Modelling potential environmental impacts of science activity in Antarctica

David O’Sullivan

Geospatial Stuff

University of Auckland

Fraser Morgan

Bioeconomy Science Institute

Overview

Study area and data

Hiking networks

Some outputs

Conclusions

Study area and data

GPS data

Collected using QStarz Q1100P GPS Tracking Recorders between 2016 and 2018

40+ hour battery life and simple use

Location recorded every 30 seconds

Scientists turned devices on when leaving base camp in the morning and off on return at the end of the day

The Dry Valleys

Bing maps aerial imagery

And on the ground

Yup! There’s streetview imagery

Hiking networks

Building a hiking network

Start with regular lattice across study area

Many topologies could be used

We went with hexagonal at ~107m spacing

Up \(\neq\) down

Slope is critical to movement, so attach heights to nodes and calculate slope in each direction along lattice edges

Then use a hiking function to estimate traversal times

Hiking functions

Smoothing GPS data

Curve fitting to cleaned data

Terrain-dependent vs combined

Completing and applying the network

Estimate edge traversal times based on slope, hiking function, and land cover (moraine or rock)

Make nodes and edges into a directed graph

Use various graph algorithms to find, e.g., everywhere-to-everywhere shortest paths

Some outputs

Betweenness centrality

After some exploration, we landed on betweenness centrality as a useful metric

Counts how often each node appears on shortest paths between every other pair of nodes

→ Indicator of relative likelihood of locations being visited

Radius-limited betweenness

Restrict betweenness centrality to nodes no more than some cost (i.e., time) apart

Much faster to calculate

Also… it may have more value: perhaps relevant to how people navigate in such environments

Planning a path network?

A way to reduce impact might be to plan paths

Experimental at this stage

Based on a minimum spanning tree approximation to an arborescence

Conclusions

Image: Beacon Valley, from reddit.com/rHumanForScale

Modelling potential environmental impacts of science activity in Antarctica

Overview

Study area and data

Hiking networks

Some outputs

Conclusions

Study area and data

GPS data

The Dry Valleys

And on the ground

Hiking networks

Building a hiking network

Start with regular lattice across study area

Many topologies could be used

We went with hexagonal at ~107m spacing

Up \(\neq\) down

Slope is critical to movement, so attach heights to nodes and calculate slope in each direction along lattice edges

Then use a hiking function to estimate traversal times

Hiking functions

Smoothing GPS data

Curve fitting to cleaned data

Terrain-dependent vs combined

Completing and applying the network

Estimate edge traversal times based on slope, hiking function, and land cover (moraine or rock)

Make nodes and edges into a directed graph

Use various graph algorithms to find, e.g., everywhere-to-everywhere shortest paths

Some outputs

Betweenness centrality

After some exploration, we landed on betweenness centrality as a useful metric

Counts how often each node appears on shortest paths between every other pair of nodes

→ Indicator of relative likelihood of locations being visited

Radius-limited betweenness

Restrict betweenness centrality to nodes no more than some cost (i.e., time) apart

Much faster to calculate

Also… it may have more value: perhaps relevant to how people navigate in such environments

Planning a path network?

A way to reduce impact might be to plan paths

Experimental at this stage

Based on a minimum spanning tree approximation to an arborescence

Conclusions

Hiking network concept is implicit in hiking functions but a new terrain representation

Terrain-differentiated GPS data-fitted hiking functions are novel

Potential wider application of radius-limited betweenness centrality

Next up: start a conversation with Antarctic scientists about planning

Questions?

For more

On this project dosull.github.io/antarctica

On Geospatial Stuff dosull.github.io