Dear Attila

I might suggest that you use the Muller-Dombois & Ellemberg  1974 'Arrhnatherum' tables. These have been a kind of standard for the testing of numerical procedures in many articles. You can always look in EVA or SIVIM databases (Iberian Peninsula and Macaronesia) (the last one I use a lot).

http://euroveg.org/eva-database

http://www.sivim.info/sivi/

Best.

J.

Dear Jorge,

thanks for your answer, I will look up these tables. However, I am still waiting for other ideas because I already have an example data set with Arrhenatheretalia grasslands from Hungary. It would be more convincing to include examples from contrasting vegetation types.

Best wishes,

Attila

Hi Attila,

How would you define that a vegetation type is diferent from another by the same degree? If you want to have the same degree of differentiation beforehand and than use it to test clustering algorithms, this appears to be tautological. While it is relatively easy to find 10 vegetation units, each of which has a similar distance to its nearest neighbour, it seems to be unrealistic to find 10 vegetation units that have the same average distance too each other.

Anyway, if you need 10 plots from 10 different dry grassland associations (i.e. from my understanding a similar degree of differentiation) with highly standardised sampling, I can provide such from dry grasslands in NE Germany (own PhD) or from various regions in the Palaearctic (EDGG Expedition data) or from the Succulent Karoo in South Africa (https://www.researchgate.net/publication/228108623_A_first_formal_classification_of_Hardeveld_vegetation_in_Namaqualand_South_Africa). If you are interested I would need to check whether all of those mentioned have >= 10 x >=10, but mostly this criterion should be met.

Best regards,

Jürgen

Article A first formal classification of Hardeveld vegetation in Nam...

Hi Jürgen,

thank you for writing!

"How would you define that a vegetation type is diferent from another by the same degree?"

I mean compositional difference that can be quantified by an appropriate dissimilarity index.

"If you want to have the same degree of differentiation beforehand and than use it to test clustering algorithms, this appears to be tautological."

I would like to test how cluster validity measures perform with high-dimensional data, when the optimal number of clusters is known to be high and the types are really well differentiated. Some validy measures most often detect few clusters (maybe 3 to 8) as the best classification solution. This can be either due to the real structure of the data often used for classification and testing validity measures (biased towards the northern temperate biomes!) or a bias in the validity measures. I would like to sort this effect out.

"it seems to be unrealistic to find 10 vegetation units that have the same average distance too each other"

Well, it seems that but a good method should work well even in this extreme case. Such situation could occur when beta-diversity is extremely high and the data is really multidimensional (practically, 9 dimensional).

So, do you think your data would be helpful for my purposes?

Best wishes,

Attila

Hi Attila,

I don't know if our dataset can be used, but if you think so, I can provide the file and help you with interpreting the content. The data is from Norway, so it is not exactly South-Africa or Australia :-)

Anyway, good luck Attila.

Best regards, Anders

Article Location of plant species in Norway gathered as a part of a ...

Dear Anders,

thank you for your response. I will contact you for more details soon.

Best wishes,

Attila