Beach water quality forecast models can be useful tools for public health protection. Despite two decades of research, there has been hitherto no comparative field validation of statistical vs mechanistic model forecasts against operational data for marine beaches. This paper reports a novel performance assessment of a three-dimensional (3D) deterministic hydrodynamic model versus a multiple linear regression (MLR) over a wide range of hydro-meteorological and pollution source conditions. The two models are used concurrently to provide continuous daily water quality forecasts for eight marine bathing beaches directly impacted by a major sewage outfall discharging 2.5 million m³/d of chemically enhanced primary treatment effluent. The predicted Escherichia coli concentrations on the beaches located in complex flow environments are studied over a period of significant changes in treatment levels and sewage flows (2011–2018). The performance of both models is found to be superior to the current advisory based purely on past observations. While the 3D model is process-based and the MLR model is data-driven, both models have comparable performance with about 70–80% accuracy. The model forecasts of E. coli concentrations are significantly correlated with field data. In general, MLR models have slightly higher overall accuracy, while the 3D model provides better prediction for the observed exceedances of water quality standard (model sensitivity). The 3D model is however indispensable in addressing issues of emergency response, setting of effluent standards and disinfection dosage optimisation. The two models serve useful complementary roles in a real time beach water quality forecast system for smart environmental management.

海滩水质预测模型可以成为公共卫生保护的有用工具。尽管进行了 20 年的研究，但迄今为止，还没有针对海洋海滩的运营数据对统计与机械模型预测进行比较现场验证。本文报告了在广泛的水文气象和污染源条件下对三维 (3D) 确定性水动力模型与多元线性回归 (MLR) 进行的新型性能评估。两个模型同时使用，为八个海水浴场提供连续的每日水质预测，这些海滩直接受到一个主要污水排放口排放 250 万立方米^/天的化学强化初级处理出水的影响。预测的大肠杆菌在处理水平和污水流量发生显着变化的时期（2011-2018 年）研究了位于复杂流动环境中的海滩上的浓度。两种模型的性能都优于纯粹基于过去观察的当前建议。虽然 3D 模型是基于过程的，而 MLR 模型是数据驱动的，但两种模型的性能相当，准确率约为 70-80%。大肠杆菌的模型预测浓度与现场数据显着相关。一般来说，MLR 模型的整体精度略高，而 3D 模型对观察到的水质超标情况（模型敏感性）提供了更好的预测。然而，3D 模型对于解决应急响应、污水标准设置和消毒剂量优化等问题是必不可少的。这两个模型在用于智能环境管理的实时海滩水质预测系统中起到了有益的互补作用。