Identify sites where species communities have changed in...

Temporal beta diversity:

Identify sites where species communities have changed in exceptional ways

SIBECOL 2019 1st Meeting of the Iberian Ecological Society & XIV AEET Meeting

Universitat de Barcelona, Tuesday 5 February 2019

Pierre Legendre Département de sciences biologiques

Université de Montréal, Canada http://www.numericalecology.com/

© Pierre Legendre 2019

Outline of the presentation

1. Objective: compare two surveys 2. The B and C components of D 3. Temporal beta diversity indices (TBI) 4. Test of the TBI indices – See the published paper (2019) 5. Summary of results from simulation study – See the published paper (2019) 6. The BCI Forest Dynamics plot

6.1. Preliminary spatial analysis 6.2. Temporal changes at BCI 6.3. A space-time interaction at BCI? 6.4. BCI habitat map 6.5. TBI indices in the swamp 6.6. B-C analysis using the TBI() function 6.7. The B-C plots 6.8. Changes in species composition from T1 to T2

7. B-C analysis: What did we learn? 8. Other TBI studies, published or submitted 9. References

Temporalbetadiversity 2

1. Objective: compare two surveys

Ecologists and geneticists often want to compare observations at several sites, repeated at two different times: T1 and T2.

In landscape ecology and genetics, the data are community composition or population gene frequencies observed at different sites and different times.

The data we have experimented with are community composition data from terrestrial and marine ecosystems.

We can refer to the variation in community composition across time as temporal beta diversity.


The general question of interest to ecologists is:

In what ways do the observations at each site differ between surveys T1 and T2?

Two particular questions are of interest:

• Are there sites with exceptional changes in community composition from T1 to T2?

• Changes in community composition can be quantified as species losses and gains. How do the species losses and gains, from T1 to T2, compare among the sites?


When comparing two surveys of a site at times T1 and T2, the relevant information is found in the two community composition vectors, containing species presence-absence or abundance data.

In species-rich communities, these vectors contain too much information to be easy to grasp. Ecologists often crunch the information into a dissimilarity index computed from the two vectors.

The method of analysis has two parts:

1. Estimate the change at each site between T1 and T2 using an appropriate dissimilarity index, and test the significance of that change.

2. Find ways of partitioning the dissimilarity information into finer indices of losses and gains of species, which may tell us something about the processes at work in the system.


2. The B and C components of D Squares

represent species

a b c

Dissimilarity = b+c

Componentsof DJ and DS

Time 1

Time 2

Jaccard dissimilarity index: DJ = (b + c)/(a + b + c) Sørensen dissimilarity index: DS = (b + c)/(2a + b + c) The b and c statistics in the numerators of these indices decompose presence-absence data into species losses (b) and gains (c) in the interval from T1 to T2.

Computing a D index from presence-absence data at a site:

6

Spec.1 Spec.4Spec.3Spec.2Time1 Time2 Time1 Time2 Time1 Time2 Time1 Time2

A = abun. common to times 1 and 2 = A1+A2+0+A4B = abundances unique to time 1 = 0+B2+B3+0C = abundances unique to time 2 = C1+0+0+0

A1 A1 A2 A2

A4 A4B2

B3

C1

Computing a D index from community abundance data at a site:

Ružička dissimilarity: DR = (B + C)/(A + B + C) Percentage difference (aka Bray-Curtis): D%diff = (B + C)/(2A + B + C)

The numerators of these indices decompose abundance data into species losses (B) and gains (C) in the interval from T1 to T2.

7

In summary, the decomposition of the numerators is into … • losses (b) and gains (c) for species presence-absence data, • or losses (B) and gains (C) of individuals-per-species for abundance data. These components draw attention to the asymmetry implicit in a comparison between times T1 and T2. When comparing sites spread on a map, no asymmetric process is involved. The distinction between loss and gain is not necessary. The Baselga and Podani indices, developed to estimate the replacement and richness/abundance differences in spatial analysis, do not emphasize the time asymmetry. They have a different role in spatial beta diversity analysis. In temporal studies, however, comparison of the loss (B) and gain (C) components is important and informative.

8

3. Temporal beta diversity indices (TBI)

Summary of previous section The dissimilarity (D) of the community composition between two observations of a site (vectors for T1 and T2) can be computed using the following dissimilarity indices, which will be used as TBI.

For abundance data:

• the Ružička index DR = (B + C) /(A + B + C)

• the percentage difference index D%diff = (B + C) /(2A + B + C).

For presence-absence data:

• the Jaccard index: DJ = (b + c) /(a + b + c) • the Sørensen index: DS = (b + c) /(2a + b + c)

9

We will compute the dissimilarity (TBIi = Di) between vector T1i in Mat.1 and vector T2i in Mat.2 for each site i :


Are there sites where the T1–T2 differences are very important?

• A test of significance was developed to identify these sites.

• The test was validated in a simulation study.

See the published paper (Legendre, Ecology and Evolution, 2019)

4-5. Test of the TBI indices


6. The BCI Forest Dynamics plot

Basic data about the BCI Forest Dynamics Plot

l  Country: Panama l  Location: Barro Colorado Island in Lake Gatun in the Panama canal l  Coordinates: 09°09’ N, 79°51’ W l  Plot size: 50 ha (1000 × 500 m), divided in 1250 (20 × 20 m) quadrats l  Average precipitation: 2581 mm / yr l  Average temperature: 27°C l  120 – 160 m above sea level l  8 surveys: 1982+, 1985, 1990, 1995, 2000, 2005, 2010, 2015 l  Number of tree species in the 8 censuses combined: 325 l  207 719 trees ≥ 1 cm DBH in the 2015 survey (297 species) l  In the 1985 census,

Ø  Mean number of species per 20 × 20 m quadrat: 53.5 Ø  Range of values of n. species per 20 × 20 m quadrat: [20, 84]


Local contributions to beta diversity (LCBD indices; Legendre & De Cáceres, 2013) were computed across the BCI forest plot (1250 quadrats, 20 × 20 m) for the 1985 and 2015 surveys.

=>

Temporalbetadiversity

6.1. Preliminary spatial analysis

14

PLCBD ≤ 0.05 Padjusted ≤ 0.05

0 200 400 600 800 1000

0100

200

300

400

500

LCBD indices, BCI 1985, 20x20

Easting (m)

Nor

thin

g (m

)Map of LCBD indices at the scale of (20 m × 20 m) quadrats

15

What is special at the quadrats that have significant LCBD?

0 200 400 600 800 1000

0100

200

300

400

500


Easting (m)

Nor

thin

g (m

)

Secondary (younger) forest ~ 100 years old

LCBDindices,BCI1985,20x20

Swamp

Clif

f St

ream

The swamp near the centre of BCI (photo P. Legendre, 04 February 2015). Community changes in the swamp: answer in the next section of the talk. 17

PLCBD ≤ 0.05 Padjusted ≤ 0.05

0 200 400 600 800 1000

0100

200

300

400

500


Easting (m)

Nor

thin

g (m

)

Why is it that the swamp quadrats no longer have significant LCBD indices in 2015?

1985 1990 1995 2000 2005 2010 2015

200000

220000

240000

Change in number of tree stems across censuses

Censuses

N.tree.stems

Changes in BCI tree communities have taken place along the years.

6.2. Temporal changes at BCI

From 1985 to 2015, 34343 tree stems (or 14.9%) were lost at BCI. 19

An anova method to test a space-time interaction in the absence of replication was developed by Legendre, De Cáceres & Borcard (2010). It is implemented in function stimodels() of the adespatial package in R.

Data from four BCI census were analysed as a test case in that paper, from 1982-83 to 1995.

A significant space-time interaction was reported across the four censuses –

• for the whole community of 315 species in the (20 × 20 m) quadrats

• and for 43% of the individual species tested separately.

A significant space-time interaction means that the spatial structure of the data has changed significantly among the censuses (multivariate or univariate data).

6.3. A space-time interaction at BCI?

20

We checked that this was still the case for the 7 censuses conducted between 1985 and 2015.

Indeed, significant space-time interactions were detected for the (20 × 20 m) and the (100 × 100 m) quadrats.

We will now study these differences in more detail between censuses #2 (1985) and #8 (2015) for the whole plot, then for each habitat type separately.


0 200 400 600 800 1000

0100

200

300

400

500

Habitat map, BCI, 20x20

Easting (m)

Nor

thin

g (m

)

Group 1Group 2

Group 3

Group4

Group 5

Gr 5Group 6

Gr 7

Gr 7

Harms et al. (2001) BCI habitat classification Group 1 – Old forest, Low plateau Group 2 – Old forest, High plateau Group 3 – Old forest, Slope Group 4 – Old forest, Swamp Group 5 – Old forest, Streamside Group 6 – Young forest Group 7 – Mixed habitats

6.4. BCI habitat map

..

In the swamp (30 quadrats), two quadrats (circles with red rims) had significant TBI indices between the 1985 and 2015 censuses, after Holm correction for 30 simultaneous tests. Three more quadrats (circles with green rims) were significant before the Holm correction but not after. Circle radii are proportional to TBI values.

TBIvaluesGold:C>B(gains)White:B=CGrey:B>C(losses)Redrim:p.adj≤0.05Greenrim:p.TBI≤0.05,p.adj>0.05Blackrim:notsignificant

6.5. TBI indices in the swamp

Temporal beta diversity

200 300 400 500

150

200

250

300

350

TBI map 1985-2015, group 4 (swamp), abundance data

Easting (m)

Nor

thin

g (m

)

150m

145m

140m

145m

23

6.6. B-C analysis using the TBI() functionAs an example, we will analyse the changes in the swamp (30 quadrats) between 1985 and 2015.

# Load the TBI functionout <- TBI(BCI.20.2[group.4,], BCI.20.8[group.4,], method="%diff", nperm=9999, test.t.perm=TRUE)

out$BCD.mat B/(2A+B+C) C/(2A+B+C) D=(B+C)/(2A+B+C) ChangeSite.1 0.22135417 0.1276042 0.3489583 – Site.2 0.14325069 0.2231405 0.3663912 + Site.3 0.15151515 0.4377104 0.5892256 + Site.4 0.12334802 0.2687225 0.3920705 + Site.5 0.05230769 0.2769231 0.3292308 + Site.6 0.13095238 0.3452381 0.4761905 + Site.7 0.13962264 0.2490566 0.3886792 + Site.8 0.05500000 0.6150000 0.6700000 + [...]Site.29 0.03896104 0.6926407 0.7316017 + Site.30 0.17064846 0.2559727 0.4266212 +

The function then displays summary statistics:

out$BCD.summarymean(B/den) mean(C/den) mean(D) B/(B+C) C/(B+C) Change 0.137376 0.34327 0.480646 0.2858153 0.7141847 +

A test of significance (paired t-test) of the difference between the vectors of C/den (gains) and B/den (losses) is also computed (parametric and non-parametric p-values):

out$t.test_B.C mean(C-B) Stat p.param p.perm p<=0.05Paired t.test 0.205894 6.031982 1.459043e-06 0.0001 *

The results indicate that for group 4 (30 quadrats), there has been a significant gain of species abundances between the 1985 and 2015 censuses.

Temporal beta diversity 25

out$TBI # The TBI values for each of the n = 30 quadrats of group 4 [1] 0.3490 0.3664 0.5892 0.3921 0.3292 0.4762 0.3887 0.6700 0.3492 0.5111 0.5205 0.4527 0.6597 0.3909 0.5144 0.7453 0.4713 0.3744 0.4500 0.5294 0.6000 0.4776 0.4609 0.3675 0.2764 0.5660 0.4531 0.5301 0.7316 0.4266

out$p.TBI # The associated p-values (permutational test) [1] 0.9579 0.9250 0.0587 0.8662 0.9754 0.4988 0.8713 0.0033 0.9534 0.3202 0.2738 0.6182 0.0051 0.8721 0.3066 0.0001 0.5223 0.9150 0.6429 0.2401 0.0443 0.4948 0.5722 0.9262 0.9975 0.1114 0.6166 0.2272 0.0003 0.7396

out$p.adj # The p-values adjusted for multiple tests (Holm correction)

[1] 1.000 1.000 1.000 1.000 1.000 1.000 1.000 0.092 1.000 1.000 1.000 1.000 0.138 1.000 1.000 0.003 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 0.009 1.000

which(out$p.adj <= 0.05) # Which sites are TBI-significant?[1] 16 29

The output also includes:


0 200 400 600 800 1000

0100

200

300

400

500

Habitat map, BCI, 20x20

Easting (m)

Nor

thin

g (m

)

Group 1Group 2

Group 3

Group4

Group 5

Gr 5Group 6

Gr 7

Gr 7

Harms et al. (2001) BCI habitat classification Group 1 – Old forest, Low plateau Group 2 – Old forest, High plateau Group 3 – Old forest, Slope Group 4 – Old forest, Swamp Group 5 – Old forest, Streamside Group 6 – Young forest Group 7 – Mixed habitats

The BCI habitat map shown again

Note that Group 1 is Gold Group 4 (swamp) is Red

Zone where losses dominate (B > C)

Zone where gains dominate (C > B)

Larger D (D = B+C)

B-C plot for abundance data

Green line: B = C.

The red line goes through the centroid of the points, parallel to the green line.

Here the red line is below the green, showing that losses in species abundances dominated gains in BCI (1250 quadrats) from 1985 to 2015.

0.0 0.2 0.4 0.6

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

B-C plot, 20x20, 1985-2015, abundance data

Species losses (B)

Spe

cies

gai

ns (C

)

6.7. The B-C plots

Presence-absence data: similar B-C plot results

B-C plots for habitats 1–6 (abundances)

Squares: gains dominate.

Circles: losses dominate.

Gains dominate group 4 (swamp) (red line above green).

Losses dominate groups {1, 2, 3, 5, 6} (green line above red).

Pres-abs. data: similar B-C plots

0.0 0.1 0.2 0.3 0.4 0.5

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

B-C plot, abundances, habitat gr. 1

Species losses (B)

Spe

cies

gai

ns (C

)

0.0 0.1 0.2 0.3 0.4 0.5

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7


Species losses (B)

Spe

cies

gai

ns (C

)

0.0 0.1 0.2 0.3 0.4 0.5

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7


Species losses (B)

Spe

cies

gai

ns (C

)

0.0 0.1 0.2 0.3 0.4 0.5

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7


Species losses (B)

Spe

cies

gai

ns (C

)

0.0 0.1 0.2 0.3 0.4 0.5

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7


Species losses (B)

Spe

cies

gai

ns (C

)

0.0 0.1 0.2 0.3 0.4 0.5

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7


Species losses (B)

Spe

cies

gai

ns (C

)

In habitat 4 (the swamp), numbers of species and abundances-per-species increased in most quadrats during that period.

Why?

We know that at BCI, the period 1982–1992 included several extreme dry seasons; since then, there have been few such drought events. High tree growth rates and death rates were observed during the drought but not since then (Condit et al.2017).


We studied in more detail the demogaphic changes at BCI on the quadrats of the swamp habitat in the 1985–2015 interval.

To that end, we analysed the changes in the 209 species found in the swamp in 1985 and 2015, across its 30 quadrats, using paired t-tests.

The tests were carried out with 9999 random permutations of the values, in each quadrat, between T1 and T2. A Holm correction for multiple testing was applied to the resulting p-values.


6.8. Changes in species composition from T1 to T2

31

The first 7 species noticeably increased in abundances.

They also all increased in occurrence.

The following species increased significantly1 in abundances (number of stems) in the quadrats of the swamp, from T1 to T2:

Abund.T1 Abund.T2 Diff(T2-T1)

alibed 10 45 +35eugega 15 111 +96guargu 39 132 +93simaam 21 67 +46tab2ar 49 148 +99tri2tu 132 322 +190xyl1ma 2 38 +36bactma 245 65 –180

1 Holm-corrected p-values ≤ 0.05, 209 tests. Same for the parametric and permutational p-values.

In th

e sw

amp

Abund.T1 Abund.T2 Diff(T2-T1)

alibed 160 193 +33Eugega 472 965 +493guargu 383 474 +91simaam 745 910 +165tab2ar 681 830 +149tri2tu 3724 2421 –1303xyl1ma 121 415 +294bactma 16 6 –10

Results for Group 1 (old forest surrounding the swamp): 6 of the 7 species that increased in abundance in the swamp also increased in Group 1 in the same time period.

The most abundant of these 7 species (tri2tu) actually decreased in abundance in Group 1 between 1985 and 2015, whereas it increased in Group 4.

In g

roup

1


Explanation of the change in species composition in Group 4

Species that were not adapted to dry conditions decreased in abundance in Group 1 during the drought. They created gaps in the canopy.

These gaps allowed other species that were adapted to dryer conditions to occupy these gaps and increase in abundance in Group 1.

These species also migrated into the adjacent swamp (Group 4). This was facilitated by the fact that the swamp had become dryer during the drought.


7. B-C analysis: What did we learn?

1. Analysis of the B and C components brings us to the heart of the mechanisms by which communities change through time:

losses (b) and gains (c) of species, losses (B) and gains (C) of individuals of the various species.

2. B-C analysis is especially interesting in species-rich communities where we cannot examine the changes in each species individually.


3. B-C analysis can also be applied to subgroups of sites (e.g. habitat types in BCI), as shown in this presentation.

It could also be applied to specific groups of species that are known to react differently to environmental stressors, e.g. different size classes, or species of different origins, for example: temperate, transitional, boreal trees (Brice et al., 2019).

4. With this talk, I am hoping to encourage ecologists to try this type of analysis.

New ideas of how to use the results will emerge as ecologists experiment with this method on data from different ecosystems.


5. TBI analysis and B-C plots are useful to identify exceptional sites in space-time ecological surveys carried out to study the effects of natural and anthropogenic changes to ecosystems.

Such studies are presently carried out by teams of ecologists around the world. They are collecting data over land, in lakes and in the oceans to assess the effects of climate change on natural communities and other types of biodiversity data.

Researchers would like to identify the sites where important changes have taken place. They can then focus their attention onto these sites and seek what has been going on there, why community composition has changed in an exceptional way at these sites.

The TBI method was designed for this type of research.


8. Other TBI studies, published or in progress

• Marine data: Field experiment on a Pacific atoll (atmospheric test of a Hydrogen bomb in Fangatofa in 1968). Molluscs, 36 species. Legendre & Salvat (2015). Published in Proceedings B.

• Freshwater diatom communities 150 years apart in sediment cores, paleoecology. 169 lakes throughout the USA. Winegardner et al. (2017). Published in Global Ecology and Biogeography.

• Freshwater fish communities in streams across France, 332 sites, 1980-2012. Kuczynski et al. (2018). Published in Global Ecology and Biogeography.

• Brackish data: benthos in Chesapeake Bay, USA. 25 brackish sites, fall surveys of 2005 and 2008 compared, 52 species. In the MS describing the TBI method, Legendre (2019). Published in Ecology and Evolution.


• Forest ecology: Changes in community composition in Québec forests between a historical (1970–1980) and a contemporary (2000–2016) period. Brice et al. (2019).

• Freshwater: Changes in fish community compositions in the St. Lawrence River across five decades. Study in progress.

• Another forest ecology study is ongoing, using surveys done in the Fushan Forest Dynamics plot (25 ha) in Taiwan. We are comparing the surveys completed in 2004 and 2014. Sun, Chen & Legendre (in prep.)


9. References

Brice, M.-H., K. Cazelles, P. Legendre & M.-J. Fortin. 2019. Disturbances amplify tree community responses to climate change in the temperate-boreal ecotone. Global Ecology & Biogeography 28:1668–1681.

Condit, R., Pérez, R., Lao, S., Aguilar, S., Hubbell, S.P. 2017. Demographic trends and climate over 35 years in the Barro Colorado 50 ha plot. Forest Ecosystems 4: 17.

Edgington, E. S. 1995. Randomization tests. 3rd edition. Marcel Dekker, New York. Harms, K.E., R. Condit, S. P. Hubble and R. B. Foster. 2001. Habitat associations of

trees and shrubs in a 50-ha neotropical forest plot. Journal of Ecology 89: 94–959. Kuczynski, L., P. Legendre & G. Grenouillet. 2018. Concomitant impacts of climate

change, fragmentation and non-native species have led to reorganization of fish communities since the 1980s. Global Ecology and Biogeography 17: 213–222.

Legendre, P. 2019. A temporal beta-diversity index to identify sites that have changed in exceptional ways in space-time surveys. Ecology and Evolution 9: 3500–3514. https://doi.org/10.1002/ece3.4984.

Legendre, P. & M. De Cáceres. 2013. Beta diversity as the variance of community data: dissimilarity coefficients and partitioning. Ecology Letters 16: 951–963.


Legendre, P., M. De Cáceres & D. Borcard. 2010. Community surveys through space and time: testing the space-time interaction in the absence of replication. Ecology 91: 262–272.

Legendre, P. & Condit, R. 2019. Spatial and temporal analysis of beta diversity in the Barro Colorado Island forest dynamics plot, Panama. Forest Ecosystems 6: 1–11. https://doi.org/10.1186/s40663-019-0164-4.

Legendre, P. & L. Legendre. 2012. Numerical ecology, 3rd English edition. Elsevier Science BV, Amsterdam.

Legendre, P., & B. Salvat. 2015. Thirty-year recovery of mollusc communities after nuclear experimentations on Fangataufa atoll (Tuamotu, French Polynesia). Proceedings of the Royal Society B 282: 20150750.

Winegardner, A. K., P. Legendre, B. E. Beisner & I. Gregory-Eaves. 2017. Diatom diversity patterns over the past ~ 150 years across the conterminous United States: identifying mechanisms behind beta diversity. Global Ecology and Biogeography 26: 1303–1315.


End of section


Identify sites where species communities have changed in...

Documents

Transcript of Identify sites where species communities have changed in...