I have a community assembly dataset with 299 species at 15 sites. Im interested in correcting for the effect of spatial autocorrelation on beta-diversity (or species turnover). Dataset here.
There is an obvious effect of spatial distance between the sites and community dissimilarity. I have a completed a mantel correlogram to show this effect but now I’m a bit stuck on how to account for this effect or correct for beta diversity analyzes. And more importantly how does the effect of distance between sites compare to my environmental effect (precip)
#load libraries library (vegan) library (fossil) #load datasets spp.list<-read.csv('Spatial Autocorrelation.csv', header=T) rownames(spp.list)=spp.list$Site sites.xy= spp.list[,(2:3)] precip= spp.list[,(4)] spp.list<-spp.list[,(5:303)] #check for spatial autocorrelation and environmental correlation using Mantel Test #build a species dissimilarity matrix dist.mat=vegdist (spp.list) #build a geographic distance matrix for sites geo.dist.mat=earth.dist(sites.xy) #build a distance matrix for precip values precip.dist <-dist (precip) #identify the correlations between two matrices Matel Correlelogram correlog<-mantel.correlog( dist.mat, geo.dist.mat, nperm=999 ) summary(correlog) correlog plot (correlog) 
#partial Mantel Test including both distance and precipitation differences library (ecodist) natt.pmgram<-pmgram (dist.mat, geo.dist.mat, precip.dist, nperm=999) plot (natt.pmgram) 
Im not sure how to interpret these correlograms. Does anyone have any suggestions/explanations? and can these analyses be used to correct for the obvious spatial auto-correlation issues in my dataset?
Best Answer
The first figure shows that species dissimilarity in the first two distance classes are significantly positively spatially correlated, whereas the forth distance class is negatively spatially correlated, with Holm's correction. However, I'd be a little concerned about fitting so many distance classes for the relatively few sites you seem to have (15, right?), are the bins adequately represented? The part I am not sure about is how you can have distance and precipitation simultaneously, with more distance classes (fig.2) than in distance alone (fig.1); because of that I was waiting to see if you get other answers, alas (at least for now), it may be too narrow of a topic.
Depending on whether you use univariate measure of beta-diversity or the entire dissimilarity matrix, you could use simple detrending or a combination of Moran Eigenvector Maps (MEM) with variance partitioning to analyze the relative importance of precipitation and spatial autocorrelation. Again, depending on your hypothesized correlation structure, asymmetric EM could be used: Blanchet et al. 2008 http://adn.biol.umontreal.ca/~numericalecology/Reprints/Blanchet_et_al_2008-EcoMod.pdf
Similar Posts:
- Solved – Interpretation of Mantel r correlations
- Solved – Interpretation of Mantel r correlations
- Solved – How to compute a measure of distance between sites with continuous variables
- Solved – NMDS from Jaccard and Bray-Curtis identical. Is that a bad thing
- Solved – Is it possible to use Hellinger distance for environmental variables