水保所知识产出(1956---)知识库

The benefits of the Taguchi method have been demonstrated in diverse fields. However, its applicability to soil
erosion experiments at the slope scale under various conditions remains unclear. Using 6 published datasets that
included the dependent variables of erosion rate (D), runoff rate (R), sediment concentration (C), flow velocity
(V) and transport capacity (Tc), we compared the results from the full-factorial design method with those from
the Taguchi and orthogonal design methods to validate the applicability of the Taguchi method to simulated soil
erosion experiments at the slope scale under different conditions. The statistical parameters of the dependent
variables from the orthogonal design and Taguchi method were very close to those from the full-factorial design
for all dependent variables. The trends of the main effects based on different factor levels were consistent for 35
out of 45 sets from the full-factorial design and Taguchi method and for 27 sets from the orthogonal design. The
optimum conditions for 10 and 7 dependent variables (out of 14) for the Taguchi and orthogonal design methods
were the same as those for the full-factorial design, respectively. In addition, 13 and 8 dependent variables (out
of 14) for the Taguchi and orthogonal design methods had the same factor rank order of percentage contribu-
tions as the full-factorial design method, respectively. Based on a univariate analysis of variance, the evaluating
indicators for predictive power, including the determination coefficient, Nash-Sutcliffe efficiency, relative root
mean squared error, mean absolute percentage error and Thiel inequality coefficient, indicated that the Taguchi
method predictions were better than the orthogonal design predictions. Overall, the Taguchi method could
obtain more reliable conclusions for soil erosion than the orthogonal design method at the slope scale. These
findings suggest that the Taguchi method could be successfully applied to soil erosion experiments and could
better replace the full-factorial design method at the slope scale compared with the orthogonal design method.