You can apply both multivariate and univariate statistical methods. Multivariate methods such as NMDS will allow you to group study sites according to their similarities in species composition/abundance. If you have a specific hypothesis to test, you can draw univariate information from your species matrix (such as species diversity, or species richness, or evenness) and then test your hypothesis with an univariate repeated measures analysis of variance (ANOVA). However, there are assumptions (such as distribution normality and variance homogeneity) that must be met for ANOVA use, and in case you do not meet these assumptions, many authors would recommend the use of non-parametric equivalents to ANOVA (such as Kruskal-Wallis test). If you intend to use univariate tests, I would strongly recommend you read the book "Experiments in ecology: their logical design and interpretation using analysis of variance", from A.J. Underwood, published in 1999 by Cambridge University Press. For multivariate analyses, the software Primer (http://www.primer-e.com/) is a must. All the best! 

Dear bro

I would like to agree to da Gama answer, further you can calculate the diversity indics like richness eveness etc., if you have any any question regards diversity indices and  multivariate and univerate anylsis please connect me.

All the best

With Regards

Dear Da gama and S Prakash..Thank you very much for the response. The data I have is not on the community structure describing the diversity, density, abundance of corals and other organisms. I have a time series data on the percentage cover of different benthic forms like corals, macroalgae, dead corals, sand rubbles etc. The Univariate and Multivariate analysis will help to discriminate the study sites based on their community composition and their relative abundance. But with this data, my intention is to analyse the change in the percent cover of benthic forms over time. When I go through the literature, I read about autocorrelation analysis which will help in predicting the cyclic or periodic patterns for any time series data. Pls see if you can help me in this.

Thank you

First I would suggest that you plot all the data in graphs, with percent cover of each species over time. Then you can observe if there were conspicuous shifts in cover over time, and if so, formulate testable hypothesis concerning these shifts and analyse the data through hypothesis testing statistics. As far as I understood, you need first of all of a good descriptive statistics such as graphs. If you plot central tendency data (means or medians) with their respective dispersions (e.g., standard deviations, errors, etc.) you may also have a clear portrait of shifts in dispersion of data, as sometimes communities may not vary on average, varying in dispersion instead. There is an issue of non-independence when you analyse percent cover of several species on the same sampling areas, e.g., quadrats, as of course, as you mentioned, when macroalgae abundance increases, consequently the other species will decrease.

Da Gama is correct - full data analysis first BUT with your initial research questions in mind.

Time series are likely to show non-linear patterns, therefore try GAM. Percentage cover will be a strictly positive distribution (cannot be less than zero) so applying classical analyses such as ANOVA is pointless. A time series is a series of repeated measures at one or several locations, you will have autocorrelation in your data so not all your observations are independent. I'm guessing you have plots or transects or some other discrete sampling area, probably replicated. If so you need to apply models that deal with such nested data, again lack of independence of observations from plots within an area of reef needs to be accounted for. So, I would envisage using use GAMMs with non-gaussian distributions for your response variable precentage cover. See Zuur et al 2007, Analysing Ecological Data; Zuur et al 2009, Mixed Models and extensions in Ecology; Zuur et al 2012, A beginners guide to Generalized Additive Models in R.

You can use mutlivariate analyses as per Prakash's advice even if you dont have species data. You can treat each benthic form as a 'species' and then you arrange your data so that each siteXtime is one row- e,g site a time 1, site b time 1, site a time 2, site b time 2 etc. and then your columns will be 'species' or benthic forms and the data will be % cover.. The software should be able to compare sites over time- I use CANOCO and you could use something like principal response curves to monitor change over time- you would need to chose one site as the base line (x axis) against which all other sites would vary over time- this method works best when the different sites are not extremely different at time 1. You could also probably do it the other way round and treat the different benthic forms as the different 'sites' and monitor how they change over time using the principal response curves. 

I agree with above answers mentioned mixed models and data exploration. For time series, you may consider it is regular or irregular spaced time series. The application of GAM depends on the result of data exploration in case of non-linear effect. If data do not conform to normal distribution, then GAMM or GLMM may be applicable as your response variable is expressed as percentage.

Tania & Xiaoguang @ Thank you very much for the clear response. I prefer using Principle response curve to monitor the changes over time and let me check the reliability of the results

Tania @ Dear madam, I dont have CANOCO software with me and I have only PRIMER software for doing all the possible ecological analysis. I shall feel grateful If you can help me doing this analysis and in return I can share the authorship with you in my publication. If you agree, please send me your personal mail ID to my mail and we can discuss further. My mail Id: [email protected]

Hi,

Are you at the stage where what you need is to test whether there are significant trends in percent cover for the different life forms? If that is the case, you could use a multivariate version of the Mann-Kendall test, described in: http://www.slu.se/PageFiles/70727/Manual.pdf

Best wishes

Jose @ Thank you for the reference. I will look in to it.

Hi Jose

That URL is broken do you have the pdf it is supposed to link to?

cheers