太子河干流总氮与氨氮水质参数反演及时空变化研究  

Study on the Inversion and Spatio-temporal Variations of Total Nitrogen and Ammonia Nitrogen Water Quality Parameters in the Mainstream of Taizi River

作　　者：吴奇[1] 宫福征 白伟桦 史凯丰 王丽学[1] 陈福江 WU Qi;GONG Fu-zheng;BAI Wei-hua;SHI Kai-feng;WANG Li-xue;CHEN Fu-jiang(College of Water Conservancy,Shenyang Agricultural University,Shenyang 110866,China;Liaoning Heku Management Service Center of Liaoning Hydrological Bureau,Shenyang 110003,China;Dandong Hydrology Bureau of Liaoning Province,Dandong 118001,China)

机构地区：[1]沈阳农业大学水利学院,辽宁沈阳110866 [2]辽宁省水文局河库管理服务中心,辽宁沈阳110003 [3]辽宁省丹东水文局,辽宁丹东118001

出　　处：《生态与农村环境学报》2024年第8期1017-1028,共12页Journal of Ecology and Rural Environment

基　　金：国家自然科学基金(52009078);中国博士后科学基金(2021M693863);辽宁省科学研究经费项目(LJKMZ20221005)。

摘　　要：太子河是辽宁省的重要水系之一。随着对水体环境的日益重视,太子河流域水环境质量逐年好转,其研究方向逐渐由“治理为主”转向“监控为主、治理为辅”,提升太子河水质参数监测的时效性成为更为紧迫的任务。为了对太子河总氮(TN)和氨氮(NH_(4)^(+)-N)水质指标进行实时监测,基于太子河流域2014—2019年10个监测断面的数据,通过剖析Landsat 8遥感影像不同波段反射率的组合与监测断面实测水质数据的线性关系,结合反向传播(BP)神经网络模型的优化,建立了TN和NH_(4)^(+)-N浓度的水质参数反演模型,并反演了2014—2019年太子河干流TN和NH_(4)^(+)-N浓度的时空分布规律。结果表明,构建的BP神经网络优化模型预测效果较好,TN和NH_(4)^(+)-N的决定系数(R^(2))分别为0.777和0.550,均方根误差(RMSE)为1.464和0.667 mg·L^(-1),适用于对流域TN和NH_(4)^(+)-N的反演。2014—2019年TN和NH_(4)^(+)-N水质参数整体趋势向好,NH_(4)^(+)-N浓度处于GB 3838—2002《地表水环境质量标准》中Ⅲ类水标准,TN浓度常年处于Ⅴ类水质标准。太子河流域TN和NH_(4)^(+)-N浓度空间差异性较大,上游区域水质较好,中游区域水质明显下降,下游小林子至唐马寨区域水质较差。综上所述,基于BP神经网络算法优化的太子河流域水质反演是可行的,对时空两个尺度的反演结果可信且时效性较强。The Taizi River is one of the most important water systems in Liaoning Province.With the increasing attention to the water environment,the water environment quality of Taizi River Basin is improving year by year,and its research direction is gradually shifting from"governance"to"monitoring and governance".It has become a more urgent task to improve the timeline of water quality parameter monitoring of Taizi River.As such,the objective of this study is to implement a real-time monitoring system that can precisely track the levels of total nitrogen(TN)and ammonia nitrogen(NH_(4)^(+)-N)in the Taizi River Basin.This study evaluated the relationship between different Landsat 8 remote sensing image bands and water quality data from 10 monitoring sites over a five-year period(2014-2019)by utilizing the analysis of linear correlation.As a result,a highly optimized BP neural network model was established.Through this model,the spatio-temporal distribution of TN and NH_(4)^(+)-N in the main stream of Taizi River from 2014 to 2019 was successfully inverted,and the prediction efficiency of the model was significantly high.The BP neural network model demonstrated a superior accuracy,with a coefficient of determination(R^(2))of 0.777 and 0.550,respectively,and a root mean square error(RMSE)of 1.464 and 0.667 mg·L^(-1)for TN and NH_(4)^(+)-N,respectively.The findings of this study indicate that the TN and NH_(4)^(+)-N water quality parameters in the Taizi River Basin exhibited an overall positive trend from 2014 to 2019.Specifically,the overall NH_(4)^(+)-N concentration was in ClassⅢof the Environmental Quality Standards for Surface Waters,whereas TN concentration remained consistently at ClassⅤwater standards throughout the year.Analysis of the spatial distribution of TN and NH_(4)^(+)-N highlighted significant variability across the region.Elevated water quality was observed in the upper reaches,followed by a reduction in quality in the mid-sections,and severely degraded water quality in the lower reaches,ranging from X

关 键 词：机器学习模型 遥感 总氮 氨氮 太子河 

分 类 号：X832[环境科学与工程—环境工程]