Auto Calibration

What is it?

Auto calibration is used to calibrate and validate SWAT model outputs. The auto calibration tool uses Non-Dominated Sorting Genetic Algorithm II (NSGA-II) developed by Ercan and Goodall (2016)*. Auto calibration tool at SwaNET minimizes three objective functions: 1-NSE (Nash-Sutcliffe efficiency), absolute value of PBIAS (percent bias) and RSR (root mean square error to observations standard deviation ratio). In this method, a set of initial parent population (input parameter values) are selected and the SWAT output for each parent is evaluated against the objective functions using user uploaded observed values. Best parents are selected and are then subjected to reproduction, crossover and mutation. The process continues until the number of iterations are reached. Most significant calibration parameters for stream flow, sediment, nitrogen and phosphorus can be selected and optimized. After the auto-calibration process is complete, the results can be plotted and desired input parameters can be applied to the SWAT model.
3D view

How to use it:

Observed streamflow data

User can either use the observed streamflow file already uploaded to the server or can upload a new observed streamflow file.
To use the observed streamflow file already uploaded to the server, click "Use Uploaded File". A list of already uploaded files appears. Click on the desired file name.
To upload a new observed streamflow file, check the box next to 'Import' and click 'Import'. Browse to the desired observed streamflow file. Observed streamflow file is a CSV file. Example of observed streamflow data is shown below.
Observed:Kalamazoo-Daily
1985,1,1,0
1
0,49.5544
1,58.332608
2,66.261312
3,70.225664
.....
Line 1: Comment. If the observed data is monthly, user must specify. Example: xxx:yy-Monthly.
Line 2: Observation start date (Year, Month, Day, Hour).
Line 3: Multiplier to convert observed values to SWAT unit. This is useful if the observed values are not in SWAT units.
Line 4 and below: Day after start date, observed value. If there is any missing value, don't include it.

Viewing the uploaded data

Click 'Plot data' button to create

Selecting subbasins for calibration

After selecting calibration data file, select subbasin from the list above 'Subbasin/Reach'. Click 'Select' to select and add subbasin-observed data pair to the table on the right. User can select multiple subbasin-observed data pair for calibration. To remove the selected pair from the list, click 'Clear' button on the right.

Select SWAT output to compare

SwaNET allows users to perform auto-calibration of SWAT model by comparing observed data against any SWAT subbasin/reach level outputs. Hence, user should know what kind of observed data is being used (e.g. stream flow, sediment, nitrogen).
Under 'Select output to compare', select whether the observed file corresponds to reach or subbasin level output. Once selected, corresponding outputs are listed on the right. Select appropriate SWAT output.

Calibration options

Calibration period: select the duration of calibration period.
Validation period: select the duration of validation period. Calibration and validation period must be continuous.
Warm-up years: enter the number of warm-up years for SWAT. Usually it is 1-2 years.
Calibration/validation time step: select either monthly or daily. If the uploaded observed data is monthly, select monthly calibration. Daily data is not aggregated to monthly values. .

Parameter selection

Under 'Auto calibration parameters', select parameters to optimize. Stream flow, sediment, nitrogen and phosphorus calibration parameters are listed in order. Once selected, click 'Add'. The selected parameter with its SWAT range and operation is added to the list below. Please note, some SWAT parameters are replaced with user value (Replace) while others percentage of original values are added/subtracted (Add(%)). Once added, parameter ranges can be modified.

Auto calibration options

Population size: Initial population size. Must be an even number.
Number of generations: Number of generations to run genetic algorithm.
Cross-over probability: This is the probability that cross-over happens in the algorithm. Ranges from 0.5-1.0.
Cross-over type: This defines how cross-over occurs. In simple cross-over, genes are divided into segments and replaced, while in uniform cross-over, genes are randomly selected and replaced.
Number of bits: This represents how many bits assigned to each variable(parameters).
Mutation probability: This represents the likelihood of mutation of each gene. Ranges from 0-1.
Seed: This is the seed for generating random number in the model. Ranges between 0 and 1.
Sampling interval: Number of Latin Hypercube Sampling intervals for creating initial population.
Read from last population: If checked, the model starts from the last population created in previous runs.

*Ercan, M. B., & Goodall, J. L. (2016). Design and implementation of a general software library for using NSGA-II with SWAT for multi-objective model calibration. Environmental Modelling & Software, 84, 112-120.

Load/run calibration

To load the results from last calibration run, click 'Load latest results'. If no calibration results are available, this button is disables.
To run calibration, click 'Run Auto Calibration'.
Auto calibration can be stopped any time by clicking 'Stop Autocalibration'.

Once complete, a new window pops up, showing calibration results.