38
WHY ARE ELLENBERG INDICATOR VALUES SO GOOD EXPLANATORY VARIABLES? David Zelený

Why are ellenberg indicator values so good explanatory variables?

  • Upload
    others

  • View
    13

  • Download
    0

Embed Size (px)

Citation preview

WHY ARE ELLENBERG INDICATOR VALUES

SO GOOD EXPLANATORY VARIABLES?

David Zelený

ELLENBERG INDICATOR VALUES

3 2 6 6 5 2

CALCULATION OF MEAN ELLENBERG

INDICATOR VALUES

EIV-reaction 1 2 3

Mycelis muralis 6 1 0 0

Moehringia trinervia 7 0 1 1

Mercurialis perennis 7 1 0 1

Lathyrus vernus 4 0 1 0

Myosotis sylvatica 7 1 1 0

Milium effusum 5 0 0 1

Melica nutans 3 1 1 0

Melampyrum pratense 2 0 1 1

Myosotis ramosissima 1 1 1 0

Lychnis viscaria 2 0 0 1

Melittis melissophyllum 3 0 1 0

CALCULATION OF MEAN ELLENBERG

INDICATOR VALUES

EIV-reaction 1 2 3

Mycelis muralis 6 1 0 0

Moehringia trinervia 7 0 1 1

Mercurialis perennis 7 1 0 1

Lathyrus vernus 4 0 1 0

Myosotis sylvatica 7 1 1 0

Milium effusum 5 0 0 1

Melica nutans 3 1 1 0

Melampyrum pratense 2 0 1 1

Myosotis ramosissima 1 1 1 0

Lychnis viscaria 2 0 0 1

Melittis melissophyllum 3 0 1 0

4.8

mean

CALCULATION OF MEAN ELLENBERG

INDICATOR VALUES

EIV-reaction 1 2 3

Mycelis muralis 6 1 0 0

Moehringia trinervia 7 0 1 1

Mercurialis perennis 7 1 0 1

Lathyrus vernus 4 0 1 0

Myosotis sylvatica 7 1 1 0

Milium effusum 5 0 0 1

Melica nutans 3 1 1 0

Melampyrum pratense 2 0 1 1

Myosotis ramosissima 1 1 1 0

Lychnis viscaria 2 0 0 1

Melittis melissophyllum 3 0 1 0

mean EIV: 4.8 3.9 4.6

CALCULATION OF MEAN ELLENBERG

INDICATOR VALUES

EIV-reaction 1 2 3 4

Mycelis muralis 6 1 0 0 0

Moehringia trinervia 7 0 1 1 1

Mercurialis perennis 7 1 0 1 1

Lathyrus vernus 4 0 1 0 0

Myosotis sylvatica 7 1 1 0 0

Milium effusum 5 0 0 1 1

Melica nutans 3 1 1 0 0

Melampyrum pratense 2 0 1 1 1

Myosotis ramosissima 1 1 1 0 0

Lychnis viscaria 2 0 0 1 1

Melittis melissophyllum 3 0 1 0 0

mean EIV: 4.8 3.9 4.6 4.6

mean EIV inherits information about

compositional similarity between plots

RANDOMIZATION OF EIVS AMONG SPECIES

3 2 6 6 5 2

RANDOMIZATION OF EIVS AMONG SPECIES

3 2 6 6 5 2

CALCULATION OF MEAN RANDOMIZED

ELLENBERG INDICATOR VALUES

EIV-reaction 1 2 3 4

Mycelis muralis 6 1 0 0 0

Moehringia trinervia 7 0 1 1 1

Mercurialis perennis 7 1 0 1 1

Lathyrus vernus 4 0 1 0 0

Myosotis sylvatica 7 1 1 0 0

Milium effusum 5 0 0 1 1

Melica nutans 3 1 1 0 0

Melampyrum pratense 2 0 1 1 1

Myosotis ramosissima 1 1 1 0 0

Lychnis viscaria 2 0 0 1 1

Melittis melissophyllum 3 0 1 0 0

CALCULATION OF MEAN RANDOMIZED

ELLENBERG INDICATOR VALUES

EIV-reaction 1 2 3 4

Mycelis muralis 7 1 0 0 0

Moehringia trinervia 5 0 1 1 1

Mercurialis perennis 4 1 0 1 1

Lathyrus vernus 2 0 1 0 0

Myosotis sylvatica 3 1 1 0 0

Milium effusum 2 0 0 1 1

Melica nutans 7 1 1 0 0

Melampyrum pratense 3 0 1 1 1

Myosotis ramosissima 7 1 1 0 0

Lychnis viscaria 6 0 0 1 1

Melittis melissophyllum 1 0 1 0 0

CALCULATION OF MEAN RANDOMIZED

ELLENBERG INDICATOR VALUES

EIV-reaction 1 2 3 4

Mycelis muralis 7 1 0 0 0

Moehringia trinervia 7 0 1 1 1

Mercurialis perennis 5 1 0 1 1

Lathyrus vernus 3 0 1 0 0

Myosotis sylvatica 2 1 1 0 0

Milium effusum 6 0 0 1 1

Melica nutans 2 1 1 0 0

Melampyrum pratense 7 0 1 1 1

Myosotis ramosissima 4 1 1 0 0

Lychnis viscaria 3 0 0 1 1

Melittis melissophyllum 1 0 1 0 0

CALCULATION OF MEAN RANDOMIZED

ELLENBERG INDICATOR VALUES

EIV-reaction 1 2 3 4

Mycelis muralis 6 1 0 0 0

Moehringia trinervia 4 0 1 1 1

Mercurialis perennis 3 1 0 1 1

Lathyrus vernus 3 0 1 0 0

Myosotis sylvatica 5 1 1 0 0

Milium effusum 7 0 0 1 1

Melica nutans 7 1 1 0 0

Melampyrum pratense 1 0 1 1 1

Myosotis ramosissima 7 1 1 0 0

Lychnis viscaria 2 0 0 1 1

Melittis melissophyllum 2 0 1 0 0

CALCULATION OF MEAN RANDOMIZED

ELLENBERG INDICATOR VALUES

EIV-reaction 1 2 3 4

Mycelis muralis 6 1 0 0 0

Moehringia trinervia 4 0 1 1 1

Mercurialis perennis 3 1 0 1 1

Lathyrus vernus 3 0 1 0 0

Myosotis sylvatica 5 1 1 0 0

Milium effusum 7 0 0 1 1

Melica nutans 7 1 1 0 0

Melampyrum pratense 1 0 1 1 1

Myosotis ramosissima 7 1 1 0 0

Lychnis viscaria 2 0 0 1 1

Melittis melissophyllum 2 0 1 0 0

Mean RANDOMIZED EIV: 5.6 4.1 3.4 3.4

EIV-reaction 1 2 3

Mycelis muralis 6 1 0 0

Moehringia trinervia 7 0 1 1

Mercurialis perennis 7 1 0 1

Lathyrus vernus 4 0 1 0

Myosotis sylvatica 7 1 1 0

Milium effusum 5 0 0 1

Melica nutans 3 1 1 0

Melampyrum pratense 2 0 1 1

Myosotis ramosissima 1 1 1 0

Lychnis viscaria 2 0 0 1

Melittis melissophyllum 3 0 1 0

Mean EIV: 4.8 3.9 4.6

EIV-reaction 1 2 3

Mycelis muralis 6 1 0 0

Moehringia trinervia 4 0 1 1

Mercurialis perennis 3 1 0 1

Lathyrus vernus 3 0 1 0

Myosotis sylvatica 5 1 1 0

Milium effusum 7 0 0 1

Melica nutans 7 1 1 0

Melampyrum pratense 1 0 1 1

Myosotis ramosissima 7 1 1 0

Lychnis viscaria 2 0 0 1

Melittis melissophyllum 2 0 1 0

Mean RANDOMIZED EIV: 5.6 4.1 3.4

EIV-reaction 1 2 3

Mycelis muralis 6 1 0 0

Moehringia trinervia 7 0 1 1

Mercurialis perennis 7 1 0 1

Lathyrus vernus 4 0 1 0

Myosotis sylvatica 7 1 1 0

Milium effusum 5 0 0 1

Melica nutans 3 1 1 0

Melampyrum pratense 2 0 1 1

Myosotis ramosissima 1 1 1 0

Lychnis viscaria 2 0 0 1

Melittis melissophyllum 3 0 1 0

Random variable: 4.6 4.8 3.9

THREE TYPES OF VARIABLES:

DATA USED FOR ANALYSES

Dataset 1

94 vegetation plots

forest vegetation in Vltava river

valley

measured soil pH

Dataset 2

1000 vegetation plots

forest vegetation

randomly selected from Czech

National Phytosociological

Database

INFORMATION ABOUT COMPOSITIONAL

SIMILARITY AMONG PLOTS INHERITED INTO

measured pH calculated mean EIV for soil

reaction

r , P - results of Mantel’s test of correlation between two dissimilarity matrices

plot 1 plot 2 plot 3 plot 4

plot 2 0.33

plot 3 0.34 0.37

plot 4 0.35 0.22 0.42

plot 5 0.84 0.84 0.76 0.82

plot dissimilarity

Δ measured pH

plot 1

5.10

plot 2

4.09

plot 3

4.10

plot 4

4.15

plot 2 4.09 1.01

plot 3 4.10 1.00 0.01

plot 4 4.15 0.95 0.06 0.05

plot 5 5.35 0.25 1.26 1.25 1.20

Bray-Curtis

distance

INFORMATION ABOUT COMPOSITIONAL

SIMILARITY AMONG PLOTS INHERITED INTO

measured pH calculated mean EIV for soil

reaction

r , P - results of Mantel’s test of correlation between two dissimilarity matrices

INFORMATION ABOUT COMPOSITIONAL

SIMILARITY AMONG PLOTS INHERITED INTO

mean randomized EIV for soil

reaction random variable

r , P - results of Mantel’s test of correlation between two dissimilarity matrices

EIVS AS EXPLANATORY VARIABLES IN CCA

vegetation

Ecological

knowledge

(Ellenberg)

explanatory

variable dependent

variable

Circularity of

reasoning

Species

composition

Calculated

mean EIV

COMPARISON OF MEASURED PH AND

CALCULATED EIV FOR SOIL REACTION

++

+

+

++

++

+

+

+

+

+

+

+

+

+

+++

+

+

+

+

+

+

+

+

+

++

+

+

+

+

+

+

++

+

+

+

+

+

+

+

+ ++++

+

+

+++

++

++

+

+

+

+

+

+

+

++

+

+

+

+ +

+

+

+

+

+

++

+

+

+

+

+

+

+

++

+

++

+

2

3

4

5

6

7

3.5 4.0 4.5 5.0

Measured soil pH

Me

an

Elle

nb

erg

re

actio

n

Data: dataset 1 – river valley

CCA: COMPARISON OF MEASURED PH AND

CALCULATED EIV FOR SOIL REACTION

0

1

2

3

4

5

real pH Ellenberg reaction

Exp

lain

ed

va

ria

bili

ty [

%]

CCA: COMPARISON OF MEASURED PH AND

CALCULATED EIV FOR SOIL REACTION

0

1

2

3

4

5

real pH Ellenberg reaction

Exp

lain

ed

va

ria

bili

ty [

%]

1.1

%

CCA: COMPARISON OF MEASURED PH AND

CALCULATED EIV FOR SOIL REACTION

2.0

%

2.0

%

0

1

2

3

4

5

real pH Ellenberg reaction

Exp

lain

ed

va

ria

bili

ty [

%]

EIVS CORRELATED WITH DCA SCORES

vegetation

Circularity of

reasoning

correlation

Species

composition

sample scores

on DCA axis

Calculated

mean EIV

Ecological

knowledge

(Ellenberg)

MEAN EIVS CORRELATED WITH DCA SCORES

-2 -1 0 1 2

-2

-1

0

1

2

DCA1

DC

A2

Light Temp

Cont Moist

Nutr

React

DCA1 DCA2

Light +++ +++

Temp ++ +++

Cont ++ +++

Moist - - - n.s.

Nutr - - - n.s.

React - - - n.s.

Tab.: significance of

Pearson’s correlation

coefficient

MEAN EIVS CORRELATED WITH DCA SCORES

sample scores on DCA axis

mean E

IV

information

about

compositional

similarity

MEAN RANDOMIZED EIV CORRELATED

WITH DCA SCORES

Mean randomized EIV

inherits information about compositional similarity among plots

carry no ecological information

more than 50% are significantly (p < 0.05) correlated with the first DCA axis!

mean EIV

mean randomized EIV

random variable

0

10

20

30

40

50

60

DCA1 DCA2 DCA3 DCA4

Sig

nific

ant corr

ela

tions [%

]

REGRESSION OF SPECIES RICHNESS ON MEAN

EIVS

10

20

30

40

50

2 3 4 5 6

Nu

mb

er

of

sp

ecie

s

Mean EIV for soil reaction

R2 = 0.30

p < 0.001

information

about

compositional

similarity

REGRESSION OF SPECIES RICHNESS ON MEAN

EIVS

10

20

30

40

50

2 3 4 5 6

Nu

mb

er

of

sp

ecie

s

Mean EIV for soil reaction

R2 = 0.30

p < 0.001

mean EIV

mean randomized EIV

random variable

Almost 40% of

significant

regressions !

0

10

20

30

40

50

Species richness

Sig

nific

ant re

gre

ssio

ns [%

]

Dependent variable: species richness

Explanatory variables: mean EIV

measured variables

USE OF MEAN EIVS IN REGRESSION AND

CLASSIFICATION TREES

Moist <> 5.82266

React <> 4.19643

14.4 7 obs

1 Nutr <> 5.02273

SOILDPT <> 1.325

RALTRIV <> 0.6

29.7 6 obs

2

24 6 obs

339.6 5 obs

4

pH.H <> 4.265

ASPSSW <> 80

RELPOS <> 0.5

21.7 7 obs

5

26.1 7 obs

6

sute <> 0.5

22.6 5 obs

7

17.7 6 obs

8

28.1 9 obs

9

Moist <> 6.54378

35.7 6 obs

10

41.3 7 obs

11

REGRESSION TREES – VARIABILITY EXPLAINED

BY MEAN RANDOMIZED EIV

0

5

10

15

20

Expla

ined v

ariabili

ty [%

]

mean randomized EIV

1 2

random variable

1 2

SUMMARY

mean Ellenberg indicator values inherits information about

compositional similarity among plots

use in CCA (as explanatory variables)

circularity of reasoning

unrealistically high explained variability

use in DCA (correlation with DCA axis)

circularity of reasoning less obvious, but still present

unrealistically high correlation coefficients

~ 50 % probability of significant result even in case of no

ecological meaning

SUMMARY

correlation with species richness

unrealistically high correlation coefficients and higher probability

of significant results

use in regression trees

when mixing mean EIVs with measured variables, mean EIVs will

perform as better predictors

unrealistically high explained variability

randomized values

mean randomized EIV

REGRESSION OF MEAN EIV WITH 1ST AXIS OF DCA MODIFIED MONTE-CARLO PERMUTATION TEST

R2

De

nsity

0

10

20

30

40

50

60

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18

0

10

20

30

40

50

60

0.112

0.166

R2 threshold

for p < 0.05

Monte-Carlo distribution of R2

0

20

40

60

80

100

Temp Cont Light Moist Nutr React

Sig

nific

an

t re

su

lts [

%]

REGRESSION OF MEAN EIV WITH 1ST AXIS OF DCA MODIFIED MONTE-CARLO PERMUTATION TEST

mean EIV

mean randomized EIV

Data: dataset 2 – 100 plots

randomly selected from database

CONCLUSIONS

for any analysis with mean EIV: be careful with testing the

significance of relationship

for DCA: do not test the significance of correlation between mean

EIV and plot scores on DCA axes - or use modified Monte-Carlo test

for correlation with species richness or other vegetation-derived

variable: expect unrealistically high correlation coefficient and higher

probability of getting significant result

for regression and classification trees: do not mix mean EIV with

measured variables, if dependent variable is derived from species

composition (species richness, classification)

ACKNOWLEDGEMENT

to Lubomír Tichý, Milan Chytrý and Ching-Feng Li from Department of Botany &

Zoology, Masaryk University, for comments and recommendations

this study was supported by long-term research plan MSM 0021622416

Thank you for your

attention!