enter image description hereI have collected data on mangrove tree species density and composition from two habitat categories (sample groups) by using belt transects of various lengths between 100 x 10 M to 500 x 10 m. (3 & 2 samples at each sample group; n=5). data is similar to the attached image. The length of transects was varied because of the limited habitat availability in each sample group. Replicate data were collected from the same transects after 8 years. Now I want to compare the species richness and composition between the temporal replicates of each sample groups, but I'm a bit skeptical that due to the various sizes of samples I can't directly use the data for running some analysis like Mann-Whitney U test or Multi-response permutation procedure to compare the tree densities and species composition between the temporal replicates of sample groups. Kindly suggest a better way to analyze the data and how can I normalize the data for the sample area before running my analysis? Thanks in advance. - Nehru


Where you have longer transects, do you have data stored at a finer scale? For example, can you split your 500x10m transects into 5 100x10m transects (or whatever your smallest transect was)? If so, you could use rarefaction, so that each area is sampled the same amount. This essentially involves randomly discarding some of the sampling from your more fully-sampled sites until all the sites have comparable sampling.

Rarefaction is conservative, but it does result in you losing quite a bit of your data. An alternative method is to use species accumulation curves - these use the results of your sampling to estimate the 'unseen' diversity of each site, so that then you can then compare the "total" richness. The R function specpool includes methods for carrying this out.

  • $\begingroup$ Unfortunately, we have no such fine-scale breakup of the transects. $\endgroup$
    – Nehru
    Jun 11 '20 at 9:07
  • 1
    $\begingroup$ In that case species accumulation curves will be more useful. Of course, you will need to bear in mind that your estimated total richness produced by these curves are just estimates - you should show the uncertainty around these estimates in your write-up. $\endgroup$
    – rw2
    Jun 11 '20 at 9:31
  • $\begingroup$ If you use count data for the composition analysis, you can fit generalized linear models with an offset (with poisson and negative binomial distribution). Thanks to the logarithm properties, this offset allows you to take into account the sampling effort. However the relation between the offset and the response variable is linear with this method, so in your case species accumulation curves might be a better choice. $\endgroup$ Jun 11 '20 at 13:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.