Report - ICP Forests

3. Intensive Monitoring 75
Missing values and correspondence between vegetation and abiotic data
After the exclusion of records with deficient soil data, there are three records without
coordinates (and hence, without climate zone and EMEP deposition estimates). These were also
excluded from the analysis. After these exclusions and the replacement of missing soil data by
estimates, 616 usable records remain with soil, 'other', and EMEP deposition data, and 265 with
soil, 'other', bulk and throughfall deposition data.
Annex 3.4.3 gives the correlation matrix of all variables used in the CCA analysis of last relevé
per plot using EMEP deposition estimates. Other analyses have been carried out on different
subsets of the data and therefore have slightly different correlation matrices; for the correlation
between the EMEP estimates and measured bulk and throughfall deposition see Annex 3.4.2.
On the vegetation side, species with less than three occurrences and relevés with less than three
species were excluded from the multivariate analysis of the last relevé per plot, to avoid a very
heterogeneous data set. This resulted in 598 usable records, of which 477 had usable abiotic data
(excl. bulk and throughfall deposition). These 477 relevés had 170 species (after the exclusion
of very rare ones). A separate analysis was done with the bulk and throughfall deposition data
instead of the EMEP estimates; for this analysis there were 181 usable records containing 114
species. For the analysis of the vegetation change the first and the last relevé of each plot were
used if the interval between them was seven years or more; this yielded 161 plots, of which 138
had usable abiotic data, and 60 bulk deposition and throughfall data. In this operation rare
species or species-poor relevés were not excluded because the linear statistical methods used
here are less sensitive to heterogeneity.
3.4.3 Statistical methods
The statistical methods used are similar to those used in 2002 (De Vries et al. 2002). The effect
of the environmental variables on the vegetation of the last relevé of each plot was assessed by
canonical correspondence analysis (CCA). The exclusion of very rare species (< 3 occurrences)
and downweighting of rare species resulted in a gradient length GL 1 = 7.5 and explained
variance λ 1 / Σλ = 4.6% for the first axis at 477 samples and 170 species which was judged
acceptable. The change in the vegetation was determined as (%cover in last relevé) minus
(%cover in first relevé) per species. The significance of this change was determined by ordinary
linear statistics. The relation between the change in the vegetation and the environmental
variables was analysed for all species together using RDA (= the linear form of CCA = the
canonical form of PCA). Ellenberg values were determined as the unweighted mean over all
species (incl. the rare species but excl. relevés with < 3 species with a known Ellenberg value).
The relation between the change in Ellenberg values and environmental variables was assessed
using multiple regression. All multivariate operations were carried out by the program
CANOCO v 4.53, all univariate operation by the program GENSTAT v 12.1.
3.4.4 Results
CCA analysis of last relevé per plot
Possible observer effects were assessed by determining the unique contributions ('TMVs') of the
countries and the 'real' environmental variables (Table 3.4.4-1). Out of a total of 25% variance
that can be explained anyway, 5% is uniquely due to the countries. As both the geographical
coordinates and climate zones are among the environmental variables this 'county effect' is most
probably caused by methodological differences. Estonia, France, the Netherlands, Ireland and
Italy are the most deviant countries (in that order) and their effect is significant even after
accounting for the effect of all environmental variables. Therefore the countries were used as
covariables in the subsequent analysis. Table 3.4.4-2 gives the result of the forward selection.
Like in De Vries at el. (2002), the variables were included in the model stepwise, at each step