partitioning {ecospace}

Use Partitioning Rule to Simulate Ecological Diversification of a Biota.

Description

Implement Monte Carlo simulation of a biota undergoing ecological diversification using the partitioning rule.

Usage

partitioning(
  nreps = 1,
  Sseed,
  Smax,
  ecospace,
  method = "Euclidean",
  rule = "strict",
  strength = 1
)

Arguments

nreps

Vector of integers (such as a sequence) specifying sample number produced. Only used when function is applied within lapply or related function. Default nreps=1 or any other integer produces a single sample.

Sseed

Integer giving number of species (or other taxa) to use at start of simulation.

Smax

Maximum number of species (or other taxa) to include in simulation.

ecospace

An ecospace framework (functional trait space) of class ecospace.

method

Distance measure to use when calculating functional distances between species. Default is Euclidean using stats::dist. Gower or any other value uses Gower distance (using FD::gowdis). Presence of factor or ordered factor character types forces use of Gower distance.

rule

The partitioning implementation to use in the simulation. Default strict implements the 'minimum distant neighbor' rule; relaxed implements the "maximum nearest neighbor" rule. See 'Details' below for further explanation.

strength

Strength parameter controlling probability that partitioning rule is followed during simulation. Values must range between strength = 1 (default, rules always implemented) and strength = 0 (rules never implemented).

Details

Simulations are implemented as Monte Carlo processes in which species are added iteratively to assemblages, with all added species having their character states specified by the model rules, here the 'partitioning' rule. Simulations begin with the seeding of Sseed number of species, chosen at random (with replacement) from either the species pool (if provided in the weight.file when building the ecospace framework using create_ecospace) or following the neutral-rule algorithm (if a pool is not provided). Once seeded, the simulations proceed iteratively (character-by-character, species-by-species) by following the appropriate algorithm, as explained below, until terminated at Smax.

Partitioning rule algorithm: Measure distances between all pairs of species, using Euclidean or Gower distance method specified by method argument. Use either of the following rules to identify the position of each additional species.

strict (minimum distant neighbor) rule

Identify the maximum distances between all pairs of species (the most-distant neighbors); the space to be partitioned is the minimum of these distances. This implementation progressively fills in the largest parts of the ecospace that are least occupied between neighboring species, and eventually partitions the ecospace in straight-line gradients between seed species.

relaxed (maximum nearest neighbor) rule

Identify nearest-neighbor distances between all pairs of species; the space to be partitioned is the maximum of these distances. This implementation places new species in the most unoccupied portion of the ecospace that is within the cluster of pre-existing species, often the centroid.

In both rules, each new species is created as a resampled combination of the character states of the identified neighbors. If multiple pairs meet the specific criteria, one of these pairs is chosen at random. Ordered, multistate character partitioning (such as ordered factors or order numeric character types) can include any state equal to or between the observed states of existing species. The probability of following the partitioning rule is determined by the strength parameter. Default strength = 1 always implements the rule, whereas strength = 0 never implements it (essentially making the simulation follow the neutral rule.)

Each newly assigned character is compared with the ecospace framework (ecospace) to confirm that it is an allowed state combination before proceeding to the next character. If the newly built character is disallowed from the ecospace framework (i.e., because it has "dual absences" [0,0], has been excluded based on the species pool [weight.file in create_ecospace], or is not allowed by the ecospace constraint parameter), then the character-selection algorithm is repeated until an allowable character is selected. When simulations proceed to very large sample sizes (>100), this confirmatory process can require long computational times, and produce "new" intermediate species that are functionally identical to pre-existing species. This can occur, for example, when no life habits, or perhaps only one, exist that forms an allowable intermediate between the selected neighbors.

Partitioning rules tend to produce ecospaces displaying linear gradients between seed species (in the strict implementation) or concentration of life habits near the functional centroid (in the relaxed implementation). Additional details on the partitioning simulation are provided in Novack-Gottshall (2016a,b), including sensitivity to ecospace framework (functional trait space) structure, recommendations for model selection, and basis in ecological and evolutionary theory.

Value

Returns a data frame with Smax rows (representing species) and as many columns as specified by number of characters/states (functional traits) in the ecospace framework. Columns will have the same data type (numeric, factor, ordered numeric, or ordered factor) as specified in the ecospace framework.

Note

A bug exists within FD::gowdis where nearest-neighbor distances can not be calculated when certain characters (especially numeric characters with values other than 0 and 1) share identical traits across species. The nature of the bug is under investigation, but the current implementation is reliable under most uses. If you run into problems because of this bug, a work-around is to manually change the function to call cluster::daisy using metric = "gower" instead.

The function has been written to allow usage (using lapply or some other list-apply function) in 'embarrassingly parallel' implementations in a high-performance computing environment.

Author(s)

Phil Novack-Gottshall [e-mail]

References

Bush, A. and P.M. Novack-Gottshall. 2012. Modelling the ecological-functional diversification of marine Metazoa on geological time scales. Biology Letters 8: 151-155.

Novack-Gottshall, P.M. 2016a. General models of ecological diversification. I. Conceptual synthesis. Paleobiology 42: 185-208.

Novack-Gottshall, P.M. 2016b. General models of ecological diversification. II. Simulations and empirical applications. Paleobiology 42: 209-239.

See Also

create_ecospace, neutral, redundancy, expansion

Examples

# Create an ecospace framework with 15 3-state factor characters
# Can also accept following character types: "numeric", "ord.num", "ord.fac"
nchar <- 15
ecospace <- create_ecospace(nchar = nchar, char.state = rep(3, nchar),
  char.type = rep("factor", nchar))

# Single (default) sample produced by partitioning function (with strength = 1 and
#   "strict" partitioning rules):
Sseed <- 5
Smax <- 40
x <- partitioning(Sseed = Sseed, Smax = Smax, ecospace = ecospace, rule = "strict")
head(x, 10)

# Plot results, showing order of assembly
# (Seed species in red, next 5 in black, remainder in gray)
# Notice the 'strict' partitioning model produces an ecospace with life-habit gradients
#   between seed species
seq <- seq(nchar)
types <- sapply(seq, function(seq) ecospace[[seq]]$type)
if(any(types == "ord.fac" | types == "factor")) pc <- prcomp(FD::gowdis(x)) else
  pc <- prcomp(x)
plot(pc$x, type = "n", main = paste("Partitioning model,\n", Smax, "species"))
text(pc$x[,1], pc$x[,2], labels = seq(Smax), col = c(rep("red", Sseed), rep("black", 5),
  rep("slategray", (Smax - Sseed - 5))), pch = c(rep(19, Sseed), rep(21, (Smax - Sseed))),
  cex = .8)

# Same, but following "relaxed" partitioning rules:
# Notice the 'relaxed' partitioning model only fills in the ecospace between seed species
x <- partitioning(Sseed = Sseed, Smax = Smax, ecospace = ecospace, rule = "relaxed")
if(any(types == "ord.fac" | types == "factor")) pc <- prcomp(FD::gowdis(x)) else
  pc <- prcomp(x)
plot(pc$x, type = "n", main = paste("Partitioning model,\n", Smax, "species"))
text(pc$x[,1], pc$x[,2], labels = seq(Smax), col = c(rep("red", Sseed), rep("black", 5),
  rep("slategray", (Smax - Sseed - 5))), pch = c(rep(19, Sseed), rep(21, (Smax - Sseed))),
  cex = .8)

# Change strength parameter so rules followed 95% of time:
x <- partitioning(Sseed = Sseed, Smax = Smax, ecospace = ecospace, strength = 0.95, rule = "strict")
if(any(types == "ord.fac" | types == "factor")) pc <- prcomp(FD::gowdis(x)) else
  pc <- prcomp(x)
plot(pc$x, type = "n", main = paste("Partitioning model,\n", Smax, "species"))
text(pc$x[,1], pc$x[,2], labels = seq(Smax), col = c(rep("red", Sseed), rep("black", 5),
  rep("slategray", (Smax - Sseed - 5))), pch = c(rep(19, Sseed), rep(21, (Smax - Sseed))),
  cex = .8)

# Create 5 samples using multiple nreps and lapply (can be slow)
nreps <- 1:5
samples <- lapply(X = nreps, FUN = partitioning, Sseed = 5, Smax = 40, ecospace)
str(samples)