Experts Share How DeepSee Boosted Research Efficiency During Ocean Expeditions | HackerNoon

Table of Links

Abstract and 1 Introduction

2 Related Work

3 Methodology

4 Studying Deep Ocean Ecosystem and 4.1 Deep Ocean Research Goals

4.2 Workflow and Data

4.3 Design Challenges and User Tasks

5 The DeepSea System

• 5.1 Map View
• 5.2 Core View

5.3 Interpolation View and 5.4 Implementation

6 Usage Scenarios and 6.1 Scenario: Pre-Cruise Planning

• 6.2 Scenario: On-the-Fly Decision-Making

7 Evaluation and 7.1 Cruise Deployment

7.2 Expert Interviews

7.3 Limitations

7.4 Lessons Learned

8 Conclusions and Future Work, Acknowledgments, and References

7.2 Expert Interviews

After the cruise, we conducted individual semi-structured interviews (1 − 3 hours long each) with five domain experts in deep ocean research (P1 − 5). These experts were the same collaborators involved in the design process and carried out the initial deployment [40]. Very few scientists have access to the deep ocean sample data we utilized, limiting the number of potential candidates able to properly evaluate DeepSee. However, in line with prior design studies [15, 18, 33], it is common for visualizations that tackle very specific problems to be validated by only a few users.

Our five collaborators were a team of graduate students, postdocs, and research scientists broadly interested in fieldwork at the intersection of geology, geochemistry, and biology in the deep ocean. The multidisciplinary team has worked together over several years on logistics around planning and executing cruises, collecting and organizing prior and current cruise data, and analyzing samples obtained during cruises. Two collaborators self-identified as female and three self-identified as male. We organized the interview questions and domain experts’ feedback around three broad questions we directly asked each collaborator. We then conducted inductive thematic analysis [7] of the feedback to identify emergent themes that were discussed amongst all authors.

(1) Research – “How did using DeepSee support your fieldwork research goals?”

(2) Workflow – “How did using DeepSee improve the science return of samples?”

(3) Teamwork – “How did using DeepSee help you with your tasks in your role on the team?”

(1) In pre-cruise planning, fluid interaction and aggregation between multiple types and scales of data collected long-term can help researchers visually discover more insights.

We first explored how visualizations supported fieldwork-driven research. In pre-cruise planning, several capabilities should be prioritized in system design, including data aggregation across multiple disparate sources, comparing multiple data types, fluidly interacting between data scales [16], and keeping humans in the analysis loop through data simulations. These capabilities can enable serendipitous discoveries from analyzing data collected over a long period of time. When asked who in the research community could benefit the most from DeepSee, P5 responded: “It’s really fieldwork driven. If someone has been cataloging a field site for a decade, like a marine site, I would absolutely recommend it.”

To support these claims, our collaborators provided several examples of visualizations playing a pivotal role in supporting their research goals. The team found directly comparing the Map View with the Core View or Interpolation View helped them discover “holes” in their sampling. For P4, this rich history of sampling helped them familiarize themselves with a new hydrothermal vent site: “‘Oh, these were hot cores, so this must be a hot area’… I can identify areas to sample before I [return to a field site].” These features introduced potential for making serendipitous discoveries in the future, as P4 described while demonstrating how they would use DeepSee: “I didn’t realize Sulfide was so high in this area! Why is that? …Well, it looks like based on ANME distributions [in the Core View] …these taxa are in this area and very abundant!” As a microbial ecology lab, comparing interactions between geochemical and biological phenomena at multiple scales is also vital. P3 specifically felt that DeepSee fit in a role that “enabled spatial thinking for the middle range of analysis” between the microscopic and global level. P2 described the value of DeepSee in supporting human-in-the-loop sampling scenarios: “Many times we can’t be [at a site] or make plans in time. Having a tool like this helps you come up with a plan backed by data and communicate these plans. We can overcome language barriers. Almost like a virtual reality simulation!”

(2) When planning fieldwork expeditions, integrating and visualizing 2D/3D data together can increase the science return on limited samples.

In supporting the team’s research goals, we also found that DeepSee contributed to fieldwork research workflows in several unique ways. We found that visualization of complex data during pre-cruise planning was the most useful to the team. To support this, the ability to visualize data in a single interface helped experts orient themselves to the field site, perform gradient-based analysis across the area, and seamlessly explore trends between 2D and 3D. This ultimately increased the long-term value or science return on using sample data to drive fieldwork research workflows in the future. Beyond the visualizations, P4 also found that preparing data for testing in DeepSee catalyzed useful data work for the long-term, such as “formatting and standardizing data collection, and planning on what columns we need.”

P5 summarized an example DeepSee workflow used in the lab, highlighting how questions involved overlaying multiple data types: “We first find a site that we’re interested in, a vent for example, then zoom in to those samples and select them. Then, we’ll go to the Core View and look at the geochemical and microbial taxa values for the cores. We’ll filter out different core types… then look at the bar charts [in the Core View]. For example, we might see levels that are higher than normal of ANME, so we switch to a seafloor photomosaic map [in the Map View] and maybe we don’t see any visible surface expression. This then leads us to ask what the geochemistry is in this area, so we switch to Sulfide [in the Core View]… Time is precious, so this [ability to toggle between data] helps with deciding what data we want to collect from this site again and which we might not need.” Comparing DeepSee with existing map tools, P1 reported being more comfortable and efficient with ArcGIS, R and Python, while P5 reported they preferred using the Map View to “quickly orient myself to where samples are and to test the area over where we wanted to interpolate.” P4, in charge of data preparation, summarized both the startup cost of DeepSee and the payoff for their team in the long run around catalyzing data aggregation: “There was initially a higher amount of work to consolidate the data… Once we were past that main push, it was easier to use DeepSee… Now we can see patterns in the old data that were hidden in an Excel sheet. If I have a hypothesis and need to show the data to test that hypothesis, DeepSee makes it possible in just a few clicks.” Analogously, P5 described a new use case for DeepSee “as a repository of information… new researchers that come into the lab can look at [previous] data we’ve collected from a site right away.”

(3) Simple, intuitive and modular visualizations can help different fieldwork team members rapidly solve a diversity of specific, directed research tasks.

Finally, because we designed for a single team with a rich diversity of roles and responsibilities, we investigated how DeepSee impacted communication and teamwork. We found that the team dynamic can be characterized by many diverse goals with limited time and resources, where individual wants and needs illuminate unique use cases that make modular tools like DeepSee valuable. Having multiple visualizations with simple and intuitive controls helped our collaborators address specific tasks, such as filtering and selecting specific sites of interest, performing predictive analysis of potential data trends for deciding where to sample, and learning a single software instead of multiple tools.

While one of the lab’s main research goals is to study deep ocean microbial ecology, every team member has unique questions they want to answer. For example, P2 told us that “DeepSee lets me preselect sample sites in a meaningful way. I can design samples I would like based on the data we’ve gotten back.” P2 further found value in the predictive capabilities of the Interpolation View: “If we think a gradient exists in a certain area, we can fill in the data and see that… DeepSee can maximize time by really targeting questions about exploration to hone in on specific data when there is a history of data.” For P3, a key improvement was in creating a single interface for visualizing data, “reducing the need to learn additional software such as ARCGIS.” During cruises, rapidly changing environments and unyielding time constraints can create high-pressure situations. P5 described a particular cruise environment prior to the initial deployment of DeepSee: “When we went to Pescadero years ago, the site was dynamic. The coordinates from the ROV are not very accurate. There’s a lot of stress aligning previous bathymetry maps to the current location. You’re looking at previous log books and looking at screens in the control room and talking to the pilots of the ROV.” Multiple team members agreed that simplicity and familiarity with DeepSee were critical in such situations.