Community Assembly

1
Community Assembly

• A pervasive theme in community ecology is that
  the species composition of a community is
  governed by deterministic assembly rules
• Typically these rules emphasize the importance of
  interspecific interactions (e.g. niche overlap,
  body size distributions)

2
Community Assembly

• In this section we will focus on assembly rules
  that predict the presence or absence of
  particular species combinations

3
Community Assemblylaboratory evidence?

• The best evidence laboratory studies
• Gilpin et al. (1986) examined the structure of
  Drosophia communties.
• When communities were established with 10 (of 28
  considered) species, the subsequent stable
  community was always fewer than four species
• 2101024 initial combos, lt12 persist

4
Community Assemblyfield studies

• Ant communities in Florida mangroves
• Two primary species, limited by island size
  formed a checkerboard pattern
• Two secondary species, limited by presence of
  primary species

5
Diamonds Assembly Rules

• Diamond (1975) popularized the study of community
  assembly in a detailed account of the
  distribution of 141 land-bird species on New
  Guinea and its satellite islands in the Bismark
  Archipeloago

6
Diamonds Assembly Rules

• 1) if one considers all the combinations that
  can be formed from a group of related species,
  only certain ones of these combinations exist in
  nature
• 2) these permissible combinations resist
  invaders that would transform them into forbidden
  combination

7
Diamonds Assembly Rules

• 3) a combination that is stable on a large or
  species-rich island may be unstable on a small or
  species-poor island
• 4) on a small or species-poor island a
  combination may resist invaders that would be
  incorporated on a larger or more species-rich
  island

8
Diamonds Assembly Rules

• 5) some pairs of species never coexist, either
  by themselves or as part of a larger
  combination
• 6) some pairs of species that form an unstable
  combination by themselves may form part of a
  stable larger combination
• 7) some combination that are composed entirely
  of stable sub-combinations are themselves
  unstable

9
Diamonds Assembly Rules

• Although not explicitly stated, the rules infer
  competition (forbidden combinations)
• Some of the rules are so general it has been very
  difficult to make them operational

10
Diamonds Assembly Rules

• In 1979, Conner and Simberloff attacked Diamonds
  study suggesting Rules 2,3,4,6, and 7 were either
  tautologies or restatements or other rules
• Rules 1 and 5 are identical, just differing on
  related species

11
Diamonds Assembly Rules

• Rule 5 describes a chekerboard pattern of species
  occurrences, which is perhaps the simplest of
  Diamonds assembly rules.
• The rule for a complete checkerboard pattern is
  very stringent two species may never co-occur
  (99 of 100)

12
Diamonds Assembly Rules

• Checkerboard distribution of two Macropygia
  cuckoo-dove species in the Bismarck Archipelago

13
Diamonds Assembly Rules

• But, is it really that surprising?
• With (2141) 9,870 possible species pairs, 7
  pairs showing exclusive distributions may not be
  surprising
• Because Diamond did not publish original data,
  Conner and Simberloff used other data

14
Null Assembly ModelsR-mode analyses

• They constrained the observed presence-absence
  matrix subject to the following three
  constraints
• 1) row totals of RMatrix were maintained
• Constraint maintains the differences between
  species in their frequency of occurrence

15
Null Assembly Models

• 2) column totals of the RMatrix were maintained
• Constraint maintained differences among islands
  in the number of species they contained

16
Null Assembly Models

• 3) for each row, species occurrences were
  restricted to those islands for which total
  species richness fell within the range occupied
  by the species
• Constraint maintained the observed incidence
  function for each species (it could not occur in
  assemblages larger or smaller than those observed)

17
Null Assembly Models

• Although the constraints were too much for
  matrices with a large number of widespread
  species, recent advances in randomization
  algorithms have overcome this shortcoming

18
Null Assembly ModelsConnor and Simberloff
19
Null Assembly ModelsMatthews

• Matthews (1982) analyzed the occurrence of 13
  minnow species distribution in six streams of the
  Ozark watershed
• Although some species pairs that never
  co-occurred in watershed were morphologically and
  ecologically similar, the observed number of
  checkerboard pairs matched the predictions of the
  null model (although assumed binominal
  distribution)

20
Null Assembly ModelsCriticisms

• The dilution effect because CS analyzed
  confamilial groups or entire avifaunas,
  competitive effects were no apparent
• Diamonds choice of examples suggested that the
  ecological guild was the correct unit of measure
  (although guild identification is not always easy
  or apparent)

21
Null Assembly Models

• For example, Graves and Gotelli (1993) tested the
  significance of checkerboard distributions in
  mixed-species flocks of Amazonian forest birds
• Results No difference for the entire assemblage
  of flocking or for guilds
• Only a difference when analysis was restricted to
  congeneric species within feeding guilds

22
Table 7.3
23
Null Assembly ModelsCriticisms

• Effects of randomization constraints the 3
  constraints of CS were severe and made it less
  likely that the null hypothesis would be rejected
• For example, relaxing the incidence function
  constraint, the New Hebrides matrix revealed a
  significant negative association

24
Null Assembly ModelsCriticisms

• Also, does the assumptions of CS have their
  flaws? What if the incidence frequency is
  actually influenced by competition? (or some
  other force)
• How would you test this?
• Compare archipelagoes with varying numbers of
  competitors and see if their occurrence frequency
  varies

25
Null Assembly ModelsCriticisms

• Also, some have claimed that is circular to
  constrain marginal totals, because the marginals
  also reflect interspecific competition
• If true, a separate analysis for determining the
  total number of island occurrences is a separate
  hypothesis and requires a separate null model
  (however, competition may not be only factor in
  island distribution)

26
Null Assembly ModelsCriticisms

• Should marginal constrains be incorporated into
  null model at all?
• View 1 co-occurrence patterns are nonrandom,
  given the observed sample of species and
  islands (appropriate)
• View 2 the randomization is viewed as a model of
  community colonization in the absence of
  competition (not appropriate)

27
Null Assembly ModelsCriticisms

• Significance tests CS compared the observed and
  expected distributions with a chi-squared test
• May not be appropriate due to constraints of
  marginal totals (non-linear)

28
Other Null Models

• Wright and Biehl (1982) suggested a
  shared-island test for detecting unusual
  species co-occurrences
• For each species pair, they calculated the tail
  probability of finding the observed number of
  co-occurrences, but with RC transposed

29
Wright and Biehl (1982)

• Advantage directly pinpoints particular species
  pairs that show aggregated or segregated
  distributions (however a few pairs can unduly
  influence statistics)
• Problem assumes all sites are equivalent, thus
  confounds species-site associations with the
  effects of species interactions

30
Analyzing /- Matrix

• Two modes of analysis Q-mode and R-mode
• Q-mode analysis assesses the similarity of
  different columns, indicating how similar sites
  are in the species they contain
• R-mode compares the rows of the matrix and
  indicates how similar species are in the set of
  islands they occupy

31
Q-mode biogeograpy

• How to quantify the degree of similarity between
  2 islands?
• Biogeographers have developed such tools as
  Jaccards Index (0-1)
• J Nc / (N1 N2 Nc)
• But it lacks a statistical distribution. So what?
• What would your null model be?

32
A simple colonization model (0)

• Johnson (unpublished 1974 presentation) used the
  number of shared species as a simple index of
  similarity between sites and then asked what
  should be the number of shared species under the
  simplest colonization model (Null 0)
• Ess mn / P
• (Two islands with m n species, P in the
  equiprobable source pool)

33
Small-island Limitation (Null 1)

• Habitat availability might be responsible for the
  fact that most sites shared more species than
  expected compared with Null Hypothesis 0
• In particular, species may be missing from small
  islands (lacking appropriate habitat)
• Ess mn / Pn (where Pn is of sp. in pool of
  larger island (mn)

34
Island Limitation

• There could also be a size restriction, but from
  the other direction
• Islands could be too big, not allowing for
  supertramp species to persist
• To incorporate this constraint, you could limit
  your source pool to only those species which
  occur on islands of a particular size or larger

35
Island Limitation

• The probability of occurrence is influenced by
  community size, island area, or attributes (e.g.
  distance) and can be incorporated as an
  incidence function

36
Nonrandom Dispersal (Null II)

• Null 0 assumes colonization is identical
• If colonization is stochastic, species still
  would be expected to occur at different
  frequencies on islands because they differ in
  their abilities to disperse and persist
• However, the attributes related to disperal and
  persistence (body size, population size,
  geographic range size) are difficult to assess

37
Nonrandom Dispersal (Null II)

• What to do?
• So one option is to use the occurrence
  distribution to weight species (circular?)
• However, marginal constraints do not determine
  the occurrence pattern itself
• Constraints can be absolute or probabilistic
  (more later)

38
Problems with Q-mode

• Competition may not be being assessed as pairwise
  island comparisons because many are between
  islands that have the same species sets.
  Consequently, it would fail to detect a
  significant checkerboard effect
• Second, because the pairs of islands are not
  independent, it is not appropriate to ask whether
  more than 5 of the pairs are significantly
  different from expectation

39
Summary of Q- and R-mode

• Q-mode appears strong to test for biogeographic
  grouping (similarities)
• R-mode is better to assess species interactions
  (i.e. competition) at sites shared in common

40
Gilpin and Diamond

• Gilpin and Diamond (1982) developed their own
  R-mode analysis
• For species i on island j, they calculated the
  probability of occurrence as
• Pij RiCj / N
• Where R is the row total for species i and C is
  the column total for island j and N is the grand
  total

41
Gilpin and Diamond

• Next, they calculated the expected overlap for
  each species pair by summing the product of these
  probability across all islands
• Observed and expected overlaps for each species
  pair were standardized and then compared with a
  chi-squared test

42
Gilpin and Diamond

• If the null hypothesis of independent placement
  were true, the histogram of normalized deviates
  would follow a normal distribution with unusual
  aggregation at the right and unusual segregation
  at the left
• Upon re-testing the New Hebridean birds, no new
  differences were found

43
Gilpin and Diamond

• However, the original Bismarck data, they found a
  strong excess of positive association and a weak
  excess of negative associations (but overall
  placed less emphasis on competitive interactions
  dictating community structure)
• Importance introduced idea that marginal totals
  (min max) were expected values, not absolute
  constraints

44
Gilpin and Diamond

• How? In different runs of a stochastic model, we
  would not expect each island to support precisely
  the observed number of species, or each species,
  to always occur with its observed frequency.
• In fact, putting a cap on species numbers could
  be interpreted as a competitive cap or limit
• Instead, islands are treated as targets
  independently by species with some variance about
  the expected species number in the null model

45
Summary of R-mode Analysis

• The controversy over R-mode analysis reduces to
  four issues
• 1) Which species and which islands should be
  analyzed?
• Issues such as source pools, colonizations
  potential, habitat availability should be
  considered before any analysis is conducted

46
Summary of R-mode Analysis

• 2) Which metric should be used?
• What is the proper way to quantify nonrandomness
  and species associations in the /- matrix
• Since there are many different kinds of
  structure in a /- matrix, we will utilize five
  different metrics

47
Summary of R-mode Analysis

• 2) Which metric should be used?
• A) the number of species combinations
• If assembly rules are operative, there should be
  fewer species combinations than expected
• B) the number of checkerboard distributions
• Is the most testable of the Diamond Rules and
  represent the strongest form of species
  competition (complete species repulsion)

48
Summary of R-mode Analysis

• 2) Which metric should be used?
• C) the checkerboardness index of Stone and
  Roberts (1990)
• Measures the overall tendency for species pairs
  to co-occur. May reveal competitive pairs, but
  not occuring in a perfect checkerboard
• D) the togetherness index of Stone and Roberts
  (1992)
• Measures overall tendency of species to co-occur
  (although both positive and negative are
  possible)

49
Summary of R-mode Analysis

• 2) Which metric should be used?
• E) Schulters Variance Ratio (1984)
• A modified version of checkerboardedness this
  measure does not constrain column totals
• Different patterns of negative covariation may be
  revealed by comparing the variance ratio to null
  model predictions

50
Summary of R-mode Analysis

• 3) Which simulation procedure should be used?
• If we accept that CS were correct in that
  neither islands nor species are equiprobable,
  this should be reflected in the null model
• Connor and Simberloff (RxC too constrained)
• Gilpin and Diamond (RxC expectations)

51
Summary of R-mode Analysis

• 3) Which simulation procedure should be used? Two
  alternatives
• Gotelli and Graves (R total fixed, C totals
  probabilistic)
• Observed frequency of each species is fixed and
  sites are treated as targets the probability
  of occurrence of each sp. at each site is
  proportional to the total number of species at
  that site. Thus S will vary, but will on average,
  be arranged similarly to observed rankings

52
Summary of R-mode Analysis

• 3) Which simulation procedure should be used? Two
  alternatives
• Gotelli and Graves (RxC totally probabilistic)
• Less constrained, allows R and C to be
  probabilistic). The placement is not simulated,
  but rather N species occurrences across the
  entire matrix
• Specifically, it the cell probability that a
  species (Ri/N) and selecting the site (Cj/N) will
  occur simultaneously thus the cell probability
  is (RiCj / N2), hence the most likely occurrence
  will be the most common species on the most
  species-rich island and vice-versa

53
Grouping Options

• In EcoSim, you can group by Guild (row)
• This analysis expects a data matrix in which each
  species is classified into a single guild
• Guild designations are in the second column
• The simulation reshuffles the guild labels to
  different species (e.g. reorganizes guilds)

54
Grouping Options

• You can also group by Region (columns)
• Region designations are given in the second row
  of the matrix
• The simulation does not alter the structure of
  the matrix, but reshuffles the region labels
  among the different sites

55
Grouping Options

• Another option in the Guild Analysis for EcoSim
  is that of favored states.
• This approach tests the hypothesis of Fox that
  species are added sequentially to a community so
  that different functional groups or guilds are
  represented as evenly as possible

56
Favored States

• Communities are classified as favored or
  unfavored.
• EcoSim reshuffles the guild labels then examines
  each column of the matrix and designates it as
  favored or unfavored

57
(No Transcript)
58
Incidence Functions

• Concept introduced by Diamond (1975) to describe
  the probability of occurrence of a species with
  respect to ordered site characteristics, such as
  species number

59
Incidence Functions

• The x-axis is the number of species on the island
  and the y axis is the proportion of islands in a
  given size class that were occupied by the species

60
Incidence Functions

• High-S species occurred mostly on large,
  species-rich islands, whereas the much less
  common supertramp species showed the opposite
  pattern

61
Incidence Functions

• Gilpin and Diamond (1981) explored the connection
  between the incidence function and the
  equilibrium theory of island biogeography
• The IF represents the time that a species
  occupies islands of a particular size class
  (early succession species occur briefly)
• Paradigm was a lack of competition with each
  species having a species-specific colonization
  and extinction rates

62
Incidence Functions

• The IF may also simply reflect the distribution
  of habitat types among islands
• For example, high-S species may be habitat
  specialists and those specialized habitat may
  only exist on larger islands

63
Incidence Functions

• We can use null models to clarify what the proper
  interpretations of the IF should be
• Whittam and Siegel-Causey (1981) examined Alaskan
  seabird colonies using IF

64
Incidence Functions

• They found examples of both high-S species (CM)
  as well as supertramps (GWG)

Frequency of Occurrence
Species Richness
65
Incidence Functionsother implications

• IF analysis can be used to identify unusual
  minimum area requirements for particular species
• Just looking at the charts may not be enough as
  small islands may be missing certain species due
  to the small likelihood of random settlement

66
Incidence Functions

• Schoener and Schoener (1983) expanded Diamonds
  IF idea to go beyond island area or species
  richness.
• One can order sites by any number of criteria,
  and then the occurrence of species tested against
  this ordering (i.e. Mann-Whitney U test a
  measure of the strength of ordering)

67
Example

• Schoener and Schoener examined 76 species of
  birds on 521 small islands in the Bahamas (as
  well as other vertebrate groups)
• They also measured area, isolation, habitat
  availability and vegetation structure

68
Occurrence Sequence

• Lizards are perfectly ordered
• Resident birds are highly structured
• Migrant birds are more haphazard

69
Results

• Species occurrences were predictable, although
  different groups followed different assembly rule
• Lizards and resident birds were ordered with
  respect to island area, migrant birds were more
  related to island isolation
• The occurrence of both lizards and birds cold be
  predicted by vegetation and habitat structure

70
Implications

• Some checkerboards will only be detected when
  habitat differences among sites are measured and
  incorporated into the analysis
• When distributions of species are with respect to
  site characteristics, the less the patterns will
  conform to a simple checkerboard pattern
• An alternative is nested species patterns