# fit a model of tag loss using a CMR database. This function fits a model of tag loss using a CMR database. The names of parameters can be: Model Pfaller et al. (2019): - **Left tag lost when 2 are present**: `D1_L2`, `D2D1_L2`, `D3D2_L2`, `A_L2`, `B_L2`, `C_L2`, `delta_L2` - **Right tag lost when 2 are present**: `D1_R2`, `D2D1_R2`, `D3D2_R2`, `A_R2`, `B_R2`, `C_R2`, `delta_R2` - **Left tag lost when 1 is present**: `D1_L1`, `D2D1_L1`, `D3D2_L1`, `A_L1`, `B_L1`, `C_L1`, `delta_L1` - **Right tag lost when 1 is present**: `D1_R1`, `D2D1_R1`, `D3D2_R1`, `A_R1`, `B_R1`, `C_R1`, `delta_R1` - **One tag lost when 2 are present**: `D1_2`, `D2D1_2`, `D3D2_2`, `A_2`, `B_2`, `C_2`, `delta_2` - **One tag lost when 1 is present**: `D1_1`, `D2D1_1`, `D3D2_1`, `A_1`, `B_1`, `C_1`, `delta_1` `pA`, `pB` and `pC` are the daily probabilities of tag loss with `pA=-logit(A)`, `pB=-logit(B)` and `pC=-logit(C)` . `delta` is used as: `p = p + delta`. Nothe that `delta` can be negative Tag loss rate is `pA` at day 1 Tag loss rate changes gradually from `pA` to `pB` that is reached at day `D1` Tag loss rate is `pB` from day `D1` to day `D2=D1+D2D1` Tag loss rate changes gradually from `pB` to `pC` that is reached at day `D3=D2+D3D2` When parameters from Rivalan et al. (2005) are used: - **One tag lost when 2 are present**: `a0_2`, `a1_2`, `a2_2`, `a3_2`, `a4_2`, `delta_2` - **One tag lost when 1 is present**: `a0_1`, `a1_1`, `a2_1`, `a3_1`, `a4_1`, `delta_1` When parameters from Casale et al. (2017) are used: Model I - **One tag lost when 2 are present**: `CasaleModelIc_2` - **One tag lost when 1 is present**: `CasaleModelIc_1` Model II - **One tag lost when 2 are present**: `CasaleModelIIa0_2`, `CasaleModelIIa1_2`, `CasaleModelIIa4_2` - **One tag lost when 1 is present**: `CasaleModelIIa0_1`, `CasaleModelIIa1_1`, `CasaleModelIIa4_1` Model III - **One tag lost when 2 are present**: `CasaleModelIIIa0_2`, `CasaleModelIIIa1_2`, `CasaleModelIIIa4_2` - **One tag lost when 1 is present**: `CasaleModelIIIa0_1`, `CasaleModelIIIa1_1`, `CasaleModelIIIa4_1` Model IV - **One tag lost when 2 are present**: `CasaleModelIVa0_2`, `CasaleModelIVa1_2`, `CasaleModelIVa2_2`, `CasaleModelIVa3_2`, `CasaleModelIVa4_2` - **One tag lost when 1 is present**: `CasaleModelIVa0_1`, `CasaleModelIVa1_1`, `CasaleModelIVa2_1`, `CasaleModelIVa3_1`, `CasaleModelIVa4_1` Model V - **One tag lost when 2 are present**: `CasaleModelVa0_2`, `CasaleModelVa1_2`, `CasaleModelVa2_2`, `CasaleModelVa3_2`, `CasaleModelVa4_2` - **One tag lost when 1 is present**: `CasaleModelVa0_1`, `CasaleModelVa1_1`, `CasaleModelVa2_1`, `CasaleModelVa3_1`, `CasaleModelVa4_1` If only one parameter is fitted, method must be "Brent" and `upper` and `lower` parameters must be set up with finite values. model_before can be ""par['a0_1']=par['a0_2'];par['a1_1']=par['a1_2']". model_after can be "p1=p2" ```r Tagloss_fit( data = stop("A database formated using Tagloss_format() must be used"), fitted.parameters = NULL, fixed.parameters = NULL, model_before = NULL, model_after = NULL, control = list(trace = 1, maxit = 10000), method = "Nelder-Mead", lower = -Inf, upper = Inf, hessian = FALSE, mc.cores = detectCores(all.tests = FALSE, logical = TRUE), groups = NULL ) ``` ## Arguments - `data`: An object formated using Tagloss_format - `fitted.parameters`: Set of parameters to be fitted - `fixed.parameters`: Set of fixed parameters - `model_before`: Transformation of parameters before to use Tagloss_model() - `model_after`: Transformation of parameters after to use Tagloss_model() - `control`: Control parameters to be send to optim() - `method`: optim() method - `lower`: Lower value for parameter when Brent method is used - `upper`: Upper value for parameter when Brent method is used - `hessian`: Does the hessian matrix should be estimated - `mc.cores`: Number of cores to use for parallel computing - `groups`: Number of groups for parallel computing ## Returns Return a list object with the model describing tag loss. ## Description This function fits a model of tag loss using a CMR database. The names of parameters can be: Model Pfaller et al. (2019): - **Left tag lost when 2 are present**: `D1_L2`, `D2D1_L2`, `D3D2_L2`, `A_L2`, `B_L2`, `C_L2`, `delta_L2` - **Right tag lost when 2 are present**: `D1_R2`, `D2D1_R2`, `D3D2_R2`, `A_R2`, `B_R2`, `C_R2`, `delta_R2` - **Left tag lost when 1 is present**: `D1_L1`, `D2D1_L1`, `D3D2_L1`, `A_L1`, `B_L1`, `C_L1`, `delta_L1` - **Right tag lost when 1 is present**: `D1_R1`, `D2D1_R1`, `D3D2_R1`, `A_R1`, `B_R1`, `C_R1`, `delta_R1` - **One tag lost when 2 are present**: `D1_2`, `D2D1_2`, `D3D2_2`, `A_2`, `B_2`, `C_2`, `delta_2` - **One tag lost when 1 is present**: `D1_1`, `D2D1_1`, `D3D2_1`, `A_1`, `B_1`, `C_1`, `delta_1` `pA`, `pB` and `pC` are the daily probabilities of tag loss with `pA=-logit(A)`, `pB=-logit(B)` and `pC=-logit(C)` . `delta` is used as: `p = p + delta`. Nothe that `delta` can be negative Tag loss rate is `pA` at day 1 Tag loss rate changes gradually from `pA` to `pB` that is reached at day `D1` Tag loss rate is `pB` from day `D1` to day `D2=D1+D2D1` Tag loss rate changes gradually from `pB` to `pC` that is reached at day `D3=D2+D3D2` When parameters from Rivalan et al. (2005) are used: - **One tag lost when 2 are present**: `a0_2`, `a1_2`, `a2_2`, `a3_2`, `a4_2`, `delta_2` - **One tag lost when 1 is present**: `a0_1`, `a1_1`, `a2_1`, `a3_1`, `a4_1`, `delta_1` When parameters from Casale et al. (2017) are used: Model I - **One tag lost when 2 are present**: `CasaleModelIc_2` - **One tag lost when 1 is present**: `CasaleModelIc_1` Model II - **One tag lost when 2 are present**: `CasaleModelIIa0_2`, `CasaleModelIIa1_2`, `CasaleModelIIa4_2` - **One tag lost when 1 is present**: `CasaleModelIIa0_1`, `CasaleModelIIa1_1`, `CasaleModelIIa4_1` Model III - **One tag lost when 2 are present**: `CasaleModelIIIa0_2`, `CasaleModelIIIa1_2`, `CasaleModelIIIa4_2` - **One tag lost when 1 is present**: `CasaleModelIIIa0_1`, `CasaleModelIIIa1_1`, `CasaleModelIIIa4_1` Model IV - **One tag lost when 2 are present**: `CasaleModelIVa0_2`, `CasaleModelIVa1_2`, `CasaleModelIVa2_2`, `CasaleModelIVa3_2`, `CasaleModelIVa4_2` - **One tag lost when 1 is present**: `CasaleModelIVa0_1`, `CasaleModelIVa1_1`, `CasaleModelIVa2_1`, `CasaleModelIVa3_1`, `CasaleModelIVa4_1` Model V - **One tag lost when 2 are present**: `CasaleModelVa0_2`, `CasaleModelVa1_2`, `CasaleModelVa2_2`, `CasaleModelVa3_2`, `CasaleModelVa4_2` - **One tag lost when 1 is present**: `CasaleModelVa0_1`, `CasaleModelVa1_1`, `CasaleModelVa2_1`, `CasaleModelVa3_1`, `CasaleModelVa4_1` If only one parameter is fitted, method must be "Brent" and `upper` and `lower` parameters must be set up with finite values. model_before can be ""par['a0_1']=par['a0_2'];par['a1_1']=par['a1_2']". model_after can be "p1=p2" ## Details Tagloss_fit fits a model of tag loss using a CMR database. ## Examples ```r ## Not run: library(phenology) # Example data_f_21 <- Tagloss_format(outLR, model="21") # model fitted by Rivalan et al. 2005 par <- c(a0_2=-5.43E-2, a1_2=-103.52, a4_2=5.62E-4, delta_1=3.2E-4) pfixed <- c(a2_2=0, a3_2=0, a2_1=0, a3_1=0) model_before <- "par['a0_1']=par['a0_2'];par['a1_1']=par['a1_2'];par['a4_1']=par['a4_2']" o <- Tagloss_fit(data=data_f_21, fitted.parameters=par, fixed.parameters=pfixed, model_before=model_before) plot(o, t=1:1000, model="cumul") plot(o, t=1:1000, model="1") plot(o, t=1:1000, model="2", add=TRUE, col="red") # Same data fitted with new model par <- c(D1_1 = 100.15324837975547, A_1 = 5.9576927964120188, B_1 = 8.769924225871069, B_2 = 8.2353860179664125) pfixed <- c(D2D1_1 = 2568, D3D2_1 = 2568, D2D1_2 = 2568, D3D2_2 = 2568) o_4p_p1p2 <- Tagloss_fit(data=data_f_21, fitted.parameters = par, fixed.parameters = pfixed, model_before = "par['C_1']=par['B_1']; par['A_2']=par['A_1']; par['C_2']=par['B_2']; par['D1_2']=par['D1_1']", hessian=TRUE) # Without the N20 the computing is much faster data_f_21_fast <- subset(data_f_21, subset=(is.na(data_f_21$N20))) par <- c('D1_2' = 49.78891736351531, 'D2D1_2' = 1059.3635769732305, 'D3D2_2' = 12.434313273804602, 'A_2' = 5.2238379144659683, 'B_2' = 8.0050044071275543, 'C_2' = 8.4317863609499675, 'D1_1' = 701.80273287212935, 'D2D1_1' = 0.010951749100596819, 'D3D2_1' = 3773.6290607434876, 'A_1' = 205.42435592344776, 'B_1' = 9.9598342503239863, 'C_1' = 6.7234868237164722) o <- Tagloss_fit(data=data_f_21_fast, fitted.parameters=par, hessian = TRUE) plot(o, model="1", col="red") plot(o, model="2", col="blue", add=TRUE) legend("topright", legend=c("2->1", "1->0"), lty=1, col=c("blue", "red")) ## End(Not run) ``` ## References Rivalan, P., Godfrey, M.H., Prévot-Julliard, A.-C., Girondot, M., 2005. Maximum likelihood estimates of tag loss in leatherback sea turtles. Journal of Wildlife Management 69, 540-548. Casale, P., Freggi, D., Salvemini, P., 2017. Tag loss is a minor limiting factor in sea turtle tagging programs relying on distant tag returns: the case of Mediterranean loggerhead sea turtles. European Journal of Wildlife Research 63. Pfaller JB, Williams KL, Frick MG, Shamblin BM, Nairn CJ, Girondot M (2019) Genetic determination of tag loss dynamics in nesting loggerhead turtles: A new chapter in “the tag loss problem”. Marine Biology 166: 97 doi 10.1007/s00227-019-3545-x ## See Also Other Model of Tag-loss: `Tagloss_L()`, `Tagloss_LengthObs()`, `Tagloss_cumul()`, `Tagloss_daymax()`, `Tagloss_format()`, `Tagloss_mcmc()`, `Tagloss_mcmc_p()`, `Tagloss_model()`, `Tagloss_simulate()`, `logLik.Tagloss()`, `o_4p_p1p2`, `plot.Tagloss()`, `plot.TaglossData()` ## Author(s) Marc Girondot marc.girondot@gmail.com

Tagloss_fit function