Using the Taguchi experimental design for assessing within-field variability of surface run-off and soil erosion risk

Water erosion is one of the soil degradation processes driven by environmental and field factors such as rainfall intensity, slope gradient, dynamics of vegetation cover, soil characteristics, and management practices. Most of the studies assess the separate contribution of these factors under controlled conditions. However, there is a lack of adequate knowledge regarding the complex interactions between prevailing factors and soil erosion processes under heterogeneous field conditions. This study investigated 16 combinations of 5 factors at 4 levels of each factor on the soil erosion process using Taguchi's fractional factorial experiment design, identifying the factor combinations resulting in maximum sediment yield, runoff, organic carbon, and nitrogen losses. We considered the factors: Soil organic matter and silt content (SiltOM), vegetation cover (VC), slope steepness (SS), rainfall intensity (RI), and depth to a loamy layer (DLL). The interactive effects of these factors and their combinations were visualized from the analysis of signal-to-noise (S/N) responses. Results indicated that interactions between the selected factors and soil erosion processes exist and multiple linear regression models were developed to predict sediment yields, runoff, carbon, and nitrogen losses at the sub-field scale. Results revealed that 1) RI with 40.6% showed the highest contribution to sediment yield followed by SS (23.8%), VC (17.74%), SiltOM (14.77%), and DLL (3.17%), indicating a strong rainfall-erosion relationship; 2) the combination of levels of factors generating highest sediment yield was determined; 3) A simple multiple linear regression model developed for predicting local sediment yield showed the highest agreement with field observations (R² = 82.5%). The findings suggest that Taguchi design could be used reliably for modeling soil erosion at field and sub-field scales. Using local calibration data such models have great potential for soil erosion risk assessments at the field scale, especially in areas where contributing factors and factor levels change at small spatial scales.