calibration and validation
The total numbers of observed storm-runoff events in watershed XDG for
calibration and validation were 47 and 29, respectively. For the
calibration dataset, the optimized parameters for the tested methods are
given in Table 2. Table 3 presents a comparison of the overall
performance of the models using these datasets based on statistical
indexes.
4.1.1
The original SCS-CN method
Figure
2 shows the predicted runoff values plotted against the corresponding
measurement values for the calibration and validation datasets. The
original SCS-CN method under-predicted
the
large and over-predicted as well as under-predicted some small
storm-runoff events (Fig. 2a). The statistical indexes of
linear regression in Table 3 also
confirm this. A similar pattern either calibration or validation
sub-datasets was observed, with regression line slopes of 0.83 and 0.77
and NSE values of -118.64% and -182.60% (Table 3),
respectively, which are indicative of the poor performance of the
conventional SCS-CN method.
4.1.2
The Huang et al. (2006) and Huang et al. (2007) method
Huang et al.(2006) and Huang et al.(2007) method incorporated the slope
and soil moisture condition factors into the standard SCS-CN method,
respectively. The prediction of large storm-runoff events was improved
with a higher slope of 0.90 using the Huang et al. (2006) method as
compared with the original SCS-CN method (0.80) (Fig. 2b). However, it
also over-predicted some small storm-runoff events when taken slope
factor account into the original SCS-CN method. Therefore, from Table 3
can be inferred that the Huang et al. (2006) method showed little
improvement for the full dataset as confirmed by the value of theNSE which decreased from -148.17% (original SCS-CN method) to
-177.63%.
Table 3 also shows that the NSE value (-9.8%) of Huang et
al.(2007) method during the validation period was much improved as
compared to the SCS-CN (-118.64%) and Huang et al.(2006) method
(-133.00%), and it was visualized that the over-predicted by the latter
two methods of some small storm-runoff events were mitigated. However,
the regression slope of Huang et al.(2007) method deviates more from the
1 : 1 line than the original SCS-CN and Huang et al.(2006) method, which
suggested the former still underestimated large and overestimated as
well as underestimated small runoff events exist (Fig. 2c).