skip to main content


Title: Building on a human-centred, iterative, and agile co-design strategy to facilitate the availability of deep ocean data
Abstract

Current information on the status and trends of ocean change is needed to support effective and responsive management, particularly for the deep ocean. Creating consistent, collaborative and actionable mechanisms is a key component of the Deep Ocean Observing Strategy, a program of the United Nations Decade of Ocean Science for Sustainable Development. Here, we share an iterative, agile, and human-centred approach to co-designing datastreams for deep-sea indicators that serves stakeholders, including US National Marine Sanctuaries, presented as a four-phase project roadmap initially focused on the Monterey Bay National Marine Sanctuary, and then generalized to other areas such as the US West Coast, offshore wind development areas, and managed marine spaces globally. Ongoing efforts to provide key physical, biogeochemical, biological, and ecosystem variables for California's Marine Protected Areas are informing this co-design process. We share lessons learned so far and present co-design as a useful tool for (1) assessing the availability of information from deep ecosystems, (2) ensuring interoperability, and (3) providing essential information on the status and trends of indicators. Documenting and sharing this co-design strategy and scalable four-phase roadmap will further the aims of DOOS and other initiatives, including the Deep Ocean Stewardship Initiative and Challenger 150.

 
more » « less
Award ID(s):
2114717
NSF-PAR ID:
10370851
Author(s) / Creator(s):
; ; ; ; ; ; ;
Publisher / Repository:
Oxford University Press
Date Published:
Journal Name:
ICES Journal of Marine Science
Volume:
80
Issue:
2
ISSN:
1054-3139
Format(s):
Medium: X Size: p. 347-351
Size(s):
["p. 347-351"]
Sponsoring Org:
National Science Foundation
More Like this
  1. In this paper, we outline the need for a coordinated international effort toward the building of an open-access Global Ocean Oxygen Database and ATlas (GO 2 DAT) complying with the FAIR principles (Findable, Accessible, Interoperable, and Reusable). GO 2 DAT will combine data from the coastal and open ocean, as measured by the chemical Winkler titration method or by sensors (e.g., optodes, electrodes) from Eulerian and Lagrangian platforms (e.g., ships, moorings, profiling floats, gliders, ships of opportunities, marine mammals, cabled observatories). GO 2 DAT will further adopt a community-agreed, fully documented metadata format and a consistent quality control (QC) procedure and quality flagging (QF) system. GO 2 DAT will serve to support the development of advanced data analysis and biogeochemical models for improving our mapping, understanding and forecasting capabilities for ocean O 2 changes and deoxygenation trends. It will offer the opportunity to develop quality-controlled data synthesis products with unprecedented spatial (vertical and horizontal) and temporal (sub-seasonal to multi-decadal) resolution. These products will support model assessment, improvement and evaluation as well as the development of climate and ocean health indicators. They will further support the decision-making processes associated with the emerging blue economy, the conservation of marine resources and their associated ecosystem services and the development of management tools required by a diverse community of users (e.g., environmental agencies, aquaculture, and fishing sectors). A better knowledge base of the spatial and temporal variations of marine O 2 will improve our understanding of the ocean O 2 budget, and allow better quantification of the Earth’s carbon and heat budgets. With the ever-increasing need to protect and sustainably manage ocean services, GO 2 DAT will allow scientists to fully harness the increasing volumes of O 2 data already delivered by the expanding global ocean observing system and enable smooth incorporation of much higher quantities of data from autonomous platforms in the open ocean and coastal areas into comprehensive data products in the years to come. This paper aims at engaging the community (e.g., scientists, data managers, policy makers, service users) toward the development of GO 2 DAT within the framework of the UN Global Ocean Oxygen Decade (GOOD) program recently endorsed by IOC-UNESCO. A roadmap toward GO 2 DAT is proposed highlighting the efforts needed (e.g., in terms of human resources). 
    more » « less
  2. Obeid, Iyad Selesnick (Ed.)
    Electroencephalography (EEG) is a popular clinical monitoring tool used for diagnosing brain-related disorders such as epilepsy [1]. As monitoring EEGs in a critical-care setting is an expensive and tedious task, there is a great interest in developing real-time EEG monitoring tools to improve patient care quality and efficiency [2]. However, clinicians require automatic seizure detection tools that provide decisions with at least 75% sensitivity and less than 1 false alarm (FA) per 24 hours [3]. Some commercial tools recently claim to reach such performance levels, including the Olympic Brainz Monitor [4] and Persyst 14 [5]. In this abstract, we describe our efforts to transform a high-performance offline seizure detection system [3] into a low latency real-time or online seizure detection system. An overview of the system is shown in Figure 1. The main difference between an online versus offline system is that an online system should always be causal and has minimum latency which is often defined by domain experts. The offline system, shown in Figure 2, uses two phases of deep learning models with postprocessing [3]. The channel-based long short term memory (LSTM) model (Phase 1 or P1) processes linear frequency cepstral coefficients (LFCC) [6] features from each EEG channel separately. We use the hypotheses generated by the P1 model and create additional features that carry information about the detected events and their confidence. The P2 model uses these additional features and the LFCC features to learn the temporal and spatial aspects of the EEG signals using a hybrid convolutional neural network (CNN) and LSTM model. Finally, Phase 3 aggregates the results from both P1 and P2 before applying a final postprocessing step. The online system implements Phase 1 by taking advantage of the Linux piping mechanism, multithreading techniques, and multi-core processors. To convert Phase 1 into an online system, we divide the system into five major modules: signal preprocessor, feature extractor, event decoder, postprocessor, and visualizer. The system reads 0.1-second frames from each EEG channel and sends them to the feature extractor and the visualizer. The feature extractor generates LFCC features in real time from the streaming EEG signal. Next, the system computes seizure and background probabilities using a channel-based LSTM model and applies a postprocessor to aggregate the detected events across channels. The system then displays the EEG signal and the decisions simultaneously using a visualization module. The online system uses C++, Python, TensorFlow, and PyQtGraph in its implementation. The online system accepts streamed EEG data sampled at 250 Hz as input. The system begins processing the EEG signal by applying a TCP montage [8]. Depending on the type of the montage, the EEG signal can have either 22 or 20 channels. To enable the online operation, we send 0.1-second (25 samples) length frames from each channel of the streamed EEG signal to the feature extractor and the visualizer. Feature extraction is performed sequentially on each channel. The signal preprocessor writes the sample frames into two streams to facilitate these modules. In the first stream, the feature extractor receives the signals using stdin. In parallel, as a second stream, the visualizer shares a user-defined file with the signal preprocessor. This user-defined file holds raw signal information as a buffer for the visualizer. The signal preprocessor writes into the file while the visualizer reads from it. Reading and writing into the same file poses a challenge. The visualizer can start reading while the signal preprocessor is writing into it. To resolve this issue, we utilize a file locking mechanism in the signal preprocessor and visualizer. Each of the processes temporarily locks the file, performs its operation, releases the lock, and tries to obtain the lock after a waiting period. The file locking mechanism ensures that only one process can access the file by prohibiting other processes from reading or writing while one process is modifying the file [9]. The feature extractor uses circular buffers to save 0.3 seconds or 75 samples from each channel for extracting 0.2-second or 50-sample long center-aligned windows. The module generates 8 absolute LFCC features where the zeroth cepstral coefficient is replaced by a temporal domain energy term. For extracting the rest of the features, three pipelines are used. The differential energy feature is calculated in a 0.9-second absolute feature window with a frame size of 0.1 seconds. The difference between the maximum and minimum temporal energy terms is calculated in this range. Then, the first derivative or the delta features are calculated using another 0.9-second window. Finally, the second derivative or delta-delta features are calculated using a 0.3-second window [6]. The differential energy for the delta-delta features is not included. In total, we extract 26 features from the raw sample windows which add 1.1 seconds of delay to the system. We used the Temple University Hospital Seizure Database (TUSZ) v1.2.1 for developing the online system [10]. The statistics for this dataset are shown in Table 1. A channel-based LSTM model was trained using the features derived from the train set using the online feature extractor module. A window-based normalization technique was applied to those features. In the offline model, we scale features by normalizing using the maximum absolute value of a channel [11] before applying a sliding window approach. Since the online system has access to a limited amount of data, we normalize based on the observed window. The model uses the feature vectors with a frame size of 1 second and a window size of 7 seconds. We evaluated the model using the offline P1 postprocessor to determine the efficacy of the delayed features and the window-based normalization technique. As shown by the results of experiments 1 and 4 in Table 2, these changes give us a comparable performance to the offline model. The online event decoder module utilizes this trained model for computing probabilities for the seizure and background classes. These posteriors are then postprocessed to remove spurious detections. The online postprocessor receives and saves 8 seconds of class posteriors in a buffer for further processing. It applies multiple heuristic filters (e.g., probability threshold) to make an overall decision by combining events across the channels. These filters evaluate the average confidence, the duration of a seizure, and the channels where the seizures were observed. The postprocessor delivers the label and confidence to the visualizer. The visualizer starts to display the signal as soon as it gets access to the signal file, as shown in Figure 1 using the “Signal File” and “Visualizer” blocks. Once the visualizer receives the label and confidence for the latest epoch from the postprocessor, it overlays the decision and color codes that epoch. The visualizer uses red for seizure with the label SEIZ and green for the background class with the label BCKG. Once the streaming finishes, the system saves three files: a signal file in which the sample frames are saved in the order they were streamed, a time segmented event (TSE) file with the overall decisions and confidences, and a hypotheses (HYP) file that saves the label and confidence for each epoch. The user can plot the signal and decisions using the signal and HYP files with only the visualizer by enabling appropriate options. For comparing the performance of different stages of development, we used the test set of TUSZ v1.2.1 database. It contains 1015 EEG records of varying duration. The any-overlap performance [12] of the overall system shown in Figure 2 is 40.29% sensitivity with 5.77 FAs per 24 hours. For comparison, the previous state-of-the-art model developed on this database performed at 30.71% sensitivity with 6.77 FAs per 24 hours [3]. The individual performances of the deep learning phases are as follows: Phase 1’s (P1) performance is 39.46% sensitivity and 11.62 FAs per 24 hours, and Phase 2 detects seizures with 41.16% sensitivity and 11.69 FAs per 24 hours. We trained an LSTM model with the delayed features and the window-based normalization technique for developing the online system. Using the offline decoder and postprocessor, the model performed at 36.23% sensitivity with 9.52 FAs per 24 hours. The trained model was then evaluated with the online modules. The current performance of the overall online system is 45.80% sensitivity with 28.14 FAs per 24 hours. Table 2 summarizes the performances of these systems. The performance of the online system deviates from the offline P1 model because the online postprocessor fails to combine the events as the seizure probability fluctuates during an event. The modules in the online system add a total of 11.1 seconds of delay for processing each second of the data, as shown in Figure 3. In practice, we also count the time for loading the model and starting the visualizer block. When we consider these facts, the system consumes 15 seconds to display the first hypothesis. The system detects seizure onsets with an average latency of 15 seconds. Implementing an automatic seizure detection model in real time is not trivial. We used a variety of techniques such as the file locking mechanism, multithreading, circular buffers, real-time event decoding, and signal-decision plotting to realize the system. A video demonstrating the system is available at: https://www.isip.piconepress.com/projects/nsf_pfi_tt/resources/videos/realtime_eeg_analysis/v2.5.1/video_2.5.1.mp4. The final conference submission will include a more detailed analysis of the online performance of each module. ACKNOWLEDGMENTS Research reported in this publication was most recently supported by the National Science Foundation Partnership for Innovation award number IIP-1827565 and the Pennsylvania Commonwealth Universal Research Enhancement Program (PA CURE). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the official views of any of these organizations. REFERENCES [1] A. Craik, Y. He, and J. L. Contreras-Vidal, “Deep learning for electroencephalogram (EEG) classification tasks: a review,” J. Neural Eng., vol. 16, no. 3, p. 031001, 2019. https://doi.org/10.1088/1741-2552/ab0ab5. [2] A. C. Bridi, T. Q. Louro, and R. C. L. Da Silva, “Clinical Alarms in intensive care: implications of alarm fatigue for the safety of patients,” Rev. Lat. Am. Enfermagem, vol. 22, no. 6, p. 1034, 2014. https://doi.org/10.1590/0104-1169.3488.2513. [3] M. Golmohammadi, V. Shah, I. Obeid, and J. Picone, “Deep Learning Approaches for Automatic Seizure Detection from Scalp Electroencephalograms,” in Signal Processing in Medicine and Biology: Emerging Trends in Research and Applications, 1st ed., I. Obeid, I. Selesnick, and J. Picone, Eds. New York, New York, USA: Springer, 2020, pp. 233–274. https://doi.org/10.1007/978-3-030-36844-9_8. [4] “CFM Olympic Brainz Monitor.” [Online]. Available: https://newborncare.natus.com/products-services/newborn-care-products/newborn-brain-injury/cfm-olympic-brainz-monitor. [Accessed: 17-Jul-2020]. [5] M. L. Scheuer, S. B. Wilson, A. Antony, G. Ghearing, A. Urban, and A. I. Bagic, “Seizure Detection: Interreader Agreement and Detection Algorithm Assessments Using a Large Dataset,” J. Clin. Neurophysiol., 2020. https://doi.org/10.1097/WNP.0000000000000709. [6] A. Harati, M. Golmohammadi, S. Lopez, I. Obeid, and J. Picone, “Improved EEG Event Classification Using Differential Energy,” in Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium, 2015, pp. 1–4. https://doi.org/10.1109/SPMB.2015.7405421. [7] V. Shah, C. Campbell, I. Obeid, and J. Picone, “Improved Spatio-Temporal Modeling in Automated Seizure Detection using Channel-Dependent Posteriors,” Neurocomputing, 2021. [8] W. Tatum, A. Husain, S. Benbadis, and P. Kaplan, Handbook of EEG Interpretation. New York City, New York, USA: Demos Medical Publishing, 2007. [9] D. P. Bovet and C. Marco, Understanding the Linux Kernel, 3rd ed. O’Reilly Media, Inc., 2005. https://www.oreilly.com/library/view/understanding-the-linux/0596005652/. [10] V. Shah et al., “The Temple University Hospital Seizure Detection Corpus,” Front. Neuroinform., vol. 12, pp. 1–6, 2018. https://doi.org/10.3389/fninf.2018.00083. [11] F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011. https://dl.acm.org/doi/10.5555/1953048.2078195. [12] J. Gotman, D. Flanagan, J. Zhang, and B. Rosenblatt, “Automatic seizure detection in the newborn: Methods and initial evaluation,” Electroencephalogr. Clin. Neurophysiol., vol. 103, no. 3, pp. 356–362, 1997. https://doi.org/10.1016/S0013-4694(97)00003-9. 
    more » « less
  3. null (Ed.)
    International Ocean Discovery Program Expedition 368 is the second of two consecutive cruises that form the South China Sea Rifted Margin program. Expeditions 367 and 368 share the common key objectives of testing scientific hypotheses of breakup of the northern South China Sea (SCS) margin and comparing its rifting style and history to other nonvolcanic or magma-poor rifted margins. Four primary sites were selected for the overall program: one in the outer margin high (OMH) and three seaward of the OMH on distinct, margin-parallel basement ridges. These three ridges are informally labeled A, B, and C. They are located within the continent–ocean transition (COT) zone ranging from the OMH to the interpreted steady-state oceanic crust (Ridge C) of the SCS. The main scientific objectives include 1. Determining the nature of the basement within crustal units across the COT of the SCS that are critical to constrain style of rifting, 2. Constraining the time interval from initial crustal extension and plate rupture to the initial generation of igneous ocean crust, 3. Constraining vertical crustal movements during breakup, and 4. Examining the nature of igneous activity from rifting to seafloor spreading. In addition, the sediment cores from the drill sites targeting primarily tectonic and basement objectives will provide information on the Cenozoic regional environmental development of the Southeast Asia margin. Expedition 368 was planned to drill at two primary sites (U1501 and U1503) at the OMH and Ridge C, respectively. However, based on drilling results from Expedition 367, Expedition 368 chose to insert an alternate site on Ridge A (Site U1502). In total, the expedition completed operations at four sites (U1501, U1502, U1504, and U1505). Site U1503, however, was not completed beyond casing to 990 m because of mechanical problems with the drilling equipment that limited the expedition from 25 May 2017 to the end of the expedition to operate with a drill string not longer than 3400 m. New alternate Site U1504 proposed during Expedition 367 met this condition. Site U1505 also met the operational constraints of the 3400 m drill string (total) and was an alternate site for the already drilled Site U1501. At Site U1501, we cored to 697.1 m in 9.4 days, with 78.5% recovery. We also drilled ahead for 433.5 m in Hole U1501D and then logged downhole data from 78.3 to 399.3 m. In 19.3 days at Site U1502, we penetrated 1679.0 m, set 723.7 m of casing and cored a total of 576.3 m with 53.5% recovery, and collected downhole log data from 785.3 to 875.3 m and seismic data through the 10¾ inch casing. At Site U1503, we penetrated 995.1 m, setting 991.5 m of 10¾ inch casing, but no cores were taken. At Site U1504, we took 40 rotary core barrel (RCB) cores over two holes. The cored interval between both holes was 277.3 m with 26.8% recovery. An 88.2 m interval was drilled in Hole U1504B. At Site U1505, we cored 668.0 m with 101.1% recovery. Logging data was collected from 80.1 to 341.2 m. Operations at this site covered 6.1 days. Except for Site U1505, we drilled to acoustic basement, which prior to the expedition, except for Site U1501, had been interpreted to be crystalline basement. A total of 6.65 days were lost due to mechanical breakdown or waiting on spare supplies for repair of drilling equipment. At Site U1501 on the OMH, coring ~45 m into the acoustic basement sampled highly lithified sandstone to conglomerate of presumed Mesozoic age overlain by siliciclastic Eocene pre- to synrift sediments of Oligocene age and topped by primarily carbonaceous postrift sediments of early Miocene to Pleistocene age. Site U1502 on Ridge A was cased to 723.7 m. At this site, we recovered 180 m of hydrothermally altered brecciated basalts comprising sheet and pillow lavas below deep-marine sediments of Oligocene to late Miocene age. Coring was not performed within the upper 380 m (~Pliocene–Pleistocene) at Site U1502. At Site U1503 on Ridge C, 991.5 m of casing was installed in preparation for the planned deep drilling to ~1800 m, but no coring was performed due to mechanical failures, and the site was abandoned without further activity. Coring at Site U1504 on the OMH ~45 km east of Site U1501 recovered metamorphic schist to gneiss (greenschist facies) below late Eocene (?) carbonate rocks (partly reef debris) and early Miocene to Pleistocene sediments. At Site U1505, we cored to 480.15 m through Pleistocene to late Oligocene mainly carbonaceous ooze followed at depth by early Oligocene to late Eocene siliciclastic sediments. Efforts were made at every drill site to correlate the core with the seismic data and seismic stratigraphic unconformities interpreted within the Eocene to Plio–Pleistocene sedimentary sequence prior to drilling. The predrilling interpretation of ages of these unconformities was in general confirmed by drilling results. As a result of the constraints on the length of drill string that could be deployed during the later part of Expedition 368, the secondary expedition objectives addressing the environmental history of the SCS and Southeast Asia received more focus than planned because these sites are located in shallower water depths and required less penetration depth. This forced change in emphasis, however, was without fatal consequences for the primary tectonic objectives. The two expeditions together provided solid evidence for a process of breakup that included vigorous synrift magmatism as opposed to the often-favored interpretation of the SCS margin as a magma-starved margin. 
    more » « less
  4. Marine protected areas (MPAs) are a key tool for achieving goals for biodiversity conservation and human well-being, including improving climate resilience and equitable access to nature. At a national level, they are central components in the U.S. commitment to conserve at least 30% of U.S. waters by 2030. By definition, the primary goal of an MPA is the long-term conservation of nature; however, not all MPAs provide the same ecological and social benefits. A U.S. system of MPAs that is equitable, well-managed, representative and connected, and includes areas at a level of protection that can deliver desired outcomes is best positioned to support national goals. We used a new MPA framework, The MPA Guide, to assess the level of protection and stage of establishment of the 50 largest U.S. MPAs, which make up 99.7% of the total U.S. MPA area (3.19 million km2). Over 96% of this area, including 99% of that which is fully or highly protected against extractive or destructive human activities, is in the central Pacific ocean. Total MPA area in other regions is sparse – only 1.9% of the U.S. ocean excluding the central Pacific is protected in any kind of MPA (120,976 km2). Over three quarters of the non-central Pacific MPA area is lightly or minimally protected against extractive or destructive human activities. These results highlight an urgent need to improve the quality, quantity, and representativeness of MPA protection in U.S. waters to bring benefits to human and marine communities. We identify and review the state of the science, including focal areas for achieving desired MPA outcomes and lessons learned from places where sound ecological and social design principles come together in MPAs that are set up to achieve national goals for equity, climate resilience, and biodiversity conservation. We recommend key opportunities for action specific to the U.S. context, including increasing funding, research, equity, and protection level for new and existing U.S. MPAs.

     
    more » « less
  5. International Ocean Discovery Program (IODP) Expedition 357 successfully cored an east–west transect across the southern wall of Atlantis Massif on the western flank of the Mid-Atlantic Ridge (MAR) to study the links between serpentinization processes and microbial activity in the shallow subsurface of highly altered ultramafic and mafic sequences that have been uplifted to the seafloor along a major detachment fault zone. The primary goals of this expedition were to (1) examine the role of serpentinization in driving hydrothermal systems, sustaining microbial communities, and sequestering carbon; (2) characterize the tectonomagmatic processes that lead to lithospheric heterogeneities and detachment faulting; and (3) assess how abiotic and biotic processes change with variations in rock type and progressive exposure on the seafloor. To accomplish these objectives, we developed a coring and sampling strategy centered on the use of seabed drills—the first time that such systems have been used in the scientific ocean drilling programs. This technology was chosen in the hope of achieving high recovery of the carbonate cap sequences and intact contact and deformation relationships. The expedition plans also included several engineering developments to assess geochemical parameters during drilling; sample bottom water before, during, and after drilling; supply synthetic tracers during drilling for contamination assessment; acquire in situ electrical resistivity and magnetic susceptibility measurements for assessing fractures, fluid flow, and extent of serpentinization; and seal boreholes to provide opportunities for future experiments. Expedition 359 was designed to address changes in sea level and currents, along with monsoon evolution in the Indian Ocean. The Maldives archipelago holds a unique and mostly unread Indian Ocean archive of the evolving Cenozoic icehouse world. Cores from eight drill sites in the Inner Sea of the Maldives provide the tropical marine record that is key for better understanding the effects of this global evolution in the Indo-Pacific realm. In addition, the bank geometries of the carbonate archipelago provide a physical record of changing sea level and ocean currents. The bank growth occurs in pulses of aggradation and progradation that are controlled by sea level fluctuations during the early and middle Miocene, including the mid-Miocene Climate Optimum. A dramatic shift in development of the carbonate edifice from a sea level–controlled to a predominantly current-controlled system appears to be directly linked to the evolving Indian monsoon. This phase led to a twofold configuration of bank development: bank growth continued in some parts of the edifice, whereas in other places, banks drowned. Drowning steps seem to coincide with onset and intensification of the monsoon-related current system and subsequent deposition of contourite fans and large-scale sediment drifts. As such, the drift deposits will provide a continuous record of Indian monsoon development in the region of the Maldives. A major focus of Expedition 359 was to date precisely the onset of the current system. This goal was successfully completed during the expedition. The second important outcome of Expedition 359 was groundtruthing the hypothesis that the dramatic, pronounced change in style of the carbonate platform sequence stacking was caused by a combination of relative sea level fluctuations and ocean current system changes. These questions are directly addressed by the shipboard scientific data. In addition, Expedition 359 cores will provide a complete Neogene δ13C record of the platform and platform margin sediments and a comparison with pelagic records over the same time period. This comparison will allow assessment of the extent to which platform carbonates record changes in the global carbon cycle and whether changes in the carbon isotopic composition of organic and inorganic components covary and the implications this has on the deep-time record. This determination is important because such records are the only type that exists in deep time. 
    more » « less