Identifying spatial influence of urban elements on road-deposited sediment and the associated phosphorus by coupling Geodetector and Bayesian Networks

Abstract

Elevated particles and phosphorus washed from road-deposited sediment (RDS) are noteworthy causes of eutrophication in urban water bodies. Identifying how urban elements (e.g., dwellings, roads) spatially influence RDS and the associated phosphorus can help pinpoint the primary management areas for RDS pollution and therefore effectively mitigate this problem. This study investigated spatial influence of urban elements on RDS build-up load and phosphorus load in Hanyang district of Wuhan city in central China. Bayesian Networks (BNs), combined with geographical detector (Geodetector) and correlation analysis, were applied to quantify spatial association between kernel density of urban elements, RDS build-up load and phosphorus load in RDS. Results showed that (1) areas with higher density of factories related elements usually had elevated level of RDS build-up load, aluminum-bound phosphorus (Al-P), occluded phosphorus (Oc-P), organophosphorus (Or-P). Higher load of RDS associated iron-bound phosphorus (Fe-P) and apatite phosphorus (Ca-P) usually occurred where dwellings, catering, and entertainment related elements were concentrated. (2) Urban elements mainly showed positive correlation with RDS build-up load, Fe-P, Ca-P, De-P (detrital apatite phosphorus), while they chiefly showed negative correlation with Ex-P (exchangeable phosphorus), Al-P, Oc-P, and Or-P. Bus stations, dwellings, and factories related elements had relatively strong determinant power over spatial stratified heterogeneity of RDS and RDS-associated phosphorus. (3) Geodetector and correlation analysis could boost factors filtering and construction of network structures in the process of developing BNs models. The developed BNs resulted in sound prediction of <150 μm RDS build-up load and phosphorus load, given that the prediction accuracy of models ranged from 0.532 to 0.657. These findings demonstrate that urban elements are useful spatial predictors of RDS pollution, and coupling Geodetector and BNs is promising in RDS pollution prediction and supporting urban nonpoint source pollution management.