Predicting the spatial and temporal variations in habitat characteristics of Harpadon nehereus in the Yangtze River Estuary based on ensemble modeling
-
摘要: 为了评估“集成方法”能否改进物种分布模型（Species Distribution Models, SDMs）在海洋环境高度动态变化的河口区域的预测性能，本研究基于2013—2021年长江口海洋生物资源调查数据，使用8种基于不同算法的单一模型对长江口的优势物种之一龙头鱼（Harpadon nehereus）构建了栖息地生境的集成模型（Ensemble Model, EM）。结果显示：（1）所有单一模型的预测性能均优于随机分布模型，而EM具有最高的预测准确性和稳健性（受试者工作特征曲线下面积（Area Under receiver operating character Curve, AUC）=0.875；真实技巧统计值（True skill statistic, TSS）=0.650；KAPPA系数=0.560；总体精度（Overall accuracy, OA）=0.867）；（2）EM能最为准确地识别出龙头鱼的出现点和未出现点，也能清晰地区分出未采样区域适宜性水平的差异，并预测出不同模型共同的高适宜区域；（3）最后，EM能准确识别龙头鱼的关键环境需求并反映出多模型集中的变化趋势，其最适盐度、温度和化学需氧量的范围分别为2.754—30.300、28.278—30.934℃、4.605—8.080 mg/L。本研究可为长江口龙头鱼资源的可持续利用和栖息地保护工作提供更可靠的研究方法。Abstract: To evaluate whether the ‘ensemble approach’ can enhance the predictive performance of Species Distribution Models (SDMs) in dynamically changing estuarine environments, this study utilized eight single models based on different algorithms to construct an Ensemble Model (EM) for the habitat of Harpadon nehereus, a dominant species in the Yangtze River Estuary (YRE). The data used for modeling were derived from marine biological resource surveys conducted in the YRE from 2013 to 2021. The results showed that: (1) All single models outperformed the random distribution model, with the EM demonstrating the highest predictive accuracy and robustness (Area Under receiver operating character Curve, AUC=0.875; True skill statistic, TSS=0.650; KAPPA=0.560; Overall accuracy, OA=0.867); (2) The EM accurately identified both presence and absence stations of H. nehereus, clearly differentiated suitability levels in unsampled regions, and predicted areas of high suitability shared by different models; (3) Finally, the EM accurately identified the key environmental requirements of H. nehereus and reflected the central tendency across multiple models. The most suitable habitat for H. nehereus was found in waters with salinity, temperature, and chemical oxygen demand ranges of 2.754—30.300, 28.278—30.934°C, and 4.605—8.080 mg/L, respectively. This study provides a more reliable research method for the sustainable utilization and habitat protection of H. nehereus resources in the YRE.
-
图 1 长江口海洋生物资源监测调查的站点分布和底拖网调查示意图
Fig. 1 Distribution map of monitoring stations for marine biological resources in the Yangtze River Estuary
图 2 构建长江口龙头鱼栖息地模型的技术路线图（GBM：梯度提升机；RF：随机森林；ANN：人工神经网络；CTA：分类树分析；FDA：柔性判别分析；GAM：广义加性模型；GLM：广义线性模型；SRE：表面分布区分室模型；EM：集成模型）
Fig. 2 Technical framework for constructing the habitat model of Harpadon nehereus in the Yangtze River Estuary (GBM: Gradient Boosting Machine; RF: Random Forest; ANN: Artificial Neural Network; CTA: Classification Tree Analysis; FDA: Flexible Discriminant Analysis; GAM: Generalized Additive Model; GLM: Generalized Linear Model; SRE: Surface Range Envelop; EM: Ensemble Model）
图 3 各解释变量的皮尔逊相关性系数矩阵分析图（COD：化学需氧量；Depth：水深；Tem：水温；Chl a：叶绿素a；Sal：盐度；DO：溶解氧；prey：饵料生物资源密度）
Fig. 3 Pearson correlation coefficient for explanatory variables (COD: chemical oxygen demand; Tem: temperature; Chl a: chlorophyll a; Sal: salinity; DO: dissolved oxygen; prey: the stock density of prey species)
图 4 8种单一模型和集成模型（Ensemble Model, EM）的准确性评价结果
（1）AUC得分箱线图；（2）TSS得分箱线图；（3）KAPPA得分箱线图；（4）OA得分箱线图
Fig. 4 Model accuracy scores obtained by the eight single models and Ensemble Model (EM)
(1) the box plot of the AUC scores; (2) the box plot of the TSS scores; (3) the box plot of the KAPPA scores; (4) the box plot of the OA scores
图 5 2015年8种单一模型和集成模型（Ensemble Model, EM）预测的龙头鱼适生区分布与实际调查结果的对比
（1）—（4）GBM各季节预测结果；（5）—（8）RF各季节预测结果；（9）—（12）FDA各季节预测结果；（13）—（16）GLM各季节预测结果；（17）—（20）GAM各季节预测结果；（21）—（24）ANN各季节预测结果；（25）—（28）CTA各季节预测结果；（29）—（32）SRE各季节预测结果；（33）—（36）EM各季节预测结果
Fig. 5 Comparison of the distribution of suitable habitat for Harpadon nehereus predicted by the eight single models and Ensemble Model (EM) in 2015 with the actual observations
(1) — (4) seasonal predictions of GBM; (5) — (8) seasonal predictions of RF; (9) — (12) seasonal predictions of FDA; (13) — (16) seasonal predictions of GLM; (17) — (20) seasonal predictions of GAM; (21) — (24) seasonal predictions of ANN; (25) — (28) seasonal predictions of CTA; (29) — (32) seasonal predictions of SRE; (33) — (36) seasonal predictions of EM
图 6 5种单一模型和集成模型（Ensemble Model, EM）预测的各环境变量重要性雷达图（Tem：水温；Chl a：叶绿素a；Sal：盐度；COD：化学需氧量；Depth：水深；prey：饵料生物资源密度）
（1）GBM的因子重要性雷达图；（2）RF的因子重要性雷达图；（3）CTA的因子重要性雷达图；（4）GAM的因子重要性雷达图；（5）SRE的因子重要性雷达图；（6）EM的因子重要性雷达图
Fig. 6 Radar charts of the importance of each environmental variable predicted by the five single models and Ensemble Model (EM) (Tem: temperature; Chl a: chlorophyll a; Sal: salinity; COD: chemical oxygen demand; prey: the stock density of prey species)
(1) radar chart of the importance of variables in GBM; (2) radar chart of the importance of variables in RF; (3) radar chart of the importance of variables in CTA; (4) radar chart of the importance of variables in GAM; (5) radar chart of the importance of variables in SRE; (6) radar chart of the importance of variables in EM
图 7 4种单一模型和集成模型（Ensemble Model, EM）绘制的环境变量偏依赖图（Sal：盐度；Tem：水温；COD：化学需氧量）
（1）Sal的偏依赖图；（2）Tem的偏依赖图；（3）COD的偏依赖图
Fig. 7 Partial dependence plots of environmental variables obtained by the four single models and Ensemble Model (EM) (Sal: salinity; Tem: temperature; COD: chemical oxygen demand)
(1) partial dependence plot of Sal; (2) partial dependence plot of Tem; (3) partial dependence plot of COD
表 1 本研究中各单一模型的参数设置
Tab. 1 Parameter settings for each single model in this study
类别 模型 参数 值 描述 机器学习算法 梯度提升机
（Gradient Boosting Machine, GBM）n.trees 2500 生成树的数量 interaction.depth 7 交互项的最大深度 shrinkage 0.001 在构建每棵树时，收缩节点的权重 随机森林
（Random forest, RF）ntree 500 生成树的数量 mtry default 每次分割时随机抽样作为候选变量的数量 nodesize 5 每个终端节点的最少样本数 人工神经网络
（Artificial Neural Network, ANN）size NULL 神经网络的规模 decay NULL 学习率衰减因子 maxit 200 训练神经网络时的最大迭代次数 分类算法 分类树分析
（Classification Tree Analysis, CTA）minbucket 5 节点中最少的样本数 minsplit 5 最小分裂点的值 cp 0.001 控制修剪范围 柔性判别分析（Flexible Discriminant Analysis, FDA） Method Mars FDA的建模方法 add_args NULL 附加参数列表 回归算法 广义加性模型（Generalized Additive Model, GAM） algo GAM_mgcv 使用的算法和优化方法 k −1 平滑器的阶次 interaction.level 0 交互项的阶次 广义线性模型
（Generalized Linear Model, GLM）type Quadratic 模型类型 interaction.level 0 交互项的阶次 mustart 0.5 默认起始值 包络算法 表面分布区分室模型（Surface Range Envelop, SRE） quant 0.025 模型的分辨率或离散程度 
下载: 导出CSV
-
[1] 潘邵媛, 王学昉, 田思泉, 等. 长江口中华鲟保护区海洋环境监测浮标站点的优化设计[J]. 海洋学报, 2021, 43(4): 55−64.Pan Shaoyuan, Wang Xuefang, Tian Siquan, et al. The design of the stations of marine environmental monitoring buoys in the Chinese sturgeon nature reserve in the Changjiang River Estuary[J]. Haiyang Xuebao, 2021, 43(4): 55−64. [2] 杨钧渊, 陈锦辉, 钟俊生, 等. 长江口崇明东滩水域仔稚鱼种类组成及多样性[J]. 上海海洋大学学报, 2023, 32(4): 829−840. doi: 10.12024/jsou.20230204101Yang Junyuan, Chen Jinhui, Zhong Junsheng, et al. Species composition and diversity of fish larvae and juveniles in the water area of Chongming Dongtan, Yangtze River Estuary[J]. Journal of Shanghai Ocean University, 2023, 32(4): 829−840. doi: 10.12024/jsou.20230204101 [3] 潘绪伟, 程家骅. 长江口外海域龙头鱼营养生态学特征[J]. 中国水产科学, 2011, 18(5): 1132−1140.Pan Xuwei, Cheng Jiahua. Feeding ecology of Harpadon nehereus in areas adjacent to Changjiang River estuary[J]. Journal of Fishery Sciences of China, 2011, 18(5): 1132−1140. [4] Li Yu, Gao Chunxia, Chen Jinhui, et al. Spatial–temporal distribution characteristics of Harpadon nehereus in the Yangtze River Estuary and its relationship with environmental factors[J]. Frontiers in Marine Science, 2024, 11: 1340522. doi: 10.3389/fmars.2024.1340522 [5] 唐未, 王学昉, 吴峰, 等. 基于最大熵模型模拟西印度洋剑鱼栖息地的时空分布[J]. 海洋学报, 2022, 44(10): 100−108. doi: 10.12284/j.issn.0253-4193.2022.10.hyxb202210009Tang Wei, Wang Xuefang, Wu Feng, et al. Simulation of spatio-temporal distribution of swordfish habitat in the western Indian Ocean based on maximum entropy model[J]. Haiyang Xuebao, 2022, 44(10): 100−108. doi: 10.12284/j.issn.0253-4193.2022.10.hyxb202210009 [6] Le Pape O, Baulier L, Cloarec A, et al. Habitat suitability for juvenile common sole (Solea solea, L. ) in the Bay of Biscay (France): A quantitative description using indicators based on epibenthic fauna[J]. Journal of Sea Research, 2007, 57(2/3): 126−136. [7] França S, Cabral H N. Predicting fish species distribution in estuaries: Influence of species’ ecology in model accuracy[J]. Estuarine, Coastal and Shelf Science, 2016, 180: 11−20. doi: 10.1016/j.ecss.2016.06.010 [8] Austin M. Species distribution models and ecological theory: A critical assessment and some possible new approaches[J]. Ecological Modelling, 2007, 200(1/2): 1−19. [9] Jiang Rijin, Sun Haoqi, Li Xiafang, et al. Habitat suitability evaluation of Harpadon nehereus in nearshore of Zhejiang province, China[J]. Frontiers in Marine Science, 2022, 9: 961735. doi: 10.3389/fmars.2022.961735 [10] Mohammadi A, Almasieh K, Nayeri D, et al. Comparison of habitat suitability and connectivity modelling for three carnivores of conservation concern in an Iranian montane landscape[J]. Landscape Ecology, 2022, 37(2): 411−430. doi: 10.1007/s10980-021-01386-5 [11] Poulos H M, Chernoff B, Fuller P L, et al. Ensemble forecasting of potential habitat for three invasive fishes[J]. Aquatic Invasions, 2012, 7(1): 59−72. doi: 10.3391/ai.2012.7.1.007 [12] 王寇, 李博, 李爱国, 等. 夏季长江口及其邻近海域湍流特征分析[J]. 海洋学报, 2021, 43(11): 22−31.Wang Kou, Li Bo, Li Aiguo, et al. Characteristics of turbulence in the Changjiang River Estuary and its adjacent waters in summer[J]. Haiyang Xuebao, 2021, 43(11): 22−31. [13] 史赟荣, 晁敏, 沈新强. 长江口张网鱼类群落结构特征及月相变化[J]. 海洋学报, 2014, 36(2): 81−92. doi: 10.3969/j.issn.0253-4193.2014.02.009Shi Yunrong, Chao Min, Shen Xinqiang. Characteristics and monthly variations of set net fish community structure in the Changjiang River estuary[J]. Haiyang Xuebao, 2014, 36(2): 81−92. doi: 10.3969/j.issn.0253-4193.2014.02.009 [14] 吴晓丹, 宋金明, 李学刚. 长江口邻近海域水团特征与影响范围的季节变化[J]. 海洋科学, 2014, 38(12): 110−119. doi: 10.11759/hykx20140305001Wu Xiaodan, Song Jinming, Li Xuegang. Ssonal variation of water mass characteristic and influence area in the Yangtze Estuary and its adjacent waters[J]. Marine Sciences, 2014, 38(12): 110−119. doi: 10.11759/hykx20140305001 [15] 周晓英. 长江口海域表层水温变化的气候特征[D]. 青岛: 中国海洋大学, 2005.Zhou Xiaoying. Climate characteristics of sea surface temperature (SST) variation in the Changjiang Estuary[D]. Qingdao: Ocean University of China, 2005. [16] 何柄震. 长江口海域营养物质时空演变趋势及其对入海通量的响应研究[D]. 沈阳: 沈阳大学, 2024.He Bingzhen. The spatiotemporal evolution trend of nutrients in the Yangtze River Estuary and its response to the fluxes into the sea[D]. Shenyang: Shenyang University, 2024. [17] 朱宇新, 周斌, 赵俊杰. 盐水入侵对长江口盐度分布及影响研究[J]. 环境保护前沿, 2014, 4(1): 25−33. doi: 10.12677/AEP.2014.41B005Zhu Yuxin, Zhou Bin, Zhao Junjie. Research on the salinity distribution and the influence of saltwater intrusion in Yangtze River[J]. Advances in Environmental Protection, 2014, 4(1): 25−33. doi: 10.12677/AEP.2014.41B005 [18] 马金, 黄金玲, 陈锦辉, 等. 基于GAM的长江口鱼类资源时空分布及影响因素[J]. 水产学报, 2020, 44(6): 958−968.Ma Jin, Huang Jinling, Chen Jinhui, et al. Analysis of spatiotemporal fish density distribution and its influential factors based on generalized additive model (GAM) in the Yangtze River estuary[J]. Journal of Fisheries of China, 2020, 44(6): 958−968. [19] Wang Yichuan, Wu Xinghua, Zheng Leifu, et al. Modeling seasonal changes in the habitat suitability of Coilia nasus in the Yangtze River Estuary using tree-based methods[J]. Regional Studies in Marine Science, 2023, 67: 103212. doi: 10.1016/j.rsma.2023.103212 [20] França S, Costa M J, Cabral H N. Inter- and intra-estuarine fish assemblage variability patterns along the Portuguese coast[J]. Estuarine, Coastal and Shelf Science, 2011, 91(2): 262−271. doi: 10.1016/j.ecss.2010.10.035 [21] 中华人民共和国国家质量监督检验检疫总局, 中国国家标准化管理委员会. GB/T 12763.6-2007, 海洋调查规范 第6部分: 海洋生物调查[S]. 北京: 中国标准出版社, 2008: 8.General Administration of Quality Supervision, Inspection and Quarantine of the People's Republic of China, Standardization Administration of the People’s Republic of China. GB/T 12763.6-2007, Specifications for oceanographic survey-Part 6: Marine biological survey[S]. Beijing: Standards Press of China, 2008: 8. [22] 国家环境保护局. GB 11607-89, 渔业水质标准[S]. 北京: 中国标准出版社, 1990: 8.Environmental Protection Agency. GB 11607-89, Water quality standard for fisheries[S]. Beijing: Standards Press of China, 1990: 8. [23] 环境保护部. HJ 828—2017, 水质 化学需氧量的测定 重铬酸盐法[S]. 北京: 中国环境出版社, 2017: 3.Ministry of Environmental Protection. HJ 828—2017, Water quality-Determination of the chemical oxygen demand-Dichromate method[S]. Beijing: China Environmental Publishing House, 2017: 3. [24] Bacheler N M, Paramore L M, Buckel J A, et al. Abiotic and biotic factors influence the habitat use of an estuarine fish[J]. Marine Ecology Progress Series, 2009, 377: 263−277. doi: 10.3354/meps07805 [25] Luan Jing, Xu Binduo, Ji Yupeng, et al. Improving the spatial transferability of species distribution models to inform biological conservation of two piscivore fish species[J]. Biodiversity and Conservation, 2024, 33(14): 4215−4235. doi: 10.1007/s10531-024-02947-1 [26] 徐超, 王思凯, 赵峰, 等. 长江口水生动物食物网营养结构及其变化[J]. 水生生物学报, 2019, 43(1): 155−164. doi: 10.7541/2019.019Xu Chao, Wang Sikai, Zhao Feng, et al. Trophic structure of food web and its variation on aquatic animals in the Yangtze Estuary[J]. Acta Hydrobiologica Sinica, 2019, 43(1): 155−164. doi: 10.7541/2019.019 [27] Yalçın E, Gurbet R. Environmental influences on the spatio-temporal distribution of European Hake (Merluccius merluccius) in Izmir Bay, Aegean Sea[J]. Turkish Journal of Fisheries and Aquatic Sciences, 2016, 16(1): 1−14. [28] Dormann C F, Elith J, Bacher S, et al. Collinearity: A review of methods to deal with it and a simulation study evaluating their performance[J]. Ecography, 2013, 36(1): 27−46. doi: 10.1111/j.1600-0587.2012.07348.x [29] Shrestha U B, Sharma K P, Devkota A, et al. Potential impact of climate change on the distribution of six invasive alien plants in Nepal[J]. Ecological Indicators, 2018, 95: 99−107. doi: 10.1016/j.ecolind.2018.07.009 [30] Elith J, Graham C H, Anderson R P, et al. Novel methods improve prediction of species’ distributions from occurrence data[J]. Ecography, 2006, 29(2): 129−151. doi: 10.1111/j.2006.0906-7590.04596.x [31] Fielding A H, Bell J F. A review of methods for the assessment of prediction errors in conservation presence/absence models[J]. Environmental Conservation, 1997, 24(1): 38−49. doi: 10.1017/S0376892997000088 [32] Liu Canran, White M, Newell G. Measuring and comparing the accuracy of species distribution models with presence-absence data[J]. Ecography, 2011, 34(2): 232−243. doi: 10.1111/j.1600-0587.2010.06354.x [33] Hao Tianxiao, Elith J, Guillera-Arroita G, et al. A review of evidence about use and performance of species distribution modelling ensembles like BIOMOD[J]. Diversity and Distributions, 2019, 25(5): 839−852. doi: 10.1111/ddi.12892 [34] Marmion M, Parviainen M, Luoto M, et al. Evaluation of consensus methods in predictive species distribution modelling[J]. Diversity and Distributions, 2009, 15(1): 59−69. doi: 10.1111/j.1472-4642.2008.00491.x [35] Thuiller W, Georges D, Engler R. Biomod2: Ensemble platform for species distribution modelling[Z]. 2014. (查阅网上资料, 未找到本条文献出版信息, 请确认) [36] Liu Zunlei, Jin Yan, Yang Linlin, et al. Improving prediction for potential spawning areas from a two-step perspective: A comparison of multi-model approaches for sparse egg distribution[J]. Journal of Sea Research, 2024, 197: 102460. doi: 10.1016/j.seares.2023.102460 [37] Phillips N D, Reid N, Thys T, et al. Applying species distribution modelling to a data poor, pelagic fish complex: The ocean sunfishes[J]. Journal of Biogeography, 2017, 44(10): 2176−2187. doi: 10.1111/jbi.13033 [38] França S, Cabral H N. Predicting fish species richness in estuaries: Which modelling technique to use?[J]. Environmental Modelling & Software, 2015, 66: 17−26. [39] Dreiseitl S, Ohno-Machado L. Logistic regression and artificial neural network classification models: a methodology review[J]. Journal of Biomedical Informatics, 2002, 35(5/6): 352−359. [40] Li Zengguang, Ye Zhenjiang, Wan Rong, et al. Model selection between traditional and popular methods for standardizing catch rates of target species: A case study of Japanese Spanish mackerel in the gillnet fishery[J]. Fisheries Research, 2015, 161: 312−319. doi: 10.1016/j.fishres.2014.08.021 [41] 刘尊雷, 杨林林, 袁兴伟, 等. 基于集成模型的小黄鱼越冬群体适宜生境及其环境影响因素[J]. 应用生态学报, 2020, 31(6): 2076−2086.Liu Zunlei, Yang Linlin, Yuan Xingwei, et al. Overwintering distribution and its environmental determinants of small yellow croaker based on ensemble habitat suitability modeling[J]. Chinese Journal of Applied Ecology, 2020, 31(6): 2076−2086. [42] Povak N A, Hessburg P F, Reynolds K M, et al. Machine learning and hurdle models for improving regional predictions of stream water acid neutralizing capacity[J]. Water Resources Research, 2013, 49(6): 3531−3546. doi: 10.1002/wrcr.20308 [43] Dormann C F, Purschke O, Márquez J R G, et al. Components of uncertainty in species distribution analysis: a case study of the Great Grey Shrike[J]. Ecology, 2008, 89(12): 3371−3386. doi: 10.1890/07-1772.1 [44] Yang Xiaolong, Zhang Xiumei, Zhang Peidong, et al. Ensemble habitat suitability modeling for predicting optimal sites for eelgrass (Zostera marina) in the tidal lagoon ecosystem: Implications for restoration and conservation[J]. Journal of Environmental Management, 2023, 330: 117108. doi: 10.1016/j.jenvman.2022.117108 [45] Stohlgren T J, Ma P, Kumar S, et al. Ensemble habitat mapping of invasive plant species[J]. Risk Analysis, 2010, 30(2): 224−235. doi: 10.1111/j.1539-6924.2009.01343.x [46] Araújo M B, New M. Ensemble forecasting of species distributions[J]. Trends in Ecology & Evolution, 2007, 22(1): 42−47. [47] Araújo M B, Whittaker R J, Ladle R J, et al. Reducing uncertainty in projections of extinction risk from climate change[J]. Global Ecology and Biogeography, 2005, 14(6): 529−538. doi: 10.1111/j.1466-822X.2005.00182.x [48] 孙浩奇. 浙江近海龙头鱼生物学特征及其时空分布与环境因子关系[D]. 舟山: 浙江海洋大学, 2022.Sun Haoqi. A study on the biological characteristics, temporal and spatial distribution, and the relationship with environmental factors of Harpadon nehereus in Zhejiang Province coastal waters[D]. Zhoushan: Zhejiang Ocean University, 2022. [49] Lewin W C, Mehner T, Ritterbusch D, et al. The influence of anthropogenic shoreline changes on the littoral abundance of fish species in German lowland lakes varying in depth as determined by boosted regression trees[J]. Hydrobiologia, 2014, 724(1): 293−306. doi: 10.1007/s10750-013-1746-8 [50] Collins S D, Abbott J C, Mcintyre N E. Quantifying the degree of bias from using county-scale data in species distribution modeling: can increasing sample size or using county-averaged environmental data reduce distributional overprediction?[J]. Ecology and Evolution, 2017, 7(15): 6012−6022. doi: 10.1002/ece3.3115 -
计量
- 文章访问数: 7
- HTML全文浏览量: 4
- PDF下载量: 0
- 被引次数: 0
