Poor statistical measures of SUFI2 calibration

[Somsubhra chattopadhyay]

Dear SWAT-CUP users

I am calibrating my watershed with 2 USGS sites. I am running SUFI2 program for this. I have chosen 10 parameters to calibrate on monthly streamflow basis with 500 simulations for the 1st iteration. I am actually getting very poor values of R square and Nash Sutcliffe efficiency values given by the program. After the 1st iteration with 500 simulations the statistics is like

Variable                      p-factor    r-factor       R2         NS         br2           MSE           SSQR
FLOW_OUT_18          0.40         1.03         0.06       -0.92      0.0099        360.7530      19.4178
FLOW_OUT_49          0.41         1.12         0.04       -0.95      0.0060        1469.2467     86.1662


My observed flow data is for 9 years (monthly basis) and the entire simulation is for 30 years having 2 years as warm up period. So in the sufi2 swedit.def file the beginning and the ending year of simulation is my calibration period of 9 years and not the actual simulation years (is that right?).

My question is how do I make the statistics better and how to improve the calibration? If after 500 simulation the resulting statistical measures are so poor, do I need many more simulations?  After 500-1000 simulations the results are satisfactory but I don't know in this case why it is not.

Please help me to understand this.

Best
Somsubhra

--
Thanks and Regards,
Somsubhra Chattopadhyay
Graduate Research Assistant
Water Resource Research Group
Department of Civil Engineering
North Carolina A&T State University
Greensboro, North Carolina
Email: scha...@aggies.ncat.edu

[Karim]

... Although the r-factor is around 1 the p-factor is rather small around 40%. This means you have chosen bad ranges.... Look at the 95ppu and try to adjust the ranges up and down depending on how the base flows and peaks are simulated. I.e if peak too low increase CN2 etc...

Best,

Karim

[Karim]

Ps the new parameter set should partly do this for you...

On May 20, 2012, at 4:44, Somsubhra chattopadhyay <som.ag...@gmail.com> wrote:

[Somsubhra chattopadhyay]

Dr. Abbaspour

Thanks for the reply. I kept the initial parameter ranges in
accordance with the SWAT suggested range. Before calibration when I
just compared the model prediction with the observed data using SWAT
CHECK ,I found that the streamflow is over estimated. This is the
reason I just did some manual calibration and then kept the ranges in
the par_inf. text file.

10 : Number of Parameters
500 : number of simulations

r__CN2.mgt -0.1 0.1
v__ALPHA_BF.gw 0.0 1.0
v__GW_DELAY.gw 0.0 500.0
v__GWQMN.gw 500.0 5000.0
v__GW_REVAP.gw 0.1 0.2
v__ESCO.hru 0.7 1.0
v__REVAPMN.gw 0.0 500.0
v__EPCO.hru 0.0 1.0
v__ALPHA_BNK.rte 0.0 1.0
r__SOL_BD(1).sol -0.5 0.6

This was my initial ranges. Could you please let me know how can I
improve the ranges or judge that I have chosen bad ranges. Should I
incorporate more parameters with their ranges?

Thank you
Somsubhra

On May 19, 11:24 pm, Karim <k_abbasp...@yahoo.com> wrote:
> Ps the new parameter set should partly do this for you...
>

> > Email: schat...@aggies.ncat.edu

[Karim Abbaspour]

.... use the recommendations in the attached file to adjust your parameters ...

 

Best,

Karim

 

-------------------------------------------------
Dr. K.C. Abbaspour
Eawag, Swiss Federal Institute for Aquatic Science and Technology
Ueberlandstr. 133, P.O. Box 611, 8600 Duebendorf, Switzerland
email: abba...@eawag.ch
phone: +41 44 823 5359
fax: +41 44 823 5375
http://www.eawag.ch/index_EN