Sobek3 D-Flow1D calibration twin experiment

This page describes a Sobek3 D-Flow1D calibration twin experiment with OpenDA. More information about twin experiments can be found here. In this experiment the roughness values of a modelled part of the Belgian Meuse are calibrated. This is done by splitting the river over multiple branches with different roughness definitions. Those branches can have different roughness values at different positions. These roughness values can be tweaked by OpenDA for calibration with the use of exchange items. Below we first discuss how the roughness values are divided over the exchange items. For validation of the calibration steps water level values from a Sobek3 D-Flow1D model run are used as observations.

For the twin experiment we disturb the original roughness by adding 1.00. By running OpenDA these roughness values will be changed step by step each time comparing the resulting simulated water levels with the water levels from the observations. This way the roughness values should revert back to the original values from the model run that generated the water level observations which proves the calibration is working correctly.

Exchange items for roughness values

The roughness within a branch can be:

Constant
Flow dependent
Water-level dependent
Location dependent (can be split by observation point)
Water-level and location dependent
Flow and location dependent

This can be specified per section or floodplain. Note that chainages are used to locate points where roughness values change or where observation points are placed. A chainage is a distance measurement along the branch.

The details of the roughness definition are described in the Sobek3 documentation. For integration with OpenDA it is needed that all specifiable values can be changed by the algorithm through exchange items. Values on different locations within one branch can be provided in one exchange item. In the example below of flow and location-dependent roughness there will be 7 exchange items (represented by the orange lines).

When an observation point is placed the branch will be split at the last chainage before the observation point which results in 14 exchange items for the example below.

Constant roughness branch

Below the Belgian Meuse is taken as example, which is split in multiple branches. Branches BM1_10, BM1_11, BM1_14, BM1_15, and BM1_17 all have constant roughness values which are provided to OpenDA with a single exchange item each. The exchange items for branches with constant roughness follow the naming convention <sectionId>-<branchId>-x<chainage>-<roughness type> which results in Main-BM1_11-x0-Chezy for branch BM1_11.

_images/belgian_meuse_constant_roughness.png

Location-dependent roughness branch

When looking closer at branch BM1_12 it can be seen that it has three chainages with on each chainage a different roughness value. These roughness values are provided to OpenDA together in one exchange item. The exchange items for branches with location-dependent roughness follow the naming convention <sectionId>-<branchId>-x<first chainage>-<roughness type> which results in Main-BM1_12-x0-Chezy for branch BM1_12.

_images/belgian_meuse_changing_roughness.png

Location-dependent roughness branch split by observation point

When looking closer at branch BM1_13 it can be seen that it has 4 chainages with on each chainage a different roughness value. Within this branch there is an observation point named Huy at chainage 30325. This observation point will split the branch at the last chainage before or at the observation point and provide the different roughness values in two different exchange items, one item for chainages 0 and 2900 and one item for chainages 30300 and 35400. Following the naming convention <sectionId>-<branchId>-x<first chainage>-<roughness type> this results in Main-BM1_13-x0-Chezy and Main-BM1_13-x30300-Chezy.

_images/belgian_meuse_changing_roughness_split.png

Configuration setup

Directory structure and files

The application to be performed is specified in a so-called OpenDA application (.oda) file. This oda file is the top of a hierarchy of configuration files that is organized in a directory structure that is usually setup as indicated below.

<calibrate_dud.oda>
algorithm contains the configuration file(s) for the calibration algorithm
stochObserver contains the configuration files and measurement data for the so-called stochastic observer, the set of measurements and the specification of their uncertainty
stochModel contains the configuration files for the so-called stochastic model factory, that specify how model instances can be created.
- template contains the base model and the associated model files. The dimr config file belgischeMaas.xml is located here.
- bbWrapperConfig.xml describes the way the model can be accessed. It refers to
  - DimrConfigFile: belgischeMaas.xml
  - Sobek3SpatialDefinitionFile: roughness-Main.ini (contains the input for OpenDA, it specifies the parameters which can be tweaked), ObservationPoints.ini and ObservationPointsSelection.txt (specify which observation points should be used, see paragraphs about splitting further on)
  - NetcdfDataScalarTimeSeriesDataObject: observations.nc (used to compare with and validate the calibration steps results)
- bbModelConfig.xml contains file names of files specified in the wrapper.
- bbStochModelConfig.xml describes which items can be calibrated, and specifies the relation between the measurement series and the related observation point in the model.

Sobek3 D-Flow1D specific configuration

The main configuration element in bbWrapperConfig.xml which is specific for Sobek3 D-Flow1D is the <dataObject> with class name nl.deltares.openda.models.sobek3.Sobek3SpatialDefinitionFile. This is coupled with the file roughness-Main.ini which contains all roughness definitions and values spread over the different branches.

By adding an argument that refers to an observationFile the observation points in that file will be used to split spatially defined roughness over exchange items. By adding an argument that refers to an observationSelectionFile a selection can be made from the points in the observation file:

<dataObject className="nl.deltares.openda.models.sobek3.Sobek3SpatialDefinitionFile">
      <file>dflow1d/roughness-Main.ini</file>
      <id>roughness_main</id>
      <arg>observationFile=dflow1d/ObservationPoints.ini</arg>
      <arg>observationSelectionFile=../ObservationPointsSelection.txt</arg>
</dataObject>

The observation selection file simply contains a list of names of the observation points that should be included. Lines starting with # will be ignored. For example an observation selection file that just selects the observation point Huy could look like:

# selection of observation points for calibration
Huy

Exchange item configuration

To enable OpenDA to tweak the roughness values contained in the exchange items, the exchange items need to be configured in bbStochModelConfig.xml as follows:

<regularisationConstant>
      <stdDev value="2" transformation="set" initial="47"/>
      <vector id="Main-BM1_11-x0-Chezy"/>
</regularisationConstant>
<regularisationConstant>
      <stdDev value="2" transformation="set" initial="47"/>
      <vector id="Main-BM1_12-x0-Chezy"/>
</regularisationConstant>
<regularisationConstant>
      <stdDev value="2" transformation="set" initial="47"/>
      <vector id="Main-BM1_13-x0-Chezy"/>
</regularisationConstant>
<!-- Exchange item created for observation point Huy-->
<regularisationConstant>
      <stdDev value="2" transformation="set" initial="47"/>
      <vector id="Main-BM1_13-x30300-Chezy"/>
</regularisationConstant>
<regularisationConstant>
      <stdDev value="2" transformation="set" initial="47"/>
      <vector id="Main-BM1_14-x0-Chezy"/>
</regularisationConstant>
<regularisationConstant>
      <stdDev value="2" transformation="set" initial="47"/>
      <vector id="Main-BM1_15-x0-Chezy"/>
</regularisationConstant>
<regularisationConstant>
      <stdDev value="2" transformation="set" initial="47"/>
      <vector id="Main-BM1_17-x0-Chezy"/>
</regularisationConstant>

Here, stdDev represends the standard deviation for the parameter correction and initial sets the starting value of the parameter before calibration begins. transformation determines how the correction calculated by OpenDA is applied to the parameter. There are three options:

identity: correction is added directly to the current parameter value;
ln: correction is applied in logarithmic space;
set: correction replaces the parameter value.

When roughness values from exchange items that are not included here are disturbed at the initial state of the experiment, OpenDA will not be able to tweak them so they will remain disturbed which will result in odd calibration results.

Comparing measurements with model values

The observation points are used to compare measured values with values from the model. This comparison is specified by the <predictor> element in bbStochModelConfig.xml, where the id of the vector element refers to the measurements and the sourceVectorId to the model values at the location of the observation points:

<predictor>
      <vector id="BM1_11_00.water_level" sourceVectorId="Chooz.Water Level"/>
      <vector id="BM1_12_2020.water_level" sourceVectorId="Obs12A.Water Level"/>
      <vector id="BM1_12_20800.water_level" sourceVectorId="Obs12B.Water Level"/>
      <vector id="BM1_13_30325.water_level" sourceVectorId="Huy.Water Level"/>
      <vector id="BM1_13_35400.water_level" sourceVectorId="Obs13B.Water Level"/>
      <vector id="BM1_14_00.water_level" sourceVectorId="BM1_14_0.Water Level"/>
      <vector id="BM1_15_10.water_level" sourceVectorId="Obs15.Water Level"/>
      <vector id="BM1_17_00.water_level" sourceVectorId="BM1_17_0.Water Level"/>
</predictor>