Learn R Programming
phenology (version 10.1)

fitRMU: Adjust incomplete timeseries with various constraints.

Description

The data must be a data.frame with the first column being years and two columns for each beach: the average and the se for the estimate.
The correspondence between mean, se and density for each rookery are given in the RMU.names data.frame.
This data.frame must have a column named mean, another named se and a third named density. If no sd column exists, no sd will be considered for the series and if no density column exists, it will be considered as being "dnorm" (Gaussian distribution).
The aggregated number of nests and its confidence interval can be obtained using CI.RMU().
The names of beach columns must not begin by T_, SD_, a0_, a1_ or a2_ and cannot be r.
A RMU is the acronyme for Regional Managment Unit. See:
Wallace, B.P., DiMatteo, A.D., Hurley, B.J., Finkbeiner, E.M., Bolten, A.B., Chaloupka, M.Y., Hutchinson, B.J., Abreu-Grobois, F.A., Amorocho, D., Bjorndal, K.A., Bourjea, J., Bowen, B.W., Dueñas, R.B., Casale, P., Choudhury, B.C., Costa, A., Dutton, P.H., Fallabrino, A., Girard, A., Girondot, M., Godfrey, M.H., Hamann, M., López-Mendilaharsu, M., Marcovaldi, M.A., Mortimer, J.A., Musick, J.A., Nel, R., Seminoff, J.A., Troëng, S., Witherington, B., Mast, R.B., 2010. Regional management units for marine turtles: a novel framework for prioritizing conservation and research across multiple scales. PLoS One 5, e15465.
Variance for each value is additive based on both the observed SE (in the RMU.data object) and a value.
The value is a global constant when model.SD is "global-constant".
The value is proportional to the observed number of nests when model.SD is "global-proportional" with aSD_*observed+SD_ with aSD_ and SD_ being fitted values. This value is fixed to zero when model.SD is "Zero".
The value is dependent on the rookery when model.SD is equal to "Rookery-constant" or "Rookery-proportional" with a similar formula as previously described for "global".
if method is NULL, it will simply return the names of required parameters.

Usage

fitRMU(
  RMU.data = stop("data parameter must be provided"),
  years.byrow = TRUE,
  RMU.names = NULL,
  model.trend = "Constant",
  model.rookeries = "Constant",
  model.SD = "Global-constant",
  parameters = NULL,
  fixed.parameters = NULL,
  SE = NULL,
  method = c("Nelder-Mead", "BFGS"),
  control = list(trace = 1),
  itnmax = c(1500, 1500),
  cptmax.optim = 100,
  limit.cpt.optim = 1e-05,
  hessian = TRUE,
  replicate.CI = 1000,
  colname.year = "Year",
  maxL = 1e+09
)

Value

Return a list with the results from optim and synthesis for proportions and numbers

Arguments

RMU.data

A data.frame with a column Year (the name is defined in colname.year) and one to three columns per rookery defined in RMU.names

years.byrow

If TRUE, the RMU.data data.frame is organized with years in rows

RMU.names

A dataframe with one to three columns indicating name of columns for mean, standard deviation, and distribution for roockeris

model.trend

Can be Constant, Exponential or Year-specific

model.rookeries

Description temporal change in rookeries proportion. It be Constant, First-order or Second-order

model.SD

Can be Zero, Global-constant, Global-proportional or Rookery-constant. See description.

parameters

Parameters to fit

fixed.parameters

Parameters that are fixed

SE

Parameters SE for example from fitRMU_MHmcmc()

method

Methods to be used by optim()

control

List of controls for optim()

itnmax

A vector with maximum iterations for each method.

cptmax.optim

How many times optim can be ran when likelihood is better.

limit.cpt.optim

Limit to consider that likelihood is better.

hessian

If TRUE, the hessian matrix is calculated and then the standard error of parameters.

replicate.CI

Number of replicates to estimate CI of proportion for each rookery

colname.year

Name of the column to be used as time index

maxL

If an error is produced during the estimation of likelihood, replace -Ln L by this value

Author

Marc Girondot marc.girondot@gmail.com

Details

fitRMU is used to estimate missing information when several linked values are observed along a timeseries

See Also

Other Fill gaps in RMU: CI.RMU(), fitRMU_MHmcmc(), fitRMU_MHmcmc_p(), logLik.fitRMU(), plot.fitRMU()

Examples

Run this code
if (FALSE) {
library("phenology")
RMU.names.AtlanticW <- data.frame(mean=c("Yalimapo.French.Guiana", 
                                         "Galibi.Suriname", 
                                         "Irakumpapy.French.Guiana"), 
                                 se=c("se_Yalimapo.French.Guiana", 
                                      "se_Galibi.Suriname", 
                                      "se_Irakumpapy.French.Guiana"), 
                                 density=c("density_Yalimapo.French.Guiana", 
                                           "density_Galibi.Suriname", 
                                           "density_Irakumpapy.French.Guiana"))
data.AtlanticW <- data.frame(Year=c(1990:2000), 
      Yalimapo.French.Guiana=c(2076, 2765, 2890, 2678, NA, 
                               6542, 5678, 1243, NA, 1566, 1566),
      se_Yalimapo.French.Guiana=c(123.2, 27.7, 62.5, 126, NA, 
                                 230, 129, 167, NA, 145, 20),
      density_Yalimapo.French.Guiana=rep("dnorm", 11), 
      Galibi.Suriname=c(276, 275, 290, NA, 267, 
                       542, 678, NA, 243, 156, 123),
      se_Galibi.Suriname=c(22.3, 34.2, 23.2, NA, 23.2, 
                           4.3, 2.3, NA, 10.3, 10.1, 8.9),
      density_Galibi.Suriname=rep("dnorm", 11), 
      Irakumpapy.French.Guiana=c(1076, 1765, 1390, 1678, NA, 
                               3542, 2678, 243, NA, 566, 566),
      se_Irakumpapy.French.Guiana=c(23.2, 29.7, 22.5, 226, NA, 
                                 130, 29, 67, NA, 15, 20), 
      density_Irakumpapy.French.Guiana=rep("dnorm", 11)
      )
                           
cst <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
               colname.year="Year", model.trend="Constant", 
               model.SD="Zero")
cst <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
               colname.year="Year", model.trend="Constant", 
               model.SD="Zero", 
               control=list(trace=1, REPORT=100, maxit=500, parscale = c(3000, -0.2, 0.6)))
               
cst <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
               colname.year="Year", model.trend="Constant", 
               model.SD="Zero", method=c("Nelder-Mead","BFGS"), 
               control = list(trace = 0, REPORT = 100, maxit = 500, 
               parscale = c(3000, -0.2, 0.6)))
expo <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
               colname.year="Year", model.trend="Exponential", 
               model.SD="Zero", method=c("Nelder-Mead","BFGS"), 
               control = list(trace = 0, REPORT = 100, maxit = 500, 
               parscale = c(6000, -0.05, -0.25, 0.6)))
YS <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
             colname.year="Year", model.trend="Year-specific", method=c("Nelder-Mead","BFGS"), 
             model.SD="Zero")
YS1 <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
             colname.year="Year", model.trend="Year-specific", method=c("Nelder-Mead","BFGS"), 
             model.SD="Zero", model.rookeries="First-order")
YS1_cst <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
             colname.year="Year", model.trend="Year-specific", 
             model.SD="Constant", model.rookeries="First-order", 
             parameters=YS1$par, method=c("Nelder-Mead","BFGS"))
YS2 <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
             colname.year="Year", model.trend="Year-specific",
             model.SD="Zero", model.rookeries="Second-order", 
             parameters=YS1$par, method=c("Nelder-Mead","BFGS"))
YS2_cst <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
             colname.year="Year", model.trend="Year-specific",
             model.SD="Constant", model.rookeries="Second-order", 
             parameters=YS1_cst$par, method=c("Nelder-Mead","BFGS"))
               
compare_AIC(Constant=cst, Exponential=expo, 
YearSpecific=YS)

compare_AIC(YearSpecific_ProportionsFirstOrder_Zero=YS1,
YearSpecific_ProportionsFirstOrder_Constant=YS1_cst)

compare_AIC(YearSpecific_ProportionsConstant=YS,
           YearSpecific_ProportionsFirstOrder=YS1,
           YearSpecific_ProportionsSecondOrder=YS2)
           
compare_AIC(YearSpecific_ProportionsFirstOrder=YS1_cst,
           YearSpecific_ProportionsSecondOrder=YS2_cst)

# Example of different types of plots
plot(cst, main="Use of different beaches along the time", what="total", 
     ylim=c(0, 4000))
plot(cst, main="Use of different beaches along the time", what = "proportions", 
     replicate.CI=0)
plot(cst, main="Use of different beaches along the time", what = "numbers", 
     aggregate="model", ylim=c(0, 4000), replicate.CI=0)
plot(cst, main="Use of different beaches along the time", what = "numbers", 
     aggregate="both", ylim=c(0, 11000), replicate.CI=0)
     
plot(expo, main="Use of different beaches along the time", what="total")
plot(YS2_cst, main="Use of different beaches along the time", what="total")

plot(YS1, main="Use of different beaches along the time")
plot(YS1_cst, main="Use of different beaches along the time")
plot(YS1_cst, main="Use of different beaches along the time", what="numbers")

# Gamma distribution should be used for MCMC outputs

RMU.names.AtlanticW <- data.frame(mean=c("Yalimapo.French.Guiana", 
                                         "Galibi.Suriname", 
                                         "Irakumpapy.French.Guiana"), 
                                 se=c("se_Yalimapo.French.Guiana", 
                                      "se_Galibi.Suriname", 
                                      "se_Irakumpapy.French.Guiana"), 
                                 density=c("density_Yalimapo.French.Guiana", 
                                           "density_Galibi.Suriname", 
                                           "density_Irakumpapy.French.Guiana"), 
                                           stringsAsFactors = FALSE)
                                           
data.AtlanticW <- data.frame(Year=c(1990:2000), 
      Yalimapo.French.Guiana=c(2076, 2765, 2890, 2678, NA, 
                               6542, 5678, 1243, NA, 1566, 1566),
      se_Yalimapo.French.Guiana=c(123.2, 27.7, 62.5, 126, NA, 
                                 230, 129, 167, NA, 145, 20),
      density_Yalimapo.French.Guiana=rep("dgamma", 11), 
      Galibi.Suriname=c(276, 275, 290, NA, 267, 
                       542, 678, NA, 243, 156, 123),
      se_Galibi.Suriname=c(22.3, 34.2, 23.2, NA, 23.2, 
                           4.3, 2.3, NA, 10.3, 10.1, 8.9),
      density_Galibi.Suriname=rep("dgamma", 11), 
      Irakumpapy.French.Guiana=c(1076, 1765, 1390, 1678, NA, 
                               3542, 2678, 243, NA, 566, 566),
      se_Irakumpapy.French.Guiana=c(23.2, 29.7, 22.5, 226, NA, 
                                 130, 29, 67, NA, 15, 20), 
      density_Irakumpapy.French.Guiana=rep("dgamma", 11), stringsAsFactors = FALSE
      )
cst <- fitRMU(RMU.data=data.AtlanticW, RMU.names=RMU.names.AtlanticW, 
               colname.year="Year", model.trend="Constant", 
               model.SD="Zero")
}

Run the code above in your browser using DataLab