Unexpected Initial Discharge increase in WRF-Hydro 5.1.1 Simulation

[Aref Farhangmehr]

Dear All

I am a beginner in WRF-Hydro and trying to simulate river flow in a small 900 km² basin. For my first attempt, I used the default namelist (with very small difference DXRT=75 and AGGFACTRT=12) from WRF-Hydro version 5.1.1. At the beginning of the simulation, the discharge at forecast points suddenly increases and then declines (please kindly find the attached pic with red box).

I would appreciate any insights on how to resolve this issue.

Regards

Aref 

[aubrey]

Hi Aref:

This is probably because you are cold-starting the model. Something in your initial conditions (GW bucket storage, soil moisture states, etc.) is leading the model to discharge a lot of water initially before it stabilizes. We always recommend warm-starting the model, so you will want to run your simulation for some period of time (multiple years, ideally) to get stable states for the hydrologic storage components. Then if you start your simulation with spunup restart files, you will not see this jump.

Hope that helps.

Thanks!

Aubrey

[Aref Farhangmehr]

Dear Aubrey;

Many thanks for your kind and helpful response.

As a first step, we modified the zinit value in the GWBucket.nc file to a very low value (0.00001), which helped eliminate the initial high values in discharge at the forecast points. The results were more realistic, and the unexpected high discharge values disappeared.

For the warm-start approach, we used the treadmill method: running the model for the year 2007 four times, each time initializing with the restart files from the previous run. We then started the actual simulation using the final restart file after this spin-up. However, even with this setup (and with the default zinit value of 10) we still encountered an unexpected jump in discharge at the beginning of runs 3 and 4.

How long do you think is generally sufficient for a proper spin-up period in WRF-Hydro in semi-arid mountainous region? Or could there be another aspect I should look into (e.g., soil moisture, routing states, or forcing data)?

In addition, I’d like to mention that for a larger domain (approximately 3500 km²), I tested with a much lower zinit value (0.0000001) in GWBucket.nc, and again observed a sudden jump in discharge at the beginning of the runs!

Again, I really appreciate your time and insight.

Best regards,
Aref

[aubrey]

Hi Aref:

When you are restarting the model, are you specifying to initialize GW stores from the restart (GW_RESTART = 1) and (if using the lateral routing modules) initializing soil moisture and surface head states from the HYDRO_RST file (rst_typ = 1)? If so, I can't think of any reason you would see that initial pulse. You might test to make sure your model code and setup are passing a perfect restart test, where you run the model for some period of time (e.g., a few days), output restarts mid-way through the run, and then do a second run restarting from your mid-way point restart files. Your model states at the same point in time at the end of your first run should be exactly the same between the 2 simulations (in other words, you should get the same answer no matter if you ran the model straight through or stopped and restarted midway through).

Let us know how it goes.

Thanks!

Aubrey

[Aref Farhangmehr]

Hi Aubrey;

Many thanks for your suggestions and insight. We didn't turn on the GW_RESTART flag in the namelist (GW_RESTART =0) in the first attempts. while, after activating GW_RESTART flag in the namelist (GW_RESTART =1), the sudden jump at the beginning of the simulation disappeared in the second run.

Again thank you for your help and hope we learn more from you and your team.

Best regards;

Aref

[Aref Farhangmehr]

Dear Aubrey and WRFHydro Users;

 

I hope you're well.

I'm running a model but have encountered an issue where there's no baseflow contribution to streamflow. The hydrograph shows only short-term pulses immediately after rainfall events, with no sustained baseflow between them (please see the attached figure). This doesn't align with expected watershed behavior, where we should observe continuous baseflow maintaining streamflow between storms. 

I've attached the hydro and LSM namelists for reference.

Could you please advise on what might be causing this and how we might address the issue?

 

Best regards,

Aref

[aubrey]

Hi Aref (now cc'ing in user group):

To me it looks like you may not have enough precipitation. Is your precipitation data source of sufficient quality and resolution to capture the behavior in your small basin? Are your units correct? Your hydrograph does show a recession, but the volume is overall far too low.

Also, by looking at your streamflow observations, your basin has pretty consistent baseflow, indicating a steady groundwater or reservoir source of water that you may not currently have in the model.

You can generally maximize streamflow by setting your "slope" parameter to 1.0, which will drain the maximum amount of water into the groundwater buckets for release into streamflow. I have also setup the model with a deeper soil column (e.g., 10-20m) and turned off the GW baseflow parameterization (setting slope=0) to represent stable baseflow, so that is another option. It really depends on the characteristics of your watershed.

But as always - make sure you have reasonable precipitation inputs first. No amount of physics or parameter tweaks can make up for large precipitation biases.

Thanks!

Aubrey

[Aref Farhangmehr]

Hi Aubrey,

Thank you for your helpful response, and apologies for the delay in replying — we've been actively working on the issue.

Following your advice and insights from Mascaro et al. (2023) [https://doi.org/10.1111/1752-1688.13076], we investigated the possibility that most of the precipitation is infiltrating and contributing to subsurface flow, rather than surface runoff.

To explore this, we extracted the UGDRNOFF variable from the LDASOUT_DOMAIN1 files. Since UGDRNOFF is reported in mm, we converted it to mm³/s by summing the values over our basin, multiplying by the grid cell area, and dividing by 3600. The resulting time series (please see the attached plot) confirms that a significant portion of the water infiltrates into the ground, with only a small fraction contributing to streamflow in the reaches. Interestingly, when we summed the groundwater component (UGDRNOFF) with the simulated streamflow at the forecast point, the total discharge more closely matched the observed streamflow. Could you please confirm whether this interpretation is correct?

Also, could you please advise us on how to encourage more of the infiltrated water to enter the stream network, in order to increase baseflow contribution in streamflow and better match observed discharge?

Thanks again for your support and guidance — we're learning a lot from you and the user group.

Best regards,
Aref

[Aubrey Dugger]

Hi Aref:

What is your GWBASESWCRT option? If you are using option 1 or 2, the UGDRNOFF will be mapped to your channel as baseflow. If you are using option 0, that water becomes a sink and will not make it through to the channel. You can see a description here:

https://wrf-hydro.readthedocs.io/en/latest/model-physics.html#conceptual-base-flow-model-description

Thanks!

Aubrey

--

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

Aubrey Dugger

NCAR Research Applications Laboratory

Office: 303-497-8418, Cell: 310-663-5115

adu...@ucar.edu

[Aref Farhangmehr]

Thank you very much for your kind response.

We are using option 1 (exponential). How can I increase channel baseflow dut to UGDRNOOF?

Again thank you.

[Aubrey Dugger]

Hi Aref:

You might try running with option 2 (pass-through). This will push ALL of the UGDRNOFF through the groundwater basins and into their associated channels without storage or attenuation. If you are still seeing differences between streamflow and your basin-aggregated UGDRNOFF values using option 2, I would guess the mapping between your groundwater basins and your channel cells is not setup properly or you are looking at streamflow in a place on the network that is not an outlet for the full basin. If you want to share your domain and namelists I can take a look.

Thanks!

Aubrey