Sensitivity analysis of the OpenLISEM model: calibration for an unpaved road in Southern Brazil

Abstract

The sensitivity analysis of calibrated erosion models holds significant promise for improving the performance of simulations and the understanding of the dynamics of runoff and soil loss. The aim of this study was to perform the calibration and sensitivity analyses of the OpenLISEM model for an unpaved rural road in Southern Brazil, where the quantitative understanding of the impact of precipitation on water and soil loss is limited, and the improvement of the use of the method is crucial for determination of the hydrosedimentological dynamic. Nash-Sutcliffe model efficiency coefficients (NSE) were used to compare measured and simulated data series of discharge and sediment production. “One-factor-at-a-time” (OAT) and relative sensitivity coefficients were derived relative to changes of total discharge (Q), average soil loss (Qs), infiltration (I), runoff erosion (Er), splash erosion (Es), and deposition (D) against relative changes of the Manning coefficient (n), random roughness (rr), saturated hydraulic conductivity (Ksat), suction at the wetting front (ψsoil), aggregate stability (aggrstab), soil cohesion (coh), mean aggregate diameter (d50), and initial soil moisture content (θi). The model calibration produced NSE coefficients of 0.97 and 0.90 for discharge and sediment production data series, respectively. Q and Qs were mostly sensitive to random roughness and saturated hydraulic conductivity. Our results also showed the high relevance of Ksat for I, and aggrstab for D and Es. The sensitivity analysis of the model revealed significant variations of sensitivity, depending on the magnitude of the values adopted in the calibration and the used methodology.