How to do clustering on monthly collected data?

I want to cluster 10 different sites based upon different parameters like temperature, humidity, rain, etc. for monthly data collected over a period of 2 years.

1. Should I use means of all 24 data samples for each parameter or the whole data sets?

2. Moreover, these parameters are on different scale of measurement: Is standardization necessary in this case? If so, how would I do it?

3. Finally, does there exist a cutoff value for deciding of clusters number in a dendrogram?

-
What is the purpose of clustering in this case (i.e. what are you trying to find from this data)? Are you looking for sites with similar attributes over the entire 2yr period? What if some attributes match very well and others do not, would these patterns be of interest as well? –  Nick Jun 22 '12 at 15:36
i want to find similarity between these 10 sites based upon 18 parameters, i am planning to use hierarchical clustering, by wards method using euclidean distance . so my question is whether i should use whole data for 24 months for these 18 parameter or i should use means of these 24 samples for each parameter to perform clustering?? –  A. Singh Jun 22 '12 at 16:38

1. So you are undecided whether:

• to cluster 240 points in 18 dimensional space or
• cluster 10 points (means of 24 observations) in the same 18 dimensional space?

I would go for clustering 240 points because clustering 10 points in such a high dimensional space has little sense. I guess your concern is that same site might end up in different clusters for different periods. If it is a problem in your case, you can assign each site to one cluster by most frequent cluster number, for example.

2. Standardization is a good way to make all dimensions comparable, so that when you compute distance between points you do not have dominant dimensions. There is also another option - to take 1-absolute correlation as a distance, then you do not need any standardization.

3. You could decide on the number of clusters or try to estimate it by various methods, using score function for example.

-
thank you for your time!, can i take my 10 sites as variables and put all 18 parameters data below each site corrosponding to it, so i have 10 variables with 24x18 cases, then i can cluster these 10 variables(sites) based upon cases(parameter) ...is this way of clustering is logical and yeah parameters are standardized already before putting it below each variables. –  A. Singh Jun 26 '12 at 11:15
and yeah its not 240 points its 18x24 points in 10 dimensional space. –  A. Singh Jun 26 '12 at 11:17
You mix things up. The setting for clustering is this: you have n points (which you want to cluster) in d-dimensional space. In your case one point is observation vector of one site for one time in parameter space. So n=10*24 and d=18. –  danas.zuokas Jun 27 '12 at 6:19