Assessing species distributions and morphometrics in Southern Ocean diatoms using high throughput imaging and semi-automated image analysis

硅藻是南大洋(SO)的主要初级生产者，强烈影响着这一广泛的远洋栖息地的生态和生物地球化学过程。个别物种的出现，以及其中一些物种的形态特征，都与水团性质有关，并通过后者与气候变化有关。因此，群落组成和特定物种的形态计量学都被广泛用作过去气候变化的指标。另一方面，持续的气候变化影响了海洋学，并通过此影响了硅藻的分布，进而影响了食物网络结构和生物地球化学循环。研究硅藻-环境关系的中心工具是光学显微镜(LM)，无论是用于开发/应用替代物，还是用于了解当前和未来的分布范围。以传统方式实践的LM存在许多问题，包括吞吐量低、对高素质分类学专家的要求以及与数据归档/质量和重复性相关的问题。为了解决这些问题，该项目的申请者最近开始了一项合作，将高通量光学显微镜和自动图像分析方法结合在一起，用于硅藻的形态特征和组合成分评估。在这次合作的头9个月内，我们成功地建立了高通量、半自动的工作流程，用于对永久性载玻片上的硅藻标本进行显微成像；开发了硅藻瓣膜的自动检测和轮廓表征软件(正在出版中)，提供了人工交互和质量控制的可能性；测试了从硅藻图像中提取纹理特征的方法；并使用机器学习方法进行了基于上述方法提取的形态特征的硅藻分类的第一次测试。这些方法使我们能够拍摄和分析大量的硅藻载玻片，结合了自动化方法的吞吐量和人工质量控制的严格性。在这里，我们建议将这些方法应用于Hustedt硅藻研究中心广泛收集的SO样本，以表征丰富的浮游SO硅藻物种的地理分布和形态计量学，并利用统计技术将它们与环境参数联系起来。拟议的工作将使我们能够首次使用硅藻物种的存在-缺失数据来验证分布模型，并探索与多个物种的环境参数相关的形态测量趋势，这可能为开发强大的古海洋学指标开辟新的可能性。

Diatoms are the dominant primary producers of the Southern Ocean (SO), strongly influencing ecological and biogeochemical processes in this extensive pelagic habitat. Both occurrence of individual species, and morphometric features of some of them, are related to water mass properties, and, through the latter, to climatic changes. Accordingly, both assemblage composition and species-specific morphometrics are heavily used as indicators of past climatic changes. On the other hand, ongoing climate change affects SO oceanography, and through that, also diatom distributions, which in turn affect food web structure and biogeochemical cycles. The central tool in studies of diatom-environment relationships has been light microscopy (LM), whether for development / application of proxies or for learning about current and future distribution ranges. LM, practiced in the traditional way, has numerous issues, including low throughput, requirement of highly qualified taxonomic experts, and problems related data archival / quality and reproducibility. To address these issues, the applicants of this project have recently started a collaboration bringing together high throughput light microscopy and automated image analysis methods for morphometric characterization and assemblage composition assessment of diatoms. Within the first 9 months of this collaboration, we successfully established workflows for high throughput, semi-automated microscopic imaging of diatom specimens on permanent slides; developed software (under publication) for automated detection and outline characterization of diatom valves providing the possibility for manual interaction and quality control; tested the extraction of texture features from diatom images; and performed first tests using machine learning methods for the taxonomic classification of diatoms based on the morphometric features extracted by the above methods. These methods allow us to photograph and analyze large numbers of diatom slides combining the throughput of automated methods with the stringency of manual quality control. Here we propose to apply these methods to the extensive collection of SO samples of the Hustedt Diatom Study Centre to characterize the geographic distribution and morphometrics of abundant planktonic SO diatom species, and to relate them to environmental parameters using statistical techniques. The work proposed will allow us to validate distribution models using presence-absence data for the first time for diatoms species, and to explore morphometric trends related to environmental parameters across multiple species, which might open novel possibilities for the development of robust paleo-oceanographic proxies.