Recently I worked together with longtime professional friend Kars Alfrink on a small side project. Now that we’ve finished it, it is time to show what we have done.
The brief was by a friend of a friend for Sarah van Sonsbeeck, artist in residence at the Amsterdam Rijksakademie. Sarah does a lot of work in making sound —its presence and its absence— visible, tangible and experience-able in unexpected ways.
For a recent project she was going to walk through part of Monickendam with some people while both discussing their experience of the sounds and also simultaneously measuring sound levels and recording GPS coordinates. For the sake of simplicity we ignored measurement error and interference. This was not an attempt to make a scientifically accurate measurement of the dB levels (though such attempts do exist).
We would then take the data generated from these walks and plot them into a computer representation of the area and generate visualisations from that. Building an audiogeography superimposed on the physical landscape with the sound levels as experienced by somebody who would walk through the area.
A small project, limited in scope and budget, we took the challenge to see what we could make.
CabSpotting was the original very conceptual brief for this project where the streets are defined with data from a derivative activity providing a skeleton landscape similar to the physical world. Adding intensity to this mix would naturally lead to something like a physical intensity (= height) map of a landscape:
We didn’t look very much into the treatment of sound and its perception in architecture, but previous work by the artist and brief inquiries with friends would suggest that it is a niche and often ignored part of that field.
The first task at hand was to find a dB value for the entire area based on the samples given to us by Sarah. She had made some circumnavigations and cross sections of the area, but she (of course) hadn’t been everywhere.
After some tries we found out that this is a non-trivial problem which is not solved well by a naive approach. It is a common problem and its solution is similar to a simple sum of squares linear regression of a set of data points in two dimensions.
The best way (according to GIS people) to do this is called Kriging and it’s implemented in proprietary GIS packages. The open source implementations in R and C are quite hard to use and figure out in terms of input and output values.
We found a post on Stack Overflow with a fairly thorough discussion of this problem and a set of approaches which led us finally to an Inverse Distance Weighted interpolation of the data points which was fairly straight forward to implement and performant enough.
After the interpolation we would have a matrix of arbitrary detail with a smooth set of values for the entire measurement area. It is easy to color these values based on a scale to show ‘height’. Transforming these values into a 3D mesh as a landscape based representation turned out to be somewhat more complex. Kars took most of this work upon himself emulating the properties of a seasoned 3D programmer.
Processing turned out to have a library called unlekkerlib which can be used to create and output 3D models based on generated data for visual display and output into formats suitable for re-use in other 3D modelling packages or to put straight to a fablab.
Some of the work Watz himself has made with unlekker are in this Flickr set.
The end results were 2D and 3D versions of the silence landscape both from straight interpolation with some filters added to it to make it resemble a heightmap:
And a high quality raytrace of the same landscape:
The 3D and 2D models can also be fablabbed into various forms which was one of our recommendations. By changing the medium of the data and therefore the modality of the experience alternative insight may follow. We are very curious how that is going to turn out and how this will add tangibility to the data.
This was a fun project to do because of its limited scope and free form. It gave Kars and myself a good reason to work together (which worked out quite well). We took it far and wide to see what we could create using our skills and ready-to-use tools and we think the end results are pleasing.
One of our limitations is that both of us are not graphic designers. This means we may not be the best fit for a mostly aesthetically driven visualization project. Given our Tuftean sensibilities, we like to let the data speak for itself with integrity and without overmuch visual embellishment1.
Our focus as interaction designers and our skills in programming the Processing visual environment, would direct us towards more dynamic and interactive visualizations where our iterative interaction with the data builds up layers of meaning and we add controls and affordances to the final product so a user can interact with the data themselves and find their own meaning.
Collaborating with an artist is very interesting because the story and the context of her art imbue your work directly with more depth. Aligning technical possibilities and visual output with an artistic vision is however difficult and probably served better by more attunement and iteration than we could invest in this project.
- We can’t nay won’t do the bubblegum variety of infovis where visual elements are added for their own sake. ↩