Unraveling heterogeneous drivers and policy effects on urban water quality: A hierarchical Bayesian causal learning framework

The improvement of surface water quality is critical to mitigating global environmental degradation. However, the complex interactions between anthropogenic pollution and natural hydrology highlight the challenges of watershed governance. This paper proposes an integrated framework combining causal learning and Bayesian modeling to explore the driving mechanisms of urban water quality scores (WQS) from a heterogeneous perspective. A causal discovery algorithm integrating background knowledge is utilized to identify the underlying causal structure, and the hierarchical Bayesian model is employed to disentangle the heterogeneous mechanisms and transmission pathways. The policy intervention effects of governance strategies are evaluated through counterfactual scenario analysis. The results indicate that the nationwide WQS averaged approximately 0.639 over the study period, displaying a declining northwest-to-east spatial gradient. Causal structure identification reveals that factors such as agricultural non-point pollution, industrial activity, and end-of-pipe treatment capacity exhibit direct causal relationships with WQS. Agricultural non-point pollution shows the largest negative effect, followed by urbanization pressure. The driving effects exhibit significant watershed heterogeneity. End-of-pipe treatment gives high marginal returns in infrastructure-poor basins but saturates in developed basins like the Yangtze and Pearl Rivers. Counterfactual simulations show that under a 10% intervention at the national level, mitigating environmental pressures associated with urban population concentration can improve WQS through downstream pathways, while targeted reductions in agricultural non-point pollution and industrial point source emissions increase WQS by 1.0% and 0.59%, respectively. The research findings bridge causal mechanisms with spatial heterogeneity, informing the formulation of basin-specific regional water quality management policies.