I have been applying SIMHYD model in the southeast Australian catchments with the daily rainfall, streamflow and potential evapotranspiration data. Unfortunately, the calibration result varies within the negative range. I have tried 100s of combinations in changing calibration periods, parameters value etc but none of these are working. Has anyone tried SINHYD? Please share your experience. Thanks
Hi Muhury,
I didn't use this model before, but I will try to give some useful suggestion as I am also working with hydrological modelling. Can you elaborate on the input data (measured or from satellite?), what about the studies that have done in your area, did you check them? what is the typical ranges for the parameters?
The discharge is often very sensitive to the chosen parameter sets, input data (e.g. precipitation, evapotranspiration) as well as the model structure. Whether the model have different structure on it.
It is also important to understand the physical meaning and effect of each parameter. One thing can help is doing sensitivity analysis in which you specify the parameters that considerably influencing the performance of your model. Therefore, calibrating only base on these parameters starting with the most sensitive ones.
I hope these help!
Good luck
Hi Awad Mohammed Ali
Thank you for your suggestion. I am using station-gauged rainfall, streamflow and calculated potential evapotranspiration data for the calibration. I hope your suggestion will help me to improve the result. Thanks
One of the main reasons you could get a poor result using SIMHYD (or other hydrological models) is problems with the data. How large is your catchment and how many rain gauges are you using? Where did the potential evapotranspiration data come from? Finally, how good is the streamflow gauge? Have you looked at the gauging data and the rating curve?
A good way of investigating data issues is looking at the cross-correlation between rainfall and streamflow. The peak of the cross-correlation should be at a lag >=0, and have a value that is >~0.3 (preferably much higher, though how high depends on the unit hydrograph for the catchment.
Another common issue is correctness of model setup and data handling.
Hard to give concrete advice without knowing the data in use, where the model came from or its implementation (e.g., are you using a package?), but I suspect it's to do with model setup given you're getting negative values for the objective function (very suspicious!). Suggests data issues or error in model implementation.
Example results from my own work show you can get "okay" results with SYMHYD (https://connectedsystems.github.io/Streamfall.jl/dev/model_comparison/). Note, however, that the example only runs the calibration for 3 minutes.
Are you sure:
1. You are giving the model the right data in the right shape? e.g., it could be that rainfall is being passed in instead of streamflow.
2. Model parameters are being updated correctly during the model calibration process? Order of the parameters may have been mis-specified for example.
3. Is the calibration process correctly set up? Some default to minimize, others are set to maximize the objective function. On that note, what objective function are you using? Plot your "calibrated" model results against observed data if you haven't already.
Dear Barry Croke & Takuya Iwanaga thank you so much for your suggestions. The modelling results have eventually increased after using satellite derived evapotranspiration data. I have also recalculated the catchments area for the study catchments. The calibration results produced more than 0.8 NSE values. I appreciate your support and time.
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing