Metacommunties and RS

Haven't posted in a while, Thanksgiving saw a lot of friends in town. But another reason is this book, which I've been reading for a while (along with several articles, of course). Books quite interesting, and according to a few folk, this is the current authority on metacommunities. It seems to me that metacommunity theory is a bridge between non-spatial population/community ecology and the spatial landscape ecology- exactly what I'm interested in. Currently I'm reading about pitcher plant/invertibrate metacommunities, which offer an intersting system of spatially seperate habitat which is temporally variable, requiring dispersal between the patches- similar to clear cuts, disturbances, etc in forests. Good stuff. It's a well written book so far.





Other things recently read:
Lucas 2008. Hyperspectral remote sensing to assess vascular plant specices richness. Remote Sensing of the Environment
-Pretty good, but they didn't find any good predictors without first defining habitat class. Not necessarily a problem, but an additional step. I wonder if they would have had more predictive power with some unmixing.

Gillespie 2008 Assessing biodiversity from space. Progress in Physical Geography
-Great review. Still reading through the references, but the stuff on assessing alpha and beta diversity via remote sensing is extremely interesting, especially considering that models of invasive species typically only focus on abiotic environments, but should (according to several people) include info on biotic interactions- one aspect of which is indicated by the general biodiversity in the area. Plus, rapid identification of biodiversity hotspots- even on the local scale- would be extremely useful for conservation.

Tang 2007 Improving urban classification through fuzzy SMA.
-Not bad. They take mixed pixels, and instead of just doing traditional linear unmixing, they also use the mixed means and covariances in determining endmembers, sort of a "fuzzy endmember" (which is reminiscent of endmember bundles, Asner/Wessman/etc. (in fact, I'm not sure as to the actual difference at this point)). The paper didn't describe methods that well, so I'm not sure as to the mechanics, but the concept makes sense. It was unmixed based on only a few classes determined from hyperspatial info. I wonder if you could use neighboring pixel stuff in conjunction with fuzzy endmembers to improve classification.