Using Sobol' sensitivity analysis to identify important model layers of a prediction model for simulating soil nitrogen transport in surface runoff and subsurface water

Due to complex soil conditions of farmlands, a model simulating soil nitrogen transport in surface runoff and subsurface water is subject to uncertainty not only in model parameters but also in representation of soil layers in model layers. This study presents a model of nitrogen transport with the following three layers: a ponding-runoff layer, a soil mixing layer, and a underlying soil layer, which have a total of nine parameters. The model simulates concentrations of urea, ammonia nitrogen, and nitrate nitrogen in surface water and at the depths of 2 cm and 32 cm below soil surface. The main focus of this study is to identify the most important model layer that is crucial for simulating soil nitrogen transport. The global sensitivity analysis method of Sobol' is used to examine importance of the three model layers, because the Sobol' method can consider impacts of individual parameters and parameters interactions on model simulations. Results show the following: (1) the soil mixing layer is the most important one for simulating urea transport in surface water, and the underlying soil layer is the most important one for simulating NH4+-N and NO3--N transport in surface water, (2) the underlying soil layer is the most important one for simulating nitrogen transport at the depths of 2 cm and 32 cm below soil surface, and (3) parameters interactions have a negligible impact on the simulations. These results can be used for improvement of the model and reduction of model prediction uncertainty, and provide a reference for subsequent studies on soil nitrogen transport to reduce agricultural non-point sources pollution.