I have a dataset on different fish species abundance of five localities per month for a year. I also have a physico-chemical properties on each locality for that same per month for a year. Now I tried to analyze fish species compositional change with locality and want to know which physico-chemical variable plays the significant role to that species. For this I used DCA on overall species data which I organized month by species. The dca ordination gave me short axis length which was only just 1.7. Then bounced back to linear ordination method ie, DCA only. Thus I fully rely on DCA not others.
I over fit the environmental variables after permutations on the month by species dataset. Will these analyses be enough to explain description of fish species with localities and their physicochemical properties?
What else should I be done?
I am using ´R´ and `vegan` to all my analyses.
Hi Chitra,
to analyse which physico-chemical parameter effects which species I'd use CCA.
To see which phys-chem parameters explain a great deal of variation in overall species composition I'd apply distance-based lineal models with the species distance/similarity matrix as response variable and the distance matrix of phys-chem parameters as predictor variable. I've done this for my data with the program PRIMER add-on PERMANOVA (see http://www.primer-e.com/permanova.htm). The program does a varianve partitioning for the predictor variables. If you are using R you might check out the package dbstats.
For species compositional data I'd apply principal coordinates analysis (PCO or PCoA) with an appropriate distance/similarity measure instead of DCA. For recommendations concerning data transformation and distance measures see Legende & Legende: Numerical Ecology or http://ubio.bioinfo.cnio.es/Cursos/CEU_MDA07_practicals/Further%2520reading./Papers%2520on%2520ecological%2520statisticsTransformations%2520in%2520ordination%2520Legendre2001Oecologia.pdf
To analyse species composition differences between the localities I'd use permutational manova. It sounds like you have a sampling design that lacks replication from the description, i.e. there are no replicates within the different localities. Is that right? You could consider the monthly samples as replicates but then you would disregard the probable seasonal effect, which you might also want to test. In this case I'd analyse the data with permanova with a mixed 2factorial design with month as fixed factor with 12 levels and locality as random factor with 5 levels. This can probably also be done with R. Since you use R yourself you might know which package is appropriate.
To get a fast overview I'd apply nMDS as ordination method first.
Cheers
I agree with Janet. I quite often this type of data processed. The CCA allows for good results. With R, there ADE4 the library that is perfectly suited to the analysis of ecological data. There is also a ADE4list forum to ask specific questions to the designer of the library.
Dear Janet and Alain,
Thank you a lot for both of your answer. I am a little bit skeptic towards CCA because of short lengths of gradient which was only 1.7. May be RDA would be the right choice, right?
With the best regards,
Chitra
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology