基于机器学习的养殖池-红树林-海洋复合生态系统氨氮遥感反演  

Remote sensing inversion of ammonia nitrogen in aquaculture pond mangrove marine composite ecosystem based on machine learning

作　　者：莫锦英 田义超 王家乐 杜金泽 张强 张亚丽 陶进 林俊良 MO Jinying;TIAN Yichao;WANG Jiale;DU Jinze;ZHANG Qiang;ZHANG Yali;TAO Jin;LIN Junliang(School of Resources and Environment,Beibu Gulf University,Qinzhou 535011;School of Resources,Environment and Materials,Guangxi University,Nanning 530004;Guangxi Key Laboratory of Marine Environmental Change and Disaster in Beibu Gulf,Qinzhou 535011;Gulf Ocean Development Research Center,Beibu Gulf University,Qinzhou 535011;Key Laboratory of Marine Geographic Information Resources Development and Utilization in the Beibu Gulf,Beibu Gulf University,Qinzhou 535011;School of Marine Sciences,Beibu Gulf University,Qinzhou 535011)

机构地区：[1]北部湾大学资源与环境学院,钦州535011 [2]广西大学资源环境与材料学院,南宁530004 [3]广西北部湾海洋环境变化与灾害研究重点实验室,钦州535011 [4]北部湾海洋发展中心,钦州535011 [5]北部湾海洋地理信息资源开发利用重点实验室,钦州535011 [6]北部湾大学海洋学院,钦州535011

出　　处：《环境科学学报》2024年第11期415-429,共15页Acta Scientiae Circumstantiae

基　　金：国家自然科学基金项目(No.42261024);广西林业厅项目(No.桂林科研【2022】第4号);北部湾大学海洋科学一流学科项目(No.DRB003);广西高校人文社会科学重点研究基地项目(No.BHZKY2202);广西高校人文社科重大项目(No.JDZD202214);广西创新驱动发展专项项目(No.AA18118038);广西科技基地和人才项目(No.2019AC20088)。

摘　　要：氨氮(NH_(3)-N)是监测养殖污染和衡量水质优劣的主要指标之一,对水产养殖和近海生态环境产生重要影响,因此定量评估NH_(3)-N浓度有助于维护水产养殖业的健康发展和近海环境的监管保护.本研究基于Sentinel-2卫星提取的单波段B5~B8a、波段组合B1/B8和B1-B5、水体指数和植被指数,采用极端梯度提升(XGBR)、轻量级梯度提升机(LGBM)、随机森林(RF)和K最近邻(KNNR)4种机器学习模型,基于决定系数(R^(2))和均方根误差(RMSE)对不同模型的性能进行评价,并选取性能最优的模型,对茅尾海及其沿岸养殖池NH_(3)-N的浓度分布特征进行研究.结果表明:①XGBR模型性能最优,训练集R2达到0.9340,RMSE为0.0446 mg·L^(-1);测试集R2达到0.7522,RMSE为0.0249 mg·L^(-1).此外,单波段和波段组合对NH_(3)-N浓度变化较为敏感,在XGBR模型中的贡献度最高.②茅尾海2023年11月NH_(3)-N浓度(0.058 mg·L^(-1))高于2023年12月(0.055 mg·L^(-1)),呈现近岸高于远岸,河口区域出现较高浓度的空间分布特征;养殖池2023年11月NH_(3)-N浓度(0.107 mg·L^(-1))高于2023年12月(0.072 mg·L^(-1)),钦江沿岸高于茅岭江沿岸.③养殖池NH_(3)-N浓度高于茅尾海.养殖水体中NH_(3)-N通过康熙岭片区和尖山片区红树林后的降低效果要高于沙井片区.本研究说明了Sentinel-2数据结合XGBR模型在茅尾海及其沿岸养殖池NH_(3)-N浓度反演中表现出良好的性能,为海洋环境管理和海洋资源规划提供了科学依据和技术支持.Ammonia nitrogen(NH_(3)-N)is a critical indicator for monitoring aquaculture pollution and measuring water quality,which has a significant impact on aquaculture and the nearshore ecological environment.Consequently,the quantitative evaluation of NH_(3)-N concentrations is essential for maintaining the healthy development of aquaculture industry and regulatory protection of nearshore environment.This study utilized single band B5~B8a,band combination B1/B8 and B1-B5,water index and vegetation index extracted from Sentinel-2 satellite.Four machine learning algorithms were employed:extreme gradient boosting regression(XGBR),light generalized boosted regression(LGBM),random forest(RF),and k-nearest neighbor regression(KNNR).The performance of each model was evaluated with the coefficient of determination(R2)and root mean square error(RMSE),and the optimal performance was selected to analyse the concentration distribution characteristics of NH_(3)-N in Maowei Sea and its coastal aquaculture ponds.The findings indicated that①the XGBR algorithm exhibited superior performance,with a training set of R2=0.9340 and RMSE=0.0446 mg·L^(-1);test set R^(2)=0.7522 and RMSE=0.0249 mg·L^(-1).In addition,single band and band combination were most sensitive to NH_(3)-N concentration and made the highest contribution in the XGBR model.②The NH_(3)-N concentration in Maowei Sea was 0.058 mg·L^(-1)in November 2023,which was higher than that of 0.055 mg·L^(-1)in December 2023,showing a spatial distribution pattern of higher concentration near the shore compared to the far shore,and a higher concentration in the estuarine area.The NH_(3)-N concentration in the aquaculture pond was 0.107 mg·L^(-1)in November 2023 which was higher than that of 0.072 mg·L^(-1)in December 2023,and the NH_(3)-N concentration in aquaculture ponds is higher along the Qin River than along the Maoling River.③The NH_(3)-N concentration in the aquaculture pond was higher than that in Maowei Sea.The reduction effect of NH_(3)-N in aquaculture water after p

关 键 词：NH_(3)-N 机器学习算法 Sentinel-2 养殖池 红树林 茅尾海 

分 类 号：X87[环境科学与工程—环境工程] X832