Catchment

Calibration
A TUTORIAL ON
Create Soil Moisture Method Catchment .....................2
Understanding the Soil Moisture Model .................... 24
Manual Calibration 1 .............................................. 33
Manually Test the Sensitivity of the Model ................ 46
PEST Calibration .................................................... 57

November 2017

Note:
For this module you will need to have completed the previous modules (WEAP in
One Hour, Basic Tools, and Scenarios) or have a fair knowledge of WEAP (data
structure, Key Assumptions, Expression Builder, creating scenarios). To begin this
module, go to the Main Menu, select Revert to Version and choose the version
named Starting Point for Catchment Calibration Tutorial.
In this activity, we will demonstrate how to calibrate a WEAP model using the soil
moisture method for catchments introduced in the Hydrology module.
The importance of calibration cannot be overstated. Models are used to represent real-
world systems to enable researchers and decision makers to ask questions about the
future. A model that cannot adequately represent the past will not adequately represent
the future, and cannot guide decisions because the results in the model may not be the
same results that would occur in the real world. To ensure that model simulations
approximate reality, SEI calibrates its WEAP models according to several measures.

Create Soil Moisture Method Catchment

1. Understanding catchments
The Schematic View includes a shapefile of the catchment area generated using the
WEAP automatic catchment delineation mode.
A catchment is determined by elevation, where the borders of the catchment create a
rim of high elevation points that water flows cannot cross. Thus, a catchment denotes
an area of land where all surface water from rain, melting snow, or ice converges to a
single point at the areas lowest elevation, known as the pour point or discharge
point. It is this point that determines the shape of a catchment, since a point further
downstream would be lower elevation, and thus include a larger area that discharges into
it.
In this example, the catchments pour point correspond with the physical location of a
gauging station on the Tributary River. This way, the catchments modeled flow, when
calibrated, should match the historic data available in the gauge, which measures the
flow for the catchment area. This is a simplistic example because there are no additional
supplies and demands between the headflow and the gauge. For more information about
catchment area, see the Catchment Delineation Module of the WEAP Tutorial.

The catchment node Tributary Headflow is not positioned at the headflow of the river.
Because WEAP models river flow by inputs and outputs, river flow does not change in the
absence of either of these, and the catchments exact inflow location does not matter as long
as it is upstream of the gauge. However, the area of the catchment matters, since it
determines the area where precipitation falls and eventually runs off. Catchments should
always be carefully positioned to allow the model to examine the specific points of interest:
the places where modelers want to ask questions about water availability, and need accurate
information.

2. Change the Model Timestep

To calibrate our model, we have historical data available for the years 2000 to 2007.
This will be our calibration period. To be most efficient with WEAPs calculations,
change the years of the model to match our calibration period.
Go to Main Menu> General/Year and Time Steps, and change the Time
Horizon to be 2000-2007.
 When the model is calibrated using historic, the time horizon can be extended by altering
the Years and Time Steps menu. However, the initial conditions may need to be altered
when the Current Accounts Year changes.

3. Enter the data

Go to the Data View and click on Demand Sites and Catchments/Tributary
Headflow. In the window that appears, select the third option for catchment method:
Rainfall Runoff (soil moisture method).

WEAP has already calculated the area for this catchment in the real world,
but for the purpose of this activity, we will change this to imaginary data.
Under the Land use category, and the Area tab, enter the value 1500 square
kilometers.
Add forest and grassland land cover designations to your catchment.
Right click on the catchment Headflow in the Data View tree, select Add, and
type in the name of your land cover subdivisions: one for forest and one for grassland.
Make sure to right click Headflow each time.
Select the units for both to be share [of square kilometers]. Enter that the Forest
Area has 35%. (Just enter 35 without the percent sign.) Use the Remainder function
(which you can type, or look up in the Expression Builder) to tell WEAP that the
remaining area of the square kilometers will be grassland. It should read
Remainder(100).
When you have completed these steps, your data and data tree should look the same
as below.
 In the Data View, navigate to the Gauges Streamflow Data (Supply and
Resources/River/Tributary River/Streamflow Gauge/Gauge).
The streamflow gauge does not affect WEAPs calculations; its data are
historic observed streamflow from the basin represented in the model. The
value of a historic model is the ability to plug in the recorded data from the
past to compare it with the simulated data. The historic data for the gauge
can be formatted into a .csv file for upload into WEAP.
In the Gauge, under Inflows and Outflows, click on the white space under the year
2000 and select the ReadFromFile Wizard from the dropdown menu. Select the
SMMdata.csv file in your Additional Files folder.
The SMMdata.csv file has four data columns: Precipitation, Temperature,
Wind Speed and Streamflow. We need to make sure the correct information
is displayed for the streamflow gauge data.
In the lower section of the ReadFromFile Wizard, change the line in Data
Column to 4 Streamflow[CMS], and then click the Finish button.
 The text in the SMM.csv file uses a directive at the top that tells WEAP how to label the
different data columns, meaning the columns that start after the month and year format. In
the SMMdata.csv file, the text $Columns = Precipitation[mm], Temperature[C], Wind
Speed[m/s], Streamflow[CMS] tells WEAP the names and units of each data column. For
more information about the ReadFromFile Wizard, see the ReadFromFile Wizard video at
www.WEAP21.org/videos.

Reading the streamflow data from the CSV file, your screen should look like
this:
Now we must enter the remaining data in the SMM file into the climate data
for the Tributary catchment. The data that we will enter is assumed to be
representative for the whole area in the catchment, with the exception of
some land use data which is specific to certain land use types.

Meteorological stations that measure climatic values like precipitation, temperature and
windspeed measure these values only for the location of the station. However, WEAP
requires this data to characterize the whole catchment. When possible, it is best to gather
data and locations for as many stations as possible. This allows the user to interpolate
between them to calculate an average value for the whole catchment for each parameter for
each time step. In this example, the climatic data in the SMM.csv file is assume to already
be interpolated.

In the Data View, navigate to the Tributary Headflow catchment to view the climate
data. Use the same ReadFromFile Wizard and the SMM.csv file to enter the
precipitation, temperature and wind speed data into the Climate data in the Headflow
catchment.
Always be sure to double check the units in your file are the same units that
WEAP is using, and remember to indicate the correct column.
 The SMM.csv dataset does not include relative humidity. You could use the
monthly time series wizard that would provide average monthly values for
the catchment, but here we will use just one value for the whole year.
For simplicity, enter a
Humidity 35
Finally, the hydrology module needs to know the latitude of each catchment
to estimate potential evapotranspiration. Enter a value of 30.
Latitude 30 (degrees north)
Leave all the other parameters in the Land Use tab at their default values. This
gives us an initial, uncalibrated model.

4. View the Results

Run the results. We want to compare the modeled streamflow with the data
in the streamflow gauge.
In the Chart section of the results, navigate to Supply and
Resources>River>Streamflow.
 In the River Drop Down menu, change the river from Main River to Tributary
River.

In the right drop-down menu, choose Selected Tributary River Nodes and Reaches.
Select Basin 1 Runoff (which is the modeled flow from the catchment) and Gauge
(gauge) (which is the observed historic gauge data).
 As we can see, our modeled river (blue) is not simulating the high flows of
the river very well.

5. Save chart as a favorite

Before we start the calibration process, save this chart as a favorite so we can
easily return to it later.
Go to Favorites/Save Chart as Favorite, and call it Observed and Modeled
Streamflow.

6. Exploring the Calibration

In the results produced so far, there is not very good fit between the modeled
streamflow and the historic streamflow. The observed flow is much higher
than the modeled flow, though the seasonality appears to be more or less
correct. But there is a lot of room for improvement.
The time series record is one of seven charts we recommend considering in
the process of calibration.
1) The time series for the observed and modeled streamflow (generated
above)

2)
Annual Total of observed and modeled streamflow (while viewing the
time series generated above, check the Annual Total box in the Results
View)
 The annual total provides a good sense of the general fit of the model over
time. For example, here we see that our modeled flow is much lower, and
does not capture annual variations very well, especially the year of low
flow in 2006.
3) Monthly Average of observed and modeled streamflow (check the
Monthly Average box in the Results View
 The monthly average shows how the model deviates from the record, on
average, each annual cycle. Here we see that the modeled low flows of the
summer are closer to the observed values than the winter modeled high
flows, which are considerably underestimated in our model.
While viewing these results, click on the icon to export results to Excel ( ) on the
right side of your screen. Save them on your computer as Uncalibrated Monthly
Average Streamflow Results. We will review them later.
4) Exceedance Probability for observed and modeled streamflow (make sure
that neither Monthly Average nor Annual Total are checked, and
check the box at the bottom right of the Results View that says Percent
of Time Exceeded.
 The exceedance probability chart ranks each flow measurement by value,
the lowest on the right and the highest on the left, for both the modeled
and observed streamflow. The values on the x-axis show the percentage
of flows that exceed the values of flow. The modeled and observed
streamflow should show a close match of exceedance percentages for the
two flow records.
5) Relative Soil Moisture 1 (%) (Open the Results Menu and navigate to
Catchments/Relative Soil Moisure 1 (%). Make sure the Branch selected
is Tributary Headflow, not Big City. Also make sure that Percent of
Time Exceeded is no longer checked.)
 This chart does not compare the observed and modeled streamflow, but
it shows a result for the land use dynamics of the modeled catchment. The
Relative Soil Moisture 1 result denotes the soil in the upper bucket of the
catchment soil moisture models 2-bucket structure. This structure will be
described in detail in the next section. The catchments upper bucket is
affected by the land use types designated in the model as well as the
seasons. It is important the soil moisture of the upper bucket does not
demonstrate a changing trend over time. Except in very extreme
circumstances (multi-year droughts), the Soil Moisture 1 (%) should
remain stable over time, even as it displays seasonal patterns. The current
results for the upper bucket are acceptable. Notice that the most saturated
it gets is around 40% during the wet season.
6) Relative Soil Moisture 2 (%) (in the Results Menu, just below the Relative
Soil Moisture 1 (%) result).
The Relative Soil Moisture 2 (%) result shows the soil moisture in the
lower bucket of the modeled catchment. The lower bucket is deeper
down, not affected by land use types, and less affected by seasons because
of its separation from the surface. For its results, a little seasonality is
acceptable, and there should be no trend (either increasing or decreasing)
over time.
The decreasing trend seen in this lower bucket example could mean one
or more of several problems:
a. The initial condition for the soil moisture in the lower bucket is too
high.
b. Not enough water is percolating into the lower bucket
c. The water leaving the bucket too quickly, because
a. The conductivity is too fast
b. The bucket is too small, so the transport time is very short
We will explore some of these possibilities as we calibrate.
7) Catchment Land Class Inflows and Outflows (still in the catchment
results, navigate to Land Class Inflows and Outflows. View the result
as a Monthly Average, and select the bars to be stacked using the bar
icon on the right.)

This chart shows the average distribution of the water becoming available
in the catchment (values above zero) and how that water in parsed (values
below zero. We see a large amount of evaporation, which is water lost
from WEAPs accounting system. Lowering this evaporation will enable
more water to run off into the river, as discussed in the activities below.
For now, export this data to Excel and save it as Uncalibrated Monthly
Average Land Class Inflows and Outflows.
Using the information from these seven charts, we can begin to think
about changing the parameters of the model to allow WEAP to more
accurately model the streamflow.
Understanding the Soil Moisture Model
7. Physical Structure of the Soil Moisture Model
The soil moisture model for generating streamflow is a hydrological model
with several parameters. In a two-bucket system (explained below), the basic
equation for the upper bucket, in mm for each time step, can be summarized
as follows:
=
( +
+ + )

The equation for the lower bucket, in mm for each time step, is
=

These terms WEAP to calculate the volume of water entering the river as
=

( + )

The upper bucket can be divided into different land use areas within a single
catchment, while the lower bucket assumes the area of the entire catchment
node.
Here is the picture from the WEAP help menu (topic Soil Moisture
Method). You can find out more details about the calculations in the help
menu, but the general parameter operations are summarized below.
 The land-use and climate parameters of the soil moisture model determine
the water percolates into the earth, evaporates or runs off into the river. The
land use parameters can be summarized as follows:
1) When it rains, how much water can potentially be evapotranspirated by plants?
(controlling parameter: Kc)
2) Of the water that is not evapotranspirated, do the conditions on the surface,
such as leaf area and slope, encourage or discourage percolation? (controlling
parameter: runoff resistance)
3) The water that percolates into the ground moves first into the upper bucket of
the model (upper box outlined in bold black line in the figure above). There the
water is subject to several parameters:
a. How fast does water move through the upper bucket? (parameter: root
zone conductivity)
b. How much capacity (mm multiplied by the area of the land use type,
defined in Land Use/Area data) does the upper bucket have to hold
water (the mm measurement is recorded in the parameter: soil water
capacity)?
 c. As the water travels through the upper bucket, how much of it percolates
into the lower bucket, and how much of it runs off and becomes
streamflow (parameter: preferred flow direction)
4) Any water in the upper bucket that moves to the lower bucket (lower box in the
figure above, outlined in bold black line) has parameters defining how it
behaves:
a. What is the speed that the water moves through the lower bucket
(parameter: deep conductivity)?
b. What is the capacity (mm) of the lower bucket (parameters: deep water
capacity)?
In WEAP, these parameters are all controlled by a set of equations that
together determine how much water run offs to the river, thus producing the
modeled streamflow in WEAP (see the Soil Moisture Model in the WEAP
help menu). Their relationships to the model are complex, but are
summarized more simply (and less accurately) here:
Land Use Default Possible Impact on Model
Parameters Value in Values
WEAP

Kc 1 0- Increases in Kc lead to more evapotranspiration

net loss of water from WEAPs accounting
system, and loss of water that could contribute to
streamflow volume
Runoff 2 0- Increasing allows less water to immediately
Resistance become streamflow as it instead percolates into
Factor
the soil. While it is in the soil, more evaporation is
possible enabling a higher loss of water from
WEAPs accounting system, and loss of water that
could contribute to streamflow
Root Zone 20 0- Increasing will shorten travel time for any water in
Conductivity (mm/ the upper bucket (less evaporation possible, less
month) loss of water from system)
Soil Water 1000 0- Increasing will increase travel time for any water
Capacity (mm) in the upper bucket (more evaporation possible,
more loss of water from system)
Preferred 0.15 0-1 0 = all water leaving the upper bucket flows to
Flow lower bucket, and none to river
Direction
1 = all water leaving the upper bucket flows to
river, and none to lower bucket
Deep 0 0- Increasing will shorten travel time for any water in
Conductivity (mm/ the lower bucket before it flows into the river.
month) There is no evaporation in the lower bucket.
Deep Water 1000 0- Increasing will increase travel time for any water
Capacity (mm) in the lower bucket before it flows into the river

The above parameters control the land use types, but there are a few
additional parameters to consider changing.
1) Initial Z1 (Data/Demand Sites and Catchments/Tributary Headflow/Land
Use/Initial Z1. Parameter only available in Current Accounts or the first year of
your catchment modeling). In the Soil Moisture Model figure (above) of the
upper and lower buckets, the blue area shows how full each bucket is (Z1 for
the upper bucket, Z2 for the lower bucket). We need to set the initial condition
for the model, much like a reservoir, because the quantity of water already in
the bucket will impact what happens to the new water arriving (read more about
the mathematic expressions in the WEAP Help Menu). The example below
shows a model with an initial condition for January 2002 that is like too low,
given the consistency of the other years.
Relative Soil Moisture 1 (%)
Scenario: Reference, All months (12)
Arable
85
80 Artificial Areas
75 Forests
70 Hetero Agriculture\Irrigated
65 Hetero Agriculture\Unirrigated
60 Open Space
55 Pastures
50 Permanent Crops
Percent

45 Scrubs
40 Waters
35 Wetlands
30
25
20
15
10
5
0
Jan Apr Aug Dec Apr Aug Dec Apr Aug Dec Apr Aug Dec Apr Aug Dec Apr Aug Dec Apr Aug Dec
2002 2002 2002 2002 2003 2003 2003 2004 2004 2004 2005 2005 2005 2006 2006 2006 2007 2007 2007 2008 2008 2008
 Note that your value for the initial condition will only affect the model for the
first year or so after that, the model adjusts itself. This parameter has a range of
0-100%.
2) Initial Z2. (Data/Demand Sites and Catchments/Tributary Headflow/Land
Use/Initial Z2). The percent of soil moisture for the lower bucket should be
relatively stable over time, so you can look at that result to make a guess at the
initial condition. Range of 0-100%.
3) Freezing point. (Data/Demand Sites and Catchments/Headflow/Climate/
Freezing Point). The freezing point determines at what temperature rain
becomes snow, which delays any of the activities that control how the water
flows into the earth/river. This parameter is relevant only for watersheds with
snow, and it affects the model only during the months with snow. The value
(Celsius) should stay close to zero (WEAPs default value is -5 Celsius).
4) Melting point. (Data/Demand Sites and Catchments/Tributary Headflow/
Climate/Freezing Point). This parameter determines when snow turns to
water, after which it is subject to all the land use parameters. Again, this
parameter is only relevant and impactful during the winter and for watersheds
cold enough to have snow. The value should stay close to zero (WEAPs default
value is 5 Celsius).
5) Albedo is the fraction of solar radiation striking a land class that is reflected --
albedo increases as snow depth increases, but can also change over time as
snow gets older and dirtier. New snow is very white, and has higher reflectance,
whereas snow mixed with soil or dust will attack more heat, and thus have
lower albedo. In WEAP, the user can choose limits for high and low albedo
values in the system.
6) Irrigated Area. Data/Demand Sites and Catchments/Headflow/Irrigation
where you can designate any irrigated land use areas. For each land use type, is it
irrigated or not? You can only select 0% irrigated or 100% irrigated.
Catchments with irrigated areas must be connected to water supplies (a river, a
groundwater node, etc.) so they have somewhere to draw water from when
irrigation is needed.
7) Upper threshold [of irrigation]. (Data/ Demand sites and
Catchments/Headflow/Irrigation/Upper Threshold). This parameter
represents the ideal value of soil moisture for the plants. In every time step,
WEAP asks is there enough water in the soil? If there isnt enough (see
description of Lower Threshold, below), WEAP will demand water from the
connected supply to irrigate the catchment, aiming to provide enough water to
reach the upper threshold of soil moisture (if it rains in the same time step,
 WEAP will probably overshoot the target). Note that all land use types have
options for data entry of their upper threshold, but this data is only read by
WEAP if the parameter Irrigated Area (defined above) is 100% for the
respective land use type.
8) The Lower threshold [of irrigation], specifies each irrigated land use types
tolerance for dryness. If, in a time step, the soil moisture is between the upper
and lower threshold, then there is still enough soil moisture, and WEAP will
not irrigate. WEAP will only irrigate if the soil moisture is lower than the lower
threshold, and then it will irrigate as much water as it thinks it necessary to raise
the soil moisture to the upper threshold.
To understand the dynamics of the Upper and Lower Thresholds a little bit
better, consider the following situations:

Upper Lower Range of WEAPs Action

Threshold Threshold possible soil
Value Value moisture
that would
produce the
same
outcome

90% 20% 20% - 100% WEAP does nothing, the soil moisture is
tolerable for the plants

90% 20% 0-19% WEAP provides enough water to raise the soil
moisture up to 90%. Since the gap between
20%-90% is quite large, this will probably be a
large amount of water.

90% 80% 80%-100% WEAP does nothing

90% 80% 0-79% WEAP irrigates a lot of water to raise the soil
moisture to 90%. This may be a little bit, if the
soil moisture is 79%, or a lot, if the soil
moisture is 10% (which would be pretty
amazing, considering that it would have to
have dropped to that value from the previous
time step, which was supposed to have 80%
or more soil moisture.)
10% 20% 11%-100% WEAP does nothing

10% 20% 0%-9% WEAP irrigates up to 20% soil moisture

Note that the value of the percentage gap between thresholds might be the
same (for example, both 90%-90% and 10%-20% have 10% gaps), but these
situations produce very different actions in WEAP. For the 10%-20% range,
WEAP has a lot of soil moisture values when it does not need to irrigate, as
opposed to the 80%-90% range, when it the catchment will demand more
water to keep the soil moisture close to 90%.
Similarly, the size of the gap matters small gaps will produce more frequent
irrigation demand (small amounts), whereas large gaps will produce
infrequent irrigation demand of very large amounts.

Some of these parameters are profoundly affected by the model time step. For example, the
freezing and thawing of soil may occur over days, rather than months. Commented [S1]: Ask Brian to help complete this part.

8. Values for the Parameters

Now that all the calibration parameters are introduced, its important to
mention two complicating factors: the Monthly Time Series Wizard, and
differentiated land use types.
Monthly Time Series Wizard
All of these parameters exist in the default with only one value over the course
of the year. But for some of them it may make sense to change them over the
different months. This introduces an additional 12 values to alter in the
calibration process for each parameter. However, not all parameters lend
themselves to monthly values: it is surface that is most affected by seasonal
changes.
The following two variables have particularly compelling reasons to be
calibrated monthly.
Kc: crops typically have annual patterns months when they are growing a
lot (and evapotranspirating a lot of water) and winter months when they are
not growing much, or not at all. Kc values that change by month may make
sense.
 Runoff Resistance Factor: The foliage and groundcover changes over the
course of the year, and the landcover might discourage runoff (a higher
Runoff Resistance Factor value) during some time steps more than others,
when the water runs off more easily.
To calibrate based on the monthly values, examining the Monthly Averages
will show which months have too low streamflow, and which months have
too high streamflow.
Different Land Use Types
We also have the option to pick different values for different land use types.
Not all variables change according to land use type for instance, Freezing
and Melting temperatures are the same for the whole catchment, and the
capacity and conductivity of the lower bucket of the Soil Moisture Model is
too deep to be affected by the land use differences on the surface.
Knowing how land use types might differ in their hydrologic characteristics
requires a good understanding of the area, and ideally measurements, if they
are available. The more land use types a model has, the more helpful it will
be to structure the key assumptions for the calibration process.

9. Using Key Assumptions for Large Models

SEI modelers have found that it can be cumbersome to edit values within
catchments, especially when working with multiple catchments that might all
have the same values for different land use types. To ease the calibration
process for large models, we recommend structuring your catchment
parameters within Key Assumptions.
 As you can see above, it is very easy to edit the different Root Zone
Conductivity values for the different land use types, without navigating to
each of the catchments individually (because, in the process of creating these
Key Assumptions, the catchments were linked to these values.)
Also note that unknown values can refer to a known value (here the value
Arable Root Zone Conductivity that is used for the subsequent branches),
such that changing that value will change all the subsequent values.
This allows editing land use parameters quickly as you move through the
calibration process.

10. Calibrating the Model: Some tips

Now that you know which parameters to calibrate, where they are, and how
to judge the success of the calibration, you are ready to start calibrating. You
may find it helpful to start with the surface parameters and calibrate
downwards in the catchment buckets. Keep a visual record of your results
charts by copying and pasting them into another document so that you can
compare them and understand the impacts of the numerical changes that you
make.
The calibration process should follow these steps:
1) Change parameter values
2) Run Results
3) Copy chart into separate document so you can view it while you change the
parameters
 4) Repeat until you think it has improved
5) Quantify the statistic parameters (explained below)
6) You might want to save your versions as your experiment with changing the
parameter values. You can go to Area/Save Version to save the ones you think
are good. That way, if your calibration efforts produces something that is not an
improvement, you can always go back to a previous version. I recommend
saving versions without the results because results take up a lot of memory, and
they can always be recalculated.
7) If youre satisfied, you can stop. If you think you can do better, keep going.
The example in this module demonstrates three iterations of this process: 2 manual
calibrations, and 1 using WEAPs automatic calibration program PEST.

Manual Calibration 1
We will enter initial estimates for the hydrologic parameters related to the two
land use types in the Tributary Headflow catchment.

11. Enter the Data

Precipitation is a key parameter in producing streamflow in WEAP. Other
parameters characterize how much precipitation becomes run-off in the
system. This model uses a monthly timestep, so some parameters associated
with soil moisture accounting are not meant to have rigorous physical
meaning; rather they are used to help describe the general hydrologic
response of the watershed. Precipitation should come from the historic
record (represented in our SMM.csv file), but the land-use parameters can be
altered within the calibration process to enable the modeled streamflow
results to better approximate the observed streamflow data. The following
steps are used in the calibration for the Tributary Headflow catchment to
improve the models simulation of the system streamflow.
The figure below shows the mean monthly observed runoff and average
precipitation for the year 2000-2007.
 Monthly Average ofHistoric Precipitation
and Streamflow 2000-2007
160
140
120
100
80
60
40
20
0

Precipitation (mm) Streamflow (CMS)

The hydrography reveals the strong correlation between precipitation and

runoff. The observed runoff suggests a small deep storage capacity, evidenced
by very low baseflows during the dry season. This suggests large seasonal
fluctuations of the top bucket storage, and thus very large seasonal variations
of soil water availability from very wet (high values of z1) to very dry (low
values of z1). This is indeed what we already saw in our result Relative Soil
Moisture 1 (%).
Assuming z1 has a value less than 100%, consistent with the existing data, the
flow to the river is the sum of three results produced in the rainfall-runoff
model: surface runoff (water running off from the surface, does not percolate
into the upper bucket), interflow (water that percolates through the upper
bucket, but does not percolate into the lower bucket), and baseflow (water that
percolates through both the upper and lower bucket).
Recall that the models initial results show that the current simulation does
not have enough water running off, so parameter adjustments should be
made to increase one or more of these types of runoff.
The hydrograph shown above suggests that the system is dominated by the
surface runoff component, since runoff is so highly correlated with
precipitation. From the structure of the Soil Moisture Model, we know that
surface runoff is a function of z1, the runoff resistance factor, precipitation, and
irrigation parameters.
Of these, z1 is calculated by WEAP for all timesteps after the initial timestep
(partially as a function of soil water capacity), precipitation comes from the
precipitation record, and irrigation is calculated by WEAP according to the
upper and lower thresholds entered by the user. The user has direct control
over one parameter affecting surface runoff: the runoff resistance factor.
WEAP gives a default value of 2 for the runoff resistance factor of all landuse
types when a catchment is first added to a model. To increase water runoff
to the river, this value should decrease (for less resistance). But land use also
plays a role.
Assume that for this models land use, the surface runoff from forest (35%
of the area) is usually less than from grasslands (65% of the area), perhaps
because the grasslands are more likely to include areas with slopes. To
characterize this difference, the runoff resistance factor for the forest land use
type is higher than from grassland. We will pick values for the two land use
types in Current Accounts:
Runoff Resistance Factorforest= 3
Runoff Resistance Factorgrassland= 1
The data inputs both differentiate between land use types and increases the
runoff for the majority of the catchment land area (the grasslands). By
increasing the direct runoff, there is less opportunity for evaporation, which
means the catchment will convert more precipitation to river flow, rather than
evaporation.
The changes made to the runoff resistance factors will increase the volume
of surface runoff in the model. However, interflow and baseflow are also
important contributors to the river volume. The highest month for
streamflow, according to an average of the flow gauge record for the years
2000-2007, is January (49.7 CMS on average). The lowest is July (1.6 CMS on
average), which is likely dominated by baseflow, since there is very little
precipitation at that time.
Lets consider how to change the interflow, which is impacted by the
parameters root zone conductivity and preferred flow direction and z1. Of these, the
users directly control root zone conductivity and preferred flow direction.
WEAPs default value for preferred flow direction for both land use types is 0.15,
which directs only 15% of the water from the upper bucket into the river as
interflow (the remaining 85% percolates into the lower bucket). To increase
the interflow, the user can direct more water from the upper bucket to flow
into the river by increasing the values of the preferred flow direction for both land
use types. Change the values of the preferred flow direction to 0.7 for forest
and 0.6 for grassland, respectively, to reflect the fact that horizontal flows
dominate, and the forested area generates more horizontal flow than
grassland.
Preferred Flow Direction
Forest = 0.7
Grassland = 0.6
Note that this directs less water to the lower bucket soil moisture, which was
already showing a trend of decline. We will deal with this elsewhere.
Another way to decrease evaporation, and increase the amount of
precipitation converted to river runoff is to reduce the travel time in the upper
bucket, where evaporation occurs. Though we already have estimates for Root
zone conductivity, we can also reduce the distance the water must travel by
reducing the Soil water capacity. The default value in WEAP is 1000 mm.
Decrease this value to the following estimations, which may or may not be
correct.
Soil water capacityforest = 600 mm
 Soil water capacitygrassland = 500 mm

In our initial results for Average Monthly Land Class Inflows and Outflows,
we saw that the highest value for interflow was in March (0.7 million
m3/month). But we also know from our Monthly Average Streamflow
Results that during the wet season (roughly December-March) the modeled
streamflow was too low by between 45.9 million m3/month (December) and
109.6 million m3/month (January). To obtain these results, subtract the
modeled streamflow from the observed streamflow in the Excel file Monthly
Average Streamflow Results.
Since hydrograph shows that the river runoff is well correlated with the
precipitation, much of this wet-season river volume should be surface run-
off. However, we will also assume a small, average monthly interflow
component of 12 million m3 during the wet season. This assumption might
be wrong, but it is a starting point that is considerably higher than the original
modeled value for interflow during the wet season (between 0.17 MCM in
December and 0.70 MCM in March) and considerably lower than the total
difference between modeled and observed streamflow during the wet season.
The value 12 million m3 in an area of 15 km2 translates to an equivalent depth
of 8mm that becomes interflow:
1 2 1000
12 106 3 = 8
1500 2 106 2
We will plug this value into the soil moisture model equation that calculates
interflow, and use it to calculate root zone conductivity.
= ( . . ) 1 2
Where interflow = 8mm
Preferred flow direction = 0.7 (forest) and 0.6 (grasslands)
z1 = Recall from our original results for Relative Soil Moisture 1 (%) that the
relative storage of the upper layer during the wet period reached about 40%.
However, we have made the upper bucket much smaller, so we should expect
this value to be larger, all things equal. For this calculation, use a value of 70%
of the maximum soil water holding capacity. However, do not change the
parameter Initial z1 with the default value of 30% in Current Accounts,
because it will adjust itself as the model runs it is only the initial value.
Plugging these values into the above equation provides an estimate for Root
Zone Conductivity (k1) for forest as 23.3 mm/month because
k1 8 /(0.7 * (0.70) 2 ) . Enter 23.3 mm/month here for Forest Root Zone
Conductivity and, using the same equation but a value of 0.6 for the Preferred
Flow Direction, 27.2 mm/month for Grassland. The calibration exercise will
help suggest how to further change them.
Root Zone Conductivity (k1)forest = 23.3 mm/month
Root Zone Conductivity (k1)grassland = 27.2 mm/month
Continuing to the parameters in the lower bucket, we can start by calculating
the baseflow for our system.
In the previous step of this activity, the result for Relative Soil Moisture 2 (%)
showed a declining trend, and one of the possible reasons was that the
modeled lower bucket capacity (default = 1000 mm) was too large. Therefore,
we will halve it for this iteration of the calibration.
Deep Water Capacity = 500 mm
Considering the observed streamflow data in the hydrography, the low flow
conditions are in June, July, August and September. July has an average value
of 1.568 CMS, which would put the total July volume at about 4.2 million m3,
which we will consider baseflow. If you assume that the entire watershed
uniformly supplies this baseflow and divide this low summer volume by the
watersheds total area (1500 km2), you can estimate the equivalent average
baseflow runoff depth as 2.8 mm.
1 2 1000
2.8 = 4.2 106 3
1500 2 106 2
We can relate a baseflow runoff depth to the Deep Conductivity (k2) and the
square of the layers relative storage (z2) using the equation from the soil
moisture method:
= 2 2 2
Looking at the result Relative Soil Moisture 2 (%), it is difficult to guess stable
average, long-term relative storage of this bottom layer (z2). Perhaps it has
changed now that the Deep Water Capacity is halved. Here we will estimate
30% for z2.
Using the previously estimated values for average baseflow runoff and z2, we
can reformulate the equation to k2 2.7 /(0.30)2 or 30 mm/month. In WEAP,
increase the default value (20 mm/month) to 30 mm/month for the Deep
Conductivity (k2). This may encourage the lower bucket to run off water faster,
which is also encouraged by the fact that we made the bucket smaller.
Deep Conductivity (k2) = 30 mm/month
 Lastly, we will consider data for Kc by assuming that in this area, both the
forest and grasslands are assumed perennial and evergreen. So Kc, the
evaporation coefficient (or crop coefficient in agricultural context) data can
be constant. For crops, libraries such as the FAOs single crop coefficient
(http://www.fao.org/docrep/x0490e/x0490e0b.htm) are useful. Note that
WEAP also has explanations of various parameters in the Help menu.
As stated above, Kc may be a good candidate to vary by month. However,
leaving it at the default value may also be a good option if no data is available.
Kc =1 (default value)

12. View the Results

Run the results and look at your favorite chart that you made earlier.
Go to Favorites in the Main Menu and select Observed and Modeled Streamflow

This looks much better than before. The high flows are higher, though not
quite high enough, and the low flows are still good.
Select Annual Total for the results
The model is still underpredicting streamflow overall, and the annual trends
are still not following the observed data very well.
Select Monthly Average for the results
This chart shows us that the model is also underestimating streamflow each
month, although it this error is more pronounced during peak flow months.
This may have implications about how we are modeling immediate runoff.
Finally, the flow duration curve can be displayed
After deselecting Monthly Average, select Percent of Time Exceeded at the
bottom of the results view.
In this chart we can see that the model calculated that streamflows are larger
than 10 MCM about 56% of the time, whereas the observed data shows that
58% of the time, the streamflows are larger than 10 MCM. For a higher flow,
like 75 MCM, the model estimates that around 2% of the flows are higher,
whereas the data shows that the number should actually be about 25% of the
flows. Again we see that our modeled data is underestimating the streamflow.
The Relative Soil Moisture 1 (%) still shows a stable seasonal trend:
And the Relative Soil Moisture 2 (%) declines quickly, but could be stabilizing
around 20%. This suggests that it might be a good idea to change the initial
condition, but it could also be that one of the other parameters is still too
high.
Evaluate the following statistical parameters.
The Nash-Sutcliffe Efficiency (NSE) coefficient is commonly used in
hydrologic modeling to evaluate how well modeled stream flow matches the
observed streamflow. The idea value is 1, value between 0.5-1 are considered
acceptable (although some modelers will argue that values should exceed 0.7),
and values below 0 say that the simulated flow is worse than if the model had
used the average value of observed streamflow for each timestep.

The percent bias (PBIAS) as a measure of the models ability to match the
total volume of flow. An ideal value is zero, but Bias of plus or minus 25%
of the observed streamflow is considered acceptable.
The root mean squared error to the standard deviation (RSR) as a measure of
how much the simulated flows deviated from the observed hydrographs. An
ideal value is zero, but values less than 0.7 is considered acceptable.

The ratio of simulated versus observed flow standard deviation (SDR) as a

measure of how well the simulated flows match the flow variability within the
historical record. An ideal value is 1, but values between 0.9-1.1 are
considered acceptable.

We obtain the following value: NSE = -0.27

We obtain the following value: PBIAS = -70%
The ratio of the root mean squared error to the standard deviation (RSR) as
a measure of how much the simulated flows deviated from the observed
hydrographs. An ideal value is zero, but a value of less than 0.7 is considered
satisfactory.
We obtain the following value: RSR = 1.1276
Last, SDR measures how well the simulated flows match the flow variability
within the historic record. An optimal value would be 1, but values between
0.9-1.1 are considered satisfactory.
We obtain the following value: SDR = 1.79
There is a lot more work to be done to improve this calibration.
Create a version of the model and save it as NSE = -0.27
Manually Test the Sensitivity of the Model
Having viewed the results, we have an idea of what to change.
First, because our streamflow is still too low, we will decrease evaporation by
decreasing the Runoff Resistance Factor by 20%. This will direct more water
to runoff to the river immediately. The values were 3 and 1 for forest and
grassland, respectively. We will edit them using math operations (showing
that they will be 80% of their previous value) so we can see in WEAP how
the new values were calculated.
In the Data View, multiply Forest and Grassland Runoff Resistance Factor values
by 0.8.

We can also decrease evaporation by speeding up the travel time in the upper
bucket. We will start by reducing the values for the Soil Water Capacity, by
60%. This lowers the amount of water that can be stored and will encourage
more interflow and should increase the peak flows and help a little with the
PBIAS that showed underestimation. Soil Water Capacity can also make
WEAP more responsive to changes on annual time scale.
In the Data View, multiply Forest and Grassland Soil Water Capacity values by
0.4.

We will also increase the soil water conductivity, k1 by 30% to encourage

faster flow from the top bucket.
Multiply the Soil Water Conductivity values by 1.3.
Lastly, we want the Relative Soil Moisture 2 (%) result to be stable overtime.
We saw that the lower bucket was losing water too quickly, so one response
might be to slow down the travel time.
Multiply the Deep Water Conductivity by 0.4.
13. View the Results
Run the results and compare the Manual Calibration 1 scenario with the
Manual Calibration 2 scenario with our small changes using the monthly
average streamflow of both scenarios.
In the Results View, go to the Favorite Chart for Observed and Modeled Streamflow.
For the most part, we have succeeded in increasing the modeled streamflow.
However, its hard to tell which parameter, soil water capacity or deep water
conductivity, had the larger effect. Users seeking a deeper understanding can
examine each of these parameters individually
This calibration is much improved compared to the first one we saw.
Interestingly, the problems are not always the same throughout the years, for
example, some years the peak flows are over-predicted, and other years they
are under-predicted. Some years the low flows are very well matched, where
as other years the model under-predicts the low flows.
Click on the box to see the Monthly Average for the observed and modeled streamflow
We also note that the autumn months show a closer fit, on average, then the
rest of the model. This could suggest an opportunity to edit some variables
by month, such as the Runoff Resistance Factor, or the Kc values for the land
use types.
We can also verify that the Relative Soil Moisture 1 (%) is still stable over time
(you will need to uncheck Monthly Average)
The Relative Soil Moisture 2 (%) also appears to be steady, though the results
divide the moisture by land use types. These are inevitably different, because
they are percolating water differently.
Next we will calculate the statistics to see quantify how our model has
improved.
Return to the Favorite Chart for Observed and Modeled Streamflow, and calculate the
statistics.
NSE = 0.86 an excellent calibration value (it is greater than 0.7)
%Bias = -18% - within the preferred margins of 25%
RSR = 0.37 below 0.7 and relatively close to the ideal value of 0
SDR = 1.14 just above the preferred threshold of 1.1. So the model could
be representing flow variability better. This is possibly something that a
Monthly Time Series wizard of the Runoff Resistance Factor or Kc parameter
could improve.
Overall, we see that we have significantly improved our model for all
indicators.
 There are always more improvements to be made, but this would be an
acceptable stopping point.

14. Compensatory Variables

Our model calibration has been successful, but we would be right to wonder
about the all the other factors that we did not change. Could we have obtained
similar results by altering different variables?
As an exercise, we can explore what another set of variables might have
looked like.
Return to the Data View and delete the multiplication factor from the Deep
Conductivity parameter, so the value is once again 30 mm/month.

We had changed this variable with the hopes of providing stability for our
result Relative Soil Moisture 2 (%), by slowing down how quickly the water
left the catchment, and keeping the soil moisture around its initial value of
30%.
Instead, we will assume that the water still has a fast travel time, but the initial
soil moisture value was lower.
Change the variable for Initial Z2 to 19%

Because the soil moisture exerts a non-linear impact on travel time in the
lower bucket (higher values of soil moisture cause the bucket to expel water
faster), specifying a lower soil moisture value in conjunction with a faster
conductivity rate can still produce stability in the lower bucket soil moisture.
Run the results. View the results for Relative Soil Moisture 2 (%).
The land use types again diverge, but they still look more stable than when
we first manually calibrated the model.
Did this alternation in the characterization of our lower bucket change
statistical results?
Return to the Favorite Chart for Observed and Modeled Streamflow, and calculate the
statistical results.

Result Before Alteration After Alteration

NSE 0.86 0.86

%Bias -18.65% -18.70%

RSR 0.37 0.37

SDR 1.14 1.14

 The similarity of the results show that while the different parameter values
changed individual results (the soil moisture in the lower bucket over the
course of the model), the overall model was not greatly impacted.
When parameters have different combinations but the same general
calibration results, those parameters are called compensatory. In WEAP,
there are many examples of compensatory parameter combinations.
However, the issue at stake remains the same is the model capturing the
dynamics of the streamflow in response to climatic factors?s
In some instances, experienced modelers may be able to consider the model
and determine that one set of compensatory parameter values is more realistic
than another. In another case, like this one, either could be equally good, and
since they are effectively indistinguishable in terms of the results produced,
either can be used within the WEAP model. As the phrase goes in English,
there are many ways to skin a cat.

PEST Calibration
This exercise uses a built-in calibration feature in WEAP called PEST.
We will use PEST to automatically calibrate the Deep Water Capacity and Root
Zone Conductivity (k1) parameters for forest and grassland.
On WEAPs Main Menu, go to Advanced>PEST Calibration.

This brings up WEAPs interface to PEST.

We can now choose which parameters to calibrate within the Tributary
Headflow catchment to allow WEAP to approximate the streamflow
observations. We had some success changing the soil water capacity, but what
other parameters might help produce higher streamflow during the right time
of year.
In the box Parameters to Calibrate, click + Add.
In the window that appears, starting with the Branch drop-down menu, navigate to
Demand Sites and Catchments \ Headflow. Then in the Variable Menu, choose Deep
Water Capacity.
For the values, add:
Lower Bound 0
Upper Bound 1500
Initial Value 500
 Click Save.
Next, add another variable by clicking Add and navigating to the Branch drop-
down menu, navigate to Demand Sites and Catchments \ Tributary Headflow \ Forest.
In the Variable drop down menu, select the root zone conductivity variable, which is
specific to the land use type.
Change the Parameter Name Abbreviation to Root_Zone_F.
This is because WEAP needs to differentiate between the same parameter for
different land use types (F is for forest).
For the values, add:
Lower Bound 5
Upper Bound 50
Initial Value 27.5
Lastly, using the same process, add the root zone conductivity for grasslands, saving the
Parameter Name Abbreviation as Root_Zone_G.
Parameter Abbreviated Lower Upper
name Bound Bound

Deep Water Capacity Deep_Water_C 0 1500

(mm)

Root Zone Conductivity, Root_Zone_F 5 50

Forest

Root Zone Conductivity Root_Zone_G 5 50

In the PEST window, select the Streamflow Gauge as Gauge from our model.
Note that you can also select the years or months (for monthly time series
Wizard values) to calibrate to.
Under Scenarios to calibrate, it says Reference but it will also modify
the Current Accounts unless you uncheck the box Modify parameters in
Current Accounts also.
 Click on Build Files and Run PEST.
[Then go get yourself some coffee, because this will take several minutes].
PEST will be run iteratively to find parameter values that best fit the
streamflow. Keep in mind that the parameter values may not always make
physical sense, especially in relation to each other. For example, it could
suggest that the soil water capacity of the grasslands was five times higher
 than for forests. Usually a combination of manual and PEST-calibration can
work well.
The model may run for >10 iterations (several minutes) till finally it stops
and the final values of the PEST-calibrated parameters are shown, in the
Scenario Explorer View. Hover the mouse cursor over the slider bars for the
data variables at the top of the Scenario Explorer to see the actual values
chosen by PEST.
Look at the values for the model.
Root Zone Conductivityforest = 41.3
Root Zone Conductivitygransslands = 50
Deep Water Capacity = 43.386461
The Root Zone Conductivity for grasslands is at the threshold value that we
provided, begging the question of whether that should have been larger.
Regardless, the travel time in the upper bucket is much shorter. Similarly, by
reducing the lower bucket to such a small capacity, the travel time there is
significantly decreased.
Explore the results. Look at the Favorite Chart (notice that PEST will generate its own
favorite charts when it runs). Is the PEST calibration better than manually calibrated
results?

Calibration Ideal value Manual Manual PEST

Measure Calibration 1 Calibration 2

NSE 1 -0.27 0.86 0.87

PBIAS 0 -70% -18.65 -16.44

RSR 0 1.12 0.37 0.36

SDR 1 1.79 1.14 1.13

All of our statistical indicators have improved, although it is important to note

that this is not always the case. Any users that deploy PEST repeatedly will
quickly discover that it does not always produce results that improve the
calibration.
Additionally, check the results for Relative Soil Moisture 1 (%) and Relative
Soil Moisture 2 (%). If these results are not stable over time, there is still more
work to be done despite the satisfactory statistical measurements.
Note that many of the results depend upon ranges and initial values. WEAP
users can experiment with the calibration values by combining manual and
automatic calibration processes to produce the best results.
The calibration process benefits from knowledge of the system, as much data
is available, and a large amount of patience, diligence, and repetition. Real
world results are complicated by many factors that cannot be captured with
WEAP, especially within monthly timesteps, so it is always important to judge
what aspects of the model can and should be captured. Experience will
provide more insight over time, this exercise has been an introduction to the
general concept, theory and execution of a streamflow calibration.