Assistance with run performance after first calibration

[ERICK RAUL OLVERA PRADO]

Dear WRF-Hydro community,

This is a follow-up mail about a WRF-Hydro run conducted in a catchment within the Papaloapan river watershed in Mexico (please see Study_Area_Q.pdf file).

I have simulated june 2012, the period of time between the vertical dashed lines in the time series of observed streamflow (same attached file), and would like to 

have your input about the performance of the model after some tests where a few parameters were changed, since I believe the results can be improved.

I attach a figure of the simulated and observed streamflow at the catchment outlet (Qmod_exp_14_15_Qobs_B28143.jpeg). It shows two runs, one in which

the bucket model is turned off (Exp. 1) and another where it is turned on (Exp. 2), with the exponential bucket option and Expon=1.  Some of the common 

characteristics for both runs are:

• Horizontal resolution of 1 km for the LSM and 0.5 km for the routing model.

• “Steepest Descent” for surface and subsurface routing option.

• “Gridded routing” for channel routing option.

• Hotstart from a run for the same period (2nd cycle)

• REFDK = 2.0E-6

• REFKDT = 3

• SLOPE = 1

• RETDEPRTFAC = 1

• OVROUGHRTFAC = 1

After some tests, I consider this setup performs better, further adjustment of parameters results in an improvement of the volume but deterioration of the time

to peak of the hydrograph, or vice versa. For reference, I also attach a file (Rainfall_WRF_CHIRPS_Junio_2012.pdf) with the comparison of the input rainfall 

from a WRF reanalysis (10 km resolution) vs the observational database CHIRPS. The figures show the temporal mean over the domain (upper panel) and 

the spatial mean (lower panel) computed within the catchment (red polygon). I am performing a new WRF reanalysis with nested domains so the resolution 

can increase to 1km for the input rainfall, however I consider that the lower resolution is not very bad and could still throw acceptable results.

Any input about how the baseflow and general behaviour of the hydrograph can better match observations would be appreciated.

Many thanks in advance.

Best,

Erick Olvera

Grupo Ingeniería Hidrológica
Universidad Autónoma Metropolitana-Iztapalapa
Edificio T, Cubículo 208
https://erickolvera.github.io/

[aubrey]

Hi Erick:

I'm sure others will have comments, but a few quick points:

1) When you run with GWBASESWCRT = 0 and you have the Noah-MP "slope" parameter > 0, you will not conserve mass. In this setup, water that drains from the bottom of the soil column (at a rate scaled by the "slope" parameter) will become a sink (i.e., it does not have any way to make it into the channel). There are some real-world conditions where you may want this behavior (e.g., if you are draining into karst or cracked bedrock that may not be directly connected to the river network), but you should use this option carefully since it does not conserve mass.

2) A "slope" value of 1.0 (maximum) means your soil column will drain (vertically) very quickly. Again, this might be intentional based on the landscape conditions in your basin, but just want to make sure. Your ET in this setup might be quite low. Sometimes instead I have seen our automated calibration procedure push for a small "slope" value to keep water in the soil column, where ET demand is high and you end up reducing the streamflow bias through higher ET. Again, worth a quick gut-check on whether the partitioning makes sense for your region.

3) Given the large positive bias in the model simulation as compared to the observations, which you are partially mitigating when you are using the GWBASESWCRT = 0 option (which creates a loss), it is worth making sure there are no major upstream reservoirs or diversions that might be impacting the downstream flows.

Looking forward to other folks' thoughts on this.

Thanks!

Aubrey