a paradox between PBIAS and scatter plot

[Yuan Lifeng]

Hello, everyone, 

        I am simulating daily streamflow by SWAT-CUP with SUFI-2. the value of PBIAS is -33.1% in calibration, according Gupta et al. 1999, positive values of PBIAS indicate model underestimation bias, and negative values indicate model overestimation bias, that means the swat model overpredict daily streamflow during calibration period, is that right?  However, the result of scatter plot is different with above statistical results because the trend line of scatter plot lies below 1:1 line, showing the simulation is underpredicted. Does anyone can help me to understand this question? thanks a lot. 

 Regards,

LIFENG