Several data wrangling and analysis limitations were expressed during the interviews. A common issue discussed was prerequisite knowledge needed to format the data; e.g., P5 faced challenges “inputting DNA sequence information. Knowing how to input data and the centralization of that data is challenging. It’s often that one person ends up inputting data.” While combining multiple data types in a single interface was a design goal, P1 lamented a reduction in dimensionality: “Certain data types might have different data descriptions and having multiple columns and reshaping tables is a challenge… Do I want to try to fit existing data into a certain tool like DeepSee?” P1 also wanted taxonomic data stratified hierarchically, a feature common to many biological analysis systems: “There could be very rare individual units that are technically different taxa. This creates very sparse data where abundance is very small for most of the columns. We needed a way to pick and find different subsets of taxa based on those hierarchies.” Finally, P4 wanted to see “a photo or snapshot of what a core actually looked like. An in-situ photo would allow a person to see what the sea floor looked like, what other cores were around it.”

\ The Interpolation View provided a contextualized way to view the data, as most other similar programs visualize gradients in 2D without spatial context between the sampled cores. However, there was a lack of confidence in the accuracy of the interpolation methods, as natural neighbors and linear gradients are unlikely to reflect spatially heterogeneous in-situ environmental processes. Similarly, distance to the nearest sample as a proxy for uncertainty was not accurate enough to use for in-situ decision making. Still, P2 suggested potential applications of Value-Suppressing Uncertainty Palettes (VSUPs) [11] in the future as the team develops more accurate uncertainty quantification: “We can see where the data is not strong and potentially sample to strengthen our predictions.” Specifically, the researchers noted that the modular nature of DeepSee will make incorporating more sophisticated interpolation methods (e.g., a Gaussian process such as Kriging [36]) and uncertainty measures (e.g., variance in the posterior predictive distribution of the Gaussian process) in the future straightforward.

\ One deployment goal that was not validated during DeepSee’s maiden expedition was whether data could be input in real time to visualize, as many of the relevant geochemical and biological parameters could not be measured while on the cruise, requiring further analysis on land. P3 reflected on this challenge, saying: “We talked about using DeepSee on the fly during the research expedition to load data right away. So far, that’s been impractical. We don’t have geochem or bio data to put in DeepSee while at sea. We also haven’t had a cruise to test whether these challenges were due to personnel on board or based on limitations with the DeepSee interface. Everyone was overextended on the expedition and no one had time to do additional data entry on the computer.” However, the team felt this helped them understand the time and energy allotment needed to use DeepSee for real-time data integration and visualization. P3 further discussed how they envision DeepSee being used on future cruises: “In the future, we’ll have a full team of scientists, and we’ll also have a manned submersible taking cores… We’ll have time to load the day’s cores into DeepSee to help plan for the next day… The important objective is having more people to do the same number of tasks… I could see this becoming one of the evening tasks: take the day’s coordinates and load them into DeepSee.” We synthesize design guidance for supporting real-time decision making shipboard in the next section.

:::info This paper is available on arxiv under CC BY 4.0 DEED license.

:::

:::info Authors:

(1) Adam Coscia, Georgia Institute of Technology, Atlanta, Georgia, USA ([email protected]);

(2) Haley M. Sapers, Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, California, USA ([email protected]);

(3) Noah Deutsch, Harvard University Cambridge, Massachusetts, USA ([email protected]);

(4) Malika Khurana, The New York Times Company, New York, New York, USA ([email protected]);

(5) John S. Magyar, Division of Geological and Planetary Sciences, California Institute of Technology Pasadena, California, USA ([email protected]);

(6) Sergio A. Parra, Division of Geological and Planetary Sciences, California Institute of Technology Pasadena, California, USA ([email protected]);

(7) Daniel R. Utter, [email protected] Division of Geological and Planetary Sciences, California Institute of Technology Pasadena, California, USA ([email protected]);

(8) John S. Magyar, Division of Geological and Planetary Sciences, California Institute of Technology Pasadena, California, USA ([email protected]);

(9) David W. Caress, Monterey Bay Aquarium Research Institute, Moss Landing, California, USA ([email protected]);

(10) Eric J. Martin Jennifer B. Paduan Monterey Bay Aquarium Research Institute, Moss Landing, California, USA ([email protected]);

(11) Jennifer B. Paduan, Monterey Bay Aquarium Research Institute, Moss Landing, California, USA ([email protected]);

(12) Maggie Hendrie, ArtCenter College of Design, Pasadena, California, USA ([email protected]);

(13) Santiago Lombeyda, California Institute of Technology, Pasadena, California, USA ([email protected]);

(14) Hillary Mushkin, California Institute of Technology, Pasadena, California, USA ([email protected]);

(15) Alex Endert, Georgia Institute of Technology, Atlanta, Georgia, USA ([email protected]);

(16) Scott Davidoff, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California, USA ([email protected]);

(17) Victoria J. Orphan, Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, California, USA ([email protected]).

:::

Feed: Hacker Noon - Medium

View: Original article

Tags: applications distribution energy small technology