5. HYDROSALINITY MODEL APPLICATION IN THE
5.5 Model Set-up
5.5.3 Model Calibration
The accurate parameterization of environmental models is a critical component of their successful application. The direct determination of parameter values is however often not feasible as they lack clear physical meaning or due to the costs associated with field measurements (Fischer et al., 2009). Thus, the estimation of model parameter values is often a process whereby model response is compared to observed data in a trial and error process. The user adjusts model parameters within a specific range at pre-defined intervals and this may be done manually or using automatic techniques. Janssen and Heuberger (1995) recommend that the model calibration process combines both manual and automatic techniques. A comparison of the simulated and observed streamflow is often used to evaluate the process. It is usually suggested that, if possible, calibration first be performed on an annual basis, before progressing to a monthly and eventually daily time interval. It is also recommended that the calibration process follow a model initialisation period, during which model parameters are assumed to adjust to reasonable starting values. The length of this initialisation period and the subsequent calibration period is dependent on the availability of input data and the objective of the study.
The criteria which were used to evaluate the model performance during the calibration process were the Absolute Volume Error (AVE), the Nash-Sutcliffe Efficiency (Nash and Sutcliffe, 1970), the coefficient of determination (r2) and the Index of Agreement (IOA; Willmott, 1981). This is in accordance with recommendations made by Janssen and Heuberger (1995), Krause et al. (2005) and Wagener et al. (2003).
Manual Calibration
Manual model calibration is a process whereby the user manually adjusts model parameters within a specific range at pre-defined intervals. During this process the user also identifies which parameters exert the greatest influence on simulation outputs, i.e. sensitivity analysis. These sensitive parameters are subsequently used in an automatic calibration process.
145 Besides being able to manually edit the text files with input parameters (Chapter 5.5.2) and variables, the interface of the model (Figure 5.15) allows the user to fine tune the calibration by changing certain parameters. The parameters were adjusted using a trial and error process with the objective being to compare simulated and observed runoff. The parameters were adjusted within a justifiable range and at certain intervals, until optimal efficiencies of model performance (statistical indicators) were observed. The manual calibration process is illustrated in Table 5-7.
Figure 5.15. The Graphical User Interface of the JAMS/J2000-NaCl hydrological model.
Table 5-7 The Manual Calibration Procedure and Results
Variable Definition Values range in the calibration Interval used in the calibration End value J2KNaClSoilLayer
Beta_NaCl Percolation coefficient for
inorganic salt 0 - 1 0.1 0.2
Deposition factor (kg ha-1)
Inorganic salt rainfall deposition factor per mm
rainfall
0 – 1.5 0.1 0.8*
Radiation Tab
Longitude of time-zone
center (dec. °) - - - 18
East or West of Greenwich - - - East
Daily or hourly time steps - - - Daily
Parameter a for Angstroem
formula - - -
0.25 (default) Parameter b for Angstroem
146 Variable Definition Values range in the calibration Interval used in the calibration End value Interception Tab a_rain
Maximum storage capacity of the interception storage per m2 of leaf area for rain
0 - 10 0.5 0.15
J2KProcessLayeredSoilWater Tab
soilMaxDPS (mm) Maximum depression
storage capacity 0 - 10 1 3
soilPolRed
Polynomial reduction coefficient for the computation of actual ET
0 - 100 5 80
soilLinRed
Linear reduction coefficient for the computation of actual
ET 0 - 10 1 0 (PolRed or LinRed are alternatives; PolRed was used)
soilMaxInfSummer (mm) Maximum infiltration in the
summer half year 0 - 200 5 30
soilMaxInfWinter (mm) Maximum infiltration in the
winter half year 0 - 200 5 70
soilImpGT80
Relative infiltration capacity of areas with a sealed grade
> 80%
0 - 1 0.05 0.25
soilImpLT80
Relative infiltration capacity of areas with a sealed grade
< 80%
0 - 1 0.05 0.75
soilDistMPSLPS
Calibration coefficient for allocation of infiltration to
LPS and MPS
0 - 10 1 5
soilDiffMPSLPS
Calibration coefficient for diffusion amount of MPS to
LPS
0 - 10 0.05 0.6
soilOutLPS Calibration coefficient for
outflow from the LPS 1 - 10 1 9
soilLatVertLPS
Calibration coefficient for allocation of LPS runoff to
lateral (interflow) and vertical (percolation)
components
0 - 10 0.5 10
soilMaxPerc (mm) Maximum percolation in the
time step 0 - 2000 1 14
geoMaxPerc (mm)
Maximum percolation in the time step (into semi-
consolidated rock)
0 - 2000 2 2
soilConcRD1 Recession coefficient for
overland flow 0 - 10 1 10
soilConcRD2 Recession coefficient for
interflow 0 - 10 1 9
kdiff_layer Layer MPS diffusion factor 0 - 10 0.1 0.1
BetaW Water use distribution
parameter for transpiration 0 - 100 10 10
J2KProcessGroundwater Tab
initRG1 Initial groundwater storage
in RG1 0 - 1 0.1 0
initRG2
Initial groundwater storage
147 Variable Definition Values range in the calibration Interval used in the calibration End value gwRG1RG2dist
Calibration coefficient for
water allocation to
percolation
0 - 10 1 1
gwRG1Fact Factor for runoff
contribution from RG1 0 - 10 0.5 2
gwRG2Fact Factor for runoff
contribution from RG2 0 - 10 0.5 4.5
gwCapRise Capillary rise coefficient 0 - 1 0.1 0.4
NaCl_concRG1 Initial salt concentration in
RG1 per HRU 0 – 10 1 1
NaCl_concRG2 Initial salt concentration in
RG2 per HRU 0 - 10 1 1
J2KProcessreachRouting Tab
flowRouteTA Flood routing coefficient 0 - 100 10 10
* Based on the rainfall salt concentration range (14 - 125 mg L-1) presented by Flügel (1995) and the annual total rainfall presented in Table 5-3, the deposition factor ranges between 0.13 and 1.5.
The “Regionalisation” tab (Figure 5.15) is used to access a screen where information pertaining to the regionalisation of weather variables is entered:
• Number of closest stations for regionalization: Number (n) of stations that are used for the calculation of the climate input values of an HRU.
• Power of inverse distance weighting (IDW) function for regionalization. A higher power results in less influence from distant points and vice versa. The most commonly used value is 2.
• Elevation correction on/off: Activation of the elevation correction.
• r-sqr threshold for elevation correction: Threshold for elevation correction of data. If the coefficient of determination of the regression relationship between the station values and the elevations of the station is less than the threshold, no elevation correction is carried out.
These settings can be determined for each input variable (i.e. minimum temperature, maximum temperature, mean air temperature, precipitation, absolute humidity and sunshine duration) individually.
Automatic Calibration
Automatic model calibration of the JAMS/J2000-NaCl model is supported by a semi-automated assistant that guides the user through the calibration procedure, i.e. OPTAS (Fischer et al., 2009). OPTAS runs on a computing cluster of the Department of Geoinformatics, Hydrology and Modelling at the Friedrich Schiller University Jena (Germany). This does not occupy any local computing resources and allows a calibration of up to four models at the same time. A list of all available parameters for calibration is available, from which the user should make a selection. A list of objective functions (e.g. Nash-Sutcliffe Efficiency) is also available. The user can either select a single criterion or several criteria. When selecting several criteria, a multi-criteria optimization problem is created which differs significantly from a (common) one-criterion optimization problem regarding its solution characteristics.
OPTAS incorporates several optimization methods. The Shuffle Complex Evolution (SCE; Duan et al., 1992) method was employed in this study. The SCE method was developed especially for parameter optimization applications in hydrological models. The method has illustrated its
148 effectiveness, robustness and efficiency in numerous studies, particularly in the field of hydrology (Fischer et al., 2011). The SCE method handles the optimization problem as a natural evolutionary process (Fischer et al., 2009). The SCE method is described by Fischer et al. (2009) as a population of samples, which each represent one solution. The population may be divided into complexes that evolve independently. New samples are created through the formation of new sub-complexes. The new samples, however, need to satisfy certain criteria before they are added to the population whereby they are superseding the current “worst” sample. After some iteration, the complexes are joined. The process of complex segmentation and new sub-complex formation is repeated until no further improvement of the sample “goodness of fit” can be achieved. The SCE algorithm exhibits good convergence for a variety of problems, i.e. a sufficient number of model iterations has a fairly high probability to converge to a global optimum (Fischer et al., 2009). The commonly used “goodness-of-fit” measure, i.e. NSE (Duan et al., 2006) between observed and simulated values was used to evaluate model performance (objective function).
The automatic calibration process was run for the period 01/01/2000 to 31/12/2010. The period 01/01/2000 – 31/12/2008 was used as an initialization period and the period 01/01/2009 – 31/12/2010 was used as the calibration period. The manual calibration processes identified 18 parameters which have a significant influence on model results (Table 5.7), which were subsequently used in the automatic calibration process. The parameter ranges used in the automatic calibration process is also shown in Table 5-8.
Table 5-8 Parameters Selected for Automatic Calibration
Variable Definition
Values range in the calibration
Starting
value End value
a_rain
Maximum storage capacity of the interception storage per m2 of leaf area for rain
0 - 10 0.15 0.26
soilMaxDPS (mm) Maximum depression
storage capacity 0 - 10 3 4.26
soilPolRed
Polynomial reduction coefficient for the computation of actual ET
0 - 100 80 74.47
soilMaxInfWinter (mm) Maximum infiltration in the
winter half year 0 - 200 70 89.86
soilImpLT80
Relative infiltration capacity of areas with a sealed grade
< 80%
0 - 1 0.75 0.75
soilOutLPS Calibration coefficient for
outflow from the LPS 1 - 10 9 9.02
soilLatVertLPS
Calibration coefficient for allocation of LPS runoff to
lateral (interflow) and vertical (percolation)
components
0 - 10 10 9.34
soilMaxPerc (mm) Maximum percolation in the
time step 0 – 2000 14 306.55
geoMaxPerc (mm)
Maximum percolation in the time step (into semi-
consolidated rock)
0 - 2000 2 173.03
soilConcRD1 Recession coefficient for
overland flow 0 - 10 10 9.80
soilConcRD2 Recession coefficient for
interflow 0 - 10 9 7.19
149 Variable Definition Values range in the calibration Starting
value End value
gwRG1RG2dist
Calibration coefficient for
water allocation to
percolation
0 - 1 1 0.90
gwRG1Fact Factor for runoff
contribution from RG1 0 – 10 2 3.00
gwRG2Fact Factor for runoff
contribution from RG2 0 - 10 4.5 3.20
gwCapRise Capillary rise coefficient 0 - 1 0.40 0.47
flowRouteTA Flood routing coefficient 0 - 100 10 1.38
cbWallhoehe Contour bank height (m) 0 - 1 0.64 0.62