Table 3 Summarized values of criteria in 6 test watersheds.
Related Figures (10)
Figure 1. Location of the Yangtze River Basin in China and distribution of the 120 study watersheds control stations. where f(-) is the objective function; A, is the noise that adds up to parameter 0,; N is the number of samples; and Siis averaged from Sij of N samples. A large Si corresponds to high sensitivity of 0;. Sensitivity analysis. Global sensitivity analysis (GSA) increasingly evolved in the assessment of hydro- logical models due to the development of computer technology*’. In the previous literature and corresponding application, many different GSA methods have been compared and proposed in recent years**“°. LH-OAT, RSA, and GLUE are commonly used model sensitivity analysis techniques. In this paper, the LH-OAT method’ is applied to analyze the sensitivity of parameters in the constructed model, and it maintains the robustness of the LH method and the accuracy of the OAT method. Loop work is operated in LH-OAT, that is, for parameter i in sample j, the relative sensitivity index S; is calculated from (12,13) Figure 2. Distribution plot of the Mean of Squared Residuals and explanation degree over the change of Mtry. Figure 3. Plots of the relation between model error and Ntree. Figure 4. Importance ranking plots of all 16 geomorphological parameters from the RF model. Figure 5. Box plots of the sensitivity index S; of A and Lo, by the constructed model. Table 2. Comparison among v,, v,, and v,, and the UH relative peak error (URPE) results in 6 test basins. Figure 6. Fitting plots between v, and v, at regression stage (a) and validation stage (b). Figure 7. Comparison of GIUHs derived from v,, v, and v, in 6 watersheds in different area size. Figure 8. Comparison among the observed and the calculated runoff hydrographs from GIUH-v, and GIUH- v, in 6 watersheds.
