc(pars,r,t,et,f)
%
% INPUT
% pars contains

37
 RRMT User Manual

% cmax bucket size (mm)

% state(1) initial bucket moisture content (mm)
% [set to 999 to start with full bucket]
%r rainfall (mm/dt)
%t temperature (deg. C)
% et potential evapotranspiration (mm/dt)
%f flow (mm/dt)
%
% OUTPUT
% er effective rainfall (mm/dt)
% state bucket moisture content (mm/dt)
% ae actual evapotranspiration (mm/dt) [equals potential]
% pars contains as above
%
% T. Wagener, Imperial College, March 2000

This is a simple bucket evaporating at potential rate. Effective rainfall is pro-

duced with a simple overflow mechanism and the evapotranspiration is at po-
tential rate.

3.3.6 Penman Version – SMA_IC1

% SMA MODULE: TWO STORE MODEL BSED ON PENMAN DRYING CURVE
%
% [er,dd,ae,pars]=sma_ic1(pars,r,t,et,f);
%
% INPUT
% pars contains
% root constant (mm)
% bypass (mm), default value is 0.15
% initial deficit (mm)
% drying curve parameter (-), default value is 0.08
%r rainfall (mm/dt)
%t temperature (deg. C)
% et potential evapotranspiration (mm/dt)
%f flow (mm/dt)
%
% OUTPUT
% er effective rainfall (mm/dt)
% dd=[d1 d2]
% d1 deficit upper store (mm)
% d2 deficit lower store (mm)
% ae actual evapotranspiration (mm/dt)
% pars contains as above
%
% T. Wagener, Imperial College, June 1999
%
% Reference: Jolley 1995 PhD Dissertation Imperial College, Civil Eng. Dep.

This structure is identical with the Penman model structure described earlier.
However, some of the usually fixed parameters can be calibration in this ver-
sion (see help text in green).

3.3.7 No Soil Moisture Accounting Module – SMA_NON

% SMA MODULE: NON
% I.E. EFFECTIVE RAINFALL IS EQUAL TO RAINFALL,
% USEFUL FOR TESTING ROUTING ALGORITHMS
%
% [er,dummy,dummy,pars]=sma_non(pars,r,t,et,f)
%
% INPUT
% pars contains
% nothing
%r rainfall (mm/dt)
%t temperature (deg. C)
% et potential evapotranspiration (mm/dt)
%f flow (mm/dt)
%

38
 RRMT User Manual

R AE

U1
CM AX
U2
U

capacity
storage
c S

1 F(c) 0

Figure 14 Probability distributed SMA component.

% OUTPUT
% er effective rainfall (mm/dt)
% dummy empty array
% dummy empty array
% pars contains as above
%
% T. Wagener, Imperial College, October 1999

Selecting this function leads to the rainfall input directly send to the routing
component.

3.3.8 Probability distribution of soil moisture stores – SMA_PD3 and

SMA_PD4
% SMA MODULE: PROBABILITY DISTRIBUTED MODEL 3
%
% [er,dd,ae,pars]=sma_pd3(pars,r,t,et,f);
%
% INPUT
% pars contains:
% cmax maximum storage capacity of catchment (mm)
% b degree of spatial variability of storage capacity in catchment
% c(1) initial critical capacity (mm)
%r rainfall (mm/dt)
%t temperature (deg. C)
% et potential evapotranspiration (mm/dt)
%f flow (mm/dt)
%
% OUTPUT
% er effective rainfall (mm/dt)
% dd=[S]
% S total storage in catchment (mm)
% ae actual evapotranspiration (mm/dt), here pot. or avail. moisture content
% pars contains as above
%
% T. Wagener, Imperial College, May 2001
%
% Reference: Moore (1999)

and
% SMA MODULE: PROBABILITY DISTRIBUTED MODEL 4
%
% [er,dd,ae,pars]=sma_pd4(pars,r,t,et,f);
%
% INPUT
% pars contains:
% cmax maximum storage capacity of catchment (mm)
% b degree of spatial variability of storage capacity in catchment
% c(1) initial critical capacity (mm)
%r rainfall (mm/dt)

39
 RRMT User Manual

%t temperature (deg. C)
% et potential evapotranspiration (mm/dt)
%f flow (mm/dt)
%
% OUTPUT
% er effective rainfall (mm/dt)
% dd=[S]
% S total storage in catchment (mm)
% ae actual evapotranspiration (mm/dt), linear relation to moisture content
% pars contains as above
%
% T. Wagener, Imperial College, May 2001
%
% Reference: Moore (1999)

A distribution of soil moisture storages with different capacities can be used to

account for heterogeneity in the catchment. Such a structure can for example
account for differences in response from near-stream zones, which will be
saturated quickly, and those far away, which will only contribute little to runoff.
A probability distribution can be used if the actual distribution of those zones
within the catchment is not considered. The most commonly used is the
Pareto distribution. Moore (1999) provides a good review of this type of SMA
structure. Two different versions are used here. The difference is that the first,
SMA_PD3, evaporates at potential rate as long as moisture is available, while
the second, SMA_PD4, calculates the actual evapotranspiration as a linear
function of the actual moisture content. The equations used are described in
detail in Wagener et al. (2001, HESS, see introduction for reference). A more
thorough treatment of the mathematical background is provided by Moore
(1999).

After rainfall is added to the component, i.e. its capacity, then the part of the
capacity exceeding the critical capacity (cmax) is the first contribution to the
effective rainfall, ER1 (see Figure 8)

u1k = max( rk − (c max − c k −1 ), 0)

The first part of the effective rainfall, u1, is subtracted from the rainfall and the
remaining part is added to the soil moisture store and redistributed between
the stores based on a Pareto distribution

sk = s max⋅ (1 − ((1 − (min((ck / c max),1.0)))^ (b + 1)))

Those stores that overflow produce the second part of the effective rainfall,
u2.

u 2 k = max(rk − ( sk − sk −1 ), 0)

For SMA_PD3, actual evapotranspiration, ae, is always equal to the potential

rate, pe, while the following relationship is used in SMA_PD4

aek = [1 − (( s max − sk ) / s max)^ be] ⋅ pek

where be is set to 1 for a linear relationship.

40
 RRMT User Manual

3.4 ROUTING (R) MODULES

3.4.1 Conceptual Reservoir – R_CRES

% R MODULE: CONCEPTUAL RESERVOIR
%
% [cal_f,structure,r_par]=r_cres(r_par,er,structure);
%
% INPUT
% r_par contains:
% r_par(1) residence time (T)
% r_par(2) non-linearity coefficient (-)
% er effective rainfall (mm/dt)
% structure structure array
%
% OUTPUT
% cal_f calculated flow (mm/dt)
% structure structure array
% r_par contains as above
%
% T. Wagener, A. Sincock, Matthew Lees, Imperial College, March 2000

This module is a single conceptual reservoir which can either be used on its
own, or it can be called from other functions like r_2par and r_3par, which
combine a number of conceptual reservoirs in parallel manner. One reservoir
is usually sufficient for large time steps (at least weekly) or when a baseflow
component is absent. The most common structure uses two reservoirs in par-
allel (r_2par), one to represent quick and the other to represent slow catch-
ment responses.
A storage function can be defined to describe the relationship between
outflow of the reservoir and the amount of water stored,

S (t ) = a ⋅ Q n (t )

where S(t) is the storage [L] at time t, Q(t) is the outflow [L/T] at time t, a is the
storage coefficient [L1-nTn], and n is the coefficient of non-linearity [-].
Additionally a mass balance equation describes the change in storage S(t)/dt
[L] as the difference between inflow (u(t) [L/T]) and outflow (Q(t) [L/T]) rates,

S (t )
= u (t ) − Q(t )
dt

The reservoirs can be linear, i.e. r_par(2)=1, or non-linear. A Runge-Kutta

scheme is used to solve the non-linear reservoir. The storage coefficient a is
always used in the form of rt = 1/a, even if the parameter does not just have
the dimension of time as it is the case for all non-linear models. It really repre-
sents a residence time only in the linear case.

3.4.2 Two conceptual reservoirs in parallel - R_2PAR

% ROUTING MODULE: 2 CONCEPTUAL RESERVOIRS IN PARALLEL
%
% [cal_f,structure,r_par]=r_2par(r_par,er,structure);
%
% INPUT
% r_par contains:
% rt1 residence time first reservoir (T)

41
 RRMT User Manual

% n1 nonlinearity coefficient first reservoir (-)

% rt2 residence time second reservoir (T)
% n2 nonlinearity coefficient second reservoir (-)
% alpha fraction of er going through 1 store
% er effective rainfall (mm/dt)
% structure structure array
%
% OUTPUT
% cal_f calculated streamflow (mm/dt)
% structure structure array
% r_par contains as above
%
% T. Wagener, Imperial College, March 2000

3.4.3 Three conceptual reservoirs in parallel - R_3PAR

% ROUTING MODULE: 3 CONCEPTUAL RESERVOIRS IN PARALLEL
%
% [cal_f,structure,r_par]=r_3par(r_par,er,structure);
%
% INPUT
% r_par contains:
% rt1 residence time first reservoir (T)
% n1 nonlinearity coefficient first reservoir (-)
% k2 residence time second reservoir (T)
% n2 nonlinearity coefficient second reservoir (-)
% rt3 residence time third reservoir (T)
% n3 nonlinearity coefficient third reservoir (-)
% a fraction of er going through first store (-)
% b fraction of er going through second store
% er effective rainfall (mm/dt)
% structure structure array
%
% OUTPUT
% cal_f calculated streamflow (mm/dt)
% structure structure array
% r_par contains as above
%
% T. Wagener, Imperial College, March 2000

3.4.4 Leaky Aquifer Model Structure – R _LEAK

% R MODULE: LEAKY AQUIFERS
%
% [cal_f,structure,r_par]=r_leak(r_par,er,structure);
%
% INPUT
% r_par contains:
% r_par(1) residence time rt1 (T)
% r_par(2) residence time rt2 (T)
% r_par(3) residence time rt3 (T)
% r_par(4) lower threshold h1 (mm)
% r_par(5) upper threshold h2 (mm)
% er effective rainfall (mm/dt)
% structure structure array
%
% OUTPUT
% cal_f calculated flow (mm/dt)
% structure structure array
% r_par contains as above
%
% T. Wagener, Imperial College, July 2000
%
% References: e.g. Moore RJ 1999. In Casale R & Margottini C (eds.).
% Floods and landslides: Integrated risk assessment. Springer.

This module can be used when the catchment contains a leaky aquifer, i.e.
part of the runoff is through the groundwater and therefore not contributing to
runoff. A similar structure can be found in Moore (1999), used there to model
snow. Parameters to be optimised are the two residence times and a thresh-

42
 RRMT User Manual

effective rainfall

q3=(1/k 3)(h-h2)

q2=(1/k 2)(h-h1)

h2 h1
q1=(1/k 1)h

Figure 15 Leaky aquifer routing module.

old value. Only q1 is leakage from the catchment, while q2 and q3 are contrib-
uting to the effective rainfall.

3.4.5 No Routing Component – R_NORO

% ROUTING MODULE: NO ROUTING
%
% [cal_f,structure,r_par]=r_noro(r_par,er,structure);
%
% INPUT
% r_par contains:
% nothing
% er effective rainfall (mm/dt)
% structure structure array
%
% OUTPUT
% cal_f calculated flow (mm/dt), equals er
% structure structure array
% r_par contains as above
%
% T. Wagener, Imperial College, April 1999

This module can be used if no routing is required. This could for example be
the case when large time steps are used (at least > 1 month). The calculated
flow then equals the effective rainfall.

3.4.6 Macro-pore approach, single reservoir

% R MODULE: QUICK ROUTING COMPONENT INCLUDING MACRO-PORES
%
% [cal_f,structure,r_par]=r_mapo(r_par,er,structure);
%
% INPUT
% r_par contains:
% k(t) time constant top outlet (dt)
% k(b) time constant bottom outlet (dt)
% h threshold (mm)
% er effective rainfall (mm/dt)
% structure structure array
%
% OUTPUT
% cal_f calculated flow (mm/dt)
% structure structure array
% r_par contains as above
%
% T. Wagener, Imperial College, September 2001
%
% References: e.g. Moore RJ 1999. In Casale R & Margottini C (eds.).
% Floods and landslides: Integrated risk assessment. Springer.

43
 RRMT User Manual

This is a conceptual reservoir with an additional outlet, defined by the thresh-

old h and the time constant k(t). The structure is identical to the R_LEAK
component when the top outlet is taken away (see Fig. 15).
The additional outlet, above a threshold, adds another quick flow component
which is only active when the catchment is very wet. Accounting for, e.g.
macro-pore flow.

3.4.7 Macro-pore approach, parallel structure

% ROUTING MODULE: 2 CONCEPTUAL RESERVOIRS IN PARALLEL INCLUDING MACRO-PORES
%
% [cal_f,structure,r_par]=r_2pmp(r_par,er,structure);
%
% INPUT
% r_par contains:
% k(qt) time constant top outlet (dt)
% k(qb) time constant bottom outlet (dt)
% h threshold quick reservoir (mm)
% ks time constant slow reservoir (dt)
% ns nonlinearity coefficient slow reservoir (-)
% %(q) fraction of er going through quick reservoir
% er effective rainfall (mm/dt)
% structure structure array
%
% OUTPUT
% cal_f calculated streamflow (mm/dt)
% structure structure array
% r_par contains as above
%
% Th. Wagener, Imperial College, September 2001

Here, the R_MAPO structure is placed in a parallel structure with a ‘traditional’

conceptual reservoir. The additional reservoir is supposed to account for the
slow response of the catchment.

3.5 OBJECTIVE FUNCTION (OF) MODULES

Objective functions (OF, also called cost or loss functions) are commonly
used to judge the performance of a simulation, usually combined with visual
inspection of the simulated hydrograph. A specific OF has to be defined when
an automatic search algorithm is utilized.
All OF files start with the abbreviation OF, continued by a number and the
name of the objective function. The OF are selected by typing their corre-
sponding number in the 'Objective function' field of the Main GUI. More than
one OF can be selected. The numbers of different OF have to be separated
by a blank.
Any OF can be added by the user as long as its name is consistent with the
convention used and the user adds the OF names to the file of_names.m. All
OF have to be minimized with respect to the parameter set θ . Some of the
OF are therefore transformed with respect to their usual form so that their op-
timum value is now zero.

WATCH OUT: Missing flow values are often a problem. In the RRMT, (ob-
served) negative flow values are ignored. Therefore, set all missing values to
(small) negative values. In this way, they do not influence the resulting objec-
tive function values at all.

44
 RRMT User Manual

The following equation conventions are used in the OF equations:

o observed variable
c calculated variable
i time step
N total number of time steps available
θ parameter set

3.5.1 Nash Sutcliffe Efficiency – OF_1_NSE

The Nash-Sutcliffe Efficiency (NSE) was introduced by Nash and Sutcliffe
(1970, Garrick et al., 1978) to allow the comparison of models applied to dif-
ferent catchments and to assess the value of additional or improved model
components. It ranges from minus infinity to one, which is also the optimum
value. A detailed analysis of the measure can be found in Legates and
McCabe (1999). It is defined as
N

∑ (o
i =1
i − c i (θ ))
2

NSE = 1.0 − N

∑ (o
i =1
i − o)
2

However, all objective functions in the RRMT have their optimum at a mini-
mum value. The objective function calculated is therefore
NSE* = 1 − NSE
From the first equation it can easily be seen that an NSE value of zero indi-
cates that the mean of the observed flow is an equally good predictor than the
simulated flow sequence.
NSE puts higher emphasis on fitting the peaks of a hydrograph due to the use
of squared residuals in its calculation.

3.5.2 Coefficient of Determination – OF_2_R2

The coefficient of determination R2 measures the proportion of the total vari-
ance in the observations that can be explained by the model (Legates and
McCabe, 1999). The measure has a range between zero and one, with one
being the optimum value.
2
⎡N ⎤
∑
⎢ (oi − o )(c i (θ ) − c (θ ))⎥
R 2 = N⎣ ⎦
i =1
N

∑i =1
∑
(oi − o )2 (ci (θ ) − c (θ ))2
i =1

In order to minimize the value, the following adjustment to the measure is

made,
R 2* = 1 − R 2
See Legates and McCabe (1999) for more details on this measure.

3.5.3 Root Mean Squared Error – OF_3_RMSE

The Root Mean Squared Error (RMSE) is defined as

45
 RRMT User Manual

∑ (o − ci (θ ))
1
RMSE =
2
i
N i =1

It has the advantage that its values are in the same units as the data. Again,
the use of squared values emphasises the fitting of high flows.

3.5.4 Absolute Bias - OF_4_ABIAS

Brazil (1988, p. 62) defines the absolute bias as follows
N

∑ (o
i =1
i − c i (θ ))
ABIAS = N

∑o
i =1
i

in this form it is also called volume error (Seibert, 1999, p19). The measure is
implemented in OF_3_ABIAS.M in this form.
The measure can also be defined as mean error (Sorooshian et al., 1998) and
is then written as
N

∑ (o − ci (θ ))
1
BIAS = i
N i =1

3.5.5 Heteroscedastic Maximum Likelihood Estimator – OF_5_HMLE

Sorooshian and Dracup (1981) introduced a Heteroscedastic Maximum Likeli-
hood Estimator (HMLE) OF to account for the fact that the residuals usually
increase with increasing flow values. This violates the assumption of a con-
stant variance underlying for example the RMSE. They therefore define a
weighted measure based on a Box-Cox transformation (Kottegoda and
Rosso, 1997). The measure is calculated as follows
N

∑ w (o − ci (θ ))
1 2
i i
N i =1
HMLE = 1/ N
⎡ N ⎤
∏
⎢ wi ⎥
⎣ i =1 ⎦
with the weights w being calculated as
2 (λ −1)
wi = f i

where f i is the expected true flow value. Sorooshian and Gupta (1995, p.29)
suggest to use the observed flow value as the expected true value. However,
they warn that this will lead to a biased transformation parameter λ . λ has to
be optimised in order to get a constant residual variance.
The following procedure (taken from Sorooshian and Gupta, 1995, p30) to es-
timate λ is developed by Duan (1991):

Rd
R= −1 = 0
Rh

46
 RRMT User Manual

n
Rd = ∑ wi ε i2
i =1

n
Rh = ∑ wi ε i2 ai
i =1

ln f i
ai =
ad

1 n
ad = ∑ ln f i
n i =1

1
Rd
HMLE (θ ) = n
exp[2(λ − 1)ad ]

where
o observed flow
c calculated flow
n length of time series
θ parameter set
i time step
ε model residuals
w weight
f expected true flow value (the observed flow is used here)
λ unknown transformation parameter which stabilises variance

The parameter λ is optimised using an iterative procedure called 'Golden

Search' (Forsythe et al., 1976), implemented in the fmin function in Matlab.
This is very time consuming and the measure is therefore available as an op-
tion when accessing the MCAT through the visualisation GUI.

3.5.6 RMSE Segmentation A – OF_6_FDA, OF_7_FQA, OF_8_FSA

This segmentation scheme was introduced by Boyle et al. (2001; see also
Wagener et al., 2001b) in order to avoid the loss of information when aggre-
gating the model residuals into a single value and to make the automatic cali-
bration of hydrological models more conform with manual procedures. The
segmentation is shown in Figure 16A.

47
 RRMT User Manual

DRIVEN NON
QUICK DRIVEN
QUICK
F NON-DRIVEN
F
L QUICK
L
MEAN
O O
FLOW
W W
MEAN
NON- FLOW / 2
DRIVEN DRIVEN SLOW DRIVEN NON DRIVEN
SLOW SLOW
TIME
TIME
A B
Figure 16 The two segmentation schemes implemented. Scheme A was introduced by
Boyle et al. (2001), while scheme B is found to be more suitable for catchments in the UK.

The objective functions are a function of the different response modes of the
hydrological system. The approach is based on the reasonable assumption
that the behaviour of the catchment is inherently different during periods
“driven” by rainfall and periods without rainfall. Further, the periods immedi-
ately following the cessation of rainfall and dominated by interflow can be dis-
tinguished from the later periods that are dominated by baseflow. The stream-
flow hydrograph can, therefore, be partitioned into three components (Figure
16A), "driven" (QD), "non-driven quick" (QQ), and "non-driven slow" (QS). The
time steps corresponding to each of these components are identified through
an analysis of the precipitation data and the time of concentration for the wa-
tershed. The time steps with non-zero rainfalls, lagged by the time of concen-
tration for the catchment, are classified as driven. Of the remaining (non-
driven) time steps, those with streamflow lower than a certain threshold value
(e.g., mean of the square-root of the flows) are classified as non-driven-slow,
and the rest are classified as non-driven-quick. The model performance during
these three periods (QD, QQ, and QS) is estimated by calculating the RMSE
(FD, FQ, FS) separately over each period.

3.5.7 RMSE Segmentation B – OF_9_FDB, OF_10_FQB, OF_11_FSB

Segmentation A is difficult to apply when rainfall is occurring very frequently
since it results in a very scattered segmentation. Scheme B is therefore pro-
posed (see Figure 17). This scheme is only based on the flow. Initial attempts
to introduce a threshold below which rainfall is neglected in the sense of
scheme A were not satisfactory.

48
 RRMT User Manual

Example Segmentation Scheme B

2

1.8

1.6

1.4

1.2
streamflow

1st derivative
1 zero line derivative
Driven
0.8
Non-driven Quick
0.6
Non-driven Slow

0.4

0.2

5 10 15 20 25 30 35 40 45 50 55
time step

Figure 17 Example of a segmented time-series using scheme B. High and low driven flows
are not separated in this figure though.

Driven flow (FD) occurs in this scheme when the observed flow is increasing,
plus an additional time-step to include the peaks, while time steps are classi-
fied as non-driven flow (FQ and FS) when the flow is decreasing. The split be-
tween FQ and FS is done using a threshold, here the mean flow. The classifi-
cation is based on the gradient (using the gradient command in Matlab) of the
observed streamflow hydrograph. The RMSE for each period is calculated.
OF_17_FDBL and OF_18_FDBH allow a further division of the driven re-
sponse during periods of high and of low flows. The division is based on a
threshold equal to half of the mean flow.

3.5.8 RMSE High Flows – OF_12_FH

Only (observed) flow values above a selected threshold are considered to cal-
culate this measure. The threshold can be defined in the gui_main window.

3.5.9 RMSE Low Flows – OF_13_FL

Only (observed) flow values below a selected threshold are considered to cal-
culate this measure. The threshold can be defined in the gui_main window.

3.5.10 RMSE Medium Flows – OF_14_FM

Only (observed) flow values between the selected thresholds are considered
to calculate this measure. The thresholds can be defined in the gui_main win-
dow.

3.5.11 RMSE Warm Up Period – OF_15_FWU

Only (observed) flow values during the first X% of the total calibration period
(defined in % warm up field as a fraction, i.e. %/100) are considered to calcu-
late this measure. It can be used to calculate initial values.

49
 RRMT User Manual

3.5.12 Deviation of Runoff Volumes – OF_16_DRV

Another possible OF is the comparison of calculated and observed flow vol-
umes. This measure has an optimum at 1.
N

∑ c (θ )
i =1
i
DRV = N

∑oi =1
i

3.5.13 RMSE Segmentation B – OF_17_FDBL and OF_18_FDBH

Further division of the driven periods with respect to the selected threshold
value as described earlier.

3.5.14 Box-Cox Transformation

This transformation is useful to reduce the influence of larger errors, usually
related to higher flows, on some OFs. If the selected OF uses squared residu-
als, e.g. NSE or RMSE, then the resulting fit will often be biased towards high
flows. One can therefore transform the data to gain a more balanced fit, which
is done in a smarter way in the HMLE. The effect of the transformation will
also be visible in the residuals, the effect of increasing variance with increas-
ing flow values, i.e. the so-called heteroscedascity, can often be reduced. The
equation of the Box-Cox transformation is (Kottegoda and Rosso, 1997,
p.381)

⎧ yλ −1
⎪ for λ ≠ 0
y (λ ) = ⎨ λ
⎪⎩ ln( y ) for λ = 0

A suitable value for λ streamflow time-series is 0.3 (Hoshin Gupta, University

of Arizona, Personal Communication). A logarithmic transformation will be
performed in case of the RRMT if the user selects a 0, and a square-root
transformation in case a 2 is selected.

3.6 OFF-LINE (OL) FUNCTIONS

3.6.1 OF_FLOWCONVERT
This function converts flow from one unit into another.
% DATA MANIPULATION/ANALYSIS MODULE: converts flowdata to different units
%
% function outflow = flowconvert(inflow,iunit,ounit,area)
%
% INPUT
% inflow : flow data in the format avaliable
% iunit : unit of the input data
% ounit : unit of the output data
% area : catchment area [km2]
%
% OUTPUT
% outflow: flow data in the unit required
%
% Possible units: m3/s [1], mm/d [2], mm/h [3]

50
 RRMT User Manual

3.6.2 OL_SUBAVER
This function converts data from a smaller time step to a larger one by calcu-
lating averages.
% DATA MANIPULATION/ANALYSIS: sub-samples flow
% data with mean
% value
%
% function y = subaver(u,dt)
%
% INPUT
% u data vector
% dt sub-sampling interval
%
% OUTPUT
% y new data vector

3.6.3 OL_SUBSUM
This function calculates the sum of data over a defined period.
% DATA MANIPULATION/ANALYSIS: sub summation
%
% function y = subsum(u,dt)
%
% INPUT
% u data vector
% dt subsampling interval
%
% OUTPUT
% y new data vector

51
 RRMT User Manual

4 EXPERIMENTS

4
EXPERIMENTS

52
 RRMT User Manual

Figure 18 Rainfall-runoff experiment results.

4.1 PENMAN MODEL – SMA_PEN.M

Jolley (1995) conducted two experiments to investigate the differences in
model response due to changes in the parameter values. These experiments
are repeated here.

RAINFALL-RUNOFF SENSITIVITY EXPERIMENT

The model was run with different rainfall input values over a period of one day
to show the effect of different storage sizes on the model response. The input
data are given in table 1. Figure 13 shows the results of the calculations.

Table 9 Parameters and fluxes used in the RR experiment.

PARAMETER/FLUX VALUE COMMENT

R 1 to 100 mmd-1 Rainfall

PE 5 mmd-1 Potential ET
Smax1 100 mm Size upper store
Smax2 1000 mm Size lower store
φ 0.15 Bypass parameter
g 0.08 Drying curve parameter
D1 1 to 100 mm Initial deficit upper store
D2 0 mm Initial deficit lower store

53
 RRMT User Manual

Figure 19 ET experiment results.

HER (ER, Effective Rainfall) is not generated until R (Rainfall) exceeds PE

(Potential Evapotranspiration) or the AWC (Available Water Capacity = Smax1
+ Smax2) has been satisfied. While the AWC is not satisfied, HER is only gen-
erated by the diversion mechanism and HER asymptotically approaches φ R
as R increases. As the initial deficit increases, the threshold value of R, at
which AWC is satisfied, increases. Once the AWC has been satisfied, HER
increases rapidly and asymptotically approaches R as the initial deficit repre-
sents an ever decreasing proportion of R (Jolley, 1995, 404).

EVAPOTRANSPIRATION EXPERIMENT
Jolley (1995, p404) run the Penman model using constant input time series
and a range of Smax1 values during the second experiment. The experiment
shows the model response during dry periods.

Table 10 Parameter and fluxes used in the ET experiment.

PARAMETER/FLUX VALUE COMMENT

R 0 mmd-1 Rainfall
PE 10 mmd-1 Potential ET
Smax1 0 to 100 mm Size upper store
Smax2 1000 mm Size lower store
φ 0.15 Bypass parameter
g 0.08 Drying curve parameter
D1 50 mm Initial deficit upper store
D2 0 mm Initial deficit lower store

54
 RRMT User Manual

Figure 20 Pareto distribution of storage capacity. The probability density function is

shown at the top, the distribution function at the bottom.

The curve of evapotranspiration efficiency (AE divided by PE) against Smax1

(Figure 14) reflects the threshold effect of Smax1. Smax1 separates the soil
moisture limited condition from the freely available soil moisture condition. If
Smax1 is less than the D1 then evapotranspiration continues at the potential
rate and the efficiency equals unity. In between these two conditions
evapotranspiration and efficiency increase linearly with Smax1. There is an
upper limit to AE during a single step imposed by the available soil moisture
which is not illustrated in Figure 14 (Jolley, 1995, p406).

4.2 PROBABILITY DISTRIBUTED MOODEL – SMA-PDx

The effect of different shape parameters, b, on the Pareto distribution is

shown in Figure 16 (see also Moore, 1999).

55
 RRMT User Manual

n = 1 (linear)
n = 0.5
n = 1.5
input

Figure 21 Response of linear reservoir to a unit impulse using different residence times (or
time constants) is shown at the top. The influence of different values for the non-linearity
coefficient n on the response is visualised at the bottom.

4.3 CONCEPTUAL LINEAR RESERVOIR – R-CRES

The effect of using different residence times for a linear store can be seen in
Figure 17(top). A unit-input signal is routed through the reservoir. A larger
value of k leads to a delayed release of the input from the reservoir.
The influence of different values for the non-linearity coefficient n is demon-
strated in Figure 17(bottom).

4.4 TWO CONCEPTUAL LINEAR RESERVOIRS IN PARALLEL – R-2PAR

Two linear stores in parallel are often used to simulate different response
times of a system. The quick response in a catchment can usually be associ-
ated with hydrological processes like surface runoff or near surface

56
 RRMT User Manual

Figure 22 Response of a parallel linear reservoir system to a unit impulse using different
combinations of residence times.

flows (interflow), while the slow response can be associated with the subsur-
face or groundwater (baseflow) in the hydrological system.
Figure 18 shows the response of two parallel linear reservoirs to a unit im-
pulse, using different residence times.

PARAMETER VALUES
Top Middle Bottom
Residence time quick response 0.1 1.0 4.0
Residence time slow response 5.0 25.0 50.0
% flow through quick response 0.5 0.5 0.5

57
 RRMT User Manual

5 REFERENCES

5
REFERENCES

58
 RRMT User Manual

Bentura P.L.F. and Michel C. 1997. Flood Routing in a Wide Channel with a Quadratic Lag-
and-Route Method. Hydrological Sciences Journal, 42(2), 169-189.
Beven, K.J. 1998. Generalised Likelihood Uncertainty Estimation (GLUE) package v. 1.0 –
User manual. University of Lancaster, UK.
Beven K.J. and Binley A. 1992. The future of distributed models: Model calibration and uncer-
tainty prediction. Hydrological Processes, 6, 279-298.
Brazil, L.E. 1988. Multilevel calibration strategy for complex hydrologic simulation models.
PhD Dissertation, Colorado State University, Fort Collins, Colorado.
Camacho L.A. 1997. Extension of the QUASAR Water Quality Model to Incorporate Dead
Zone Mixing Concepts. MSc Thesis, Civil and Environmental Engineering Depart-
ment, Imperial College of Science, Technology and Medicine, London, unpublished.
Chow Ven Te, Maidment D.R. and Mays L.W. 1988. Applied Hydrology. McGraw-Hill, USA.
Clarke R.T. 1994. Statistical Modelling in Hydrology. John Wiley & Sons Ltd., UK.
Duan Q., Gupta V.K. and Sorooshian S. 1992. Effective and efficient global optimization for
conceptual rainfall-runoff models. Water Resources Research, 28(4), 1015-1031.
Duan, Q., Gupta V.K. and Sorooshian S. 1993. A Shuffled Complex Evolution Approach for
Effective and Efficient Global Minimization. Journal of Optimization Theory and Ap-
plications, 76(3), 501-521.
Duan, Q., Sorooshian, S. And Gupta, V.K. 1994. Optimal use of the SCE-UA global optimisa-
tion method for calibrating catchment models. Journal of Hydrology, 158, 265-284.
Etter D.M. 1997. Engineering problem solving with Matlab - 2nd Edition. Prentice-Hall, USA
Evans J.P. and Jakeman A.J. 1998. Development of a simple, catchment-scale, rainfall-
evapotranspiration-runoff model. Environmental Modelling & Software. To be pub-
lished
Evans, J.P., A.J. Jakeman, 1998. Development of a Simple, Catchment-scale, Rainfall-
evapotranspiration-runoff Model. Environmental Modelling & Software. To be pub-
lished
Forsythe G.E., Malcolm M.A. and Moler C.B. 1976. Computer Methods for Mathematical
Computations. Prentice-Hall, USA
Gan T.Y. and Biftu G.F. 1996. Automatic Calibration of Conceptual Rainfall-runoff Models:
Optimization Algorithms, Catchment Conditions, and Model Structure. Water Re-
sources Research, 32(12), 3513-3524.
Garrick M., Cunnane C. and Nash J.E. 1978. A criterion of efficiency for rainfall-runoff models.
J. Hydrol., 36, 375-381.
Goldberg D.E. 1989. Genetic Algorithms in Search, Optimization, and Machine Learning. Ad-
dison-Wesley Publishing Company, Inc., USA
Green C.S. 1979. An improved Subcatchment Model for the River Dee. Institute of Hydrology,
Wallingford, Rep. No. 58.
Hanselman D. and Littlefield B. 1998. Mastering matlab 5 - A comprehensive tutorial and ref-
erence. Prentice Hall, USA
Hughes D.A. and Murrell H.C. 1986. Non-linear Runoff Routing – A Comparison of Solution
Methods. J Hydrol., 85, 339-347.
Jakeman A.J. and Hornberger G.M. 1993. How much complexity is warranted in a rainfall-
runoff model? Water Resources Research, 29(8), 2637-2649.
Jakeman A.J., Littlewood I.G. and Whitehead P.G. 1990. Computation of the instantaneous
unit hydrograph and identifiable component flows with application to two small up-
land catchments. J Hydrol., 117, 275-300.
Jakeman A.J., Chen T.H., Post D.A., Hornberger G.M., Littlewood I.G. and Whitehead P.G.
1993. Assessing Uncertainties in Hydrological Response to Climate at Large Scale.
Macroscale Modelling of the Hydrosphere. Proceedings of the Yokohama Sympo-
sium, IAHS Publication, 214, 37-47.
Jolley T.J. 1995. Large-scale hydrological modelling – The development and validation of im-
proved land-surface parameterisations for meteorological input. Ph.D. Dissertation,
Imperial College of Science, Technology and Medicine. London, UK, Unpublished
Kleissen F.M., Beck M.B. and Wheater H.S. 1990. Identifiability of conceptual hydrochemical
models. Water Resources Research, 26(2), 2979-2992.
Laurenson E.M. 1964. A Catchment Storage Model for Runoff Routing. J Hydrol., 2, 141-163.
Lees M.J., Price N., Wheater H.S. and Peach D. 1998. A Rainfall-runoff Simulation Model for
South Galway, Ireland. In BHS 1998. Hydrology in a changing environment – Vol-
ume III, 93-104.

59
 RRMT User Manual

Legates D.R. and McCabe G.J. Jr. 1999. Evaluating the use of "goodness-of-fit" measures in
hydrologic and hydroclimatic model validation. Water Resources Research, 35(1),
233-241.
Littlewood I.G. and Jakeman A.J. 1992. Methods of Hydrological Basin Comparison. Robin-
son M. (ed.). Methods of Hydrological Basin Comparison. Proceedings of the fourth
conference of the European network of experimental and representative basins, re-
port no. 120, 99-111.
Luce C.H. and Cundy T.W. 1994. Parameter Identification for a Runoff Model for Forest
Roads. Water Resources Research, 30(4), 1057-1069.
Mander and Greenfiled (1978) - see Jolley (1995)
Marsh T.J. 1999. Long River Flow Records. In Herschy R.W. (ed.). Hydrometry: Principles
and Practice. John Wiley & Sons Ltd., 161-180.
McCuen R.H. and Snyder W.M. 1986. Hydrologic modelling - Statistical methods and applica-
tions. Prentice-Hall, USA
Mein R.G., Laurenson E.M. and McMahon T.A. 1974. Simple Non-linear Model for Flood Es-
timation. J. Hydraul. Div., Am. Soc. Civ. Eng., 100(HY11), 1507-1518.
Moore, R.J. 1999. Real-time flood forecasting systems: Perspectives and prospects. In
Casale R. and Margottini C. (eds.). Floods and Landslides: Integrated risk assess-
ment. Springer-Verlag, Berlin, p. 147-189.
Moore R.J. 1992. Hydrological models - Nonlinear storage models. International Course for
Hydrologists, International Institute for Hydraulic and Environmental Engineering,
Institute of Hydrology, Wallingford, UK
N.E.R.C. 1979. Flood Studies Report. Vols. I – V, National Environment Research Council,
London, UK.
Nash J.E. and Sutcliffe J.V. 1970. River flow forecasting through conceptual models. I. A dis-
cussion of principles. J. Hydrol., 10, 282-290.
Penman H.L. 1949. The dependence of transpiration on weather and soil conditions. J Soil
Science, 1, 74-89
Rientjes T.H.M. and Boekelman R.H. 1998. Hydrological models. Lecture Notes, Civil Engi-
neering Department, Delft University of Technology, The Netherlands, Unpublished.
Rodda J.C., Downing R.A. and Law F.M. 1976. Systematic Hydrology. Newness-
Butterworths, London, UK
Sherratt D.J. 1985. Regional soil moisture modelling. Ph.D. Dissertation, Civil Engineering
Department, Imperial College of Science, Technology and Medicine, UK
Sherratt D.J. and Wheater H.S. 1984. The use of surface resistance - soil moisture relation-
ships in soil water budget models. Agricultural and Forest Meteorology, 31, 143-157
Sorooshian, S., Gupta, H.V. and Bastidas, L.A. 1998. Calibration of hydrologic models using
multi-objectives and visualisation techniques. Final Report on Project Proposal No.
EAR-9418147, University of Arizona, USA.
Sorooshian S. and Gupta V.K. 1995. Chapter 2 - Model Calibration. In Singh V.P. (ed.). Com-
puter Models for Watershed Hydrology. Water Resources Publications, USA, 23-68.
Sorooshian S., Duan Q. and Gupta V.K. 1993. Calibration of Rainfall-runoff Models: Applica-
tion of Global Optimization to the Sacramento Soil Moisture Accounting Model. Wa-
ter Resources Research, 29(4), 1185-1194.
Tanakamara H. 1995. Parameter Estimation for the Tank Model using Global Optimization.
Transactions of the JSIDRE, 178, 103-112.
Tanakamara H. and Burges S.J. 1996. Application of Global Optimization to Parameter Esti-
mation of the Tank Model. In: Proceedings of the International Conference on Water
Resources and Environmental Research: Towards the 21st Century, Kyoto, Japan,
29-31 October 1996, 2, 39-46.
Wagener, T., Lees, M.J. and Wheater, H.S. 2001a. A framework for the development and ap-
plication of parsimonious hydrological models. To appear in Singh, Frevert and
Meyer (Eds.) Mathematical models of small watershed hydrology – Volume 2. Wat.
Resour. Publ. LLC, USA.
Wagener T., Boyle D.P., Lees M.J., Wheater H.S., Gupta H.V., Sorooshian S. 2001. A
framework for the development and application of hydrological models. Hydrology
Earth System Sciences, 5(1), 13-26.
Wagener T., McIntyre N., Lees M.J., Wheater H.S. and Gupta H.V. 2003. Towards reduced
uncertainty in conceptual rainfall-runoff modelling: Dynamic identifiability analysis.
Hydro-logical Processes , 17, 2, 455-476.

60
 RRMT User Manual

Wagener T., Lees M.J., Wheater H.S. 1999. A generic rainfall-runoff modelling toolbox. Eos.
Trans. AGU, 80(17), Fall Meet. Suppl., F203.Wheater H.S., Jakeman A.J. and
Beven K.J. 1993. Chapter 5 - Progress and directions in rainfall-runoff modelling. In
Jakeman A.J., Beck M.B. and McAleer M.J. 1993. Modelling change in environ-
mental systems. John Wiley & Sons Ltd, UK, 101-132.
Wilcox B.P., Rawls W.J., Brakensiek D.L. and Wright J.R. 1990. Predicting runoff from range-
land catchments: A comparison of two models. Water Resources Research, 26,
2401-2410.
Ye W., Bates B.C., Viney N.R., Sivapalan M., Jakeman A.J. 1997. Performance of conceptual
rainfall-runoff models in low-yielding ephemeral catchments. Water Resources Re-
search, 33(1), 153-166.
Ye, W., A.J. Jakeman, P.C. Young, 1998. Identification of improved Rainfall-runoff Models for
an Ephemeral Low-yielding Australian Catchment. Environmental Modelling & Soft-
ware, 13, 59-74.
Young P.C. 1992. Parallel Processes in Hydrology and Water Quality: A Unified Time-Series
Approach. J. IWEM, 6, 598-612.
Young P.C. 1985. The instrumental variable method: A practical approach to identification
and system parameter estimation. In Barker H.A. and Young P.C. (ed.). Identifica-
tion and system parameter estimation. 7th IFAC/IFORS Symposium, York, UK, 3-7
July 1985, 1-15.
Young, P.C., 1984. Recursive Estimation and Time-Series Analysis – An Introduction.
Springer-Verlag, New York.
Young P.C. and Beven K.J. 1991. Computation of the instantaneous unit hydrograph and
identifiable component flows to two small upland catchments. Comment. J Hydrol-
ogy, 126, 389-396.
Young, P.C., K.J. Beven, 1990. Comments on Computation of the Instantaneous Unit Hydro-
graph and Identifiable Component Flows with Application to two Small Upland
Catchments by A.J. Jakeman et al. Journal of Hydrology, 117, 275-300.
Young, P.C., K.J. Beven, 1994. Data-based Mechanistic Modelling and the Rainfall-flow Non-
linearity. Environmetrics, Vol. 5, 335-363.

61