Merge pull request #5 from poissonconsulting/reviewpdf

Edits to paper

.gitignore

@@ -19,3 +19,4 @@ output-20[0-9][0-9]-[0-9][0-9]*
 
 report/
 inst/doc
+paper.docx

paper/paper.bib

@@ -87,7 +87,7 @@ @article{mizuki2020
 
 @article{newman2023,
 	title = {State{-}space models for ecological time{-}series data: Practical model{-}fitting},
-	author = {Newman, Ken and King, Ruth and Elvira, {Víctor} and De Valpine, Perry and McCrea, {Rachel S.} and Morgan, Byron J. T.},
+	author = {Newman, K.B. and King, R. and Elvira, V. and De Valpine, P. and McCrea, R.S. and Morgan, B.J.T.},
 	year = {2023},
 	month = {01},
 	date = {2023-01},
@@ -100,6 +100,19 @@ @article{newman2023
 	langid = {en}
 }
 
+@book{newmanmodelling2014,
+	title = {Modelling population dynamics: model formulation, fitting and assessment using state-space methods},
+	isbn = {978-1-4939-0977-3 978-1-4939-0976-6},
+	shorttitle = {Modelling population dynamics},
+	url = {http://dx.doi.org/10.1007/978-1-4939-0977-3},
+	abstract = {This book gives a unifying framework for estimating the abundance of open populations: populations subject to births, deaths and movement, given imperfect measurements or samples of the populations. The focus is primarily on populations of vertebrates for which dynamics are typically modelled within the framework of an annual cycle, and for which stochastic variability in the demographic processes is usually modest. Discrete-time models are developed in which animals can be assigned to discrete states such as age class, gender, maturity, population (within a metapopulation), or species (for multi-species models). The book goes well beyond estimation of abundance, allowing inference on underlying population processes such as birth or recruitment, survival and movement. This requires the formulation and fitting of population dynamics models. The resulting fitted models yield both estimates of abundance and estimates of parameters characterizing the underlying processes.},
+	language = {English},
+	urldate = {2015-07-10},
+	author = {Newman, K.B. and Buckland, S.T. and Morgan, B.J.T. and King, R. and Borchers, D.L. and Cole, D.J. and Besbeas, P. and Gimenez, O. and Thomas, L.},
+	year = {2014},
+	keywords = {stock-productivity, sgpd, mwst, no access-OR},
+}
+
 @article{magoun2011,
 	title = {Integrating motion{-}detection cameras and hair snags for wolverine identification},
 	author = {Magoun, Audrey J. and Long, Clinton D. and Schwartz, Michael K. and Pilgrim, Kristine L. and Lowell, Richard E. and Valkenburg, Patrick},
@@ -211,7 +224,7 @@ @book{mcelreath2016
 	address = {Boca Raton}
 }
 
-@book{schaub_integrated_2022,
+@book{schaubintegrated2022,
 	title = {Integrated population models: theory and ecological applications with {R} and {JAGS}},
 	isbn = {978-0-323-90810-8},
 	shorttitle = {Intergrated population models},

paper/paper.md

@@ -1,5 +1,5 @@
 ---
-title: 'bisonpic Software: A suite of R packages to derive wood bison population parameters from remote camera image series.'
+title: 'bisonpicsuite: A set of R packages to estimate wood bison population parameters from remote camera data.'
 authors:
 - name: Nicole Hill
   orcid: 0000-0002-7623-2153
@@ -14,73 +14,79 @@ authors:
 affiliations:
 - index: 1
   name: Poisson Consulting, Canada
-date: 29 April 2024
+date: 28 October 2024
 bibliography: paper.bib
 tags:
   - bisonpictools
+  - runbisonpic
+  - shinybisonpic
   - camera
   - R
 ---
 
 # Statement of Need
 
-Knowledge of population status and trend is integral to the effective conservation and management of wildlife populations. 
-It is particularly advantageous to anticipate future trends. 
-To this end, demographic ratios (i.e Calf:Cow ratios) are routinely used in wildlife management as a readily observable measure of productivity and for forecasting population trajectories [@fuller2007; @wittmer2005; @bender2006]. 
-More complex state-space population modelling approaches [@buckland2004; @paterson2019; @mizuki2020; @newman2023] can be additionally used to derive estimates of survival and fecundity. 
-Both approaches require classified (by age and sex) counts of individuals in a herd.
+Knowledge of population status and trends is integral to the effective conservation and management of wildlife populations. 
+To this end, demographic ratios (e.g., Calf:Cow ratios, Fig. 1) are routinely used in wildlife management as a readily observable measure of productivity and for forecasting population trajectories [@fuller2007; @wittmer2005; @bender2006]. 
+More complex state-space population modeling approaches [@buckland2004; @paterson2019; @mizuki2020; @newman2023] can also be used to estimate survival and fecundity.
+To understand the age composition of a population, both of these approaches require classified (by age and sex) counts of individuals.
 
 In remote areas, estimates of herd size and composition are typically obtained from aerial surveys. 
-However, distinguishing animal age and sex can be challenging, particularly in forested environments, if animals flee or hide from aircraft. 
+However, distinguishing animal age and sex can be challenging, particularly in forested environments if animals flee or hide from aircraft. 
 Remote cameras present an alternative method for obtaining classified counts. 
-Wildlife cameras have been utilized for purposes including species occupancy, density, behaviour, and to identify individuals through district markings [@magoun2011; @steenweg2016; @caravaggi2017; @green2020; @nakashima2020; @singh2022]. 
-Although different age and sex classes can be readily distinguished in remote camera photos for many ungulate species [@laskin2020], we are not aware of any published studies or software using cameras to derive population parameters from classified counts of herds for animals without individual markings.
+Wildlife cameras have been used to estimate species occupancy, density, behaviour, and to identify individuals through distinct markings [@magoun2011; @steenweg2016; @caravaggi2017; @green2020; @nakashima2020; @singh2022]. 
+Although different age and sex classes can be readily distinguished in remote camera photos for many ungulate species [@laskin2020], it is rare that classified count data are used to estimate population parameters without using individual markings.
 
-# Summary
-
-We present a method utilizing classified counts from remote cameras to evaluate wood bison herd demographics.
-We modeled demographic ratios, survival and productivity using an integrated population model (IPM) to combine stage-structured information from multiple data sources to describe demographic states and transitions [@schaub_integrated_2022]. A Gaussian process regression [@mcelreath2016] accounts for the spatial and temporal correlation structure of the camera observations.
-This novel approach requires an initial population estimate but does not require individual identification and could be applied non-invasively to a wide array of difficult to survey species to estimate the key parameters that drive population dynamics.
+![](ratio-plot.png "Figure 1. Data exploration plot from the shinybisonpic app, showing calf:cow ratios over time from individual remote cameras in the Ronald Lake Wood Bison range in northeast Alberta. Ratios are shown with camera trap ID, date of observation, study year, season, and group size. A ratio of 0 represents a group of entirely cows, while an infinite ratio (Inf) represents a group of entirely calves.")
 
-![](bisonpicwriteup-diagram.png "Figure 1. Overview of the bisonpic suite of tools.")
+*Figure 1. Data exploration plot from the shinybisonpic app, showing calf:cow ratios over time from individual remote cameras in the Ronald Lake Wood Bison range in northeast Alberta. Ratios are shown with camera trap ID, date of observation, study year, season, and group size. A ratio of 0 represents a group of entirely cows, while an infinite ratio (Inf) represents a group of entirely calves.*
 
-*Figure 1. Overview of the bisonpic suite of tools.*
+# Summary
 
-This method is implemented using three connected R packages, bisonpictools provides underlying functionality to clean, process, model, and visualize data. 
-The other two R packages are apps that provide a user-friendly interface to bisonpictools. 
-The first app is shinybisonpic. 
-This web-based app allows users to upload and explore the data by viewing the locations of cameras and the ratios of selected sex-age groups. 
-The second app is runbisonpic which is run locally. 
-This app allows users to run a model to calculate the abundance by class, total abundance, survival, fecundity, and various sex-age ratios. 
-Bisonpictools, shinybisonpic and runbisonpic were developed for Alberta Environment and Parks to enable remote game cameras to monitor the herd composition of wood bison.
+We present a method utilizing classified counts from remote cameras to evaluate wood bison (*Bison bison athabascae*) herd demographics.
+We modeled demographic ratios, survival, and productivity using a Bayesian integrated population model (IPM) to combine stage-structured information from multiple data sources and estimate demographic states and transitions [@schaubintegrated2022]. 
+The data included the classified counts from camera trap observations, and census and proportion of calves estimates from aerial surveys.
 
-# Features
+The counts of classified individuals from each camera trap observation were represented by a series of binomial distributions, as opposed to a multinomial distribution, which allowed us to account for individuals that were classified by age, but not by sex.
+The binomial distributions informed both the counts of each age class (calf, yearling, adult) and corresponding sex ratios.
+The probabilities for each of the binomial draws were informed by the estimated proportions of individuals in each age and sex class in the population on a given date, which was modeled using a Birth-Age-Survival (BAS) subprocess formulation of a population projection model [@newmanmodelling2014].
 
-Data Standardization
+For example, the count of calves at the $i^{th}$ camera trap observation was modeled as follows: 
 
--   Uploaded data is put through a quality control process which detects values that are not allowed to reduce errors in the following steps.
+$$C_i \sim \text{Binomial}(N_i, p_{C_i})$$
 
-Location Mapping
+where $C_i$ is the number of calves, $N_i$ is the total group size, and $p_{C_i}$ is the sum of the expected proportions of male and female calves on the date of the $i^{th}$ observation.
 
--   Users can explore the camera locations on an interactive map which helps to spatially verify location data are accurate.
+The sex ratio of calves was modeled as follows:
 
-Demographic Ratios
+$$F0_i \sim \text{Binomial}\Biggl(F0_i + M0_i, \frac{p_{F0_i}}{p_{F0_i} + p_{M0_i}}\Biggr)$$
 
--   Each sex-age group can be individually selected which gives users the freedom to explore ratios for any combination of sex-age groups.
+where $F0_i$ is the number of female calves, $M0_i$ is the total number of male calves, and $p_{F0_i}$ and $p_{M0_i}$ are the expected proportion of female and male calves on the date of the $i^{th}$ observation, respectively.
 
-Abundance Class, Abundance Total, Survival, Fecundity, and Ratios Estimates
+The population projection model also estimated the fecundity rate, the proportion of fecund cows aged two and older, the annually-varying survival rates for each class, and can produce derived estimates of several key population ratios (Fig. 2).
+The information from the aerial surveys was integrated into the model, helping to inform the total number of individuals (i.e., the sum of the class-wise abundances) and the proportion of calves (i.e., $p_{F0_i} + p_{M0_i}$) on the dates of the aerial surveys.
+A Gaussian process regression [@mcelreath2016] accounted for the spatial and temporal correlation structure of the camera trap observations.
+Collectively, these methods could be applied non-invasively to a wide array of difficult to survey species to estimate key parameters that drive population dynamics.
 
--   Complex custom model that only has a single parameter users must learn how to tune which makes the method accessible for users of various skill levels.
--   Downloadable analysis object which allows power users to generate their own plots or perform further analysis.
+![](model-plot.png "Figure 2. Example of a prediction plot from the runbisonpic app, showing estimated population ratios for the Ronald Lake Wood Bison herd, by study year. M0 and F0 are male and female calves, M1 and F1 are male and female yearlings, Calf and Yearling represent all individuals within those age classes including those with unknown sex, M2 and M3 represent male two- and three-year-olds, MA represents males aged four and older, and FA represents females aged two and older.")
 
-Documentation
+*Figure 2. Example of a prediction plot from the runbisonpic app, showing estimated population ratios for the Ronald Lake Wood Bison herd, by study year. M0 and F0 are male and female calves, M1 and F1 are male and female yearlings, Calf and Yearling represent all individuals within those age classes including those with unknown sex, M2 and M3 represent male two- and three-year-olds, MA represents males aged four and older, and FA represents females aged two and older.*
 
--   User Manual vignette that walks users through the detailed steps and options.
--   Basic instructions are easily accessible within the app.
+This method is implemented using four related R packages.
+The underlying functionality to check, clean, process, model, and visualize data is provided by `bisonpictools`.
+The other two R packages are apps that provide a user-friendly interface to `bisonpictools`. 
+The first app is `shinybisonpic`, a web-based app that allows users to upload and explore the data by viewing the locations of cameras and the ratios of selected sex-age groups (e.g., Fig. 1). 
+The second app is `runbisonpic`, a locally-run app that allows users with various skill levels to run a model to estimate the abundance by class, total abundance, survival and fecundity rates, and various sex-age ratios (e.g., Fig. 2).
+The `bisonpicsuite` package loads the three other packages.
+The software suite was developed for Alberta Environment and Protected Areas to use remote game cameras to estimate the composition, status, and trends of the Ronald Lake Wood Bison herd.
 
 # Limitations
 
-The model is slow to run and can take over 5 hours to complete running.
+- The model can take over 5 hours to run.
+- Key assumptions of the integrated population model include:
+  - There is no grouping structure beyond what is accounted for by the covariance.
+  - Every stage is equally detectable during a camera trap event.
+  - Small and large groups are equally detectable.
 
 # Acknowledgements