Development of a rainfall-runoff-soil erosion model for railway embankment steep slopes

Railway embankments in Queensland are experiencing increased maintenance costs due to rainfall-induced erosion and sedimentation problems. The erosion problem can be reduced through establishing grass cover on embankment and cutting slopes. Therefore, it is crucial to quantify the effects of grass cover on runoff and erosion. In general, most of the models (e.g., European Soil Erosion Model, Kinematic Soil Erosion and Runoff Model 2, Water Erosion Prediction Project) are event based, developed for large to small agricultural and urban catchments, and they need large volumes of data which are usually not available. Moreover, these models are developed for landscapes with gentle slopes (< 10%) (Sajjan et al., 2012). Although there are publications on runoff and soil loss from catchment scale areas, limited research has been done on railway formation environments where steep slopes (> 50%) prevail. Also, they are complex models requiring many parameters which are often not easy to measure in this study environment. Event-based models need information about antecedent soil moisture conditions. Continuous simulation is becoming a viable choice to the traditional event-based methods because the antecedent moisture prior to the runoff producing rainfall event can be predicted using simulated runoff from previous events.

The objective of this research is to develop a continuous rainfall-runoff-soil erosion model to investigate grass cover effects on runoff and soil loss on railway formation steep slopes to help quantify the scale of the problems using minimal parameters. It utilises the Saint-Venant continuity and momentum equations for overland flow, and a modified Green-Ampt model for infiltration on steep slopes. Finally, an erosion model component, consisting of two different methods, one being Modified Universal Soil Loss equation (MUSLE) based and the other Steep Slope Erosion Dynamics (SSED), was incorporated to allow simulation of soil loss for different grass cover percentages. The efficiency of the model was evaluated by graphical plots of observed runoff versus predicted runoff, percentage error (PE) and also by the Nash-Sutcliffe efficiency (NSE) values.

Several field experiments have been carried out to calibrate the proposed rainfall-runoff model that takes into account the initial soil moisture. There were 14 batter plots, each 10 m wide, around 7 m length and with different grass cover percentages. The site is located at the Gregory Erosion project field trials site near Blackwater, Queensland, Australia. Rainfall and runoff information were collected at 1-minute intervals for the period 1998 - 2010. Soil loss data were collected at different sampling intervals depending upon the timing and severity of rainfall events.

During the calibration, it was found that the hydraulic conductivity (Ks) and initial moisture content are the most important parameters for accurate estimation of runoff. It is observed that the values of Ks for 0%, 50% and 100% grass cover were found to range from 0.13 to 0.46 (mean of 0.3), 4 to 6 (mean of 5.03) and 20 to 25 (mean of 22.56) mm/hr respectively, and in general runoff decreases exponentially with increasing grass cover percentage. The Ks value increases with increasing grass cover percentage. The model has successfully predicted runoff and soil loss from the plots for majority of the cases, with NSE values varying between 0.43 and 0.99 for 0% grass cover plots, 0.06 and 0.97 for 50% grass cover plots, and -0.42 to 0.94 for 100% grass cover plots. Average PE values for soil loss were 25% for 0% grass cover plots for the SSED method and 31% for the MUSLE-based method. For 50% grass cover plots, the PE values were 22% and 28% respectively for the SSED and MUSLE based methods. For 100% grass cover plots, the PE values were 28% and 69% respectively for the SSED and MUSLE-based methods. With the aid of an antecedent soil moisture parameter, which varies with grass cover percentage, continuous simulation of runoff can be carried out using either long records of observed 1-minute timescale rainfall or the rainfall derived from a stochastic rainfall model.