Biomes 1: Digitising old maps
Digitisation efforts
As mentioned in the project blurb, some of the intial work has been on digitising older maps. This is helpful if one wants to do do spatial analyses like mapping the climatic envelope of the classified regions.
Namibia
Rutherford and Westfall's 1986 work, I believe, was the first to use predictive modelling to map biomes in South Africa. It's a very good reference on its own. Subsequent to this, the same approach was used by Irish (1994) (The Biomes of Namibia, as determined by objective categorisation). As the biome work in South Africa had been developed further, I didn't see the need to use Rutherford and Westfall's work directly but the Namibian data is clearly still relevant. The result can be seen below.
Those familiar with other efforts to map the biomes of Namibia (Giess, SAFARI project etc.) will see overall similarity but I believe the resolution of the biome edges is most "dynamic" in this case.
Botswana
The 1991 FAO map of the vegetation of Botswana is the most detailed I have seen other than maybe Wild and Barbosa's map of the whole Zambesiaca region, but I think that it is more detailed when looking at Botswana alone. It's also an incredibly aesthetically pleasing map and I wanted to pay homage to them in some way.
I used an interesting digitisation technique that I have not tried before for this map. Previously, and where the scale is not so big, it is simpler to manually draw in the polygons on top of the georeferenced image. In this case it would take forever.
Recognising that the boundaries are the most important thing to digitise (i.e. that the attributes can be easily added later), I used a tablet to trace the outlines, thereby leaving a crisp version of the map without major compression artifacts. I suppose it may even have been possible to translate the vector version of this into geometry in some way but it's beyond me at this point.
As part of some other work I developed an algorithm for digitising rasters specific to that project, but I used some elements of it here. The essence is to use RGB to HIS processing, leaving in this case binary values for border or no border. This can then be polygonised and refined with a Voronoi approach to smoothen the pixelated edges and fill the width of the borders. I will probably write it up in detail at some point but I wanted to touch on it here. The end result worked out really well and saved ages of clicking. Much more satisfying and zen to trace the outlines...
Wrap
This was just a quick note on these maps as I wanted to present the work. I think I might do some maps I found of Zimbabwe at some point but it is worth noting that Wild and Barbosa's map seems to be the most detailed for this region. The overall approach I am taking, however, will benefit from inputs on all scales.
DW