Friday, April 22, 2016

Consider Thiessen polygons as an alternative to heat maps

Recently a colleague asked me to make a heat map for her animal trapping results. I agreed, but then quickly realized that the task was more difficult than it first seemed. In this particular case the individual trapping stations were only 10 meters apart and the distance among trapping grids was 500 meters. It struck me as inappropriate to try to interpolate across such a long distance when the variability within a grid was so great. I wasn't really a fan of just displaying the points either. Somehow they just came across looking messy and imprecise. The solution that I came up with was Thiessen polygons. This is probably the simplest form of interpolation, yet it is extremely powerful visually. In the resulting maps you are still able to see the fine-scale detail of individual trapping locations yet still make out the broad-scale patterns. In this map I've left off the legend to maintain the anonymity for the person that I made the map for. I think that it pretty clearly illustrates the point that I'm trying to make.

Another nice feature of Thiessen polygons is that they are an exact interpolator. That means that the model produces the exact value at the location of the data point. This is not the case with the majority of interpolators, including kriging.

Lesson: if considering heat maps or interpolation look into Thiessen polygons first. It may be the simplest yet best tool for the job.


  1. A doubt. Did you mean to say that the rectangles you made are an output of voronoi mapping or the colour classification inside each polygon is made of voronoi?. With out a legend i find it difficult to comprehend your idea

  2. Hi Sreenath. I left off the legend in this case because that collaborator hasn't published her paper yet. You are correct that having a legend is an integral part of any map. The colors represent things such as abundance or diversity of animals trapped. Each Thiessen polygon represents a trapping location (point). These points were arrayed as a grid, but weren't 100% evenly speaced. A map of points was busy and distracting. I like the Thiessen polygons because it represents each point, even zoomed out, and makes the clusters of high values and low values very apparent.