摘要
为探索利用卫星遥感监测填海施工区悬浮物浓度的有效方法,通过大量现场实测的水质样本和卫星遥感数据对以传统回归算法与神经网络方法建立的反演模型的精度进行了比较分析.研究发现,采用神经网络方法得到的反演结果与现场实测水样的相关性最高,其相关系数达到0.95,平均误差仅为30%;而传统回归模型的反演精度偏低,平均误差高达140%,这可能是由填海施工区复杂的光学特征造成的.利用神经网络模型对工程海域悬浮物浓度连续5年的遥感监测结果表明:该海域的悬浮物浓度呈季节性变化,冬季相对较高,夏季相对较低;在空间上呈现距施工区较近处浓度较高、较远处浓度较低的分布格局.
This paper aimed to explore an effective way to monitor suspended particulate matter concentration(SPMC)in sea reclamation area(SRA)with satellite remote sensing.Traditional regression algorithms and an artificial neural network(ANN)were used to develop the retrieval model for SPMC in SRA.Through the validation with in-situ water samples,the ANN model has a superior performance with a coefficient determination(R2)of 0.95 and mean relative error(MRE)of 30%,while other models have an inferior performance with MRE of 120%.This phenomenon is mainly caused by the optical characteristics of SRA.Using the developed ANN model,the retrieval results revealed that,the SPMC in the SRA during 2010~2015 changes seasonally with higher in winter and lower in summer;the higher SPMC is found around the construction site,while the lower values were found in the far offshore regions.The distribution pattern of SPMC in the SRA is mainly caused by the reclamation construction,wind force and tidal current directions.
作者
宋南奇
王诺
吴暖
林婉妮
SONG Nanqi;WANG Nuo;WU Nuan;LIN Wanni(Institute of Bohai Sea,National Marine Environmental Monitoring Center,Dalian 116023,China;State Environmental Protection Key Laboratory of Marine Ecological Environment Restoration,Dalian 116023,China;college of Transportation Engineering,Dalian Maritime University,Dalian 116026,China)
出处
《应用基础与工程科学学报》
EI
CSCD
北大核心
2020年第5期1108-1121,共14页
Journal of Basic Science and Engineering
基金
国家自然科学基金项目
国家海洋软科学项目(JJYX201612-1)
辽宁省科学技术计划项目(012220008)。
关键词
填海
悬浮物
遥感
监测
神经网络
时空分布
reclamation
suspended particulate matter
remote sensing
monitoring
neutral network
space-time distribution