19.01.2014 Views

Report - ICP Forests

Report - ICP Forests

Report - ICP Forests

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

76 3. Intensive Monitoring<br />

the one that leads to the highest increase in explained variance with the constraint that variables<br />

with a correlation |R| > 0.5 with variables already in the model are skipped. Variable selection<br />

was stopped when none remained that could significantly (P < 0.05) improve the fit of the<br />

model. The resulting 'minimal' model explains 12.5% variance which is quite usual in this type<br />

of ecological data. pH is the most important explanatory variable, which is also usual in this<br />

type of data. There is a small, but highly significant effect of deposition quantified as the<br />

EMEP-estimated NO 3 deposition for 2000. The model is summarised in Table 3.4.4-3. The<br />

results strongly agree with those of De Vries at al. (2002) with the traditional factors (in the<br />

order: tree layer, soil, climate) as the most important explanatory variables, and c. 5% of the<br />

variance in the fitted values explained by deposition. However, in contrast to the 2002 analysis<br />

this effect of deposition is solely due to N-deposition and not partly to e.g. seawater ions.<br />

Table 3.4.4-1: Percentage explained variance due to countries and 'real' environmental variables<br />

source<br />

TMV<br />

uniquely due to countries 5,0%<br />

uniquely due to environmental variables 14,1%<br />

undetermined 6,3%<br />

total variance explained 25,4%<br />

Table 3.4.4-2: Result of forward selection of environmental variables to explain the vegetation of the last relevé of<br />

each plot, using EMEP estimates to quantify deposition and using the countries as covariables. Eigenvalues: λ 1 =<br />

0.259, λ 2 = 0.24, λ 3 = 0.188, λ 4 = 0.125, Σλ = 11.739, Number of plots = 477, number of species = 170. Rare species<br />

are downweighted. F = (regression mean square with this term - regression mean square without this term) / error<br />

mean square; P = probability of this, or a higher F-value under the null hypothesis as determined on the basis of 999<br />

bootstrap samples.<br />

Variable<br />

compartment<br />

F P percentage<br />

explained<br />

variance<br />

pH organic 8,1 0,001 1,70%<br />

mediterr. oak tree 6,79 0,001 1,45%<br />

temperate oak tree 5,95 0,001 1,28%<br />

Pinus sylv+nigra tree 4,83 0,001 1,02%<br />

Fagus tree 4,42 0,001 0,94%<br />

CEC mineral 2,84 0,001 0,60%<br />

N/C organic 2,46 0,005 0,51%<br />

Latitude climate 2,44 0,001 0,43%<br />

NO3 (2000) deposition 2,36 0,001 0,51%<br />

Longitude climate 2,28 0,001 0,51%<br />

coniferous 'other' tree 2,23 0,003 0,43%<br />

deciduous 'other' tree 2,13 0,061 0,43%<br />

Ca organic 2,1 0,002 0,43%<br />

Atlantic South climate 1,96 0,008 0,43%<br />

Age tree 1,89 0,003 0,34%<br />

Atlantic North climate 1,85 0,005 0,34%<br />

K organic 1,87 0,011 0,43%<br />

Boreal climate 1,72 0,012 0,34%<br />

P organic 1,7 0,007 0,34%<br />

Altitude 1,36 0,092 0,26%<br />

N_C_min 1,23 0,173 0,26%<br />

(further terms not given)<br />

SUM if P < 0.05 12,44%

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!