dentlinger.Rnw

\documentclass[sn-basic,NameDate,pdflatex]{sn-jnl}

\usepackage{mathptmx}
%\usepackage{lmodern}
\usepackage{amsmath,amssymb,amsfonts}
\usepackage{manyfoot}
\usepackage{booktabs}

\usepackage{parskip}
\usepackage[nolists]{endfloat}

\raggedright
\hbadness=10000
\vbadness=10000

\setlength{\tabcolsep}{4pt}

\usepackage{setspace}
\onehalfspacing

\usepackage{datetime2}
\DTMsetstyle{ddmmyyyy}

\bibpunct{(}{)}{;}{a}{}{,}

\usepackage{lineno}
\linenumbers

\usepackage{tocloft}
\cftpagenumbersoff{figure}
%\renewcommand{\cftdotsep}{\cftnodots}
\hypersetup{linktoc=none}

%\newcommand{\okina}{`}
\DeclareRobustCommand{\okina}{%
  \raisebox{\dimexpr\fontcharht\font`A-\height}{%
    \scalebox{0.8}{`}%
  }%
}

\newcommand{\Hawaii}{Hawai\okina i}
\newcommand{\Oahu}{O\okina ahu}
\newcommand{\Milolii}{Miloli\okina i}

\newcommand{\orig}[1]{{\color{red}(#1)}}
\newcommand{\xorig}[1]{{\color{olive}(#1)}}

\title{Four decades of green turtle (\emph{Chelonia mydas}) strandings on \Hawaii\ Island (1983--2022): Causes and trends
}
\author[1]{\fnm{Skylar} \sur{Dentlinger}}

\author*[2]{\fnm{Karla J.} \sur{McDermid}}%, orcid.org/0000-0002-7663-6545}
%\equalcont{These authors contributed equally to this work.}

\author*[3]{\fnm{Grady} \sur{Weyenberg}}%, orcid.org/0000-0001-6128-1772}
%\equalcont{These authors contributed equally to this work.}

\author[4]{\fnm{Laura M. R.} \sur{Jim}}
\author[5]{\fnm{Marc R.} \sur{Rice}} %, orcid.org/0009-0008-5951-0749}
\author[6]{\fnm{George H.} \sur{Balazs}}

\affil[1]{\orgdiv{Department of Marine Science}, \orgname{University of \Hawaii\ at Hilo},
\orgaddress{\city{Hilo}, \state{Hawaii}, \postcode{96720}, \country{USA}; Current address: \orgdiv{Rosenstiel School of Marine and Atmospheric Science}, \orgname{University of Miami}, \orgaddress{\city{Miami}, \state{Florida}, \postcode{33149}, \country{USA}.}} E-mail: {skylarde@hawaii.edu}}

\affil[2]{\orgdiv{Department of Marine Science}, \orgname{University of \Hawaii\ at Hilo},
\orgaddress{\city{Hilo}, \state{Hawaii}, \postcode{96720}, \country{USA}. 
\href{https://orcid.org/0000-0002-7663-6545}{orcid.org/0000-0002-7663-6545}
$^*$Correspondence: E-mail: {mcdermid@hawaii.edu}}}

\affil[3]{\orgdiv{Department of Mathematics}, \orgname{University of \Hawaii\ at Hilo}, 
\orgaddress{\city{Hilo}, \state{Hawaii}, \postcode{96720}, \country{USA}. 
\href{https://orcid.org/0000-0001-6128-1772}{orcid.org/0000-0001-6128-1772}
$^*$Correspondence: E-mail: {gradysw@hawaii.edu}}}

\affil[4]{\orgdiv{Sea Turtle Research Program}, \orgname{\Hawaii\ Preparatory Academy}, 
\orgaddress{\city{Kamuela}, \state{Hawaii}, \postcode{96743}, \country{USA}. E-mail: {laura.jim@hpa.edu}}}

\affil[5]{\orgdiv{Sea Turtle Research Program}, \orgname{\Hawaii\ Preparatory Academy}, 
\orgaddress{\city{Kamuela}, \state{Hawaii}, \postcode{96743}, \country{USA}. 
\href{https://orcid.org/0009-0008-5951-0749}{orcid.org/0009-0008-5951-0749}
E-mail: {mrice@hpa.edu}}}

\affil[6]{\orgname{Golden Honu Services of Oceania}, 
\orgaddress{\city{Honolulu}, \state{Hawaii}, \country{USA}. E-mail: {itsahonuworld@hotmail.com}}}

\date{\today}

\miscnote{Note to publisher: The Hawaiian letter \okina okina is typeset in this manuscript as an open quote, e.g. \Hawaii, \Oahu, \Milolii. In electronic publication, it should be rendered as unicode U+02BB MODIFIER LETTER TURNED COMMA.}

<<setup, include=FALSE>>=
mm2in <- 0.0393701
twocol <- 80 * mm2in
bigcol <- 170 * mm2in
smallcol <- 119 * mm2in

make_plot_namer <- function(prefix="Figure-", n=1L, width=2){
  function(x){
    x2 <- file.path(dirname(x), paste0(prefix, formatC(n, width=width, flag="0"),".", tools::file_ext(x)))
    magick::image_write(magick::image_read_pdf(x), path=fs::path_ext_set(x2, "png"))
    magick::image_write(magick::image_read_pdf(x, density=1200), path=fs::path_ext_set(x2, "tiff"),
                         format="tiff")
    n <<- n + 1L
    if (file.rename(x, x2)) x2 else x
  }
}

knitr::opts_chunk$set(
  echo = FALSE, warning = FALSE, message = FALSE, 
  error = FALSE, ## die on compilation error
  fig.width = bigcol, fig.height = 6,
  fig.process = make_plot_namer(width=1)
)

library(tidyr)
library(dplyr)
library(magrittr)
library(ggplot2)
library(lubridate)
library(magick)
library(nnet)
library(splines)
library(scales)
library(stringr)
library(readr)
library(english)
library(tibble)

BigIslandStrandings <- read_csv("BigIslandStrandings.csv") %>% 
  mutate(Cause=case_match(Cause,
                          "boat impact" ~ "Boat impact",
                          "fibropapillomatosis" ~ "FP",
                          "line" ~ "Hook/line",
                          "human take" ~ "Human take",
                          "miscellaneous" ~ "Misc.",
                          "net" ~ "Net",
                          "shark attack" ~ "Shark attack",
                          "unknown" ~ "Unknown"
                          ))

Ntot <- nrow(BigIslandStrandings)
CoastCount <- BigIslandStrandings %>% count(Coast) %>% pivot_wider(names_from = Coast, values_from = n)
CauseCount <- BigIslandStrandings %>% count(Cause) %>% mutate(p=proportions(n) %>% percent(0.1, suffix="\\%")) %>% arrange(desc(n))

@

\abstract{
The Hawaiian population of green turtles (\emph{Chelonia mydas}) has increased since Federal and State protections were implemented in the mid 1970s, and reported stranding events have also increased. 
This study analyzed \Hawaii\ Island data: stranding location, date, size, sex, presence/absence of tumors, stranding status, as well as cause of stranding. 
A total of \Sexpr{Ntot} stranded green turtles were reported from 1983--2022: 
\Sexpr{CoastCount$East} stranded on the east (windward) coast of \Hawaii\ Island and 
\Sexpr{CoastCount$West} on the west (leeward) coast.
Strandings peaked in 2011 and 2018 and were highest from March to August. 
The most common known cause of stranding was 
hook-and-line fishing gear (\Sexpr{CauseCount$p[2]} of total strandings), 
followed by fibropapillomatosis (\Sexpr{CauseCount$p[3]} ), 
human take (\Sexpr{CauseCount$p[4]}), 
miscellaneous (\Sexpr{CauseCount$p[5]}), 
boat impact (\Sexpr{CauseCount$p[6]}), 
shark attack (\Sexpr{CauseCount$p[7]}), 
and net (\Sexpr{CauseCount$p[8]});
however, \Sexpr{CauseCount$p[1]} of strandings had no known cause.
Statistical modeling did not provide convincing evidence of temporal changes in the distribution of strandings across three consolidated cause categories: human-caused; predation, disease, and weather; and unknown.
Stranded turtles on east \Hawaii\ Island had a higher frequency of fibropapillomatosis, whereas west \Hawaii\ stranded turtles showed higher incidence of shark attacks. 
These results provide the first comprehensive analyses of stranding data from \Hawaii\ Island and provide information that can inform resource managers, policy makers, and the public about the various types and magnitudes of impacts, anthropogenic and natural, to green turtles so that mitigation measures can be put into practice.
Our findings allow for comparison with other green turtle populations worldwide.
}

\keywords{sea turtles; fishing gear entanglement; fibropapillomatosis; Hawaiian Islands; marine reptile mortality; Pacific Ocean}

\begin{document}


\maketitle
\vskip 12pt
\section{Background}
Green turtles (\emph{Chelonia mydas}) are the most abundant large marine herbivores found throughout the world and in the Hawaiian Islands. 
The Hawaiian population of green turtles that was once depleted has increased since its 1974 protection under Hawaiian Law and 1978 protection under the Endangered Species Act \citep{balazs2004thirty}. 
Green turtles migrate long distances during their lifetime, from nesting to foraging grounds \citep{balazs2015review}. 
In the Hawaiian Islands, 96\% of nesting occurs on the sand islets at French Frigate Shoals, located in the Northwestern Hawaiian Islands \citep{mtbap2022}. 
Migration patterns and complicated life history patterns cause green turtles to occupy many habitats during their lifespans including pelagic environments during their early years and during migrations, as well coastal areas in their later years \citep{balazs1980synopsis, bolten2003variation}.
Therefore, green turtles are susceptible to threats in both offshore and coastal environments \citep{bolten2003variation}.

Green turtles have experienced a long history of exploitation.
The species was used for meat by indigenous coastal people around the world, as well as by European royals in the 18th and 19th centuries \citep{witzell1994origin}.
Hawaiian green turtles have been additionally impacted by hunting at foraging grounds, by harvesting of both eggs and females at nesting grounds, and by the destruction of their nesting habitat. 
Since protection began under the Endangered Species Act, a reduction in such exploitation has been observed \citep{balazs2004thirty}.
However, large marine vertebrates, including green turtles, face other threats, and are often victims of bycatch, becoming accidentally entangled or hooked by commercial or recreational fisheries activities targeting other species \citep{lewison2004understanding}. 
Bycatch is harmful to green turtles because it can cause drowning, and internal/external injuries from hooks and line entanglements. 

Fibropapillomatosis (FP) is another major threat to sea turtle populations. 
FP is a debilitating neoplastic disease associated with a herpesvirus found in turtles worldwide \citep{jacobson1991herpesvirus, herbst1994fibropapillomatosis}.
The disease was first described in green turtles in the Florida Keys in 1938 and affects mostly immature turtles \citep{herbst1994fibropapillomatosis}. 
FP is indicated by the presence of internal, external, and oral tumors. 
Oral tumors are unique to Hawaiian green turtles and are often found in the glottis, making survival difficult \citep{work2004retrospective}.
The presence of these tumors can impact the turtles' ability to breathe, swim, dive, forage, and see \citep{perrault2021insights}. 
On \Oahu, Maui, and Kauai from 1982-2003, FP was the most common cause of stranding, defined as a turtle that has been found dead, injured, or exhibits ill health or abnormal behavior \citep{chaloupka2008cause}. 

A variety of factors, both natural and anthropogenic, can cause sea turtle strandings. 
The majority of strandings involve sea turtles that died at sea and washed ashore; however, most stranded turtles show no cause of death \citep{hart2006interpreting}. 
An unknown number of deceased turtles never reach shore.
They are eaten by scavengers, sink, and/or decompose while in currents or eddies \citep{crowder1995effects, hart2006interpreting}. 
Therefore, the number of sea turtle strandings that is recorded is likely a minimal estimate \citep{hart2006interpreting}. 
Stranding response programs can provide important insight into the health, welfare, and conservation status of sea turtle populations. Analyses of the data collected by these programs provide valuable information on mortality patterns and can aid regulatory managers \citep{crowder1995effects}. 
Although \cite{chaloupka2008cause} mentioned that 6\% of statewide strandings occurred on \Hawaii\  Island  from 1982--2003, long-term stranding data specifically from \Hawaii\ Island have not been thoroughly analyzed previously.
\Hawaii\ Island merits additional scientific scrutiny of its green turtle stranding patterns with the most up-to-date, inclusive data available  because of the island's large size (over half of total Hawaiian land area), southernmost location in the archipelago, lowest human population density, and important turtle foraging and resting areas---recently proposed as critical habitat by the US Fish and Wildlife Service and the National Oceanic and Atmospheric Administration \citep{fed2023designation, fed2023proposed}.

The knowledge gained from stranding patterns can be used to establish mitigation measures to reduce strandings and maintain healthy green turtle populations.  

In the present study, a comprehensive analysis of 39 years of \Hawaii\ Island green turtle strandings is presented to (1) identify the causes of strandings affecting green turtles around \Hawaii\  Island, (2) assess trends in strandings, and (3) identify differences and similarities between strandings in west and east \Hawaii\ Island. 

\section{Materials and Methods}

\subsection{Data Collection}
Data were collected on turtles stranded on \Hawaii\ Island ($19.5439$ N, $155.6659$ W, land area $10,430$ km$^2$ with coastal circumference of $428$ km) from 1983--2022 by members of the Pacific Islands Fisheries Science Center under the US National Marine Fisheries Service, the University of \Hawaii\ at Hilo Sea Turtle Stranding Response Team, and the \Hawaii\ Preparatory Academy Sea Turtle Research Program. 
The database used in this study was compiled from records available at \url{https://georgehbalazs.com/field-notebooks-by-george-h-balazs/hawaii/}. 
The west and east coasts of \Hawaii\ Island are different in terms of climate (the windward east coast receives much more rainfall than the leeward west coast), terrain, currents, and population, so the data used in this study were analyzed for the island as a whole, as well as by west and east coast. 
West \Hawaii\ included locations from \Milolii\ north to Kawaihae, and east \Hawaii\ included locations from South Point north to Hawi (Figure \ref{fig:map}). 

For each stranded turtle, the following information was collected: date of stranding, stranding location, stranding status (alive/dead), and cause of stranding. Data on species, sex, straight carapace length (SCL), curved carapace length (CCL), and the presence or absence of tumors indicative of fibropapillomatosis were also recorded. SCL was used in size analyses because it was reported more frequently than CCL. In cases where CCL was recorded, but not SCL, CCL was converted to SCL using the following linear regression function: $\text{SCL} = 1.245 +0.913\cdot\text{CCL}$ \citep{chaloupka2008cause}. Determination of size classes of turtles followed \cite{balazs1980synopsis}: juvenile--post hatchling to 65 cm SCL; subadult--from 65 to 81 cm SCL; adult--greater than $81$ cm SCL. 

The primary cause of stranding was based on direct observation and/or necropsy when available. Causes of stranding were classified into eight categories used previously by \cite{chaloupka2008cause}: fibropapillomatosis (FP), hook-and-line fishing gear, net and gillnet fishing gear, boat impact, shark attack, human-take, miscellaneous, and unknown.
FP strandings were turtles that had gross evidence of external tumors. 
Fishing gear strandings were identified by obvious signs of an interaction or entanglement with the particular gear (hook-and-line or net) \citep{boulon2000trends, chaloupka2008cause}. 
Boat impact strandings were recognized by the presence of a crushed carapace or deep cuts originating from propellers or hulls of boats \citep{boulon2000trends, guimares2021distribution}. 
Shark attack strandings included turtles with deep incisions or removal of soft tissue or body parts \citep{stacy2021scavenging}. 
Human-take (take is defined under the Endangered Species Act as ``to harass, harm, pursue, hunt, shoot, wound, kill, trap, capture, or collect, or to attempt to engage in any such conduct'') strandings were turtles with obvious evidence of having been butchered or poached, often accompanied with spear wounds \citep{boulon2000trends}. 
Miscellaneous strandings included turtles with natural, non-anthropogenic causes not fitting in any of the other categories (e.g., natural disasters, including weather and tsunami events; and internal diseases confirmed by necropsy), and unknown strandings were those for which no cause could be determined \citep{chaloupka2008cause}. 

\subsection{Statistical methods}
Chi-square goodness of fit tests were used to determine if there were equal proportions among months of stranding, stranding status, causes of stranding, and sex of stranded turtles for all of \Hawaii\ Island. When comparing west and east \Hawaii, contingency tables and chi-square tests of independence were used. All analyses were performed using the statistical software R version 4.2 \citep{rcore2022}. Statistical significance was accepted at $p < 0.05$. 

It is reasonable to model the occurrence of turtle stranding events as a Poisson process with a rate $\lambda$ that potentially changes through time and space. 
If strandings are classified  into groups, then there are two equivalent ways of modelling this: 
each class as an independent Poisson process with its own rate $\lambda_i$, 
or as a single overall process at rate $\lambda$ that generates an event at time $T$, and this event is then distributed to a class by a categorical random draw from some class distribution $\pi$ that potentially also depends on time $T$.

Of particular interest is the case where the categorical distribution $\pi$ does not change with time, which is equivalent to saying that the ratios between the class rates $\lambda_i$ are also constant. 
Probabilistically, the class is independent of the rate of the Poisson process. 
While the overall rate $\lambda$ at which turtle strandings are observed depends on population size and human reporting patterns, this model allows us to investigate potential changes in the cause distribution $\pi$ over time.

To this end, multinomial linear models with Poisson error structures were fit using the nnet package \citep{nnet} and model selection was carried out using Akaike Information Criterion \citep{akaike1974new}. 
These models produce a prediction function which may be interpreted as the class distribution $\pi(t)$, allowing us to compare models with and without a dependence on time.

\section{Results}

%\subsection{Stranding summary}

A total of \Sexpr{Ntot} green turtles stranded on \Hawaii\ Island from June 1983 to June 2022. 
Of those strandings, 
\Sexpr{CoastCount$West} (\Sexpr{percent(CoastCount$West/Ntot, 0.1, suffix="\\%")}) 
were located on the leeward or west coast of \Hawaii\ Island, while 
\Sexpr{CoastCount$East} (\Sexpr{percent(CoastCount$East/Ntot, 0.1, suffix="\\%")}) 
were located on the windward side or east coast of \Hawaii\ Island (Figure \ref{fig:map}, Table \ref{tab:monthside}). 

\begin{figure}[tbp]
\centering

<<map, fig.width=twocol, fig.height=twocol>>=
library(tmap)
library(sf)
hawaii.sf <- read_sf("Coastline/Coastline.shp") 


tm_shape(hawaii.sf, simplify=0.0125, bbox=c(-156.15, 18.91, -154.8, 20.27)) + 
  tm_graticules(alpha=0.4, n.x=4, n.y=4) + 
  tm_borders() + tm_compass() + tm_scale_bar() + 
  tm_fill(alpha=0.3) + 
  #tm_credits("PLACEHOLDER", position=c("center", "center"), size=1.5) + 
  tm_shape(BigIslandStrandings %>% 
             st_as_sf(coords=c("Long", "Lat"), crs=4326, na.fail=FALSE) %>% 
             filter(Site != "Mountain View")) + 
  tm_symbols(shape="Coast", title.shape="Coast", alpha=0.2, size=0.3, col="black", border.alpha=1, border.lwd =  1) + 
  tm_legend(position=c(0.95,0.95), just=c(1,1), frame=TRUE) 
@

\caption{Stranding locations and the division into eastern (windward) and western (leeward) sides of \Hawaii\ island. Coastline map courtesy of United States Geological Service (USGS) and \Hawaii\ Statewide GIS Program.}\label{fig:map}
\end{figure}

\begin{table}[tbp]
\caption{\color{black}Raw counts and proportions of stranding cause from 1983--2022 for \Hawaii\ island, separated into east and west sides.
Fibropapillomatosis is abbreviated FP.
%The distribution is significantly different between sides (Chi-square test, $X_7=70$, $p < 10^{-10}$).
}\label{tab:cause}
<<causetab>>=
cause_side_table <- BigIslandStrandings %>% count(Cause,Coast) %>% pivot_wider(names_from = Coast, values_from = n) 
cause_side_test <- cause_side_table %>% select(-Cause) %>% chisq.test()

cause_side_table %>% mutate(Total=East+West, 
                            across(-Cause, . %>% proportions %>% percent(0.1, suffix=""), .names="p_{.col}")) %>%
  relocate(Cause, East, p_East, West, p_West, Total, p_Total) %>%
  set_colnames(c("Cause", "n", "%", "n", "%", "n", "%")) %>%
  knitr::kable("latex", booktabs=T, linesep=c(rep("",7), "\\addlinespace")) %>%
  kableExtra::add_header_above(c(" ", "East"=2, "West"=2, "Total"=2))
@
\end{table}

Of the \Sexpr{Ntot} stranded turtles in the records, slightly over half had no discernable cause that could be determined (the ``unknown'' cause).
The most common known cause of stranding was hook-and-line fishing gear, accounting for about 1 in 5 strandings.
The distribution of causes is significantly different between the east and west coasts of the island (Chi-square test, $X_{\Sexpr{cause_side_test$parameter}}= \Sexpr{signif(cause_side_test$statistic, 3)}$, $p < 10^{-10}$), 
with the effect being driven most strongly by the FP and Miscellaneous categories (Table \ref{tab:cause}).

\subsection{Temporal trends}

The number of strandings on \Hawaii\ Island has fluctuated over the years but shows an overall increase over time (Figure \ref{fig:sides}). 
Across both sides of the island, strandings were less frequent in winter, November--February, than in other months (Table \ref{tab:monthside}). 
The highest totals were observed between March and August.
Raw counts of hook-and-line fishing gear strandings have steadily increased over the years, and while strandings with FP as the chief cause of stranding have remained overall low, the number of FP-caused strandings was higher after 2000 (Figure \ref{fig:cause}).
The second most common known cause of stranding in west \Hawaii\ was miscellaneous, a category that includes a significant number of strandings associated with the 2011 T\={o}hoku tsunami, while in east \Hawaii\ FP is the second leading cause (Table \ref{tab:cause}, Figure \ref{fig:cause}).
%For both west and east \Hawaii, shark attack and boat impact strandings have increased over time (Figure8) (Figure 9). 
%A reduction in human take strandings over time is seen on east \Hawaii, and a peak of miscellaneous strandings is seen in 2011 on west \Hawaii\ (Figure 8) (Figure 9). 

\begin{figure}[tbp]

\centering
<<strandings, fig.height=3>>=
(BigIslandStrandings %>% count(Year, Coast, name = "Strandings") %>% ggplot(aes(x=Year, y=Strandings, fill=Coast, color=Coast))) +
  geom_col() + geom_vline(xintercept = 2019.5, lty=2) + 
  scale_fill_grey(start=0.6, end=0.2, labels=c("East Hawaii", "West Hawaii"), aesthetics = c("color", "fill")) +
  labs(y="Number of strandings", x="Year") + 
  theme_bw() + 
  theme(legend.position=c(0.05,0.96), legend.direction = "vertical", legend.justification = c(0,1), 
         legend.background = element_rect(fill=alpha("white", 0.65)) )
@
\caption{Number of strandings from 1983--2022 for \Hawaii\ island, separated into east and west sides. Data for 2020 and beyond are incomplete due to COVID-19 disruptions to data collection.}
\label{fig:sides}
\end{figure}

\begin{table}[tbp]
\caption{\color{black}Strandings in each month for \Hawaii\ island, separated into east and west sides.}\label{tab:monthside}
<<monthside>>=
BigIslandStrandings %>% mutate(Month=month(Date, label=TRUE, abbr=TRUE)) %>% count(Month,Coast) %>%
  pivot_wider(names_from = Month, values_from=n) %>% 
  bind_rows(., summarize(., across(where(is.numeric), sum), Coast="Total")) %>%
  rowwise(Coast) %>% mutate(Total=sum(c_across(where(is.numeric)))) %>% 
  knitr::kable("latex", booktabs=T, linesep=c("","\\addlinespace"))
@
\end{table}




\begin{figure}[tbp]
\centering
<<cause_time, fig.height=5>>=
(BigIslandStrandings %>% count(Year, Cause, Coast, name="Strandings") %>% ggplot(aes(x=Year, y=Strandings, fill=Coast, color=Coast))) +
  geom_col() + 
  facet_wrap(vars(Cause), scales="fixed") + 
  geom_vline(xintercept = 2019.5, lty=2) + 
  scale_fill_grey(start=0.6, end=0.2, labels=c("East Hawaii", "West Hawaii"), aesthetics = c("fill", "color")) +
  labs(y="Number of strandings", x="Year") + 
  theme_bw() + 
  theme(legend.position=c(5/6,1/6), legend.direction = "vertical", legend.justification = c(0.5,0.5), 
         legend.background = element_rect(fill=alpha("white", 0.65)) )
@
\caption{Number of strandings from each cause, separated into east and west sides. Fibropapillomatosis is abbreviated FP. Data for 2020 and beyond are incomplete due to COVID-19 disruptions to data collection.}
\label{fig:cause}
\end{figure}



<<modelfit, include=FALSE>>=
threeClass <- BigIslandStrandings %>% 
  mutate(Cause=case_match(Cause,
                          c("Human take", "Hook/line", "Net", "Boat impact") ~ "Human",
                          c("Shark attack", "Misc.", "FP") ~ "Natural",
                          "Unknown" ~ "Unknown") %>% 
           factor(c("Human", "Unknown", "Natural")))

deg3model <- multinom(Cause~ns(Year, df=3), threeClass, Year >= 1985 & Year != 2011, trace=FALSE)
predFrame <- tibble(Year=seq(1983, 2022, by=0.25))

deg3preds <- predict(deg3model, newdata=predFrame, type="probs") %>% bind_cols(predFrame, .)

deg0model <- multinom(Cause~1, threeClass, Year >= 1985 & Year != 2011, trace=FALSE)
deg1model <- multinom(Cause~ns(Year, df=1), threeClass, Year >= 1985 & Year != 2011, trace=FALSE)
deg2model <- multinom(Cause~ns(Year, df=2), threeClass, Year >= 1985 & Year != 2011, trace=FALSE)
deg4model <- multinom(Cause~ns(Year, df=4), threeClass, Year >= 1985 & Year != 2011, trace=FALSE)
@

<<bootstrap, include=FALSE, cache=TRUE>>=
library(purrr)
bootReps <- 2000
clevel <- 0.025

bootReplicates <- rerun(bootReps,
                        multinom(Cause~ns(Year, df=3), 
                                 threeClass %>% slice_sample(prop=1, replace=TRUE), 
                                 Year>=1985 & Year != 2011,
                                 trace=FALSE) %>% 
                          predict(newdata=predFrame, type="probs") %>% 
                          bind_cols(predFrame,.)) %>% bind_rows(.id="run")

bootCI <- bootReplicates %>% group_by(Year) %>% 
  mutate(Human = 1-Human ) %>%
  summarise(across(c(Human, Natural),
                   list(Lower = ~ quantile(.x, clevel),
                        Upper = ~ quantile(.x, 1-clevel)))) %>%
  pivot_longer(-Year, names_to = c("Cause", ".value"), names_sep="_")
@



\begin{figure}[tbp]
<<modelplot, fig.height=4>>=
triclassplot <- BigIslandStrandings %>% 
  mutate(Cause=case_match(Cause,
                          c("Human take", "Hook/line", "Net", "Boat impact") ~ "Human",
                          c("Shark attack", "Misc.", "FP") ~ "Natural",
                          "Unknown" ~ "Unknown") %>% 
           factor(c("Human", "Unknown", "Natural"))) %>%
  count(Year, Cause, name="Strandings", .drop = TRUE) %>%
  group_by(Year) %>%
  mutate(Prop=proportions(Strandings)) %>% ggplot(aes(x=Year, y=Prop, color=Cause, fill=Cause))

triclassplot + geom_col() +
  geom_hline(yintercept=c(0.12, 1-0.30), lty=2, color="white") + 
  scale_fill_grey(start=0.8, end=0.2, aesthetics=c("color","fill"),
                  name="Cause", labels=c("Human", "Unknown", "Predation, disease\nand weather")) +
  geom_ribbon(aes(x=Year,ymin=Lower, ymax=Upper, group=Cause), data=bootCI, 
              inherit.aes = FALSE, alpha=0.25, fill="white", color="white") +
  geom_line(aes(x=Year, y=1-Human), deg3preds, color=1, lty=1, inherit.aes = FALSE) +
  geom_line(aes(x=Year, y=Natural), deg3preds, color=1, lty=1, inherit.aes = FALSE) +
  annotate("text", x=c(1983,1984,2011), y=0.01, label="*", size=5, color=c("black", "black", "white")) + 
  labs(y="Proportion of strandings", x="Year") + 
  theme_bw() + 
  theme(legend.position=c(0.5,0.96), legend.direction = "horizontal", legend.justification = c(0.5,1), 
         legend.background = element_rect(fill=alpha("white", 0.55)))
@
\caption{A multinomial regression fit using natural splines with 3 degrees of freedom. 95\% confidence bands are constructed by bootstrapping. Records from years with asterisks (*) are excluded from the model. The dotted white lines correspond to a model with no dependence on year. 
%Akaike Information Criterion (AIC) of various models: zero degrees of freedom, $\text{AIC}=1300.9$; 1 degree, $\text{AIC}=1302.9$; 2 degrees, $\text{AIC}=1304.7$; 3 degrees, $\text{AIC}=1301.0$; 4 degrees, $\text{AIC}=1299.7$
}\label{fig:model}
\end{figure}


To investigate changes in the relative rates of stranding causes, multinomial log-linear models were fit using date of record as a predictor. 
To reduce the variance of the fitted model parameters, the causes as recorded were consolidated into 3 categories: 
Human-caused (hook-and-line, boat impact, human take, and net); predation, disease and weather (shark attack, FP, and Misc.); and the original unknown category. 
The 2011 T\={o}hoku tsunami-related strandings, as well as the records prior to 1985 were excluded from the model fit.
The Akaike Information Criterion (AIC) is used to compare a series models of models using natural splines based on date of standing with increasing degrees of freedom.
The AIC increases going from a null model (AIC \Sexpr{signif(deg0model$AIC,5)}) with no dependence on year
to a predictor function with 2 degrees of freedom (AIC \Sexpr{signif(deg2model$AIC,5)}), 
and then slightly decreases again, so that a 4 degree of freedom model (AIC \Sexpr{signif(deg4model$AIC,5)})
has an AIC \Sexpr{signif(deg0model$AIC - deg4model$AIC,2)} smaller than the null model. 
Figure \ref{fig:model} displays a 3-degree of freedom model (AIC \Sexpr{signif(deg3model$AIC,5)}),
with confidence bands constructed using the bootstrap.
The null model is represented by dotted white lines, and apparently fits within the confidence bands of the model that includes dependence on year. 
These results show a lack of evidence that the date of stranding provides significant information about the relative rates of stranding among the three consolidated cause categories.

\subsection{Size and gender}

<<size_gender, include=FALSE>>=
side_scl_table <- BigIslandStrandings %>% group_by(Coast) %>% filter(!is.na(SCL)) %>% 
  summarize(mean=mean(SCL), sd=sd(SCL), n=n(), se=sd/sqrt(n), min=min(SCL), max=max(SCL)) %>%
  column_to_rownames("Coast")

scl_table <- BigIslandStrandings %>% filter(!is.na(SCL)) %>% 
  summarize(mean=mean(SCL), sd=sd(SCL), n=n(), se=sd/sqrt(n), min=min(SCL), max=max(SCL))

size_class <- BigIslandStrandings %>% filter(!is.na(SCL)) %>% 
  count(Class=case_when(SCL < 65 ~ "Juvenile", 
                        SCL < 81 ~ "Subadult", 
                        SCL >= 81 ~ "Adult")) %>%
  column_to_rownames("Class")

size_cause_test <- aov(SCL~Cause, BigIslandStrandings) %>% summary %>% extract2(1)

gender_table <- BigIslandStrandings %>% count(Sex) %>% column_to_rownames("Sex") %>% t %>% drop

gender_side_table <- BigIslandStrandings %>% count(Sex,Coast) %>% 
  pivot_wider(names_from = Coast, values_from = n) %>% column_to_rownames("Sex")

gender_side_test <- gender_side_table %>% chisq.test()

side_size_test <- t.test(SCL~Coast, BigIslandStrandings)

@


Stranded turtles in the records ranged from 
\Sexpr{signif(scl_table$min,3)} cm to 
\Sexpr{signif(scl_table$max,3)} cm straight carapace length (SCL), with a mean of 
\Sexpr{signif(scl_table$mean,3)} cm, across 
\Sexpr{size_class["Juvenile","n"]} juveniles, 
\Sexpr{size_class["Subadult","n"]} subadults, and 
\Sexpr{size_class["Adult","n"]} adults.
No carapace length measurement was recorded in \Sexpr{Ntot-scl_table$n} of the case reports.
Turtles stranding in east \Hawaii\ 
($\mu\pm\text{SE} = \Sexpr{side_scl_table["East", "mean"] %>% signif(3)} 
\pm \Sexpr{side_scl_table["East", "se"] %>% signif(2)}$ cm SCL, $n=\Sexpr{side_scl_table["East", "n"]}$)
%\orig{($\mu\pm\text{SE}=58.8\pm0.78$ cm SCL, range 19.8--99 cm)} 
were significantly larger
(t-test, $t_{\Sexpr{round(side_size_test$parameter)}}=\Sexpr{signif(side_size_test$statistic,3)}$,
$p=\Sexpr{signif(side_size_test$p.value, 1)}$)
%\orig{(t-test unequal variance, $t_{375}=6.32$, $p=7\times 10^{-10}$)}
than those in west \Hawaii\ 
($\mu\pm\text{SE} = \Sexpr{side_scl_table["West", "mean"] %>% signif(3)} 
\pm \Sexpr{side_scl_table["West", "se"] %>% signif(1)}$ cm SCL, $n=\Sexpr{side_scl_table["West", "n"]}$)
%\orig{($\mu\pm\text{SE} = 51.3\pm0.51$ cm SCL, range 19.8--83 cm)}. 
Figure \ref{fig:length_cause} shows SCL distributions for each cause, and while the distribution of SCL is not independent of cause (ANOVA, 
$F(\Sexpr{size_cause_test$Df})=\Sexpr{size_cause_test["Cause","F value"] %>% signif(3)}$, 
$p=\Sexpr{size_cause_test["Cause","Pr(>F)"] %>% signif(2)}$), 
the differences between the groups are small compared to the within-group variances.
The records contain 
\Sexpr{gender_table["F"]} female, 
\Sexpr{gender_table["M"]} male, and 
\Sexpr{gender_table["U"]} gender undetermined cases, also with marginally different distributions between sides of the island (Chi-square, 
$X_{\Sexpr{gender_side_test$parameter}}=\Sexpr{signif(gender_side_test$statistic,2)}$,
$p=\Sexpr{signif(gender_side_test$p.value, 2)}$). 


\begin{figure}[tbp]
<<length_cause, fig.height=3>>=
ggplot(BigIslandStrandings, aes(x=Cause, y=SCL)) + geom_boxplot() + labs(y="Carapace length (cm) ") +  theme_bw()
@
\caption{Straight carapace length (SCL) was measured in 488 records, and plotted for each stranding cause. Fibropapillomatosis is abbreviated FP. Boxplot outliers begin at 1.5 times the inter-quartile distance.}\label{fig:length_cause}
\end{figure}


\subsection{FP tumor presence/absence}

\begin{table}[tbp]
\caption{\color{black}Fibropapillomatosis tumor presence in stranded turtles by side of \Hawaii\ island.
%Tumor distribution is unequal on the two sides of the island (Chi-square test, $X_2=198$, $p < 10^{-10}$).
}\label{tab:tumor_side}
<<tumor_side>>=
tumor_side_table <- BigIslandStrandings %>% count(Coast, Tumor) %>% 
  pivot_wider(names_from = Tumor, values_from = n)

tumor_side_test <- tumor_side_table %>% select(-Coast) %>% chisq.test()

tumor_side_table %>% relocate(Coast, Present=Y, None=N, "Not Recorded"=U) %>%
  bind_rows(., summarise(., across(where(is.numeric), sum), Coast="Total")) %>%
  knitr::kable("latex", booktabs=T, linesep=c("","\\addlinespace")) %>%
  kableExtra::add_header_above(c(" ", "Tumor"=3))
@
\end{table}

As shown in Table \ref{tab:tumor_side}, 
\Sexpr{sum(tumor_side_table$N)} records indicated the absence of FP tumors,
\Sexpr{sum(tumor_side_table$Y)} records presence of a tumor, and
\Sexpr{sum(tumor_side_table$U)} records are missing this observation. 
Note that the presence of a FP tumor does not necessarily mean that the primary cause of stranding was recorded as FP. 
Tumor presence/absence is significantly associated with side of the island, with turtles stranding in east \Hawaii\ more likely to have tumors than those in west \Hawaii\ (Chi-square test, $X_2=\Sexpr{number(tumor_side_test$statistic)}$, $p < 10^{-10}$).

\subsection{Stranding status}

\begin{table}[tbp]
\caption{\color{black}Survival status of stranded turtles by cause. Fibropapillomatosis is abbreviated FP.}\label{tab:cause_status}
<<cause_status>>=
cause_status_table <- BigIslandStrandings %>% count(Cause, Status=InitialStatus) %>%
  pivot_wider(names_from = Status, values_from = n, values_fill = 0) %>%
  rename(Alive=A, Dead=D, "Not Recorded" = U) 

cause_status_test <- cause_status_table %>% select(Alive, Dead) %>% chisq.test()

cause_status_table %>% bind_rows(., summarize(., across(where(is.numeric), sum), Cause="Total")) %>%
  knitr::kable("latex", booktabs=T, linesep=c(rep("",7), "\\addlinespace"))
@
\end{table}

\begin{table}[tbp]
\caption{\color{black}Survival status of stranded turtles by month.}\label{tab:month_status}
<<month_status>>=
BigIslandStrandings %>% mutate(Month=month(Date, label=TRUE, abbr=FALSE)) %>%
  count(Month, Status=InitialStatus) %>%
  pivot_wider(names_from = Status, values_from = n, values_fill = 0) %>%
  rename(Alive=A, Dead=D, "Not Recorded"=U) %>%
  knitr::kable("latex", booktabs=T, linesep="")
@
\end{table}

<<side_status, include=FALSE>>=
side_status_table <- BigIslandStrandings %>% count(Coast, InitialStatus) %>% 
  pivot_wider(names_from = InitialStatus, values_from = n, values_fill = 0) %>%
  column_to_rownames("Coast")
side_status_test <- side_status_table %>% select(-U) %>% chisq.test()
@


Of all the stranded turtles, 
\Sexpr{sum(side_status_table$A)} stranded alive, 
\Sexpr{sum(side_status_table$D)} stranded dead, and 
\Sexpr{sum(side_status_table$U)} turtles had no stranding status reported.
Stranding status was found to be significantly associated with cause 
(Chi-square test, $X_{\Sexpr{cause_status_test$parameter}} = \Sexpr{number(cause_status_test$statistic)}$,
$p < 10^{-10}$).
More turtles stranded alive than dead because of FP, hook-and-line, and miscellaneous, while boat impact, human take, shark attack, and unknown were causes more likely to result in dead stranded turtles. 
Net fishing gear strandings showed equal numbers of turtles that stranded alive and dead (Table \ref{tab:cause_status}).
More turtles stranded alive than dead in the months of November--March, while more turtles stranded dead than alive in the months of April--October (Table \ref{tab:month_status}).
Stranding status was also found to be significantly associated with stranding location (Chi-square test,
$X_{\Sexpr{side_status_test$parameter}} = \Sexpr{signif(side_status_test$statistic, 3)}$, 
$p = \Sexpr{side_status_test$p.value %>% signif(2)}$).
West \Hawaii\ had
\Sexpr{side_status_table["West","A"]} turtles strand alive and
\Sexpr{side_status_table["West","D"]} strand dead, 
while east \Hawaii\ had 
\Sexpr{side_status_table["East","A"]} turtles strand alive and
\Sexpr{side_status_table["East", "D"]} strand dead. 

%%%%%%%%%%%%%%%
%%% DISCUSSION
%%%%%%%%%%%%%%%

\section{Discussion}

\Sexpr{str_to_sentence(english(nrow(BigIslandStrandings)))} green turtles were recorded stranded on \Hawaii\ Island in the period 1983-2022, which represents an unknown fraction of total strandings on Hawaiian shores in that time. 
Stranding programs rely on reports from the public, and are therefore dependent on the density of human activity at the shoreline as well as public knowledge of the reporting procedures.
However, if a location is regularly accessed by more than a few people, a stranding is likely to be reported, and it is reasonable to believe that this will happen independently of the variables observed in these records.
%However, strandings that are reported are probably an accurate reflection of the distribution, demographics, and causes of \Hawaii\ Island strandings. 

Strandings on \Hawaii\ Island showed an overall increase in rate between 1983 and 2022. Green turtle strandings have also increased on the other main Hawaiian Islands since 1982 \citep{chaloupka2008cause}.
One important reason for this increase is a positive one: Green turtle populations in the Hawaiian Islands have recovered significantly since their 1974 protection by the State of \Hawaii\ under Regulation 36 and their 1978 protection under the Endangered Species Act \citep{balazs2004thirty, bennett2008book}.
The increase in turtle population size will directly lead to additional observed stranding events, even if the risk to an individual turtle remains constant over time \citep{boulon2000trends}. 
Additionally, the human population increase on \Hawaii\ Island and the rise in numbers of visitors at the shoreline increase the chance of encountering and reporting a stranding. 
In general, the locations of strandings shown in Figure \ref{fig:map} reflect beaches and other shoreline areas with easy public access.
Increased public awareness of strandings and response programs and the greater use of cell phones and the internet probably have led to more reporting over time.
However, the increase in reported strandings appears to slow in the early 2000s (Figure \ref{fig:sides}), stabilizing at approximately 25--30 per year.
This trend was also noticed in studies covering the other main Hawaiian Islands \citep{chaloupka2008cause}.
In a turtle carcass drifter experiment along the shores of the Mississippi, public reporting of stranded carcasses was unexpectedly low: on popular mainland beaches, only 50\% of the stranded turtles were reported; on accessible, but more remote barrier islands, only 11.1\% of stranded carcasses were reported by citizens, and 0\% of turtle carcasses that drifted into marshes were reported \citep{cook2021use}.  
These results send a strong message that remoteness of and public accessibility to stranding areas greatly influence the  discovery of turtles, and that even "structured stranding networks with established reporting mechanisms" may be overestimating the rate of reporting by the public, which influences the conclusions that can be drawn from citizen-derived data.

There are two years post 2005 which show an unusually large number of green turtle strandings: 2011 and 2018. The peak in 2011 is associated with the March 2011 magnitude 9.0 T\={o}hoku earthquake off the coast of Japan and the subsequent tsunami, large waves, and hazardous currents that it caused around \Hawaii\ Island, and particularly its western shoreline \citep{cheung2013surges}. 
The waves and currents associated with tsunamis bring marine life onshore with them and can wash turtles inland. 
Two hawksbill turtles were reported stranded in \Hawaii\ as a result of the 2011 earthquake \citep{brunson2022three}, and a 2009 tsunami in Samoa similarly led to 52 turtles stranding on land \citep{bell2011marine}. 
The apparent downward trend of strandings after 2018 is probably not because fewer turtles stranded, but is rather due to human behavioral changes caused by the COVID-19 pandemic.
Throughout the pandemic, people in general spent much less time in public locations, and for some periods, \Hawaii\ County and State beach parks were closed for recreational use by executive decree \citep{emergencyNo2, emergencyFifth}.
Similarly, tourism to the island and state was heavily restricted.
All of these factors lead to a sharp drop in the number of people visiting \Hawaii\ Island coasts, and subsequent reports of strandings. 

The highest rates of green turtle strandings occurred during the Hawaiian spring and summer months, from March--August.
This is similar to the findings on \Oahu\ where green turtle strandings were highest from March--June \citep{chaloupka2008cause}, and for adult hawksbills in the Hawaiian Archipelago where strandings were highest from June--September \citep{brunson2022three}. 
Similarly, strandings of loggerhead, green, and leatherback turtles in Brazil were highest during the austral spring and summer seasons \citep{monteiro2016long}.
Peak sea turtle stranding months during 2010-2019 in the Gulf of Mexico were also in the spring to summer (March to August) \citep{cook2021use, howell2021}.
Strandings on \Hawaii\ Island were lowest during the months of September, November, and December, but a secondary peak in the month of October was seen.
This same peak was observed in the 2022 green turtle strandings on Maui \citep{cutt2023impact} and
\Oahu\ showed a similar secondary peak of strandings in September \citep{chaloupka2008cause}. 
Although the major Hawaiian green turtle nesting season is mid-April to September/October in the Northwestern Hawaiian Islands \citep{niethammer1997reproductive},  no seasonal variation in green turtle abundance within localized coastal Hawaiian foraging grounds has been documented (Balazs unpublished).
The higher spring/summer stranding patterns seen on \Hawaii\ Island may reflect seasonal differences in water temperature which affects carcass decomposition rates \citep{cook2021use}, periodic shifts in shoreline human activity and stranding reporting, or cyclical changes in surf, currents, and winds that can push carcasses to shore.
In the Hawaiian Islands, northeasterly trade winds are the most common weather pattern, especially in the summer; however, other weather patterns could influence turtle carcass drift, including migratory mid-latitude low pressure systems that are common October to April with about nine fronts passing over \Hawaii\ Island in a season, during which winds shift from southwesterly to northerly; Kona Storms or cold-core lows, November to April, although rare with unpredictable paths, that can cause waterspouts, torrential rain, and high surf; tropical cyclones from June to mid-November, that can bring high surf, storm surge, and strong onshore winds to \Hawaii\ Island \citep{longman2021twenty,nullet2023hawaii}; in addition, El Ni\~{n}o Southern Oscillation events that sporadically impact the Hawaiian Islands can cause weakened trade winds, less rainfall, and warmer ocean temperatures (El Niño phase) between November to April or stronger trade winds, greater rainfall, and cooler ocean temperatures (La Niña phase) \citep{longman2021impacts}.

Hook-and-line fishing gear was the most common known cause of stranding of green turtles on \Hawaii\ Island as a whole.
Fishing gear strandings show a similar qualitative pattern to the overall time series (Figure \ref{fig:cause}), increasing from 1983 to the mid 2000s and then apparently leveling off. 
\cite{chaloupka2008cause} found a similar increase of hook-and-line fishing gear strandings since 1982.
It is difficult to untangle the effects of the increased population of Hawaiian green turtles from the risk of hazard from fishing activity and gear, as both factors directly affect the rate of strandings observed.

Hawaiian green turtles are frequently reported with hooks intact and line entangled around their flippers and body. 
These interactions are often a result of lost and/or discarded fishing gear or fishers cutting the line when accidental hooking occurred, which illustrates the need for stronger management and preventatives \citep{nitta1993review}. 
Hook-and-line fishing gear strandings were also prevalent on \Oahu, Maui, and Kauai, making up the second most common cause of stranding of green turtles \citep{chaloupka2008cause}.
Similar to the findings of the present study, fishing gear was the foremost cause of stranding for green turtles on Maui in 2022, with 81\% of the total strandings showing interactions \citep{cutt2023impact}.

The number of hook-and-line strandings may be even greater than estimated. 
\cite{work2015causes} performed necropsies (postmortem autopsies) on stranded turtles throughout the Pacific and found that 48\% of foreign body ingestion cases (mostly all associated with fishing gear) showed no external sign of fishing line interactions.
Green turtle strandings resulting from interactions with fishing gear are prevalent around the world, including the U.S. Virgin Islands \citep{boulon2000trends}, Brazil \citep{guimares2021distribution}, Taiwan \citep{cheng2019twenty}, New Caledonia \citep{read2023twenty}, and Greece \citep{panagopoulos2003stranding}. 
However, unlike the line/hook entanglements on \Hawaii\ Island, in Taiwan, pond nets were the most common fishing gear causing turtle strandings over 23 years \citep{cheng2019twenty}. 
In contrast, at Samanda\u{g} Beach on the eastern Mediterranean coast of southern T\"{u}rkiye (Turkey), from 2002-2017, fishing activities caused only 7\% of the green turtle standings, while marine pollution accounted for 56\% of strandings \citep{sonmez2018sixteen}. Fibropapillomatosis was the second most common cause of stranding on \Hawaii\ Island, whereas \cite{chaloupka2008cause} found FP to be the main cause of stranding in green turtles in \Oahu, Maui, and Kauai. 

The relative rates of strandings by cause over time is of particular interest for managers and conservationists because it can indicate particular sources of danger to turtle populations. 
The overall rate of observation depends on population size and human reporting behavior in a complex way that is difficult to disentangle, but by looking at the distribution of causes over time we may be able to identify structural changes in the cause of strandings.
Although slightly over half of the stranded green turtles in this study were listed with ``unknown'' cause of mortality, these turtles still provide valuable temporal, geographic, and biological data. 
We share this predicament of unknown cause with others studying sea turtle strandings. 
For example, 50\% of strandings in New Caledonia were unknown, defined as ``no necropsies were done and no apparent cause of death by external examination'' \citep{read2023twenty}. 
\citet{chaloupka2008cause} also had high rates of strandings with unknown causes.  
In our study and others, given the circumstances of discovery (time, weather patterns, location, retrieval), condition of the animal (undetermined health and behavior prior to stranding, unspecified time of death, decomposition, or scavenging), and limited resources for extensive diagnostic procedures (necropsies, histopathology, toxicology, and microbiology), many stranded turtles remain forever in the category of unknown causes of mortality.
However, the goal should be to increase reporting of strandings by the public, to encourage detailed observations at time of discovery, and develop systematic survey programs by scientists to detect stranded turtles even in areas not frequented by the public, because robust understanding of stranding patterns and causes of mortality is key to the survival of green turtles in \Hawaii.

In this study, the record collection process kept eight categories of stranding cause, however, for modelling purposes we reduced these to three broad categories: direct intentional and accidental human causes, such as boat impacts and fishing and hunting related injuries; natural events, predation, and disease; and unknown causes.
The distribution of these three consolidated causes has been relatively stable since the early 1990s (Figure \ref{fig:model}), providing unconvincing evidence of any major shifts between the relative risks between direct human causes and the other categories.
One way of interpreting this result is that the increased numbers of strandings over time can be explained entirely by the growth in turtle populations and increases in reporting by the public. 
While keeping the proportion of human-caused strandings constant over time may be regarded as a minor conservation success story, given the significant growth in human population and coastal activity over the same time period, humans remain a significant source of danger to turtles.
There remains much room for improvement, in particular with regards to hook-and-line injuries.


The current study found that different sides of \Hawaii\ Island had different distributions of stranding cause. West \Hawaii\ Island had a higher proportion of shark attack and boat impact strandings, while east \Hawaii\ had more FP and human take strandings.
Increased shark attack strandings on west \Hawaii\ may be because of the larger population of tiger sharks found along the west coast \citep{meyer2009long}. 
Tiger sharks are well-known predators of sea turtles, and green turtles are found regularly in their stomach contents \citep{witzell1987selective, lowe1996ontogenetic}. 
West \Hawaii\ also has a large tourism industry, with many snorkel, diving, and manta ray and marine mammal watching tours operating in the same coastal waters that green turtles occupy. 
These tours, as well as commercial vessels, frequent the many shallow bays located in west \Hawaii\ that are important foraging habitats for green turtles. 
Increased boat presence accompanied with high vessel speeds, varying water depth, and times of poor visibility can all factor into a higher proportion of boat impact strandings on the west side of the island \citep{fuentes2021conservation}.

The majority of green turtles that stranded on \Hawaii\ Island were juveniles. Similarly, juveniles predominated the stranded green and hawksbill turtles throughout the Hawaiian Islands \citep{chaloupka2008cause, brunson2022three}.
Juvenile green turtles were also the most common size class stranded in New Caledonia \citep{read2023twenty}, Australia \citep{flint2015trends}, and Brazil \citep{monteiro2016long}. 
However, in T\"{u}rkiye and Taiwan, most green turtle strandings involved sub-adult and juvenile turtles \citep{sonmez2018sixteen, cheng2019twenty}.
The high proportion of juveniles stranding may be a result of increased nesting populations at French Frigate Shoals in the Northwestern Hawaiian Islands leading to an increase in juveniles moving from nesting to foraging areas \citep{balazs2004thirty}.
Juvenile turtles may be more immunologically na\"{i}ve and susceptible to environmental stressors that could contribute to stranding \citep{flint2015trends}. 

Larger turtles stranded on east \Hawaii\ Island than on west \Hawaii, despite the fact that stranded turtles with the highest SCL values were the result of shark attacks.
\cite{bornatowski2012shark} found that the probability that a green turtle in Brazil stranded with a shark bite increased with size, and \cite{chaloupka2008cause} reported the same trend for green turtles in the main Hawaiian Islands.
Smaller green turtles are also frequently attacked by sharks, but may be completely consumed and thus do not wash ashore after such event.
A spatial trend in size-classes was also reported by \cite{chaloupka2008cause}: larger turtles stranded on Maui and Kauai than on \Oahu.

There was no gender-bias of stranded green turtles on \Hawaii\ Island: male and female strandings occurred with a 1:1.06 ratio.
The lack of a gender-bias for green turtles was also shown in the main Hawaiian Islands \citep{work2004retrospective, chaloupka2008cause}. 
The present and prior studies are consistent with the 1:1 sex ratio of Hawaiian green turtles found by \cite{wibbels1993sex}. Unlike in the Hawaiian Islands, many green turtle populations around the world appear to have more females than males \citep{flint2010health, cheng2019twenty, read2023twenty}. 
Clutches of sea turtles are sensitive to temperature change, and an increase in the temperature during incubation can drastically change sex ratios of nests, leading to clutches of all females. 
As temperatures continue to rise as a result of climate change, the Hawaiian population of green turtles may eventually see the same skew seen in other locations around the world \citep{hawkes2009climate}.

More than 60\% of the stranded turtles on \Hawaii\ Island had no tumors indicative of FP.
No cases of internal FP tumors have been reported without the presence of external tumors \citep{work2004retrospective}. 
A decline in FP prevalence has been documented previously in Hawaiian green turtles. 
Twenty-one of 66 turtles observed with tumors in one summer on Maui were seen later with no tumors \citep{bennett2000photographic}, indicating regression of the FP.  
The low number of stranded turtles with FP on \Hawaii\ Island is consistent with the 2022 stranding report for green turtles on Maui, in which only one case of FP was reported \citep{cutt2023impact}.

Turtles were more likely to have FP on east \Hawaii, whereas FP was very rare on green turtles that stranded on west \Hawaii. 
The west (Kona) coast of \Hawaii\ Island had no diagnosed cases of FP for many of the years that FP was prevalent in the other Hawaiian Islands \citep{balazs1991current, aguirre2000blood, work2001immune}. 
In Florida, turtles with tumors are more likely to become entangled in fishing line, thus the higher percentage of hook-and-line strandings that occurred on east \Hawaii\ may be a result of higher FP presence \citep{foley2005fibropapillomatosis}. 
However, \cite{chaloupka2008cause} found no correlation between FP and fishing gear strandings in the other main Hawaiian Islands. 
Similar to the spatial variation in FP infection on \Hawaii\ Island, FP was more often found in \Oahu\ and Maui than on Kauai \citep{chaloupka2008cause}. 
Green turtles that stranded on the western (Gulf) coast of Florida (51.9\%) were more likely to have tumors than turtles that stranded on the eastern (Atlantic) coast (11.9\%) \citep{foley2005fibropapillomatosis}. 
In Australia, FP varied in prevalence from 0 to 11.6\% at 15 sites all along the Queensland coast \citep{jones2022spatial}. 

A variety of factors have been hypothesized for the varying prevalence of FP in different locations and may be the reason for the contrasting FP abundance on west and east \Hawaii\ Island.
For example, FP in Florida was greatest in areas with the greatest habitat degradation and pollution, most shallow water areas, and lowest wave-energy level, indicating that one or more of these conditions may affect FP \citep{foley2005fibropapillomatosis}.
In Brazil, highly urbanized areas have a higher FP prevalence than lightly urbanized areas \citep{bastos2022coastal}. 
Additionally, FP may be related to water temperature, with higher water temperatures correlated with greater FP prevalence \citep{manes2022occurrence}. 
An important factor that could contribute to the absence of FP on west \Hawaii\ is the precipitation pattern on the leeward side of the island.
The windward (east) side, experiences abundant, consistent rainfall and has large rivers and many streams \citep{juvik1998atlas}. 
Heavy rain may bring more land-based pollutants to rivers, and the discharge from these rivers located in urbanized areas may disrupt the immune system of green turtles, making them more susceptible to FP \citep{manes2022occurrence}. 
Despite the low rainfall and lack of flowing surface water in west \Hawaii, coastal waters can experience nutrient pollution via submarine groundwater discharge, which could impact green turtle health \citep{abaya2018spatial, abaya2018multi, panelo2022spatial}.
	
The relatively even distribution of turtles that stranded alive (\Sexpr{sum(side_status_table$A)}) versus dead \Sexpr{sum(side_status_table$D)} on \Hawaii\ Island in the present study is markedly different  from other research on O'ahu, Maui, and Kauai where approximately 75\% of green turtles stranded dead \citep{chaloupka2008cause} and on Taiwan where 80\% of stranded turtles were deceased \citep{cheng2019twenty}.
In the present study, stranding status was found to vary temporally, by cause, and spatially. 
Green turtles were more likely to strand alive in the winter months (November--March), and dead in the summer months (April--October).
Additionally, more turtles stranded dead than alive because of boat impacts, human take, and shark attacks, similar to other Hawaiian Islands, where boat impact and shark attack were the hazards most likely to result in a dead turtle \citep{chaloupka2008cause}.
Shark attacks often cause the loss of appendages and boat impacts usually cause damage to the head, appendages, and/or the carapace, all serious injuries that lead to significant mortality.
The present study found that turtles that stranded as a result of FP were more likely to strand alive than dead, unlike findings of \cite{chaloupka2008cause}. 
More turtles stranded dead than alive on west \Hawaii and more turtles stranded alive than dead on east \Hawaii, probably because shark attacks and boat impacts are more common on west \Hawaii, while FP is reported almost exclusively on eastern shores. 
\cite{chaloupka2008cause} found that the probability of mortality in a stranding decreased with turtle size, and this pattern is also observed in this data across the two sides of \Hawaii\ Island.

\section{Conclusions}

Despite the large percentage of unknown causes of stranding, this long-term data set provides important information on \Hawaii\ Island green turtle strandings. 
The considerable contribution of hook-and-line fishing gear to strandings emphasizes the need for additional mitigation efforts, such as barbless hooks and effective line removal techniques (\url{https://dlnr.hawaii.gov/dobor/marineanimalhotline/}). In \Hawaii, the public has a high level of awareness of sea turtles, but we have an imperative to increase the availability of information on what a person should do if a hooked, entangled, injured or stranded turtle is found. Continued monitoring of turtle strandings and careful data collection on stranded individuals are critical to the conservation of green turtles. 

\section{Acknowledgements}
We would like to thank the following individuals, agencies, and organizations that contributed to data collection, data availability, and/or mentorship over the last four decades:
Summer Martin, 
Brittany Clemans, 
Shandell Brunson (deceased), 
Shawn Murakawa,
Irene Kelly,
Jen Sims, 
Megan Lamson, 
Rebecca Ostertag, 
John Coney,
Leon Hallacher,
Walter Dudley,
Jason Turner,
NOAA Marine Turtle Biology \& Assessment Program,
NOAA Pacific Islands Fisheries Science Center, 
NOAA Pacific Islands Regional Office, 
University of  \Hawaii\ at Hilo Marine Option Program Sea Turtle Stranding Response Team,
\Hawaii\ Preparatory Academy Sea Turtle Research Program,
and the State of \Hawaii\ Department of Land and Natural Resources.

\section{Declarations}

\subsection{Author contributions}
Study conception and design by Skylar Dentlinger, George Balazs, and Karla J. McDermid. 
Data compilation, initial analysis and first draft prepared by Skylar Dentlinger.
Follow-up analyses, modeling, and final manuscript and code preparation by Grady Weyenberg.
All authors made significant contributions to editing and revision of the manuscript, provided important intellectual content, and have read and approved the finalized version.

\subsection{Competing Interests}
The authors declare that no funds, grants, or other support were received to assist in the preparation of this manuscript.
The authors have no relevant financial or non-financial interests to disclose.

\subsection{Data Availability}
All data analyzed during this study are included in this published article and its supplementary information files.

\subsection{Ethics approval}
This is an observational study compiled from publicly available records. 
No ethical approval is required.


\bibliography{turtles}

\clearpage

\listoffigures

\end{document}