social lives of common bottlenose dolphins in new zealand


Lab Reports, Networks
Common bottlenose dolphins. pixels.com

Introduction

In the new Netflix documentary, Seaspiracy, investigative journalist Ali Tabrizi sets out to research plastic pollution in the ocean and soon finds himself on a global journey discovering complex layers of socio political problems that are causing the slow death of the world’s oceans. The film concludes that the biggest pollution and animal killing problem is large scale commercial fishing, illegal fishing, bycatch and vague definitions of “sustainable.” Along the way, Tabrizi discusses how humans relate to marine animals and whether or not these animals have feelings and awareness. Dolphins are famous for their playful personalities and social interactions and it has long been thought that these social displays mean that they feel empathy for one another. Many scientists have ventured to prove this trait through visualized data of complex and tight knit social groups. The visualizations below seek to further tell the story of dolphins as social and empathetic animals. If humans can feel more empathy for marine animals, perhaps more can be accomplished to end illegal and overfishing.

Inspiration

The following images are a few examples of appealing visuals, excluding data details. The image on the left is from the Interaction Design Foundation, the use of a single color in different values and sizes keeps the information clear while still a wild-looking visualization. The black background network is from Analytics India Magazine, I don’t know what information it displays but the bright colors on a dark background are strikingly beautiful.

The journal article that provided the dataset that I used contained a few visualizations, one that is similar to a network, below left. The visualization on the right is from the Network Repository, which contains hundreds of network datasets of animals and other topics. This visualization is unclear and not visually attractive, but it provided me with an idea of what has been done with similar data.

Data

The primary software used for this project was the open graph visualization platform Gephi, gephi.org. Gephi provides a selection of datasets on their github, which includes links to other data collections such as Mark Newman’s personal compilation of network data. Newman is the guardian angel of network data. The data that he provides are already in GML format, graph modeling language, which is the most user-friendly format to be used in Gephi. Newman’s datasets are also open access, shared by the original authors or created by Newman himself.

Doubtful Sound, New Zealand. Google maps

The Dolphin social network dataset that I used was provided by David Lusseau and contains an, “undirected social network of frequent associations between 62 dolphins in a community living off Doubtful Sound, New Zealand.” I found the paper that originally published this data, The bottlenose dolphin community of Doubtful Sound features a large proportion of long-lasting associations – Can geographic isolation explain this unique trait? by D Lusseau, K Schneider, O J Boisseau, P Haase, E Slooten, S M Dawson published by Aberdeen Centre For Environmental Sustainability. The network data is simple, 62 nodes and only 159 edges. The data was collected over the course of 7 years, 1995 to 2001 excluding 1998. To make the data more meaningful I read the journal article mentioned and conducted brief research of similar dolphin social interaction studies. This supplemental knowledge aided my network visualization work.

Process

This project involved a lot of trial and error. The network relationship already in the dataset represents the association patterns of the dolphins, who they spent time with. It was listed as modularity class with 0-3, the 4 main social groups. The degree is the number of connections a single dolphin had, the most was 12. Again, this dataset is quite simple so I highlighted more variables to make it more meaningful. With information from the journal article, I created a new column on the data laboratory page and gave each dolphin a classification of F-female, M-male, or N-not listed. I have about 20 different final networks saved, all different color and size combinations. I tried to keep it simple but still show enough different information with color only being used to show sex, and size used to show degree. I first ran the data in Fruchterman Reingold then settled on Force Atlas 2.

  • Image 1
  • Image 2
  • Image 3
  • Image 4
  • Image 5
  • Image 6

In image 1, color represents social organization and node size represents degree range. I increased the degree to 5:60 to fill more space, intentionally keeping it a 1:12 ratio to reflect the actual degree range. Labels of the dolphin’s names are the same font size to avoid a mess. I then adjusted scale and gravity settings (image 2), which caused the entire network to rotate; I also changed the background to a dark color, increased the node size, and increased opacity to 60. Image 2 is the first use of the added sex variable; green is male, pink is female, and purple is unlisted. Image 3 displays the labels scaled to degree range and filtered out lesser degree ranges. I can’t decide if this visualization is more or less clear, the labels are too jumbled but highlight dolphins with higher association patterns. I set the node colors back to social organization (image 4) and played around with hues, opacity, and background color. For image 5, the labels are colored to sex and scaled to social organization. The nodes are shades of gray to no interfere with the label colors. With image 6, I made the labels uniform and colored according to sex. The nodes are colored to social organization and sized to degree ratio. This is too jumbled, but I liked the idea of coloring the labels for male, female, or unlisted, but there are too many colors and things being measured.

In Gephi, it would be helpful to see a live updated of the final visual with the background color side by side with the preview to choose colors more effectively.

These final two visualizations use color for one variable and size for only one variable. In both, color represents sex; blue is male, purple is female, and gray is unlisted. On the left, node size represents degree range and there is no distinction for social organization but it is somewhat possible to tell them apart anyway. On the right, node size represents social organization and labels are displayed for the dolphins with the top four degree ranges, 9-12. I deleted label information from the data laboratory to highlight the top four but it would be helpful for Gephi to have another way to filter labels without permanently editing the data. There is always more manipulation that can be done, but this final network visualization best displays the information.

Results

Grin, a female dolphin, had the most connections of all the dolphins and may be the leader of the third social pod. Group 3 is mostly female and group 2 is mostly male. Groups 0 and 1 are also mostly male but are closely connected to group 3. Perhaps Grin is the wise old grandmother who many gravitate to and group 2 is the young bachelor pod doing their own thing. The clear social organization over seven years show that dolphins are social animals and form bonds with one another and retain those bonds over time.

Reflection

After sharing my results with a colleague for a peer review, I edited the last two visualizations. She preferred the white backgrounds and simple presentation without all labels. This feedback encouraged be to continue pursuit of highlighting only a few labels.

This project was difficult and I didn’t even clean or manipulate my dataset. Gephi was mildly user friendly, but the data selection process is what took most of my time and sanity. Initially, I wanted to create a visualization with geographic coordinates and impose the image on top of a map, but I hit many walls in my data research and decided to do the best I could with a small dataset. I also began at least five other network visualizations with far bigger datasets, but apart from basic color and size changes I wasn’t able to show anything meaningful. The power grid data almost worked but the huge dataset didn’t show in the preview no matter how many times I pushed refresh. There were other GML datasets provided, but I felt compelled to choose a topic that interests me and that relates to previous work. With guidance and practice, I would like to pursue a network visualization of environmental data displayed on a map for my final project.

Resources

Related posts:

  1. Les Misérables: Network of Characters
  2. Marvel Superhero Social Network: A Gephi Visualization
  3. Planting Companion?