摘要
随着社会经济的发展,近岸海域污染形势严峻,陆域污染源数量众多且结构复杂,现有机理模型难以准确高效地定量解析陆海污染响应关系.本研究以沿海城市深圳为例,运用随机森林、自适应提升等7种机器学习方法,对在线监测数据、人工采样数据、气象数据等进行清洗融合和时空匹配后,分别针对大鹏湾、大亚湾和深圳湾建立陆域排放与近岸海域水体氮磷或叶绿素a浓度的响应关系模型,并通过设置陆源调控情景探索响应关系模型的应用.结果表明,大鹏湾与大亚湾的模型能够较好地捕捉近岸海域叶绿素a浓度的变化,其中随机森林算法与梯度提升算法表现最优,测试集R2在0.73以上;深圳湾模型对海域氮磷指标的拟合状况良好,随机森林算法与自适应提升算法的模拟精度较高,R2均可达0.92以上.此外,通过设置各陆域调控情景,研究发现降低西丽再生水厂出水有机物与总氮含量的措施对缓解深圳湾海域氮污染有一定效果.本研究方法可推广至其他沿海城市的陆海统筹治理,为相关决策提供科学支撑.
With socioeconomic development,pollution in nearshore waters is becoming increasingly severe.The numerous and complex structures of land-based pollution sources challenge existing mechanistic models to accurately and efficiently quantify the relationship between land and sea pollution linkages.This study used Shenzhen,a coastal city in China,as a case to construct quantitative response models between land-based discharges and the concentrations of nitrogen and phosphorus and chlorophyll-a in nearshore waters of Dapeng Bay,Daya Bay,and Shenzhen Bay.Seven machine learning algorithms,including Random Forest and Adaboost,were developed these models based on the cleaning and spatial-temporal matching of online monitoring data,manual sampling data,meteorological data and other variables.In addition,different terrestrial pollution control scenarios were set to explore the feasibility of model applications.The results indicate that the models for Dapeng Bay and Daya Bay can effectively capture the changing trends of nearshore chlorophyll-a concentrations,with Random Forest and Gradient Boosting demonstrating the best performance(R2>0.73).For Shenzhen Bay,the models show favorable fitting results for nitrogen-phosphorus indicators,with highest simulation accuracy achieved using Random Forest and Adaboost algorithms(R2>0.92).By setting several simulation scenarios of land-based regulation,measures to reduce the COD and TN in the effluent of Xili Water Reclaimed Plant are effective in alleviating nitrogen pollution in Shenzhen Bay.The proposed model can be generalized to other coastal cities for integrated land-sea management,offering scientific tools and data support for decision-making.
作者
赵晨羽
黎栩霞
徐旭
谢颖嘉
孙文郡
曾思育
ZHAO Chenyu;LI Xuxia;XU Xu;XIE Yingjia;SUN Wenjun;ZENG Siyu(School of Environment,Tsinghua University,Beijing 100084;Shenzhen Ecological and Environmental Monitoring Center of Guangdong Province,Shenzhen 518000)
出处
《环境科学学报》
北大核心
2025年第7期146-155,共10页
Acta Scientiae Circumstantiae
基金
国内-212地方政府基金项目(No.SZDL:2022000268)。
关键词
陆海统筹管理
多源数据融合
水质响应模型
机器学习
integrated land-sea management
multi-source data fusion
water quality response model
machine learning
