To test the applicability and performance for our E-flow approach a case study was selected for which a detailed reach scale Delft3D hydro-morphodynamic model and a habitat suitability model for the Ayu fish were available (Van Oorschot et al., 2021). The goal is to test if global data sources can provide the necessary input to do an ecological assessment on a fish species group. It is a reconnaissance to investigate if the uncalibrated global data and tools are able to reproduce the right order of magnitude of the ecological input variables and where additional analysis or data is needed.
Case Description
The Ayu fish is a migratory, rheophilic fish species that migrates up to the Tneryuu river in Japan to feed in summer and spawn in autumn, after which larvae drift towards the sea to grow up and migrate up the river again in spring. Most fish die after spawning. Fish life stages can be divided into spawning, egg-incubation, descending, ascending and feeding. The most critical life stages are ascending, spawning and egg-incubation. For ascending the flow velocities cannot be too high. For spawning and egg-incubation water depth and flow velocity are the dominant variables.
To compare the model, observed and global data, information was extracted and run from 1st of September 2009 to the 1st of September 2010. This was the modeled timeframe for the habitat study of the Ayu fish in the Tenryuu for Jpower. This is an average discharge year based on the flow duration curve analysis from discharge data from 2005 – 2019.
Method
The HydroMT suite was used to create a model setup for Wflow. HydroMT can delineate a river catchment when coordinates within the river system are provided. These coordinates should be within the river, preferable several km upstream of the outflow point. For the Tenryuu River http://bboxfinder.com/ was used to select the coordinates for the Tenryuu catchment [137.817478, 34.753038]. This resulted in a delineation of the whole river catchment from the source of the river in lake Suwa, up to the river mouth at the Philippine Sea (Figure 1).
Figure 1: Basemap with a DEM of the Tenryuu catchment and schematized rivers, reservoirs, lakes and gauges.
The following code is run in Jupyter notebook to setup the Wflow model. Note that an anaconda environment should be installed with a specific HydroMT setting from which the Jupyter notebook should be opened. A description can be found on Github[1]. Also a wflow_build.ini file should be in the run folder. An example file can be found on Github[2].
[1] https://github.com/Deltares/hydromt
[2] https://github.com/Deltares/hydromt_wflow/tree/main/examples
###
!hydromt –models
!hydromt build wflow "./wflow_tenryuu" "{'basin': [137.817478, 34.753038]}" -i wflow_build.ini --dd -vv
- "./wflow_tenryuu" = the folder where the main wflow model is located
- ini = the *.ini-file with all the model parameters
- --dd = make use of the Deltares databases (this is required for some areas)
###
Now the basic model setup is complete, but this has to be updated to the settings that we need. This can be done by updating the model forcing. Now, other maps can be added, the time frame can be edited and different scenarios can be created. The new settings can be updated through a 'wflow_update_forcing.ini' file. For the Tenryuu study, the precipitation data in the ‘precip_fn’ parameter should be changed from ‘Chirps’ to ‘era5’. Also, the start and end time can be edited here. We need to run one year but include a spin-up time of an additional year. Therefore, the start time is September 1st, 2008 and the end time is September 1st, 2010. A new folder needs to be manually created to store a new model setup with the updated output. The following code must be run to update the forcing.
###
import xarray as xr
# import hydromt
import hydromt
from hydromt.models import MODELS
# setup logging
from hydromt.log import setuplog
logger = setuplog("update_model_forcing", log_level=10)
# update wflow and create model setup in new folder
! hydromt update wflow wflow_tenryuu -o wflow_tenryuu_updated -i wflow_tenryuu/wflow_update_forcing.ini -v
- wflow_tenryuu = the folder in which the model is located
- wflow_tenryuu_updated = the folder where the new updated model is located
- wflow_tenryuu/wflow_update_forcing.ini = the folder with the *.ini file in which the parameters are defined to update the model
###
The automated setup creates several spatial files which are stored in the ‘staticgeoms’ folder. These can be used to plot the basins, rivers and gauges on a DEM (see example in Figure 1). Also a *.toml file is created which will be used as the input file for Wflow. This file can be manually edited to include additional parameters. For this study we need to include water depth. This can be added by writing the line h = "water_depth" in the [output.lateral.river] block. Water depth is calculated with the kinetic wave equation using a rectangular channel[1]. The river width to calculate these values is extracted from global data and also written as output. In addition, also river width needs to be in the output. width = "river_width". Also an output.vertical block needs to be added:
[1] https://deltares.github.io/Wflow.jl/dev/lateral/kinwave/#Kinematic-wave
###
[output.vertical]
precipitation = "precip"
interception = "int"
###
Additional keywords that can be added to the *.toml file can be found on Github[1]. To run the model, a separate installation of Wflow should be used. The model can be run with a *.bat file that links to the *.toml file.
To include suspended sediment in the model, a wflow-sediment model has to be created and run. This can be done by updating the wflow model by including sediment parameters.
[1] https://deltares.github.io/Wflow.jl/dev/lateral/kinwave_params/
###
!hydromt update wflow_sediment "wflow_tenryuu_updated" -o "./wflow_tenryuu_sediment" -i wflow_extend_sediment.ini -vv --dd
- "wflow_tenryuu_updated" = the wflow model with updated forcing
- ./wflow_tenryuu_sediment = the new wflow folder with sediment model settings
- ini = *.ini file with sediment settings
###
This will create a new model setup. Several things need to be adjusted and checked in the wflow_sediment.toml file
- The path_forcing in the [input] block needs to refer to the output of the model with the updated forcing (output.nc file)
- The start- end end-time need to be changed if necessary
- All the parameter names in the [input.vertical] and [input.lateral.river] should match the names given in the original output blocks of the wflow_sediment.toml file
Now the wflow model can be run and suspended sediment can be extracted.
Results
Catchment Model
From the wflow_sbm and the wflow_sediment model, data was extracted for water depth, discharge, river width and suspended sediment. Flow velocity was calculated with the following formula:
Overview of median discharges during the the Ayu life events. Left: ascending, middle: egg-incubation, right: spawning.
For each of these parameters statistics were calculated over the time period ascending, spawning and egg-incubation take place. For all variables, except suspended sediment the median values over these periods was calculated. For suspended sediment the maximum values were used. Figure 2 shows the median discharges during several Ayu life events. Note that the legend values are different and that the discharge is highest during ascending. Figure 3 shows the model results of the other parameters for ascending. At first glance, all the parameters are within the expected order of magnitude.
Figure 2: Overview of median discharges during the the Ayu life events. Left: ascending, middle: egg-incubation, right: spawning.
Figure 3: Overview of median water depth (left), median flow velocity (middle) and maximum suspended sediment concentrations (right) during ascending.
Local scale model
The catchment scale model results were compared to the Delft3D local scale model. Therefore, a specific set of coordinates was selected of points that were located in the Delft3D model domain (Figure 4). In total, 9 data points were selected that were in the modeled river and floodplain below the Funagira dam.
Figure 4: Overview of the extent of the catchment scale model and the local scale model several kilometers downstream of the Funagira reservoir.
Comparison catchment and local scale
The catchment scale wflow model is a 1D model that works on a different resolution than the local scale Delft3D model. Therefore, model results need to be compared on reach scale level and not on the level of cross sections. The first step is therefore to compare the discharge dynamics during each of the investigated Ayu life events (Figure 5).
Figure 5: Comparison of the catchment scale wflow model and the local scale Delft3D model for the ayu life events in boxplots.
Table 1: Statistics for water depth for all events and both models.
Water depth | Spawning | Egg-incubation | Ascending | |||
Statistic | Wflow | D3D | Wflow | D3D | Wflow | D3D |
Mean | 2.8 | 4.2 | 2.4 | 4.2 | 3.2 | 4.0 |
Min | 0.7 | 0.0 | 0.6 | 0.0 | 0.8 | 0.0 |
Max | 28.3 | 18.2 | 28.3 | 18.2 | 24.9 | 19.1 |
5P | 0.9 | 0.0 | 0.8 | 0.0 | 1.1 | 0.0 |
25P | 1.32 | 0.9 | 1.2 | 0.8 | 1.5 | 0.8 |
Median | 1.74 | 3.1 | 1.5 | 3.0 | 2.0 | 3.0 |
75P | 2.6 | 6.5 | 2.3 | 6.6 | 3.0 | 6.1 |
95P | 8.4 | 11.9 | 7.2 | 12.0 | 9.3 | 11.5 |
Table : Statistics for flow velocity for all events and both models.
Flow velocity | Spawning | Egg-incubation | Ascending | |||
Statistic | Wflow | D3D | Wflow | D3D | Wflow | D3D |
Mean | 1.6 | 0.3 | 1.5 | 0.2 | 1.8 | 0.5 |
Min | 0.3 | 0.0 | 0.3 | 0.0 | 0.3 | 0.0 |
Max | 6.3 | 6.5 | 6.2 | 6.5 | 5.3 | 6.5 |
5P | 0.3 | 0.0 | 0.8 | 0.0 | 0.4 | 0.0 |
25P | 1.2 | 0.1 | 1.1 | 0.1 | 1.3 | 0.1 |
Median | 1.6 | 0.1 | 1.4 | 0.1 | 1.7 | 0.3 |
75P | 2.1 | 0.4 | 1.9 | 0.3 | 2.2 | 0.7 |
95P | 3.1 | 1.1 | 2.7 | 1.1 | 3.3 | 1.6 |
Discussions and Recommendations
This case study has demonstrated that an uncalibrated wflow model using global data is able to predict discharges that are within the range of the discharges modeled with an advanced, calibrated 2D model like Delft3D. Water depth and flow velocities showed larger differences. Water depth and flow velocity are derivatives of discharge and river bathymetry. Since the wflow model assumes a rectangular shape as a cross-section, it is like comparing apples and pears when comparing these variables to the 2D results with detailed cross-sections. Also the river width in the wflow model is assumed to be equal over time. River width is determined based on a bankfull discharge and therefore only shows variation over the longitudinal gradient. These assumptions limit a detailed prediction of water depth and flow velocity. An improvement can be made by using a river width that is variable over time. Also, the assumption for the cross-sectional shape should be improved. A first step could be to assume a simple, but more realistic cross section with a sloping river bank (Figure 6, right). In this way, laterally varying water depth and flow velocities can be derived based on the discharge.
Figure 16: Left: current cross-section type in wflow, right: proposed new cross-section type to improve predictions on water depth and flow velocity.
The results of this case-study are specific for the Tenryuu basin, other basins might produce different results due to differences in the quality of global data. More case studies in different climatic regions and different types of rivers should be tested to provide a more complete overview. Additionally, for this study, only water depth and flow velocity were compared. It was also possible to model suspended sediment with the wflow_sed model. However, these values could not be compared with the Delft3D model because suspended sediment was not included. Also, based on personal communication with H. Boisgontier it became clear that the sediment model still needs calibration, mostly of the D50 value, to give proper predictions of the suspended sediment transport. Furthermore, it is recommended to investigate if other morphodynamic parameters and water quality parameters like water temperature, dissolved oxygen or toxicity are globally available and can be extracted in a similar way.
Conclusions
This case study was meant to test if global data and tools could be used to extract ecological variables as input for REACT. Several hydrodynamic variables could be modeled and extracted. This shows the potential of using global data and tools as the basis of the REACT. In the modeled area discharge could be predicted well, but water depth and flow velocity differed between the catchment model and the local model. Recommendations are to improve the prediction of these variables by using a variable river width and more realistic cross-sectional shape. In addition, it is relevant to investigate the applicability of other ecologically relevant morphodynamic and water quality parameters to be used in REACT.
