3
$\begingroup$

At first run I’ve encountered the problem since the values of modeled and observed discharge are significantly different. The correlation between them is nearly 0,2 as the modeled discharge gives much higher values.
I have already checked the precipitation file (it has some missing values which I substituted with zeros) and also did the correlation between the discharge and precipitation which showed the result nearly 0,35; while the cross correlation with three-day lag is 0,6.
In addition, the model predicts the precipitation values very accurately, though the discharge is very high.
How can I identify what is the conceptual logic behind the algorithm of inflow-outflow calculations in SWAT (or maybe refer me to the literature on the matter) so I could go through the output again and identify the potential problem?

$\endgroup$
1
$\begingroup$

SWAT model has a pretty good technical and theoretical documentation on their website : http://swat.tamu.edu/documentation/

Regarding the conceptual logic, SWAT is based on sub-watershed discretization. Inside these, it defines Hydrological Response Units (similar slope, land use and soil). Then, you simplify it on the basis of the HRU heterogeneity, you may chose between assigning to the sub-watersheds the response of the dominant HRU or define a multiple responses. At this step, SWAT knows which parameters to use in the hydrological processes you have selected and can run the simulation.

I would suggest that you read the introduction of the theoretical documentation and, assuming that surface runoff is a major contributor to your discharge, maybe the chapter 2.1 on surface runoff. You need to understand the Curve Number Method wich is used to compute runoff volumes. If you want to decrease runoff speed, you may increase the manning coefficient (friction to flow) for both overland flow or channel flow. As it seems that you have a time of concentration greater than one day, you should also check the surface runoff lag parameter [SURLAG].

But as SWAT is a sophisticated model with a lot of parameters and sensitive to scale, discretization methods and input resolution, it is hard to know what's wrong with your simulation. It is quite normal that a uncalibrated model, even physically based, provides results far from the observed discharge. To calibrate a model, you must first identify the most sensitive parameters by running a sensitivity analysis. Then, as you know what parameters will have significant impact on the simulated discharge, you can run a calibration by varying them in a physically acceptable range.

Hope this help,

D.D.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.