长江流域洞庭湖区出入湖磷通量模拟及水质预测:机器学习与传统水文模型耦合方法  

Simulation of Phosphorus Inflow and Outflow Fluxes and Water Quality Prediction in Dongting Lake Area of the Yangtze River Basin:A Coupled Approach of Machine Learning and Traditional Hydrological Modeling

作　　者：刘杰 陈前[2,3] 许妍 查悉妮[4] 张美一 辛小康[4] 唐文忠 张洪[1,5] Liu Jie;Chen Qian;Xu Yan;Zha Xini;Zhang Meiyi;Xin Xiaokang;Tang Wenzhong;Zhang Hong(State Key Laboratory of Environmental Aquatic Chemistry,Research Center for Eco-Environmental Sciences,Chinese Academy of Sciences,Beijing 100085,China;School of Civil Engineering,Southeast University,Nanjing 210018,China;Shanghai Municipal Engineering Design Institute(Group)Co.,Ltd.,Shanghai 200092,China;Changjiang Water Resources Protection Institute,Wuhan 430051,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院生态环境研究中心环境水质学国家重点实验室,北京100085 [2]东南大学土木工程学院,江苏南京210018 [3]上海市政工程设计研究总院(集团)有限公司,上海200092 [4]长江水资源保护科学研究所,湖北武汉430051 [5]中国科学院大学,北京100049

出　　处：《地球科学》2024年第11期3995-4007,共13页Earth Science

基　　金：国家重点研发计划项目(No.2023YFC3205600);国家自然科学基金资助项目(No.52270202).

摘　　要：磷通量的波动对出入湖河流生态系统的稳定有直接影响.在水质波动显著的洞庭湖区,使用SWAT模型模拟该区域的出入湖磷通量,并分析各水系区间的磷滞留情况.基于水文物理过程搭建耦合模型,再利用SWAT模型的模拟结果进行训练,并利用该模型对湖区近期水质进行情景预测.结果表明:洞庭湖区出入湖磷通量均呈明显的季节变化,其中无机磷是主要的磷形态.2012—2021年洞庭湖区平均TP入湖通量为2.94×10^(4)t/a,平均出湖TP通量为3.34×10^(4)t/a,平均出湖TP浓度为0.13 mg/L.其中,三口区间作为入湖磷通量贡献最大的区间,其TP输出浓度也最高,是污染治理的关键区域.洞庭湖区TP滞留率较低,使其成为长江下游重要的TP来源.耦合模型对于洞庭湖区的出入磷通量的输出过程模拟效果良好,NSE值均>0.8,在RPE值处于10%水平下的RMSE值均较低,实现了LSTM网络替代大型分布式物理模型的仿真建模.基于上述模型根据《洞庭湖水环境综合治理规划》对洞庭湖近期TP输出浓度进行了预测,在各区间入湖TP浓度控制在0.1 mg/L的情景下,能够实现湖区水环境规划目标.Fluctuations in phosphorus fluxes to and from the lake have a direct impact on the stability of river ecosystems.In the Dongting Lake area,where water quality fluctuates,the SWAT model was used to simulate the phosphorus inflow and outflow fluxes and to analyze the retention rate in each area.A coupled model was constructed based on hydrophysical processes,being trained by the results of the SWAT model and using the model to make scenario predictions of recent water quality in the lake area.The results showed that the inflow and outflow phosphorus fluxes in Dongting Lake showed obvious seasonal variations,with inorganic phosphorus being the main phosphorus format.From 2012 to 2021,the average TP inflow flux in Dongting Lake was 2.94×10^(4)t/a,the average TP outflow flux was 3.34×10^(4)t/a,and the average TP outflow concentration was 0.13 mg/L.Among the sub-basins,the Sankou area as the area with the largest contribution of phosphorus flux into the lake and the highest TP output concentration,should be emphasized to manage the phosphorus pollution in this area.The low TP retention rate in the Dongting Lake area may turn it into an important source of TP in the lower reaches of the Yangtze River.Coupled modeling is relatively precise for simulating the output process of river phosphorus fluxes,with all NSE values>0.8,and all RMSE values are low at the RPE value at 10%level,which realizes the simulation modeling of the LSTM network instead of the large-scale distributed physical model.Based on the above model,the recent TP output concentration of Dongting Lake area was predicted according to the“Dongting Lake Water Environment Comprehensive Management Plan”,and the scenario of controlling the TP concentration inflow into the lake in each area at 0.1 mg/L can realize the water environment planning objectives of the lake area.

关 键 词：出入湖磷通量 耦合模型 机器学习 磷滞留 水质预测 洞庭湖区 水文地质 环境地质. 

分 类 号：P64[天文地球—地质矿产勘探]