In rooperc4/TrigCamDensityEstimation: Estimates Volumetric Density for Stationary Camera Deployments

knitr::opts_chunk$set(echo = TRUE)
#knitr::opts_knit$set(root.dir = 'relative_path_to_root_from_Rmd' )
library(devtools)
library(ggplot2)
library(pander)
library(magick)
devtools::install_github("rooperc4/TrigCamDensityEstimation")
library("TrigCamDensityEstimation")

Purpose

The functions contained in this package are used to estimate densities of fish using a stationary camera platform (e.g. Williams et al. 2014). The functions follow the method of Williams et al. (2018) In short, this analysis takes a matrix of densities from individual camera frames for a species in range bins with volume estimates for each range bin. It makes a volume-weighted estimate of the detection function by fitting a logistic function. This function can then be used to estimate a density for the species in each frame. These can be averaged across frames to come up with a volumetric density estimate.

p<-image_read("Slide1.PNG")
plot(p)

Package Functions and Example Using Bocaccio Rockfish Data

As an example, we will use bocaccio rockfish data collected at 30 sites over ~24 hour deployments on the Footprint Bank. These data were collected in October 2017 aboard the R/V Bell-Shimada. In this case the columns are a deployment number, a range bin and a volumetric density in that range bin.

data(bocaccio)
pander::pandoc.table(bocaccio[1:6,], digits=3, split.tables=Inf)

To fit the model, we call the logistic model function that serves as a wrapper for the model fitting. The logistic model function is the wrapper that inputs the observations of density, volume and range and fits the logistic model. It also takes inputs of a species name and a deployment number (for ease of looping across species and deployments). Weights can also be included for weighting the AIC calculation. It outputs an estimate the y-intercept of the model (which is used to scale the results)) and a plot of the fitted logistic function (output in ggplot). The output also includes a list of fitted values (parameters, AIC, etc.).

f1<-logistic_model("Bocaccio rockfish",bocaccio$range_bin,bocaccio$density,bocaccio$volume,2,.5)
print(f1)

This function will produce a figure of the detecability function that is scaled to 1 at the nearest volume bin (in this case 0.5 m)

p<-image_read("Bocaccio rockfish_Logistic_Fit.png")
plot(p)

Here is the code for the logistic model function. It calls two other functions; logistic_fit and logistic_modlike which make the predictions and estimate the negative log-likelihood respectively.

logistic_model

The logistic_fit function takes the parameter values and calculates a predicted density based on the range. This is a logistic function (e.g. logistic growht model) The parameters from this equation are fit using negative-log-likelihood estimation (see logistic_modlike function below).

logistic_fit

The logistic_modlike function computes the negative-log-likelihood for the model based on parameter estimates (this is the function you are minimizing). For this function you can include optional weights, if for example you want to down-weight the points closest to the camera with low volume, but potentially high density. Default weights are = 1

logistic_modlike

Below is an example implementation which can be looped for multiple species and combining into a Footprint-wide abundance estimate. This example is how I calculated the biomass estimates for the UHSI project (Rooper et al. in prep). Basically, it takes the detecability function from above and corrects the densities by frame for the distance range, then averages the densities and computes out an overall density. This is then expanded to an areal estimate using the Footprint Bank area estimate for rocky habitat.

Example to Generate Abundance Estimate

#######DATA ANALYSES####################################
species_list<-c("Bocaccio rockfish")
deployment<-unique(bocaccio$deployment_ID)
species_parameters<-array(dim=c(0,5))
colnames(species_parameters)<-c("Species","start_L","start_K","x0","Intercept")


#fit the logistic model to density by range data for each species
for(i in 1:length(species_list)){
f1<-logistic_model(species_list[i],bocaccio$range_bin,bocaccio$density,bocaccio$volume,2,.5)
temp1<-data.frame(Species=species_list[i],start_L=f1[1],start_K=f1[2],x0=f1[3],Intercept=f1[9])
species_parameters<-rbind(species_parameters,temp1)
}

colnames(species_parameters)<-c("Species","start_L","start_K","x0","Intercept")
functs<-array(dim=0,3)
Range<-seq(.5,6,.1)
for(i in 1:length(species_list)){
x1<-dist_function(species_parameters$start_L[i],species_parameters$start_K[i],species_parameters$x0[i],species_parameters$Intercept[i],Range)
x1<-data.frame(Range,x1,species_list[i])
functs<-rbind(functs,x1)
}
colnames(functs)<-c("Range","Relative_density","Species")

dev.new(width=9,height=3,units="in",noRStudioGD=TRUE)
p<-ggplot(functs)+geom_line(aes(y=Relative_density,x=Range,colour=Species))+xlab("Range (m)")+
  ylab("Relative density")+geom_hline(yintercept=0.10,linetype="dashed",color="red",size=.75)
print(p)

density_by_frame<-array(dim=c(0,4))
colnames(density_by_frame)<-c("Group.1","Group.2","x","species")
for(i in 1:length(species_list)){
#  data5<-zero_fill(density_range,species_list[i])
  density_by_frame1<-aggregate(bocaccio$density,by=list(bocaccio$deployment_ID,bocaccio$frame_number,bocaccio$range_bin),FUN="sum")
  density_by_frame1$corrected_density<-frame_density(species_parameters$start_L[i],species_parameters$start_K[i],species_parameters$x0[i],species_parameters$Intercept[i],density_by_frame1$Group.3,density_by_frame1$x)
  density_by_frame1<-aggregate(density_by_frame1$corrected_density,by=list(density_by_frame1$Group.1,density_by_frame1$Group.2),FUN="mean",na.rm=TRUE)
  density_by_frame1$species<-species_list[i]
  density_by_frame<-rbind(density_by_frame,density_by_frame1)
}  
colnames(density_by_frame)<-c("deployment_ID","frame_number","density","species")  

#ABUNDANCE ESTIMATION
area<-10*10*6819*1.25

#USING ALL TIME
fd1<-aggregate(density_by_frame$density,by=list(density_by_frame$species,density_by_frame$deployment_ID),FUN="mean")
fd2<-aggregate(fd1$x,by=list(fd1$Group.1),FUN="mean")
fd3<-aggregate(fd1$x,by=list(fd1$Group.1),FUN="sd")
abundance_estimate<-merge(fd2,fd3,by="Group.1")
colnames(abundance_estimate)<-c("Species","Density","Density_SD")
abundance_estimate$Population<-abundance_estimate$Density*area
abundance_estimate$Population_SE<-sqrt(abundance_estimate$Density_SD^2*(area^2))/sqrt(30)
abundance_estimate$CV<-abundance_estimate$Density_SD/abundance_estimate$Density
pandoc.table(abundance_estimate)

References

Rooper CN, Williams K, Towler R. in prep. Estimating rockfish habitat-specific abundance and behaviour using stationary cameras in the southern California Bight.

Williams K, De Robertis A, Berkowitz Z, Rooper C, Towler R. 2014. An underwater stereo-camera trap. Methods in Oceanography 11:1-12. http://dx.doi.org/10.1016/j.mio.2015.01.003

Williams K, Rooper CN, De Robertis A, Levine M, Towler R. 2018. A method for computing volumetric fish density using stereo cameras. Journal of Experimental Marine Biology and Ecology 508:21-26. https://doi.org/10.1016/j.jembe.2018.08.001

rooperc4/TrigCamDensityEstimation documentation built on May 18, 2019, 1:27 a.m.