Quantification of multiscale links of key factors with watershed nitrogen and sediment exports based on a Bayesian modelling approach

Abstract

The identification of key influencing factors of nonpoint source pollution is important to control such pollution in watersheds. The choice of spatial scale is vital in influencing factor analysis. However, many previous analyses were only carried out on a specific scale or evaluated different topics on different scales, which could not integrate various information of different scales. In this study, a multiscale approach based on Bayesian belief networks (BBNs), which can quantify the links between watershed nitrogen and sediment exports and their key factors, was developed. The main results indicated that influencing factors have relatively higher level of importance, i.e. higher degree of interpretation, at town scale or subcatchment scale compared with 1 km2 scale. Key factor set based on the nesting of town scale and subcatchment scale (i.e., town_sub scale) predict nitrogen and sediment exports better than those based on three separate scales (1 km2 scale, subcatchment scale and town scale), which implies that multiscale nesting reflects more comprehensive information about watershed processes. Moreover, different nitrogen and sediment reduction targets were achieved through land use changes and resulting economic benefits or losses were compared across three scales (subcatchment scale, town scale and town_sub scale). A relatively more appropriate scale with the lowest economic loss was determined. The decreased cultivated land and increased woodland and waterbody areas mostly contributed to the reductions in nitrogen and sediment exports and the economic loss of each reduction target. When achieving the same nitrogen and sediment reduction targets, the economic loss resulting from the land use changes at the subcatchment scale was lowest, followed by the town_sub and town scales. This multiscale approach attempted to determine how key factors with the highest level of importance at different scales and watershed nonpoint source pollution can be connected, which would better inform the management of watershed nonpoint source pollution.