Abstract:
Abstract: During plot experiments with full factorial designs in order to understand the effects of factors such as soil type, slope gradient, slope length, rainfall characteristics, hydraulic factors, vegetation cover and human activities on soil erosion processes, the number of tests increases with the increasing number of factors and their levels, often leading to time, cost and labor consuming. Taguchi method has been evidenced as a simple, economical and efficient experiment design method in fields of engineering, environment and others, but not in soil erosion studies. This study investigated the feasibility of using Taguchi design method in studies on soil erosion under different conditions. To achieve this objective, a published dataset of mini-plot soil erosion by Meyer and Harmon in 1989 was used. This dataset included four factors and four levels of soil type (Brooksville, Dubbs, Atwood and Loring), slope gradient (5%, 10%, 20 and 30%), slope length (0.15, 0.3, 0.45 and 0.6 m) and rainfall intensity (13.6-14、26.4-27、74.3-76.3 and 112-115 mm/h). Results from Taguchi method, orthogonal design and full factorial design were compared. For comparison, the results from full factorial design were considered as measured values. Results showed that the test numbers used for Taguchi method/orthogonal design were only 6.3% of that for full factorial design, but there was no significant difference in mean of runoff and erosion rate between Taguchi method and the other methods. The maximum, minimum and variation coefficient and standard deviation for the erosion and runoff rates between the methods were slightly different. The main effect of each factor on the erosion and runoff rates obtained from Taguchi method followed the same trends with those obtained from full factorial design, but different from the orthogonal design sometimes. The three methods all could identify that rain intensity was the main contributor to slope erosion. However, the contribution rates from the three methods were different with higher value from Taguchi method than the others, probably because the Taguchi method could effectively reduce contribution of uncontrollable factors such as noise. In general, the contribution of each factor to the variation of erosion and runoff rate from Taguchi method was much closer to full factorial design method. The optimum condition obtained from Taguchi method to produce the maximum erosion was the same with that from the measured results. The determination coefficients (R2) of the predicted values by Taguchi design and the measured values were 0.993 (P < 0.001, N = 254) and 0.929 (P < 0.001, N = 254) for the runoff and erosion rates, respectively. The Nash-Sutcliffe Efficiency index between the predicted values by Taguchi design and measured values were 0.992 and 0.841 for the runoff and erosion rates, respectively. All above results indicate that the Taguchi method can replace full factorial designs in mini-plot erosion studies to produce the credible results and decrease the test numbers and the costs.