Updating the Biome Map
Over the past week I spent some time revisiting the analysis that kicked off my Great Antarctic *crawl project: biomes of Antibor.
In that original post I trained a support vector machine classifier on real-world elevation, temperature, rainfall, and biome data. The results were overall very believable, but they tended to be a bit 'blobby' and a small detail was missing.
Consider, for example, the following:
At at least a subcontinental scale, watersheds, drainage basins, and rivers -- "riparian corridors" --can support dense vegetation in otherwise arid environments. Since Antibor is a very arid continent, this is something I'd love to capture!
So, I reran the biome analysis but this time I added a map of flow accumulation as a predictor variable. The newest development release of the terra package has a speedy flow accumulation algorithm, making it very easy to adapt my previous script. I also switched from support vector machine to random forest for the machine learning algorithm because I found that it had better speed & accuracy.
The new biomes can be seen below -- areas shaded a darker green are forested. It didn't quite capture the effect of the pictures linked above, but compared to the old map it's slightly more visually complex and fractal-y. Forests at least feel slightly more in tune with the landscape.
There are still a few wonky areas: linear striations formed by flow direction artifacts, some areas where low-lying areas are classed as desert despite being near freshwater outlets. The thing is, these wonky areas are a lot smaller in scale, making them easier to integrate into the world building as local anomalies.