Communicating spatial uncertainty to non-experts using R

Organised by Laboratory of Geo-information Science and Remote Sensing

Wed 23 March 2016 13:30 to 14:00

Venue Lumen, gebouwnummer 100
Room 1

By Damiano Luzzi (Switserland)


Effective visualisation methods are important for the efficient use of uncertainty information for various groups of users. Uncertainty propagation analysis is often used with spatial environmental models to quantify the uncertainty within the information. A challenge arises when trying to effectively communicate the uncertainty information to non-experts (not statisticians) in a wide range of cases.

Due to the growing popularity and applicability of the open source programming language R, this project aimed to develop R functions to effectively communicate spatial uncertainty to non-experts. The spatial uncertainty information was generated using Monte Carlo algorithms, the output of which is represented by an ensemble of model outputs (i.e. a sample from a probability distribution). Three visualisation methods, adjacent maps, glyphs and an interactive application, were chosen for implementation. To provide the most universal visualisation tools for non-experts, a survey was conducted on a group of 12 university students. This survey assessed the effectiveness of the selected methods for visualising uncertainty in spatial variables such as elevation and land cover. The adjacent maps and glyphs were used for continuous variables. Both allow for displaying maps with information about the ensemble mean and standard deviation. Adjacent maps were also used for categorical data, displaying maps of the most probable class, as well as its associated probability. The interactive applications included a graphical user interface, which in addition to displaying the previously mentioned variables also allowed for comparison of joint uncertainties at multiple locations. The survey indicated that users could understand the basics of the uncertainty information displayed using the three methods, with all three having an approximately equal preference. The implementation of the visualisations was done via calls to the ggplot2 package. This allowed the user to provide control over the content, legend, colours, axes and titles. The interactive methods were implemented using the shiny package allowing users to activate the visualisation of statistical descriptions of uncertainty through interaction with a plotted map.
This research brings uncertainty visualisation to a broader audience through the development of tools for visualising uncertainty using open source software.