Finding hydrothermal chimneys along the southern East Pacific Rise with machine learning approaches to AUV-based sonar data

海底热液系统释放出热的、富含矿物质和微生物的液体，这些液体会影响海水的化学性质，对从气候变化到生物医学科学的方方面面产生影响。有史以来最详细和全面的羽流调查之一将在东南太平洋隆起进行。这次研究巡航将沿着300多公里长的洋中脊描绘漫射和聚焦热液排放的特征。基于声纳和photomosaic数据解析特征到一米以下，机器学习方法将从自主水下航行器声纳测绘中检测和表征热液喷口。这将建立一个通风口和热液地点的全面清单，并检查该系统自大约24年前的初步研究以来的变化。从亚米网格声纳数据自动检测可能的热液烟囱的工作流程将增强深潜测绘声纳的所有用户的基础设施，因为搜索热液喷口是这些仪器的常见用途。该研究将实现一个前端图形用户界面，它将提供给任何具有地理信息系统基本技能的人使用。为了使海底测绘的经验能够提供给更广泛的社区，将制作一个虚拟的实地考察，“寻找深海热液喷口”，让学生和公众接触到一个很少观察到但对地质和生物多样性都非常重要的环境。该项目还将培养一名对下一代海洋地球科学家有贡献的博士生。该研究的中心假设是热液烟囱分布与可观测的地质特征相关，这些相关性将为热液通量、寿命和烟囱分布从脊段到全球尺度的变化提供重要的见解。该项目计划在~15°-18°S的东南东太平洋凸起的热液喷口区收集声纳数据和照相图像。这些数据将用于开发一种机器学习方法，从亚米尺度声纳地图中检测和表征热液喷口。机器学习在寻找热液喷口方面的发展和应用，将有助于检测近24年后火山构造形态和热液喷口场的变化，其中岩浆复发间隔被认为约为7年。使用Python实现的机器学习算法的用户友好界面将使非专家能够从水深数据中绘制候选热液烟囱，以协助更有效的潜水计划和/或识别目标以进行进一步调查。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Seafloor hydrothermal systems emit hot, mineral- and microbial-rich fluids that impact seawater chemistry, with ramifications for everything from climate change to biomedical sciences. One of the most detailed and comprehensive plume surveys ever will be conducted on the South-East Pacific Rise. This research cruise will characterize both diffuse and focused hydrothermal discharge along a 300+ km length of the mid-ocean ridge. Based on sonar and photomosaic data resolving features to well under a meter in size, a machine learning method will detect and characterize hydrothermal vents from autonomous underwater vehicle sonar mapping. This will establish a comprehensive inventory of vent chimneys and hydrothermal sites, as well as examine changes in this system since its initial study about 24 years ago. Workflows for automated detection of possible hydrothermal chimneys from sub-meter gridded sonar data would enhance the infrastructure for all users of deep-submergence mapping sonars since searching for hydrothermal vents is a common use of these instruments. The study will implement a front-end Graphic User Interface that will be made available to anyone with basic skills in Geographic Information Systems to use. In order to make the experience of seafloor mapping available to an even broader community, a virtual fieldtrip, “Hunting Deep-Sea Hydrothermal Vents”, will be produced and expose students and the public to a rarely observed yet very critical environment for both geology and biodiversity. The project also will train a PhD student contributing to the next generation of marine geoscientists.The central hypothesis of the research is that hydrothermal chimney distributions correlate with observable geologic features, and that those correlations will provide important insights into the causes of variations in hydrothermal flux, longevity, and chimney distribution from the ridge segment to global scales. This project plans to collect sonar data and photomosaics over hydrothermal vent fields along the Southern East Pacific Rise from ~15°-18°S. These data would be used to develop a machine-learning method to detect and characterize hydrothermal vents from sub-meter scale sonar maps. The development and application of machine learning to find hydrothermal vents will assist in detecting changes after almost 24 years in both in volcanic-tectonic morphology as well as hydrothermal vent fields at a ridge where the magmatic recurrence interval is thought to be approximately 7 years long. A user-friendly interface to the machine learning algorithm, implemented in Python, would enable non-experts to map candidate hydrothermal chimneys from bathymetric data to assist in more efficient dive planning and/or identification of targets for further investigation.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.