Dear. T. Naulak. Yes, you may. But I do not consider such a choice as the best. MDS and nonmetric MDS is designed primarily for dimension reduction. Dimension solution means a drastic loss of information. As I understand, you are in comparison (searching for similarities). I would recommend going into specialized Multivariate Statistical Analysis techniques designed for clustering.

Depends on what you mean by "comparing". If you would like to visualize how the multivariate community composition and abundances vary across sites, then NMDS is a great tool. If instead you wish to know definitively if the different agroecosystems differ in community composition, then another multivariate tool is needed. I would suggest PERMANOVA, and restricted PERMANOVA if you have concerns regarding independence within your study design.

I think you can. When you only have dissimilarities, the non-metric approach (NM-MDS) is somewhat more

appealing, and also produces better ordination results than PCA... Irrespective of unequal areas. If you are interested in similarities, then different approach must be explored, clustering as suggested by the first respondent, Jozef.

You can use the two method, and compare results.

https://www.google.com/url?sa=t&source=web&rct=j&url=https://biostats.w.uib.no/5-4-multivariate-analysis-multidimensional-scaling-mds/&ved=2ahUKEwi84IDfjJbqAhWmC2MBHSw9BhMQFjAWegQIAhAB&usg=AOvVaw0lXGh3eWpOe69nKgKRfLlU

Goodluck