Sie sind auf Seite 1von 18

HYDROLOGIC MODEL SELECTION

Upper Thames River


Tier 2 Water Quantity Stress Assessment

May 9, 2011
7297-R1

CONTENTS

1.

INTRODUCTION .............................................................................................................. 1

2.

OVERVIEW OF MODELS ................................................................................................ 2

2.1. Model descriptions ..................................................................................................... 2


SSARR ................................................................................................................................. 2
SLURP ................................................................................................................................. 2
WATFLOOD ......................................................................................................................... 3
MIKE SHE ............................................................................................................................ 4
HEC-HMS ............................................................................................................................ 6
HSPF ................................................................................................................................... 8
GAWSER ............................................................................................................................. 9
HBV ...................................................................................................................................... 9
SWAT ................................................................................................................................. 11
2.2.

Previous model comparisons .................................................................................. 11

2.3.

Discussion ................................................................................................................. 12

3.

MODEL EVALUATION .................................................................................................. 14

3.1.

Design criteria ........................................................................................................... 14

3.2.

Evaluation scores ..................................................................................................... 15

REFERENCES ...................................................................................................................... 16

7-297

1. INTRODUCTION
Schlumberger Water Services (SWS) has initiated work on the Thames-Sydenham and
Region Tier 2 Water Budget under a contract issued in April 2009. Selection of a surface
water model is a key step in the assessment.
This report provides a brief overview of potential models in Section 2 based on a list provided
by the client plus other surface water models that are widely used. A discussion of previous
model comparisons is provided together with a discussion on the appropriate level of
complexity for a surface water model.
Evaluation criteria are presented together with scores for each model based on the
evaluation criteria. Final criteria, scores and weightings were assigned during a meeting with
the Thames-Sydenham Region on April 30, 2009.

7-297

2. OVERVIEW OF MODELS

2.1.

Model Descriptions

This section of the report provides an overview of potential models that could be applied for
the Water Budget Tier 2. These are models that have been used in Ontario for similar
applications or models that have been used worldwide.

SSARR
The Streamflow Synthesis and Reservoir Regulation (SSARR) Model was developed to
provide mathematical hydrologic simulations for systems analysis as required for the
planning, design, and operation of water control works. The SSARR Model has been further
developed for operational river forecasting and river management activities. As a general
purpose mathematical model of a river system, the SSARR Model is a useful tool for
streamflow and runoff forecasting, as well as for long term studies of the hydrology of a river
system. Numerous river systems in the United States and abroad have been modelled with
the SSARR Model by various agencies, organizations, and universities.
The SSARR model comprises a generalized watershed model and a streamflow and
reservoir regulation model. The model has not been updated for more than a decade and no
user support is available.

SLURP
SLURP (Semi-distributed Land Use-based Runoff Processes), is a basin model developed at
the National Hydrology Research Institute (NHRI) in Saskatoon. SLURP simulates the
hydrological cycle from precipitation to runoff including the effects of reservoirs, dams,
regulators, water extractions and irrigation schemes that has been referred to in more than
25 papers in international journals by different authors. The model may be used to examine
the effects of proposed changes in water management within a basin or to see what effects
external factors such as climate change or changing land cover might have on various water
users. The model may use locally-available climate data or may be run using only publicdomain data sets available on the internet. Satellite images may be used in the model for

7-297

Overview of models

land cover mapping, vegetation indices (for leaf area index and for evapotranspiration
calculations), cloud cover (to distribute precipitation), snow extent and snow water
equivalent. The model is supplied with a 200-page manual in pdf format and sample data
sets on a CD-ROM. The model makes full use of Windows capabilities with graphical data
inputs and graphical displays of outputs. These include the daily components of the
hydrological cycle and generated runoff.
The model initially divides a basin into hydrological sub-basins and then divides each subbasin into its component land covers using public-domain topographic analysis software.
The model then simulates the vertical water balance at each element of the sub-basin/land
cover matrix using daily climate data (Figure 2). The climate data may be obtained from
ground stations, from the outputs of global circulation models or numerical weather prediction
models or from on-line datasets such as the GSOD global climate data or the IWMI World
Water and Climate Atlas. Runoffs from each matrix element are then routed through each
sub-basin to the basin outlet taking account of reservoir regulation, diversions, groundwater
extractions and water exports from the basin
The latest version of SLURP incorporates the TOPAZ and SLURPAZ topographic analysis
programs (written by the USGS/University of Saskatchewan). This system derives the
topographic and land cover information needed by SLURP in a form that can be directly used
and represents a major saving in time and effort when compared to using a GIS.
SLURP allows the use of both surface water and groundwater for furrow or sprinkler irrigation
as well as river and groundwater extractions for urban or industrial use at any part of the
basin. Because SLURP is a distributed model, it can produce maps of basin-wide crop
transpiration, soil evaporation and net runoff for use in performance evaluations of irrigation
schemes or to investigate water availability.
The effects of changes in atmospheric CO2 on plant water use efficiency may be included
when evaluating the effects of climate change scenarios on water resources and on the
various water users. Similarly, the areas of different land covers may be changed any time
during a model run to allow investigation of the effects on water resources of changing both
land cover and climate over time.

WATFLOOD
WATFLOOD is an integrated set of computer programs aimed at flood forecasting and longterm hydrologic simulation using distributed precipitation data from radar or numerical
weather models. Continuous simulation can be carried out by chaining up to 100 events. A
shell program organizes all the menus used for data input and correction functions and acts
as the manager for the hydrologic model SPL and utility programs. The emphasis of the
WATFLOOD system is on making optimal use of remotely sensed data. Radar rainfall data,
LANDSAT or SPOT land use and/or land cover data can thus be directly incorporated in the
hydrologic modelling.
Compared to other hydrological modelling systems, WATFLOOD can be considered as a
data intensive system. The initial development stages of the system (1972-75) coincided with
the advent of remotely sensed data to provide land cover and meteorological data. It was
immediately apparent that such data could enhance hydrological modelling, but the models
that existed at the time could not be easily adapted to take advantage of the high spatial and

7-297

Overview of models

temporal resolution. From the onset, WATFLOOD was designed to incorporate the remotely
sensed data in an efficient manner. This led to the gridded format of all data used by
WATFLOOD, including discretization of the watershed itself. Thus, in WATFLOOD, each
"sub watershed" or grid cell is more or less the same size. Exceptions occur along basin and
sub- basin boundaries to ensure drainage areas at streamflow gauges are preserved
WATFLOOD preserves the distributed nature of a watershed hydrologic and meteorological
variability without sacrificing computational efficiency. This has been accomplished through
the use of Grouped Response Units, in which process parameters are tied to land cover, and
land cover mixes can vary from basin element to basin element.
As with all hydrological processes, the melting of snow is treated separately for each land
cover class. Snow cover depletion curves (SDC) are used to summarize the relation between
snow cover distribution and an average snow cover property, such as depth of water
equivalent for a given area. More specifically, these curves provide the amount of snow
covered area for a given depth of water equivalent for each land cover class. This is an
important factor because snow can only melt where it exists, not on the bare areas.
Also included in the model is a snow redistribution algorithm. An upper limit is set on the
water equivalent for each land cover class. For forests and glaciers this is very high, but for
barren areas and wetlands it is quite low. When falling snow increases the snow water
content beyond the limit for a class, the snow is redistributed to the nearest forest.
The snow melt process is modelled by an index method; in this case, a degree-hour based
heat input or loss. An accounting of the heat content of the snow pack allows re-freezing.
Baseflow is calculated by a two parameter, non-linear storage-discharge function. The
ground water reservoir (or lower zone storage LZS) is common to all GRUs within a
computational grid. The initial baseflow is determined from the first value of a measured
hydrograph at the basin outlet, using the non-linear storage-discharge function in reverse.
This initial baseflow contributed by each grid is found by prorating it to the total basin area of
the gauged watershed in which the grid is located. For ungauged areas, the average initial
flow per unit area of nearby streams is used.
WATFLOOD has been used with grid sizes from 1 to 25 km and for watershed areas from 15
to 1,700,000 km2. WATFLOOD is not sensitive to grid size as long as there are a sufficient
number of grids to maintain the integrity of the drainage system and preserve the variability
in the meteorological data. The system is completely modular but has a consistent data
structure throughout and has been under continuous development since 1972.
WATFLOOD provided a GUI interface that is no longer supported and is now only available
as a DOS program. The model structure is comparatively difficult to learn.

MIKE SHE
MIKE SHE includes all of the processes in the land phase of the hydrologic cycle:

Precipitation (rain or snow)

Evapotranspiration, including canopy interception

7-297

Overview of models

Overland sheet flow

Channel flow

Unsaturated sub-surface flow

Saturated groundwater flow

Within each of these, MIKE SHE offers several different approaches ranging from simple,
lumped and conceptual approaches to advanced, distributed and physically-based
approaches. For modelling the unsaturated zone, MIKE SHE calculates infiltration, actual
evapotranspiration and recharge. Input can be specified as constant values or time series
and can be distributed in space using stations (for instance, Thiessen polygons) or as cellby-cell values. The time series for each station can contain different time periods with
different, non-equidistant time steps. When running the simulation, MIKE SHE automatically
synchronizes all time steps to the time step adopted in the simulation.
The unsaturated zone is the link between surface water and groundwater. The unsaturated
zone model in MIKE SHE is a vertical soil profile model that interacts with both the overland
flow (through ponding) and the groundwater model (the groundwater table is the lower
boundary condition for the unsaturated zone). MIKE SHE offers three different approaches,
including a simple 2-layer root-zone mass balance approach, a gravity flow model and a full
Richards equation model. All three approaches require specification of certain soil properties.
The unsaturated zone model interacts with MIKE SHE's evapotranspiration model which
calculates actual evapotranspiration as a function of reference ET, soil moisture and crop
characteristics.
MIKE SHE includes a database of various soils and typical crops for different climatic regions
of the world. The soil database includes a number of transfer functions that link suction,
water content and hydraulic conductivity, but also allows the user to specify soil properties in
a tabular format. Soil profiles are distributed in space using soil maps.
Land use is described in terms of vegetation types combined with impervious/semiimpervious areas. Vegetation types and impervious/semi-impervious areas may be spatially
distributed using land-use maps. Crop rotation is made easy by defining a date-ofestablishment and then specifying a link to the crop type in the crop database. MIKE SHE
comes with a database of typical crops in different climatic regions. Users may edit the
database, add vegetation types to the database or create a new database from scratch.
MIKE SHE includes a traditional 2D or 3D finite-difference groundwater model, which is very
similar to MODFLOW. The geology is described in terms of layers or lenses with attached
hydraulic properties. Properties can be specified either on a cell-by-cell basis or by property
zones defined by polygons or grid-code files. MIKE SHE allows grid-independent geology
specification, which allows you to change the horizontal or vertical mesh in seconds.
Boundary conditions are specified for each computational layer. MIKE SHE supports
traditional groundwater boundary conditions and offers large flexibility in terms of spatial and
temporal variation of boundary conditions. Boundary conditions may be specified on a cellby-cell basis; but, typically, it is more convenient to attach boundary conditions to geometric
features such as polygons (lakes), lines (rivers) or points (pumps, injections, drains). A lake
could, for instance, be a polygon (.shp file) with an attached water level time series and
leakage coefficient. Similar to the meteorological time-series data, boundary time-series data

7-297

Overview of models

may be specified in separate time-series files that include different and non-equidistant time
steps.
MIKE SHE's river modelling component is the MIKE 11 modelling system for river hydraulics.
MIKE 11 is a dynamic, 1D modelling tool for the design, management and operation of river
and channel systems. MIKE 11 supports any level of complexity and offers simulation
engines that cover the entire range from simple Muskingum routing to the Higher Order
Dynamic Wave formulation of the Saint-Venant equations. It is approved for use by
regulatory authorities in many countries including the US, Australia and UK. MIKE 11 is also
approved by the US Federal Emergency Management Agency (FEMA) for use on projects
related to the National Flood Insurance Program. MIKE 11 can be used in combination with
MIKE SHE or as a stand-alone hydraulic modelling system.
MIKE SHE's overland-flow component includes a 2D finite difference diffusive wave
approach using the same 2D mesh as the groundwater component. Overland flow interacts
with the river, the unsaturated zone, and saturated groundwater zone.

HEC-HMS
HEC-HMS is designed to simulate the precipitation-runoff processes of dendritic watershed
systems. It is the successor to HEC-1 and provides a similar variety of options, but
represents a significant advancement in terms of both computer science and hydrologic
engineering. In addition to unit hydrograph, hydrologic and reservoir routing options,
capabilities include a linear quasi-distributed runoff transform (ModClark) for use with gridded
precipitation, continuous simulation with either a one-layer or more complex five-layer soil
moisture method, and a versatile parameter estimation option. The software is designed for
interactive use in a multi-tasking, multi-user network environment.
The program is a generalized modelling system capable of representing many different
watersheds. A model of the watershed is constructed by separating the hydrologic cycle into
manageable pieces and constructing boundaries around the watershed of interest. Any mass
or energy flux in the cycle can then be represented with a mathematical model. In most
cases, several model choices are available for representing each flux. Each mathematical
model included in the program is suitable in different environments and under different
conditions. Making the correct choice requires knowledge of the watershed, the goals of the
hydrologic study, and engineering judgment.
The program features a completely integrated work environment including a database, data
entry utilities, computation engine, and results reporting tools. A graphical user interface
allows the seamless movement between the different parts of the program. Program
functionality and appearance are the same across all supported platforms
HEC-HMS provides a variety of options for simulating precipitation-runoff processes. In
addition to unit hydrograph and hydrologic routing options similar to those in HEC-1,
capabilities currently available include: a linear-distributed runoff transformation that can be
applied with gridded (e.g., radar) rainfall data, a simple "moisture depletion" option that can
be used for simulations over extended time periods, and a versatile parameter optimization
option. The latest version also has capabilities for continuous soil moisture accounting and
reservoir routing operations.

7-297

Overview of models

The physical representation of a watershed is accomplished with a basin model. Hydrologic


elements are connected in a dendritic network to simulate runoff processes. Available
elements are: sub-basin, reach, junction, reservoir, diversion, source, and sink. Computation
proceeds from upstream elements in a downstream direction.
An assortment of different methods is available to simulate infiltration losses. Options for
event modelling include initial constant, SCS curve number, gridded SCS curve number,
exponential, and Green Ampt. The one-layer deficit constant method can be used for simple
continuous modelling. The five-layer soil moisture accounting method can be used for
continuous modelling of complex infiltration and evapotranspiration environments. Gridded
methods are available for both the deficit constant and soil moisture accounting methods.
Several methods are included for transforming excess precipitation into surface runoff. Unit
hydrograph methods include the Clark, Snyder, and SCS techniques. User-specified unit
hydrograph or s-graph ordinates can also be used. The modified Clark method, ModClark, is
a linear quasi-distributed unit hydrograph method that can be used with gridded
meteorological data. An implementation of the kinematic wave method with multiple planes
and channels is also included.
Multiple methods are included for representing baseflow contributions to sub-basin outflow.
The recession method gives an exponentially decreasing baseflow from a single event or
multiple sequential events. The constant monthly method can work well for continuous
simulation. The linear reservoir method conserves mass by routing infiltrated precipitation to
the channel.
A variety of hydrologic routing methods are included for simulating flow in open channels.
Routing with no attenuation can be modelled with the lag method. The traditional Muskingum
method is included along with the straddle stagger method for simple approximations of
attenuation. The modified Puls method can be used to model a reach as a series of
cascading, level pools with a user-specified storage-discharge relationship. Channels with
trapezoidal, rectangular, triangular, or circular cross sections can be modelled with the
kinematic wave or Muskingum-Cunge methods. Channels with overbank areas can be
modelled with the Muskingum-Cunge method and an 8-point cross section.
The Geo-spatial Hydrologic Modelling Extension (HEC-GeoHMS) is a software package for
use with the ArcView Geographic Information System. GeoHMS uses ArcView and Spatial
Analyst to develop a number of hydrologic modelling inputs. Analyzing digital terrain
information, HEC-GeoHMS transforms the drainage paths and watershed boundaries into a
hydrologic data structure that represents the watershed response to precipitation. In addition
to the hydrologic data structure, capabilities include the development of: grid-based data for
linear quasi-distributed runoff transformation (ModClark), the HEC-HMS basin model,
physical watershed and stream characteristics, and background map file.
HEC-GeoHMS provides an integrated work environment with data management and
customized toolkit capabilities, which includes a graphical user interface with menus, tools,
and buttons. The program features terrain, pre-processing capabilities in both interactive and
batch modes. Additional interactive capabilities allow users to construct a hydrologic
schematic of the watershed at stream gauges, hydraulic structures, and other control points.
The hydrologic results from HEC-GeoHMS are then imported by the Hydrologic Modelling
System, HEC-HMS, where simulation is performed.

7-297

Overview of models

The snowmelt function in HEC-HMS has recently been added and has reportedly had errors.
It is, therefore, relatively untried. The method incorporates the Distributed Snow Process
Model (DSPM) also developed by the Corps. DSPM estimates snowmelt in any sub-area of
watershed using the SSARR grid snow process model. Sub-area is usually defined as a
spatial grid cell of arbitrary size, although the grid cells usually are 1 or 2 km2 in size. The
SSARR grid model was extracted from the original SSARR model using the Snow-Band
snowmelt computation, which estimates liquid water available at the soil surface for a subarea for one-time step. The model is a temperature index model that accounts for cold
content and liquid water content of the snow. The melt factor can be a constant, or a function
of the antecedent temperature or time of year.
Although the HEC-HMS model is very flexible, one drawback is limited capability for
calibration of climate inputs. Different weights can be applied to precipitation gauges but
climate input calibration is not part of the calibration routines in the model.

HSPF
The model was developed in the early 1960s as the Stanford Watershed Model. In the
1970s, water-quality processes were added. Development of a Fortran version incorporating
several related models using software engineering design and development concepts was
funded by the Athens, Ga., Research Lab of EPA in the late 1970s. In the 1980s,
preprocessing and post processing software, algorithm enhancements, and use of the USGS
WDM system were developed jointly by the USGS and EPA. The current release is Version
11. An interactive version (HSPEXP) was developed by the USGS in the 1990s.
HSPF uses continuous rainfall and other meteorological records to compute streamflow
hydrographs. HSPF simulates interception soil moisture, surface runoff, interflow, baseflow,
snowpack depth and water content, snowmelt, evapotranspiration, ground-water recharge,
dissolved oxygen, biochemical oxygen demand (BOD), temperature, pesticides,
conservatives, fecal coliforms, sediment detachment and transport, sediment routing by
particle size, channel routing, reservoir routing, constituent routing, pH, ammonia, nitritenitrate, organic nitrogen, orthophosphate, organic phosphorus, phytoplankton, and
zooplankton. The program can simulate one or many pervious or impervious unit areas
discharging to one or many river reaches or reservoirs. Frequency-duration analysis can be
done for any time series. Any time step from one minute to one day that divides equally into
one day can be used. Any period from a few minutes to hundreds of years may be
simulated. HSPF is generally used to assess the effects of land-use change, reservoir
operations, point or nonpoint source treatment alternatives, flow diversions, etc. Programs,
available separately, support data preprocessing and post-processing for statistical and
graphical analysis of data saved to the Watershed Data Management (WDM) file.
Meteorological records of precipitation and estimates of potential evapotranspiration are
required for watershed simulation. Air temperature, dewpoint temperature, wind, and solar
radiation are required for snowmelt. Air temperature, wind, solar radiation, humidity, cloud
cover, tillage practices, point sources, and (or) pesticide applications may be required for
water-quality simulation. Physical measurements and related parameters are required to
describe the land area, channels, and reservoirs.
Values of a large number of HSPF parameters can not be obtained from field data and need
to be determined through the model calibration exercise. However, many of these

7-297

Overview of models

parameters were conceived to index properties of specific factors that influence events such
as water storage and fluxes in the land phase of the hydrologic cycle. A Graphical User
Interface (GUI) is available. WinHSPF was created for the US Environmental Protection
Agencys BASINS system.

GAWSER
The GAWSER (Guelph All-Weather Sequential-Events Runoff) model, is a deterministic
watershed modelling package based on the HYMO format. HYMO was developed by the US
Department of Agriculture in the 1970s and has been widely applied for runoff modelling.
The University of Guelph modified the HYMO model to run in continuous mode.
GAWSER has been applied widely in Ontario for planning, design, real-time flood
forecasting, and evaluating the effects of physical changes in the drainage basin. GAWSER
was originally set up for the entire Grand River watershed in 1987-1988, and has since
formed the heart of the Grand River Conservation Authoritys (GRCA) real-time flood
forecasting system. Within the last 18 years, it has been subsequently updated for
continuous water balance simulations, sediment transport, instream temperature predictions,
and climate change impact assessment.
GAWSER has been extensively calibrated, verified and validated in more than 40 watershed
modelling studies within the last 20 years for Ontario watersheds. Calibration parameters for
typical Ontario conditions are available from other studies. Recent modifications have been
made to tailor the model to Ontario conditions and requirements.
Each sub-catchment element in GAWSER is considered to be a total self-contained
hydrologic unit. This means that all the precipitation falling on a given sub-catchment is
accounted for in the computations. Infiltrated water, once the evapotranspiration amounts are
removed, can return as baseflow, so that the total outflow becomes the sum of computed
runoff, subsurface and baseflow components. Although this is an idealized situation that
facilitates hydrograph calculations, it is not always true in nature; infiltrated water may
reappear as baseflow at some other point downstream in the watershed, or flow to another
watercourse entirely. GAWSER has been modified so that infiltrated water in the normal
runoff computations (which includes recharge from detention ponds) or seeping from channel
reaches during low flow periods could be re-directed to a groundwater storage array in
program memory. At some other point downstream in the drainage network, water can be
released from this storage array.
The model structure is based on a series of logical commands which makes the program
relatively easy to learn. The program runs on a DOS interface which required input
commands to be set up in a file using a DOS or text editor. Output files are text files which
can be imported into Excel.

HBV
The HBV model is a conceptual hydrological model designed for use in mountainous
environments. It has a long history and has been used in more than 30 countries. The
original HBV model was developed at the Swedish Meteorological and Hydrological Institute
(SMHI). The model has been significantly modified over time, but the basic modelling

7-297

Overview of models

10

philosophy has been unchanged. This philosophy acknowledges that the model complexity
and data requirements must not be in conflict with the operational requirements. In summary:

The model shall be based on a sound scientific foundation.


Data demands must be met in typical basins.
The model complexity must be justified by model performance.
The model must be properly validated.
The model must be understandable by users.

Today, many versions of the HBV model exist and new codes are constantly being
developed by different groups. There is, in a sense, no such thing as The HBV model, but
rather the HBV modelling philosophy. The model structure is very similar to the UBC
Watershed Model, but arguably has had more opportunities for improvements because of the
open source code.
It was originally developed by Sten Bergstrm in the early 1970s while working at the
Swedish Meteorological and Hydrological Institute (SMHI). Over the past 35 years, the model
has been used extensively for hydrological forecasting, engineering design, and climate
change studies. Bergstrm (1995) provides a complete description of the history and
application of the model as well as details on the basic internal routines.
A Canadian version of the HBV model has been maintained by Dr. Dan Moore (UBC) since
the mid-1980s. Moore (1993) developed and tested a glacier routine for the model. In 2000,
Dr. Moore provided the source code for the model (then written in Turbo Pascal) to
Environment Canada and current model development has since been co-managed by
Environment Canada and UBC. The Canadian version is referred to as the HBV-EC
("Environment Canada") model. It has been integrated with the ENSIM preprocessor
developed by the Canadian Hydraulics Centre (CHC) and has excellent watershed
development and processing tools using GIS data and Digital Elevation Models. ENSIM also
provides very effective access to Canadian climate and streamflow data within the modelling
environment. CHC provides support for ENSIM and HBV-EC.
The HBV model was modified to match changes specified in the Lindstrm et al. (1997)
paper. Climate zone representation was added to the HBV-EC model to better represent the
lateral climatic gradients which may occur across a basin. Each climate zone is associated
with a single climate station and unique parameter values for specifying the distribution of
climate within the zone, such as temperature and precipitation lapse rates. Runoff from a
climate zone is lumped through a series of fast and slow response reservoirs of similar
configuration to those in the traditional HBV model.
Within the traditional HBV model, the snow melt factor does not vary with respect to terrain
slope or aspect. In the HBV-EC model, the snow melt factor varies as a function of aspect
and slope. Snowmelt is calculated using a temperature index method.
A major weakness of the model is that it currently does not have an integrated channel
routing component, although a 1D channel routing module is available through the ENSIM
package. The HBV model is not normally set up for multiple sub-catchments and channel
linkages requiring routing.

7-297

Overview of models

11

SWAT
SWAT (Soil Water Assessment Tool) developed at the USDA is a physically based,
distributed parameter continuous simulation model that runs on daily time step. SWAT has
been used to predict, over long periods of time, the impact of land management practices on
water, sediment, and agricultural chemical loads in large complex watersheds with varying
soils, land use, and management conditions. Major model components describe processes
associated with water movement, sediment movement, soils, temperature, weather, plant
growth, nutrients, pesticides and land management. In each spatial sub-unit, water balance
is represented by several storage volumes; e.g., canopy storage, snow, soil profile, shallow
aquifer and deep aquifer. Surface runoff is calculated using a curve number technique. The
curve number varies nonlinearly with the moisture content of the soil. Soil water processes
include infiltration, evaporation, plant uptake, lateral flow and percolation to deeper layers.
Actual ET is computed as sum of actual evaporation from soil and plants. Actual soil
evaporation is estimated by using exponential functions of soil depth and water content.
Plant water evaporation is simulated as a linear function of potential ET, leaf area index and
rooting depth, and can be limited by soil water content.
SWAT runs on a daily time step and was designed for evaluating the hydrologic effects of
different crop management practices on water quality. It was not designed for simulating
flood events.

2.2.

Previous Model Comparisons

Explicit comparisons between hydrologic models have been carried out by others. One of
the first such exercises was under the auspices of the WMO in the 1986 Intercomparison of
Models of Snowmelt Runoff. Models were tested using the same data sets for calibration
and then blind tests were run on new data sets for the same watershed. Models tested
included the UBC Watershed Model, HBV, SSARR, PRMS and the National Weather Service
model, NWSRFS. The report (WMO, 1986) concluded that it was not possible to rank the
tested models in order of performance. This was partly because the success of the
simulation, as reflected in the value of a certain criterion, depended on whether this criterion
was considered important during the calibration process. However, it was found that the
complexity of the structure of the models could not be related to the quality of the simulation
results. In other words, more complex models did not perform better than simple models.
The US National Weather Service (USNWS) recently initiated a similar project to the WMO
intercomparison project. The USNWS was particularly interested in whether distributed
models would give superior flow forecasting results to lumped models. An interesting
comment made (Reed et al 2004) is that there is no generally accepted definition for
distributed hydrologic modelling in the literature. Some rainfall-runoff models evaluated in
the study could be considered not to be true distributed models, because they simply apply
conceptual lumped modelling techniques to smaller modelling units. For the purposes of this
model review, we have used the term semi-distributed to apply to lumped models that
include the capability for distribution of some inputs and parameters.
The USNWS intercomparison project noted that the parameter estimation problem is a
bigger challenge for distributed hydrologic modelling than for lumped hydrologic modelling.
Although some parameters in conceptual lumped models can be related to physical
properties of a basin, these parameters are most commonly estimated through calibration.

7-297

Overview of models

12

Initial parameters for distributed models are commonly estimated using spatial datasets
describing soils, vegetation and land use. However, these so called physically based
parameter values are often adjusted through subsequent calibration to improve streamflow
simulations. Given that parameter adjustments are used to get better model performance,
the distinction between physically based parameters and conceptual model parameters
becomes somewhat blurred. (Reed et al 2004).
Models used for the USNWS intercomparison project included the VIC model and
WATFLOOD. As with the WMO project, there was no clear winner. It was found that the
lumped models outperformed distributed models in more cases than distributed models
outperformed lumped models, although some calibrated distributed models can perform at a
level comparable to or better than a calibrated lumped model, which for the USNWS is the
current operational standard. It was also found that gains from calibration indicate that
determining reasonable a priori parameters directly from physical characteristics of a
watershed is generally more difficult than defining reasonable parameters for a conceptual
lumped model through calibration.
BC Hydro has used the UBC Watershed Model for forecasting seasonal inflows on the
Columbia River for more than 20 years. The WATFLOOD and SLURP models were applied
to the Upper Columbia Basin and the results compared with the UBC Watershed Model. It
was found that the SLURP model achieved an average R2 of 0.89 for simulating the daily
inflows to Mica Dam over the period 1972-1994, using the weather from a high resolution
boundary layer model as input. The WATFLOOD model produced slightly better results with
the same weather input, an average R2 of 0.91. However, the R2 value for the UBC model
with weather data from only three observer stations was 0.96. (Druce 2001). It appears that
the climate inputs from the boundary layer model were less valid than calibration using three
stations, an illustration of the fact that more complexity does not necessarily provide
additional information.

2.3.

Discussion

A number of authors have concluded that simple models generally provide better results than
complex models. One of the first detailed investigations was carried out by Jakeman and
Hornberger (1993). They observed that, despite the fact that the physics governing the path
of a drop of water through a catchment to the stream involves complex relationships,
evidence indicates that the information content in a rainfall-runoff record is sufficient to
support models of only very limited complexity. With or without adjustment of rainfall for
antecedent conditions, almost always, one quick flow component and one slow flow
component are all that can be identified in streamflow records.
Kirchner (2006) notes that parameter tuning makes models more flexible, and thus makes
their behaviour less dependent on their structure. He concluded that there was a need to
develop reduced-form models with very few free parameters. These models would capture
the spatially extended character of hill slopes and catchments directly in their governing
equations, without requiring explicit spatial disaggregation and the accompanying
proliferation of free parameters.

7-297

Overview of models

13

Recent research at UBC by Zoran Micovic (2004) into the minimum watershed model
structure for representation of runoff processes has shown that very simple models are
preferable to more complex models, as increased complexity does not improve the results.
His research has shown that a model needs only two flow components (surface runoff and
groundwater) and two elevation bands. This conclusion is similar to the conclusion of
Jakeman and Hornberger (1993).
For the evaluation of models for the Tier 2 Water Budget, it is recommended that high scores
be given to models with fewer parameters with low scores for models that are highly
parameterized.

7-297

3. MODEL EVALUATION

3.1.

Selection Criteria

Model selection criteria were developed to provide a systematic basis for evaluating the
models. While the criteria and assigned scores are somewhat arbitrary, they do provide a
way of listing the key issues and how the models relate to those key issues. The model
selection criteria were developed based on a preliminary evaluation of models and our
experience in model calibration and implementation in the context of the Tier 2 Water
Budget. For each criterion, a score between 1 and 5 was assigned. To a large extent, the
scores are subjective based on discussions with users and a literature review. Nevertheless,
assigning scores to the different criteria provides a way of structuring the evaluation in a
systematic way. Attributes for a low score and a high score in each category are indicated in
Table 1.
A weighting system was used to weight criterion scores, so that all models considered can
be ranked. Weightings for each criterion were discussed and finalized at a meeting with the
Thames-Sydenham Region on April 30, 2009. Other potential uses considered for each
model included flood forecasting and water quality modelling. To avoid double counting,
some weights were assigned low values where it was felt that there was overlap between
criteria. For example, the criteria Ease of use and Transferability are similar.

Table 1: Evaluation Criteria


Score
Criteria
Snow accumulation and melt
Parameterization
Land use
Groundwater component
Technical support
Usage
Ease of use and modification
Documentation
Transferability
Software cost
Applicability for Water Budget work
Applicability for other purposes

7-297

1
Absent
High
Absent
Simplistic
Absent
Very little used
Challenging
None
Difficult to learn
High
Not very applicable
Not very applicable

5
Effective
Moderate
Effective
Effective
Good
Widely used in Ontario
Easy to use
Excellent
Easy to learn
Free
Applicable
Applicable

Weighting
4
3
3
4
2
2
4
3
1
0
5
4

Model evaluation

3.2.

15

Evaluation Scores

For each model and criterion, a brief description was assigned that captures the essence of
the criterion. These evaluation attributes were used to guide the assignment of scores as
shown in Table 2.
At the meeting on April 30, 2009, the scores and weightings were assigned interactively and
the surface water model selected.
Table 2: Final Scores and Weighting

Snow accumulation and melt


Parameterization
Land use
Groundwater component
Technical support
Usage
Ease of use and modification
Documentation
Transferability
Software cost
Applicability for Water Budget work
Applicability for other purposes
TOTAL SCORE
RANK

SSARR

HBV-EC

WATFLOOD

SLURP

HEC-HMS

GAWSER

MIKE SHE

HSPF

SWAT

Weight

5
3
3
3
1
1
1
1
1
5
1
1

5
5
5
2
2
2
5
2
2
5
1
1

5
3
5
2
3
3
2
3
3
5
3
5

5
4
4
2
1
1
3
2
2
3
3
4

3
5
5
3
3
4
4
4
4
5
3
5

5
4
4
4
3
5
4
4
4
4
4
5

5
2
5
3
5
3
3
3
3
1
4
5

4
2
5
3
4
3
2
3
3
5
3
4

5
2
5
2
4
3
2
4
3
5
3
1

4
3
3
4
2
2
4
3
1
0
5
4

71
9

103
8

119
4

107
6

135
2

148
1

133
3

114
5

105
7

The model with the highest weighted score was GAWSER, followed by HEC-HMS and MIKE
SHE. The robustness of the weighting system was tested by changing the weights within
reasonable ranges, and GAWSER remained the top-ranked model for all sensitivity tests.
GAWSER was also the top-ranked model if a weight greater than zero was assigned to
Software cost.
At the meeting on April 30, 2009, it was decided that the Tier 2 study would use the
GAWSER model for the surface water modelling component. Inclusion of as many
unregulated stream gauges as possible to verify the model calibration was encouraged at the
April 30, 2009 meeting.

7-297

REFERENCES
Druce, Donald, J. 2001. Insights from a history of seasonal inflow forecasting with a
conceptual hydrologic model. Journal of Hydrology 249, 102-112
Jakeman, A.J. and G.M. Hornberger. 1993, How much complexity is warranted in a rainfallrunoff model? Water Resources Research, 29, 2637-2649
Kirchner, James W. 2006. Getting the right answers for the right reasons: Linking
measurements, analyses and models to advance the science of hydrology. Water Resources
Research, Vol. 42, WO3SO4.
Micovic, Zoran, 2004. Minimum watershed model structure for representation of runoff
processes. Ph.D. thesis, University of British Columbia.
Reed, Seann, Victor Koren, Michael Smith, Ziya Zhang, Fekadu Moreda, Dong-Jun Seo, and
DMIP Participants. 2004. Overall distributed model intercomparison project results. Journal
of Hydrology, 298, 27-60
WMO 1986. Intercomparison of models of snowmelt runoff. WMO Operational Hydrology
Report No. 23. Geneva

7-297