Dear Etienne, dear friends,
Recently some of you have answered a message I posted in Ecolog. In
that answer Etienne suggested I go through a tutorial written by Dr.
Legendre. I went through that tutorial and would like to thank him for
indicating that.
Going through that, however, raised a few doubts that are between me
and the final data analyses of my present work. I would like to hear the
opinion some of you may have on that matter.
Here is, briefly, the case.
1 - The Hellinger transformation allowed us to contour the difficult
premises of classical RDA and CCA.
2 - Despite of this, these dbRDA and CCA still do not consider space
explicitly
3 - Except from small-scale experiments and small organisms studied very
locally, space is always a factor causing autocorrelation and
confounding the effects of environmental factors
4 - We have 36 cacti communities, each on a rocky outcrop, scattered
throug a seminatural grassland landscape. Eighteen of these communities
are embedded in 2-years old eucalyptus plantations. In our database we
have a total of 10 cacti species (gamma-diversity), the 36 sites per 10
species abundance species matrix, a matrix of environmental variables
(eucalyptus (binary), outcrop size, distance to the nearest outcrop,
size of the nearest outcrop, %shrub cover) and the spatial matrix (xy
UTM coordinates).
5 - I permormed a PERMANCOVA in R taking management (eucalyptus or not)
as the main factor and all the variables of the env matrix as
covariates. Management was significant and almos all the covariates were
not. A NMDS showed a weak division between outcrops in eucaliptus
plantation and outcrops in grasslands (very scattered).
6 - This analysis was, however, flawed, because for obvious reasons
these two groups of outcrops were clearly spatially apart in the
landscape. Including the XY coordinates as covariates made the
management to become not significant.
7 - However, the raw XY coordinates are not appropriate to include in
the analysis. I performed a varpart (for short) in R, that showed that
the environmental variables had only a small effect. And here comes the
question.
8 - The environmental matrix contains very different variables, and
management is one of them. How to consider those variables explicitly if
the partitioning approach only attributes a % contribution to them as a
group? For instance: the small effect the env matrix has is due to which
variable? Management? Or those linked to shrub cover? This is important
because there are large companies involved, and any outcome will push
the destiny of this endemic and endangered cacti community.
9 - A possible solution would be to break the env matrix into a number
of smaller matrices (management, physical properties, plant cover). But
this would complicate the outpud and also management, for instance, has
only one variable, which is binary. This cannot yield a distance-matrix
in the analysis.
Well, if you reached this point, thank you already. Any ideas are
welcome.
Best regards,
Alexandre
Dr. Alexandre F. Souza
Programa de Pós-Graduação em Biologia: Diversidade e Manejo da Vida
Silvestre
Universidade do Vale do Rio dos Sinos (UNISINOS)
Av. UNISINOS 950 - C.P. 275, São Leopoldo 93022-000, RS - Brasil
Telefone: (051)3590-8477 ramal 1263
Skype: alexfadigas
[email protected]
http://www.unisinos.br/laboratorios/lecopop
--
Esta mensagem foi verificada pelo sistema de antiv�rus e
acredita-se estar livre de perigo.
