Of Maps and Men

It's always nice when you find a gem. I just found out about a recent paper (pdf warning), published in Frontiers in Ecology, written to try and quantify what kind of human-created biomes exist, and where those biomes are found (A biome being a region characterized by distinct flora and fauna, as well as charactistic landforms, hydrology, and weather).

Figure 1: Small map of global anthropomorphic biomes. Click to see a larger, higher resolution version (*pdf warning) From Ellis, EC, Ramankutty, N. 2007. Putting people in the map: anthropogenic biomes of the world. Front Ecol Environ 2008; 6, doi: 10.1890/070062



The authors emphasize the fact that human-affected land accounts for 90% of global net primary production (energy produced by plants, over and above metabolic needs) as well as 80% of all forested territory. I don't, however, think that this is the most interesting result from this paper. Pretty much anyone with a background in modern ecology would have found those numbers unsurprising, and would have been revealed in a much simpler analysis. What the real importance of this paper stems from is how they discovered the different biome types. Rather than arbitrarily assigning different categories, they used cluster analysis (a method from statistical ecology) to discover groupings of data that "hung together" for many different variables. This does a lot to remove observer bias or researcher preconception from the final map.

This report doesn't just provide a very pretty map of human effects and variance in cultivation and settlement patterns (for real fun: a Google Maps version!). The best part of this report is the fantastic data set it provides. Anyone with even a vague interest in geographic-economic interactions, settlement patterns, or land-use patterns could use this to test any number of mechanism-driven models.

I do have a couple complaints, however: they don't specify what clustering algorithm they are using, except that it is from SPSS. There are several ways to cluster data, and depending on the structure of the data set, this can make a difference. I suspect, however, that the data they are using is likely fairly straight-forward to analyze, so the exact details of the algorithm shouldn't change the overall picture. Also, they may have published supplemental data that I just didn't find.