Channel Routing Method Differences and Parameter Calibration

[zed li]

Hello everyone,

I am writing to seek advice regarding discrepancies in my simulation results when using different channel routing methods in the model.

I have run simulations with two channel routing options:

1. Diff.Wave-gridded: Results match observed streamflow reasonably well.

2. Musk.-Cunge-reach: Results show significant deviation from observations.

To this point, I have only calibrated two parameters:

• REFKDT_DATA in GENPARM.TBL

• MannN in CHANPARM.TBL

I understand that when using Diff.Wave-gridded, the MannN values in CHANPARM.TBL are applied, while for Musk.-Cunge-reach, the Manning’s n is taken from the n variable in Route_Link.nc (currently set to a uniform default of 0.035).

My main questions are:

1. Cause of discrepancy
   Could the difference in results be attributed solely to the change in channel routing methods, or might there be other factors (e.g., model structural differences, parameter interactions, or spatial representation) that I should consider?

2. Parameter equivalence and result alignment
   If I adjust the n values in Route_Link.nc to be hydrologically equivalent to the MannN values I used in CHANPARM.TBL, should I expect the two simulations to produce similar results? Or are the routing methods fundamentally too different for parameter equivalence alone to reconcile outputs?

3. Calibration approach for n in Route_Link.nc
   What would be a recommended strategy to calibrate the n parameter in Route_Link.nc for the Musk.–Cunge-reach method?

Any guidance, references, or shared experiences would be greatly appreciated. 

Thank you in advance for your support.

Zed Li

[Juan C. Tufino]

Dear Zed,

Thank you for sharing your detailed analysis regarding the channel routing methods.

I would like to ask a quick question related to your setup: could you please let me know what forcing data you used for your WRF-Hydro simulations (e.g., atmospheric forcing source, spatial and temporal resolution)?

This information would be very helpful for comparison purposes.

Thank you in advance.

Best regards,

Juan Tufino

--
You received this message because you are subscribed to the Google Groups "wrf-hydro_users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to wrf-hydro_use...@ucar.edu.
To view this discussion visit https://groups.google.com/a/ucar.edu/d/msgid/wrf-hydro_users/059f7a4f-35e4-4503-9a8f-4c79888b1c8cn%40ucar.edu.

[zed li]

Dear Juan Tufino,

I used the GLDAS as the meteorological forcing data source, regridded to my study basin using the official script. The precipitation data were obtained through observation-based rainfall inverse distance interpolation, with a spatial resolution of 1 km and a temporal resolution of 1 hour. The GLDAS data link and the script link are provided.

GLDAS:https://disc.gsfc.nasa.gov/datasets/GLDAS_NOAH025_3H_2.1/summary?keywords=GLDAS

regridding script:WRF-Hydro® Pre-processing Tools | Research Applications Laboratory

 Best regards,

zed li

[Arezoo RafieeiNasab]

Hi Zed, 

Here is my attempt to answer your question: 

1. Cause of discrepancy
   Could the difference in results be attributed solely to the change in channel routing methods, or might there be other factors (e.g., model structural differences, parameter interactions, or spatial representation) that I should consider?

In the current test the mannings are different and the channel routing methodology is different. As you point out, both model structure and parameters are different, so I would not expect similar results. MC is routing the flow on vector based river network, while the Diffusive Wave is routing the flow on the channel grid network. Most of the time, for a given study, users try both methods to find out which one is more appropriate for the case study. A consideration is that the MC runs faster than the Diffusive method, so if the model response was close to each other for computational efficiency, I would have suggested going with MC. However in your case, it seems the DF is superior in performance. 

1. Parameter equivalence and result alignment
   If I adjust the n values in Route_Link.nc to be hydrologically equivalent to the MannN values I used in CHANPARM.TBL, should I expect the two simulations to produce similar results? Or are the routing methods fundamentally too different for parameter equivalence alone to reconcile outputs?

Even if you match the n value between the two experiments, I would still expect differences as the equations solved are different between the MC and Diffusive wave. 

1. Calibration approach for n in Route_Link.nc
   What would be a recommended strategy to calibrate the n parameter in Route_Link.nc for the Musk.–Cunge-reach method?

The PyWrfHycroCalib supports calibration for gridded routing. I do not have ready scripts to do calibration of the n in Routlink, however, it should not be difficult if you would like to do so. I have not done it before so this is not a suggestion based on previous studies, but you might want to consider tying the calibration of the n values to stream order (similar to the CHANPARM.TBL). 

I hope this helps!

Arezoo

To view this discussion visit https://groups.google.com/a/ucar.edu/d/msgid/wrf-hydro_users/9c8c9556-c140-48e4-8d72-f78d99b05532n%40ucar.edu.

--

-------------------------------------------------------------------------------------------------

My working day may not be your working day. Please do not feel obliged to reply to this email outside of your normal working hours.

-------------------------------------------------------------------------------------------------

Arezoo Rafieei Nasab, Ph.D.

Scientist V

NSF NCAR Research Applications Laboratory

are...@ucar.edu

[Juan Carlos Tufino]

Dear Zed Li,

Thank you very much for your previous reply and for sharing the details about your GLDAS-based forcing setup, as well as your use of observation-based rainfall interpolated through the inverse distance method. I really appreciate the information you provided.

I found your approach very interesting, especially the way you managed to integrate the observed rainfall data into the model. Could you please clarify how exactly you incorporated the observed precipitation into WRF-Hydro?
For instance, did you replace the GLDAS precipitation fields using one of the forcing options in the namelist (e.g., FORC_TYP = 6 or FORC_TYP = 7), or did you apply your interpolated rainfall data directly to the GLDAS dataset before running the model?

Your configuration seems to have worked very effectively, and I am truly impressed by your results.

Thank you again for sharing the details of your setup and for your helpful response.

Best regards,
Juan C. Tufino

[zed li]

Dear Juan,

To answer your question: in our configuration, we used FORC_TYP = 1, which corresponds to the default WRF-Hydro forcing format. Instead of using a separate forcing option for observed rainfall, we directly replaced the precipitation field in the GLDAS-based forcing file with our interpolated observed rainfall data before running the simulation. Essentially, we kept all other meteorological variables from GLDAS unchanged and only substituted the precipitation variable with our corrected, observation-based precipitation.

This allowed us to leverage the temporal and spatial structure of GLDAS for variables such as temperature, humidity, and radiation, while ensuring that the rainfall input was constrained by local observations.

Best regards,
Zed Li

[zed li]

Dear Arezoo RafieeiNasab,

Thank you for your reply and guidance. I am currently modifying the value of the n variable based on the order variable in the Route_Link.nc file. I will follow up with the results once the adjustments and testing are complete.

I truly appreciate your kind and helpful support throughout this process.

Best regards,
Zed Li