Abstract

Hillslope erosion causes important soil loss, contributes significantly to contaminant transport, and causes stream and lake embankment. Recent studies estimate that worldwide, the existing 45,000 dams (dams over 15 m [49 ft] high) trap more than 25% of global sediment flux and that these dams would be losing between 0.5% and 1% of their capacity annually (Syvitski et al. 2005). Additionally, the increase in soil erosion due to human mismanagement causes advancing desertification of the landscape. The aim of this work is to provide an alternative tool for decision making regarding erosion management in the Chilean "Secano" (dryland) area. Specifically, a numerical erosion model with a physical base was developed and tested. Although the model is based on previous developments, it considers rainfall splash together with sheet and rill flow in a novel way and can be easily incorporated in distributed hydrological models and/or geographical information systems. It is an event model, which divides the hillslope length into elements. Water and sediment can enter each element from upslope, while Manning's equation is used to estimate mean velocities. New data gathered in the area, the first to be obtained in a systematic manner, was available from field plots in the Secano and was used to test the model. Soil loss had been monitored for a year in standard erosion plots with three different treatments: traditional rotation (wheat/fallow), no-till rotation (wheat/legume), and natural prairie (plus one bare soil control per slope). The total event sediment loss for the treatments was predicted rather well without using parameter calibration, obtaining estimations within 35% to 165% of field measurements and reasonable root-mean square errors, given the range expected when models were applied to data from plots installed in agricultural fields. Two-way analysis of variance results suggest the model considers factors of slope and treatment well, and the results show an overall r(2) of 0.46 considering all pairs of estimated values to measurements. Predictions were better for the case of milder slopes and native vegetation (prairie). Sensitivity analysis showed that for sheet flow, the most sensitive parameters were vegetation cover and soil cohesion; while for rill flow, these were cohesion and slope. In both cases, the rill network information provided as input to the model had little effect on simulation results.