有害藻华的暴发对公共卫生、旅游业和养殖业发展、生态系统健康等造成严重危害，因此受到了广泛关注。在气候变化、富营养化和养殖业发展等多重胁迫下，有害藻华处于动态变化中，且呈现明显的区域性特点。在我国，有害藻华是一类严重的海洋生态灾害问题，不仅造成海水养殖业的巨大损失，对人类健康及生态安全也构成了潜在威胁。近年来，我国近海有害藻华呈现出多样化、有害化和小型化的演变趋势，由此带来不同的危害效应，也增加了其监测难度，亟待深入探索各海域有害藻华的演变特征和驱动因素，发展有害藻华的预测预警方法。随着对有害藻华监测和研究工作的不断深入，可用于有害藻华分析的数据类型不断增多，数据覆盖的时空尺度也在持续增加。通过数理统计、地理信息系统（geographical information system，GIS）和机器学习等数据分析、挖掘手段，可以为探究有害藻华演变特征、发展有害藻华预测预警方法提供重要支持。目前，对我国近海有害藻华演变特征的认识仍不清晰，对驱动其长期演变的关键因素认识也不够充分，同时对一些特殊类型有害藻华的监测预警手段也十分缺乏。对此，本研究选择我国近海不同海域的典型有害藻华问题，基于有害藻华相关的多源数据，应用数理统计和地理信息系统工具探究了典型海域有害藻华的演变特征，通过机器学习手段探索了有害藻华的预测方法，并尝试搭建了有害藻华数据分析和可视化平台。

针对我国渤海和黄海海域微藻形成的有害藻华（包括赤潮和褐潮）问题，收集和整理了近几十年来有害藻华的相关数据，通过数理统计和GIS分析工具，探讨了渤海和黄海海域有害藻华的演变特征。基于对渤海有害藻华发生情况（1952-2017）、环境因子（1956-2017）和社会经济类（2017）数据的整理和分析，共记录有害藻华230次，藻华影响面积超过70,000 km²，记录藻华原因种64种。基于有害藻华频次、规模和物种组成状况，结合变点检测结果，可将渤海有害藻华发生历史分为3个阶段。在此期间，有害藻华发生频次有增长趋势，但近年来趋于稳定。有害藻华呈明显的季节扩张，热点区域从渤海湾向秦皇岛近岸海域转移。渤海有害藻华的演变特征的独特性表现为典型藻华原因种的明显变化，从甲藻类赤潮生物逐渐转变为定鞭藻类和海金藻类，同时伴随着藻华原因种的多样化、有害化和小型化趋势。分析认为，渤海有害藻华的演变特征与气候变化、海水养殖业发展和富营养化等多重胁迫的综合作用相关。尝试通过多种时序模型对渤海有害藻华的发生情况进行了预测，并综合有害藻华危险度和承灾体易损度，初步评估了渤海有害藻华的风险状况，结果显示渤海有害藻华高风险区域主要出现在渤海湾和秦皇岛近岸海域。针对黄海海域的有害藻华，收集和整理了赤潮发生情况数据集（1972-2017）、绿潮数据集（2008-2017）和环境因子数据集（1970-2017）。分析结果显示，黄海海域的甲藻赤潮频次、规模、季节窗口、空间分布和原因种多样性等方面均有增加或扩展趋势；北黄海和南黄海海域的赤潮演变特征存在一定差异；从2007年起南黄海大规模暴发的绿潮灾害有可能改变赤潮的发生频次和季节窗口。

选择东海海域能够产生麻痹性贝类毒素的亚历山大藻赤潮，以及南海海域能够形成“巨型囊体（giant colony）”的球形棕囊藻赤潮作为两个典型案例，应用机器学习手段，尝试发展了基于多源数据的麻痹性贝类毒素预测和棕囊藻囊体预测方法。利用黄、东海两个航次获取的亚历山大藻赤潮相关数据集，以及南海北部湾海域八个航次获取的棕囊藻赤潮相关数据集，运用7种机器学习算法，对海域麻痹性贝类毒素和囊体是否产生进行了预测。在麻痹性贝类毒素的预测案例中，最优模型为支持向量机（Support Vector Machine, SVM），模型优化后的准确率达94%，筛选的预测预警指标分别为温度、盐度和磷酸盐。在棕囊藻囊体预测的案例中，最优模型为轻量级梯度提升机（Light Gradient Boosting Machine，LightGBM），优化后准确率达84%，可用于棕囊藻囊体预测预警的指标为聚球藻生物量、青绿素、硅甲藻黄素、温度和盐度，可选预测预警指标为叶绿素c₃、磷酸盐和微型真核藻类丰度等。

基于对有害藻华相关数据的收集、整理和分析，本研究尝试构建了一套集有害藻华多源异构数据管理、数据分析和可视化为一体的综合平台，包括数据管理平台、GIS平台和可视化平台。本平台的搭建实现了多源有害藻华数据的动态关联和存储可拓展性，集成了时空分析方法，初步实现了数据和分析产品的可视化。本研究以渤海有害藻华长期时空演变规律探究为案例，展示了有害藻华数据分析和可视化平台的应用效果。

综上，本研究应用数理统计和地理信息系统手段，探究了长时间序列渤海和黄海海域有害藻华的演变特征，揭示了渤海有害藻华原因种的显著变化，以及气候变化、海水养殖业发展和富营养化等多重胁迫综合作用对有害藻华演变的驱动；发现了黄海海域甲藻赤潮优势度的增加，以及大规模绿潮对赤潮发生频次和季节分布特征的潜在影响。通过机器学习手段，发展了基于现场观测数据预测藻毒素分布和球形棕囊藻囊体分布的方法，筛选了可用于不同类型有害藻华监测预警的指标。尝试搭建了有害藻华的数据分析和可视化平台，为有害藻华事件的研究、监测和管理提供了有力的支撑手段。

Harmful algal blooms (HABs) have severe impacts on public health, tourism, mariculture industry and ecosystems, and therefore raise great public attention. The HABs vary at regional and local scale under the joint impacts of multiple environmental stresses, such as climate change, eutrophication, mariculture development and so on. In China, as a type of severe marine ecological disaster, HABs have caused huge economic losses to mariculture industry, and even impose a threat to human health and marine ecosystems. In recent years, HABs show the trend toward more diversification, noxiousness and miniaturization in the coastal waters of China. This leads to different hazardous effects, and results in the difficulty in monitoring of HABs. Therefore, it’s necessary to explore the evolution features of HABs and the major environmental drivers, and to develop predictive methods. With the development of monitoring and research on HABs, data available for HAB analysis are increasing, along with the increasing coverage on spatial and temporal scale. It is possible to explore the evolution features of HABs and develop predictive models through methods like mathematical statistics, geographical information system (GIS) tools, machine learning and so on. So far, the evolution features of HABs, as well as the relationship between HABs and environmental factors in the China Seas, are still not well understood. Meanwhile, there are lack of methods on monitoring and prediction for some special types of HABs. In this study, statistical and GIS tools were used to analyze the evolution features of HABs in the typical sea areas based on the multi-source data related to HABs, and the machine learning methods were used to build predictive models for HABs. Moreover, a platform was established for management, analysis and visualization of data related to HABs.

The HABs formed by microalgae, mainly red tides and brown tides, in the Bohai Sea and the Yellow Sea were studied to reveal the evolution features in the two regions based on the historical records of HABs over the last several decades. In the Bohai Sea, datasets of HAB events (1952-2017), environmental variables (1956-2017) and social and economic data (2017) were compiled. There are 230 HAB events with 64 identified species affecting nearly 70,000 km² area in the Bohai Sea. The history of HABs in the Bohai Sea can be characterized into three periods based on the characteristics of frequency, scale and species composition, and assisted by the analysis of change point detection. Our analysis shows that HABs in the Bohai Sea increase their frequency over the period studied, although the increase has plateaued in the last decade. The seasonal distribution of HAB events has clearly expanded, and the main hotspot moved from Bohai Bay to coastal waters of Qinhuangdao over the three periods. A unique feature of HAB evolution is the rapid shift of typical HAB-forming microalgae from dinoflagellates to haptophytes and pelagophytes over the three periods, with a trend toward diversification, noxiousness and miniaturization. We consider that the rapid shift of HABs in the semi-enclosed Bohai Sea are related to the combined effects of climate change, mariculture development, and eutrophication. The frequency of HABs can be predicted by time series models. Considering the hazard of HABs and the vulnerability of potential targets affected by HABs, the risk of HABs in the Bohai Sea was assessed. The high-risk areas of HABs are mainly located in the Bohai Bay and the coastal waters of Qinhuangdao. In the Yellow Sea, datasets of HAB events (1972-2017), green tides (2008-2017) and environmental variables (1970-2017) were compiled. There is an increasing dominance of dinoflagellate red tides in terms of frequency, scale, seasonality, spatial distribution and species. The red tides in the northern Yellow Sea and southern Yellow Sea, however, have different evolution features. The recurrent large-scale green tides in the southern Yellow Sea from 2007 may lead to the decrease of interannual and seasonal frequency of red tides.

Focusing on paralytic shellfish toxins mainly produced by Alexandrium spp. in the East China Sea, and the “giant colony” produced by Phaeocystis globosa in the Beibu Gulf, the South China Sea, prediction methods have been developed through the approach of machine-leaning based on multi-source data. Datasets related to paralytic shellfish toxins from 2 cruises conducted in the Yellow Sea and the East China Sea, and related to Phaeocystis colony from 8 cruises conducted in the Beibu Gulf, the South China Sea, were collected and compiled. Based on these datasets, we compared 7 types of machine learning algorithms to predict the existence of paralytic shellfish toxins and Phaeocystis colony. In the case study of paralytic shellfish toxins, the optimal model is Support Vector Machine (SVM). The accuracy of the model is 94% after optimization. The predictive indicators for paralytic shellfish toxins are temperature, salinity and phosphate. In the case study of Phaeocystis colony, the optimal model is Light Gradient Boosting Machine (LightGBM). The accuracy is 84% after optimization, and the predictive indicators include biomass of Synechococcus, prasinoxanthin, diadinoxanthin, temperature and salinity, and chlorophyll c₃, phosphate and nanoeukaryote abundance could serve as optional predictive indicators.

Based on analysis of HAB datasets, an integrated platform was designed to manage, analyze and visualize different types of data related to HABs. The platform integrated the database platform, GIS analysis platform, and the visualization platform. The platform realized the dynamic join and storage scalability for HAB data, and integrated different tools ranged from temporal and spatial analysis to preliminary data visualization. A case study on the long-term temporal and spatial evolution of HABs in the Bohai Sea has been applied in the study.

In summary, we used the statistical methods and GIS tools to study the evolution features of HABs in the Bohai Sea and the Yellow Sea. The results revealed dramatic changes of HAB causative species in the Bohai Sea, and the combined impacts of multiple drivers like climate change, development of mariculture and eutrophication on the evolution of HABs. In the Yellow Sea, the increasing dominance of dinoflagellate red tides, and the potential effects of large-scale green tide on the evolution red tides, were discovered. Moreover, the machine learning approach was successfully used to predict the distribution of paralytic shellfish toxins and Phaeocystis colony. A platform was established for data management, GIS analysis and data visualization. The results showcase the potential of HAB studies through the approach of data analysis, which could support the research, monitoring and management of HABs.

第1章 绪论... 1

1.1 有害藻华概述... 1

1.1.1 有害藻华发生情况及危害效应... 1

1.1.2 有害藻华动态变化... 3

1.1.3 有害藻华影响因素... 7

1.2 有害藻华研究数据和方法基础... 10

1.2.1 数据基础... 10

1.2.2 方法基础... 14

1.3 中国近海有害藻华研究现状... 18

1.3.1 有害藻华演变特征... 18

1.3.2 有害藻华监测与预测... 20

1.4 研究目的和意义... 21

第2章 渤海有害藻华时空演变特征及其关键驱动因子... 23

2.1 前言... 23

2.2 材料与方法... 23

2.2.1 研究区域... 23

2.2.2 数据收集与整理... 24

2.2.3 数据分析方法... 27

2.3 结果... 29

2.3.1 有害藻华发生情况... 29

2.3.2 有害藻华演变特征... 33

2.3.3 环境因子与有害藻华的关系... 39

2.3.4 有害藻华预测... 42

2.3.5 有害藻华风险评估... 43

2.4 讨论... 46

2.4.1 有害藻华的演变特征... 46

2.4.2 环境因子对有害藻华的影响... 47

2.4.3 有害藻华的风险... 48

2.4.4 存在的问题和不足... 49

2.5 小结... 49

第3章 黄海有害藻华演变特征及其关键驱动因子... 51

3.1 前言... 51

3.2 材料与方法... 52

3.2.1 研究区域... 52

3.2.2 数据收集与整理... 53

3.2.3 数据分析方法... 57

3.3 结果... 57

3.3.1 有害藻华发生情况... 57

3.3.2 有害藻华演变特征... 62

3.3.3 环境因子与赤潮的关系... 65

3.4 讨论... 68

3.4.1 甲藻赤潮优势度增加... 68

3.4.2 绿潮对赤潮的潜在影响... 70

3.4.3 存在的问题和不足... 70

3.5 小结... 70

第4章 麻痹性贝类毒素和棕囊藻囊体预测方法初探... 73

4.1 前言... 73

4.2 材料与方法... 74

4.2.1 研究区域... 74

4.2.2 数据收集与整理... 75

4.2.3 数据分析方法... 79

4.3 结果... 89

4.3.1 预测模型选择... 89

4.3.2 预测模型优化与特征选择... 91

4.3.3 预测模型验证... 96

4.4 讨论... 97

4.4.1 模型评价... 97

4.4.2 预测预警指标意义... 98

4.4.3 存在的问题和不足... 99

4.5 小结... 100

第5章 有害藻华数据分析与可视化平台的设计与应用... 101

5.1 前言... 101

5.2 总体设计... 102

5.3 功能设计... 103

5.3.1 数据管理平台... 104

5.3.2 GIS平台... 104

5.3.3 可视化平台... 104

5.4 平台运行环境... 105

5.5 平台应用... 106

5.5.1 数据管理平台... 106

5.5.2 GIS平台... 107

5.5.3 可视化平台... 109

5.6 小结... 110

第6章 结论与创新点... 111

6.1 结论... 111

6.2 创新点... 112

参考文献... 115

附录  缩略语表... 137

致  谢... 139

作者简历及攻读学位期间发表的学术论文与研究成果... 143