Uncertainty and sensitivity analysis of spatial distributed roughness to a hydrodynamic water quality model: a case study on Lake Taihu, China.

Roughness is an important parameter in hydrodynamic and water quality modelling; it has direct effects on bottom shear stress which relied on sediment and vegetation. The varied roughness caused by spatial heterogeneity of sediment and vegetation may lead to uncertain simulation results. To investigate the effect of roughness uncertainty on the performance of hydrodynamic water quality models, a typical large shallow lake in China (Lake Taihu) was divided into eight areas for illustrating the effect of spatial variation of roughness on hydrodynamics and water quality. Total nitrogen (TN) was selected as the variable to calculate the uncertainty interval, and sensitive positions greatly affected by roughness as well as the appropriate range of roughness were explored by means of regional sensitive analysis (RSA). The results showed that roughness had the most significant effect on the bottom velocity. The uncertainty for water quality caused by roughness presented a striking spatial difference; the uncertainty interval for TN could be up to 1.3 mg/L. The posterior distribution of roughness was given to further narrowed the range of roughness, and the updated roughness range manifested that roughness value should be set higher in the area with thick sediment and abundant vegetation. It is of utmost importance to consider the comprehensive effects of sediment and vegetation in the determination of roughness. For certain lake areas with great water quality simulation error, the error could be effectively reduced by setting spatial distributed roughness. The optimization scheme was provided for the reasonable determination of roughness, so that the dynamic characteristic at the sediment-water interface could be represented synthetically. In this paper, the uncertainty and sensitivity of roughness in hydrodynamic water quality model are analyzed to provide reference for parameter setting of large shallow water lake model. For large scale lakes, parameters need to be modified according to the actual condition due to the spatial difference of friction coefficient at the bottom.