5.1. The effect of rainfall duration
Rainfall characteristics are the driving force for runoff generation, and different rainfall regimes could result in different runoff generation capacities for individual plot.
The 148 rainfall events of the six plots in the two watersheds (XDG and CBG) were divided into three groups using K-means clustering (Hong, 2003) based upon rainfall depth and duration to test the performance of SCS-CN method for different types of rainfall regimes (Table 4). The classification met the ANOVA criterion for a significant level (***P < 0.001) Rainfall regime 1 is the group of rainfall events with strong intensity, short duration and lower precipitation, which had the largest occurring frequency, occupying 79.73% of the total events; Rainfall Regime 3 consists of rainfall events with low intensity, long duration and infrequent occurrence (3.38%), while Rainfall Regime 2 is composed of rainfall events, which have moderate rainfall eigenvalues, i.e., lower precipitation and shorter duration than Rainfall Regime 3, but higher precipitation and longer duration than Rainfall Regime 1. In general, The mean P and D decreased in the following order: Rainfall Regime 3 >Rainfall Regime 2 >Rainfall Regime 1. This classification of rainfall regimes is consistence with Wei et al. (2007) and Fang et al.(2008), which also classified the rainfall into three rainfall regimes with 14 and 9 years of field measurements on the Loess Plateau, respectively.
Figure 6 shows the predicted runoff values plotted against the corresponding measurement values for the three rainfall regimes. The original SCS-CN method under-predicted most storm-runoff events of the Rainfall Regime 1 and over-predicted the events of Rainfall Regime 3, while the Rainfall Regime 2 seems preformed better than Rainfall Regime 1 and 3, but the points of Rainfall Regime 2 still lie scatter to 1:1 line (Fig. 6a). This is because the SCS-CN method only account for the rainfall amount but ignore the storm duration. The runoff predicted by the conventional SCS-CN method were increased as the rainfall amount increased from Rainfall Regime 1 to 3, while the storm duration from Rainfall Regime 1 to 3 is also increased thereby the measured runoff did not increase with rainfall amount monotonously, which finally made the SCS-CN method perform poorly. However, when incorporating the storm duration in the proposed method with Eq.8, under-prediction of the Rainfall Regime 1 and over-prediction of Rainfall Regime 3 were removed and also improved the performance of predicting Rainfall Regime 2 with most of the data points lie close to 1:1 as compared with the traditional SCS-CN method (Fig. 6b). The better performance indicated the storm duration plays a significant role in rainfall-runoff production and estimation, and the proposed method can account for different type of rainfall regime with varying duration. Other also confirmed that storm duration plays a vital role in both runoff generation and prediction (Fang et al., 2008; Mishra et al., 2008b; Reaney et al., 2010)