Background

Overview

Teaching: 5 min
Exercises: 0 min

Questions

• What is the Ocean Tracking Network?

• How does your local telemetry network interact with OTN?

• What methods of data analysis will be covered?

Objectives

Intro to OTN

The Ocean Tracking Network (OTN) supports global telemetry research by providing training, equipment, and data infrastructure to our large network of partners.

OTN and affiliated networks (including FACT) provide automated cross-referencing of your detection data with other tags in the system to help resolve “mystery detections” and provide detection data to taggers in other regions. OTN’s Data Managers will also extensively quality-control your submitted metadata for errors to ensure the most accurate records possible are stored in the database. OTN’s database and Data Portal website are excellent places to archive your datasets for future use and sharing with collaborators. We offer pathways to publish your datasets with OBIS, and via open data portals like ERDDAP, GeoServer etc. The data-product format returned by OTN is directly ingestible by analysis packages such as glatos and resonATe for ease of analysis. OTN offers support for the use of these packages and tools.

Learn more about OTN and our partners here https://members.oceantrack.org/. Please contact OTNDC@DAL.CA if you are interested in connecting with your regional network and learning about their affiliation with OTN.

Intended Audience

This set of workshop material is directed at researchers who are ready to begin the work of acoustic telemetry data analysis. The first few lessons will begin with introductory R - no previous coding experince required. The workshop material progresses into more advanced techniques as we move along, beginning around lesson 8 “Introduction to Glatos”.

If you’d like to refresh your R coding skills outside of this workshop curriculum, we recommend Data Analysis and Visualization in R for Ecologists as a good starting point. Much of this content is included in the first two lessons of this workshop.

Getting Started

Please follow the instrucions in the “Setup” tab along the top menu to install all required software, packages and data files. If you have questions or are running into errors please reach out to OTNDC@DAL.CA for support.

Intro to Telemetry Data Analysis

OTN-affiliated telemetry networks all provide researchers with pre-formatted datasets, which are easily ingested into these data analysis tools.

Before diving in to a lot of analysis, it is important to take the time to clean and sort your dataset, taking the pre-formatted files and combining them in different ways, allowing you to analyse the data with different questions in mind.

There are multiple R packages necessary for efficient and thorough telemetry data analysis. General packages that allow for data cleaning and arrangement, dataset manipulation and visualization, network analysis and temporo-spatial locating are used in conjuction with the telemetry analysis tool packages VTtrack, actel and glatos.

There are many more useful packages covered in this workshop, but here are some highlights:

Intro to the glatos Package

glatos is an R package with functions useful to members of the Great Lakes Acoustic Telemetry Observation System (http://glatos.glos.us). Developed by Chris Holbrook of GLATOS, OTN helps to maintain and keep relevant. Functions may be generally useful for processing, analyzing, simulating, and visualizing acoustic telemetry data, but are not strictly limited to acoustic telemetry applications. Tools included in this package facilitate false filtering of detections due to time between pings and disstance between pings. There are tools to summarise and plot, including mapping of animal movement. Learn more here.

Maintainer: Dr. Chris Holbrook, ( cholbrook@usgs.gov )

Intro to the VTrack Package

This package was designed to facilitate the assimilation, analysis and synthesis of animal location and movement data collected by the VEMCO suite of acoustic transmitters and receivers. As well as database and geographic information capabilities the principal feature of VTrack is the qualification and identification of ecologically relevant events from the acoustic detection and sensor data. This procedure condenses the acoustic detection database by orders of magnitude, greatly enhancing the synthesis of acoustic detection data. The development version of VTrack now houses a suit of functions called the ‘Animal Tracking Toolbox’ that helps users to follow a workflow to efficiently analyse acoustic telemetry data. You can find more information about the new functions in this vignette. Learn more here.

Maintainer: Dr. Vinay Udyawer ( vinay.udyawer@gmail.com )

Key Points

Intro to R

Overview

Teaching: 30 min
Exercises: 20 min

Questions

• What are common operators in R?

• What are common data types in R?

• What are some base R functions?

• How do I deal with missing data?

Objectives

First, lets learn about RStudio.

RStudio is divided into 4 “Panes”: the Source for your scripts and documents (top-left, in the default layout); your Environment/History (top-right) which shows all the objects in your working space (Environment) and your command history (History); your Files/Plots/Packages/Help/Viewer (bottom-right); and the R Console (bottom-left). The placement of these panes and their content can be customized (see menu, Tools -> Global Options -> Pane Layout).

The R Script in the top pane can be saved and edited, while code typed directly into the Console below will disappear after closing the R session.

R can access files on and save outputs to any folder on your computer. R knows where to look for and save files based on the current working directory. This is one of the first things you should set up: a folder you’d like to contain all your data, scripts and outputs. The working directory path will be different for everyone. For the workshop, we’ve included the path one of our instructors uses, but you should use your computer’s file explorer to find the correct path for your data.

Setting up R

# Packages ####
# once you install packages to your computer, you can "check them out" of your packages library each time you need them
# make sure you check the "mask" messages that appear - sometimes packages have functions with the same names!

library(tidyverse)# really neat collection of packages! https://www.tidyverse.org/
library(lubridate)
library(viridis)
library(plotly)
library(ggmap)


# Working Directory ####


setwd('YOUR-CUSTOM-PATH-HERE/workshop/data') #set folder you're going to work in
getwd() #check working directory

#you can also change it in the RStudio interface by navigating in the file browser where your working directory should be
#(if you can't see the folder you want, choose the three horizonal dots on the right side of the Home bar),
#and clicking on the blue gear icon "More", and select "Set As Working Directory".

Intro to R

Like most programming langauges, we can do basic mathematical operations with R. These, along with variable assignment, form the basis of everything for which we will use R.

Operators

Operators in R include standard mathematical operators (+, -, *, /) as well as an assignment operator, <- (a less-than sign followed by a hyphen). The assignment operator is used to associate a value with a variable name (or, to ‘assign’ the value a name). This lets us refer to that value later, by the name we’ve given to it. This may look unfamiliar, but it fulfils the same function as the ‘=’ operator in most other languages.

3 + 5 #maths! including - , *, /

weight_kg <- 55 #assignment operator! for objects/variables. shortcut: alt + -
weight_kg

weight_lb <- 2.2 * weight_kg #can assign output to an object. can use objects to do calculations

Variables Challenge

If we change the value of weight_kg to be 100, does the value of weight_lb also change? Remember: You can check the contents of an object by typing out its name and running the line in RStudio.

Solution

No! You have to re-assign 2.2*weight_kg to the object weight_lb for it to update.

The order you run your operations is very important, if you change something you may need to re-run everything!

weight_kg <- 100

weight_lb #didnt change!

weight_lb <- 2.2 * weight_kg #now its updated

Functions

While we can write code as we have in the section above - line by line, executed one line at a time - it is often more efficient to run multiple lines of code at once. By using functions, we can even compress complex calculations into just one line!

Functions use a single name to refer to underlying blocks of code that execute a specific calculation. To run a function you need two things: the name of the function, which is usually indicative of the function’s purpose; and the function’s arguments- the variables or values on which the function should execute.

#functions take "arguments": you have to tell them what to run their script against

ten <- sqrt(weight_kg) #contain calculations wrapped into one command to type.
#Output of the function can be assigned directly to a variable...

round(3.14159) #... but doesn't have to be.

Since there are hundreds of functions and often their functionality can be nuanced, we have several ways to get more information on a given function. First, we can use ‘args()’, itself a function that takes the name of another function as an argument, which will tell us the required arguments of the function against which we run it.

Second, we can use the ‘?’ operator. Typing a question mark followed by the name of a function will open a Help window in RStudio’s bottom-right panel. This will contain the most complete documentation available for the function in question.

args(round) #the args() function will show you the required arguments of another function

?round #will show you the full help page for a function, so you can see what it does

Functions Challenge

Can you round the value 3.14159 to two decimal places?

Hint: Using args() on a function can give you a clue.

Solution

round(3.14159, 2) #the round function's second argument is the number of digits you want in the result
round(3.14159, digits = 2) #same as above
round(digits = 2, x = 3.14159) #when reordered you need to specify

Vectors and Data Types

While variables can hold a single value, sometimes we want to store multiple values in the same variable name. For this, we can use an R data structure called a ‘vector.’ Vectors contain one or more variables of the same data type, and can be assigned to a single variable name, as below.

weight_g <- c(21, 34, 39, 54, 55) #use the combine function to join values into a vector object

length(weight_g) #explore vector
class(weight_g) #a vector can only contain one data type
str(weight_g) #find the structure of your object.

Above, we mentioned ‘data type’. This refers to the kind of data represented by a value, or stored by the appropriate variable. Data types include character (words or letters), logical (boolean TRUE or FALSE values), or numeric data. Crucially, vectors can only contain one type of data, and will force all data in the vector to conform to that type (i.e, data in the vector will all be treated as the same data type, regardless of whether or not it was of that type when the vector was created.) We can always check the data type of a variable or vector by using the ‘class()’ function, which takes the variable name as an argument.

#our first vector is numeric.
#other options include: character (words), logical (TRUE or FALSE), integer etc.

animals <- c("mouse", "rat", "dog") #to create a character vector, use quotes

class(weight_g)
class(animals)

# Note:
#R will convert (force) all values in a vector to the same data type.
#for this reason: try to keep one data type in each vector
#a data table / data frame is just multiple vectors (columns)
#this is helpful to remember when setting up your field sheets!

Vectors Challenge

What data type will this vector become?

challenge3 <- c(1, 2, 3, "4")

Hint: You can check a vector’s type with the class() function.

Solution

R will force all of these to be characters, since the number 4 has quotes around it! Will always coerce data types following this structure: logical → numeric → character ← logical

class(challenge3)

Indexing and Subsetting

We can use subsetting to select only a portion of a vector. For this, we use square brackets after the name of a vector. If we supply a single numeric value, as below, we will retrieve only the value from that index of the vector. Note: vectors in R are indexed with 1 representing the first index- other languages use 0 for the start of their array, so if you are coming from a language like Python, this can be disorienting.

animals #calling your object will print it out
animals[2] #square brackets = indexing. selects the 2nd value in your vector

We can select a specific value, as above, but we can also select one or more entries based on conditions. By supplying one or more criteria to our indexing syntax, we can retrieve the elements of the array that match that criteria.

weight_g > 50 #conditional indexing: selects based on criteria
weight_g[weight_g <=30 | weight_g == 55] #many new operators here!  
                                         #<= less than or equal to; | "or"; == equal to. Also available are >=, greater than or equal to; < and > for less than or greater than (no equals); and & for "and". 
weight_g[weight_g >= 30 & weight_g == 21] #  >=  greater than or equal to, & "and"
                                          # this particular example give 0 results - why?

Missing Data

In practical data analysis, our data is often incomplete. It is therefore useful to cover some methods of dealing with NA values. NA is R’s shorthand for a null value; or a value where there is no data. Certain functions cannot process NA data, and therefore provide arguments that allow NA values to be removed before the function execution.

heights <- c(2, 4, 4, NA, 6)
mean(heights) #some functions cant handle NAs
mean(heights, na.rm = TRUE) #remove the NAs before calculating

This can be done within an individual function as above, but for our entire analysis we may want to produce a copy of our dataset without the NA values included. Below, we’ll explore a few ways to do that.

heights[!is.na(heights)] #select for values where its NOT NA
#[] square brackets are the base R way to select a subset of data --> called indexing
#! is an operator that reverses the function

na.omit(heights) #omit the NAs

heights[complete.cases(heights)] #select only complete cases

Missing Data Challenge

Question 1: Using the following vector of heighs in inches, create a new vector, called heights_no_na, with the NAs removed.

heights <- c(63, 69, 60, 65, NA, 68, 61, 70, 61, 59, 64, 69, 63, 63, NA, 72, 65, 64, 70, 63, 65)

Solution

heights_no_na <- heights[!is.na(heights)] 
# or
heights_no_na <- na.omit(heights)
# or
heights_no_na <- heights[complete.cases(heights)]

Question 2: Use the function median() to calculate the median of the heights vector.

Solution

median(heights, na.rm = TRUE)

Bonus question: Use R to figure out how many people in the set are taller than 67 inches.

Solution

heights_above_67 <- heights_no_na[heights_no_na > 67]
length(heights_above_67)

Key Points

Starting with Data Frames

Overview

Teaching: 25 min
Exercises: 10 min

Questions

• How do I import tabular data?

• How do I explore my data set?

• What are some basic data manipulation functions?

Objectives

Dataframes and dplyr

In this lesson, we’re going to introduce a package called dplyr. dplyr takes advantage of an operator called a pipe to create chains of data manipulation that produce powerful exploratory summaries. It also provides a suite of further functionality for manipulating dataframes: tabular sets of data that are common in data analysis. If you’ve imported the tidyverse library, as we did during setup and in the last episode, then congratulations: you already have dplyr (along with a host of other useful packages).

You may not be familiar with dataframes by name, but you may recognize the structure. Dataframes are arranged into rows and columns, not unlike tables in typical spreadsheet format (ex: Excel). In R, they are represented as vectors of vectors: that is, a vector wherein each column is itself a vector. If you are familiar with matrices, or two-dimensional arrays in other languages, the structure of a dataframe will be clear to you.

However, dataframes are not merely vectors- they are a specific type of object with their own functionality, which we will cover in this lesson.

We are going to use OTN-style detection extracts for this lesson. If you’re unfamiliar with detection extracts formats from OTN-style database nodes, see the documentation here

Importing from CSVs

Before we can start analyzing our data, we need to import it into R. Fortunately, we have a function for this. read_csv is a function from the readr package, also included with the tidyverse library. This function can read data from a .csv file into a dataframe. “.csv” is an extension that denotes a Comma-Separated Value file, or a file wherein data is arranged into rows, and entries within rows are delimited by commas. They’re common in data analysis.

For the purposes of this lesson, we will only cover read_csv; however, there is another function, read_excel, which you can use to import excel files. It’s from a different library (readxl) and is outside the scope of this lesson, but worth investigating if you need it.

To import your data from your CSV file, we just need to pass the file path to read_csv, and assign the output to a variable. Note that the file path you give to read_csv will be relative to the working directory you set in the last lesson, so keep that in mind.

#imports file into R. paste the filepath to the unzipped file here!
#read_csv can take both csv and zip files, as long as the zip file contains a csv.

tqcs_matched_2010 <- read_csv("tqcs_matched_detections_2010.zip", guess_max = 117172) #Import 2010 detections

You may have noticed that our call to read_csv has a second argument: guess_max. This is a useful argument when some of our columns begin with a lot of NULL values. When determining what data type to assign to a column, rather than checking every single entry, R will check the first few and make a guess based on that. If the first few values are null, R will get confused and throw an error when it actually finds data further down in the column. guess_max lets us tell R exactly how many columns to read before trying to make a guess. This way, we know it will read enough entries in each column to actually find data, which it will prioritize over the NULL values when assigning a type to the column. This parameter isn’t always necessary, but it can be vital depending on your dataset.

We can now refer to the variable tqcs_matched_2010 to access, manipulate, and view the data from our CSV. In the next sections, we will explore some of the basic operations you can perform on dataframes.

Exploring Detection Extracts

Let’s start with a practical example. What can we find out about these matched detections? We’ll begin by running the code below, and then give some insight into what each function does. Remember, if you’re ever confused about the purpose of a function, you can use ‘?’ followed by the function name (i.e, ?head, ?View) to get more information.

head(tqcs_matched_2010) #first 6 rows
View(tqcs_matched_2010) #can also click on object in Environment window
str(tqcs_matched_2010) #can see the type of each column (vector)
glimpse(tqcs_matched_2010) #similar to str()

#summary() is a base R function that will spit out some quick stats about a vector (column)
#the $ syntax is the way base R selects columns from a data frame

summary(tqcs_matched_2010$latitude)

You may now have an idea of what each of those functions does, but we will briefly explain each here.

head takes the dataframe as a parameter and returns the first 6 rows of the dataframe. This is useful if you want to quickly check that a dataframe imported, or that all the columns are there, or see other such at-a-glance information. Its primary utility is that it is much faster to load and review than the entire dataframe, which may be several tens of thousands of rows long. Note that the related function tail will return the last six elements.

If we do want to load the entire dataframe, though, we can use View, which will open the dataframe in its own panel, where we can scroll through it as though it were an Excel file. This is useful for seeking out specific information without having to consult the file itself. Note that for large dataframes, this function can take a long time to execute.

Next are the functions str and glimpse, which do similar but different things. str is short for ‘structure’ and will print out a lot of information about your dataframe, including the number of rows and columns (called ‘observations’ and ‘variables’), the column names, the first four entries of each column, and each column type as well. str can sometimes be a bit overwhelming, so if you want to see a more condensed output, glimpse can be useful. It prints less information, but is cleaner and more compact, which can be desirable.

Finally, we have the summary function, which takes a single column from a dataframe and produces a summary of its basic statistics. You may have noticed the ‘$’ in the summary call- this is how we index a specific column from a dataframe. In this case, we are referring to the latitude column of our dataframe.

Using what you now know about summary functions, try to answer the challenge below.

Detection Extracts Challenge

Question 1: What is the class of the station column in tqcs_matched_2010, and how many rows and columns are in the tqcs_matched_2010 dataset??

Solution

The column is a character, and there are 1,737,597 rows with 36 columns

str(tqcs_matched_2010)
# or
glimpse(tqcs_matched_2010)

Data Manipulation

Now that we’ve learned how to import and summarize our data, we can learn how to use dplyr to manipulate it. The name ‘dplyr’ may seem esoteric- the ‘d’ is short for ‘dataframe’, and ‘plyr’ is meant to evoke pliers, and thereby cutting, twisting, and shaping. This is an elegant summation of the dplyr library’s functionality.

We are going to introduce a new operator in this section, called the “dplyr pipe”. Not to be confused with |, which is also called a pipe in some other languages, the dplyr pipe is rendered as %>%. A pipe takes the output of the function or contents of the variable on the left and passes them to the function on the right. It is often read as “and then.” If you want to quickly add a pipe, the keybord shortcut CTRL + SHIFT + M will do so.

library(dplyr) #can use tidyverse package dplyr to do exploration on dataframes in a nicer way

# %>% is a "pipe" which allows you to join functions together in sequence.

tqcs_matched_2010 %>% dplyr::select(6) #selects column 6

# Using the above transliteration: "take tqcs_matched_2010 AND THEN select column number 6 from it using the select function in the dplyr library"

You may have noticed another unfamiliar operator above, the double colon (::). This is used to specify the package from which we want to pull a function. Until now, we haven’t needed this, but as your code grows and the number of libraries you’re using increases, it’s likely that multiple functions across several different packages will have the same name (a phenomenon called “overloading”). R has no automatic way of knowing which package contains the function you are referring to, so using double colons lets us specify it explicitly. It’s important to be able to do this, since different functions with the same name often do markedly different things.

Let’s explore a few other examples of how we can use dplyr and pipes to manipulate our dataframe.

tqcs_matched_2010 %>% slice(1:5) #selects rows 1 to 5 in the dplyr way
# Take tqcs_matched_2010 AND THEN slice rows 1 through 5.

#We can also use multiple pipes.
tqcs_matched_2010 %>%
  distinct(detectedby) %>% nrow #number of arrays that detected my fish in dplyr!
# Take tqcs_matched_2010 AND THEN select only the unique entries in the detectedby column AND THEN count them with nrow.

#We can do the same as above with other columns too.
tqcs_matched_2010 %>%
  distinct(catalognumber) %>%
  nrow #number of animals that were detected
# Take tqcs_matched_2010 AND THEN select only the unique entries in the catalognumber column AND THEN count them with nrow.

#We can use filtering to conditionally select rows as well.
tqcs_matched_2010 %>% filter(catalognumber=="TQCS-1049258-2008-02-14")
# Take tqcs_matched_2010 AND THEN select only those rows where catalognumber is equal to the above value.

tqcs_matched_2010 %>% filter(monthcollected >= 10) #all dets in/after October of 2016
# Take tqcs_matched_2010 AND THEN select only those rows where monthcollected is greater than or equal to 10.

These are all ways to extract a specific subset of our data, but dplyr can also be used to manipulate dataframes to give you even greater insights. We’re now going to use two new functions: group_by, which allows us to group our data by the values of a single column, and summarise (not to be confused with summary above!), which can be used to calculate summary statistics across your grouped variables, and produces a new dataframe containing these values as the output. These functions can be difficult to grasp, so don’t forget to use ?group_by and ?summarise if you get lost.

#get the mean value across a column using GroupBy and Summarize

tqcs_matched_2010 %>% #Take tqcs_matched_2010, AND THEN...
  group_by(catalognumber) %>%  #Group the data by catalognumber- that is, create a group within the dataframe where each group contains all the rows related to a specific catalognumber. AND THEN...
  summarise(MeanLat=mean(latitude)) #use summarise to add a new column containing the mean latitude of each group. We named this new column "MeanLat" but you could name it anything

With just a few lines of code, we’ve created a dataframe that contains each of our catalog numbers and the mean latitude at which those fish were detected. dplyr, its wide array of functions, and the powerful pipe operator can let us build out detailed summaries like this one without writing too much code.

Data Manipulation Challenge

Question 1: Find the max lat and max longitude for animal “TQCS-1049258-2008-02-14”.

Solution

tqcs_matched_2010 %>%
 filter(catalognumber=="TQCS-1049258-2008-02-14") %>%
 summarise(MaxLat=max(latitude), MaxLong=max(longitude))

Question 2: Find the min lat/long of each animal for detections occurring in July.

Solution

tqcs_matched_2010 %>%
  filter(monthcollected == 7) %>%
  group_by(catalognumber) %>%
  summarise(MinLat=min(latitude), MinLong=min(longitude))

Joining Detection Extracts

We’re now going to briefly touch on a few useful dataframe use-cases that aren’t directly related to dplyr, but with which dplyr can help us.

One function that we’ll need to know is rbind, a base R function which lets us combine two R objects together. Since detections for animals tagged during a study often appear in multiple years, this functionality will let us merge the dataframes together. We’ll also use distinct, a dplyr function that lets us trim out duplicate release records for each animal, since these are listed in each detection extract.

tqcs_matched_2011 <- read_csv("tqcs_matched_detections_2011.zip", guess_max = 87131) #Import 2011 detections

tqcs_matched_10_11_full <- rbind(tqcs_matched_2010, tqcs_matched_2011) #Now join the two dataframes

#release records for animals often appear in >1 year, this will remove the duplicates
tqcs_matched_10_11_full <- tqcs_matched_10_11_full %>% distinct() # Use distinct to remove duplicates.

tqcs_matched_10_11 <- tqcs_matched_10_11_full %>% slice(1:100000) # subset our example data to help this workshop run smoother!

Dealing with Datetimes

Datetime data is in a special format which is neither numeric nor character. It can be tricky to deal with, too, since Excel frequently reformats dates in any file it opens. We also have to concern ourselves with practical matters of time, like time zone and date formatting. Fortunately, the lubridate library gives us a whole host of functionality to manage datetime data.

We’ll also use a dplyr function called mutate, which lets us add new columns or change existing ones, while preserving the existing data in the table. Be careful not to confuse this with its sister function transmute, which adds or manipulates columns while dropping existing data. If you’re ever in doubt as to which is which, remember: ?mutate and ?transmute will bring up the help files.

library(lubridate) #Import our Lubridate library.

tqcs_matched_10_11 %>% mutate(datecollected=ymd_hms(datecollected)) #Use the lubridate function ymd_hms to change the format of the date.

#as.POSIXct(tqcs_matched_10_11$datecollected) #this is the base R way - if you ever see this function

We’ve just used a single function, ymd_hms, but with it we’ve been able to completely reformat the entire datecollected column. ymd_hms is short for Year, Month, Day, Hours, Minutes, and Seconds. For example, at time of writing, it’s 2021-05-14 14:21:40. Other format functions exist too, like dmy_hms, which specifies the day first and year third (i.e, 14-05-2021 14:21:40). Investigate the documentation to find which is right for you.

There are too many useful lubridate functions to cover in the scope of this lesson. These include parse_date_time, which can be used to read in date data in multiple formats, which is useful if you have a column contianing heterogenous date data; as well as with_tz, which lets you make your data sensitive to timezones (including automatic daylight savings time awareness). Dates are a tricky subject, so be sure to investigate lubridate to make sure you find the functions you need.

Key Points

Intro to Plotting

Overview

Teaching: 15 min
Exercises: 10 min

Questions

• How do I plot my data?

• How can I plot summaries of my data?

Objectives

• Learn how to make basic plots with ggplot2

• Learn how to combine dplyr summaries with ggplot2 plots

Background

Now that we have learned how to import, inspect, and manipulate our data, we are next going to learn how to visualize it. R provides a robust plotting suite in the library ggplot2. ggplot2 takes advantage of tidyverse pipes and chains of data manipulation to build plotting code. Additionally, it separates the aesthetics of the plot (what are we plotting) from the styling of the plot (what the plot looks like). What this means is that data aesthetics and styles can be built separately and then combined and recombined to produce modular, reusable plotting code.

While ggplot2 function calls can look daunting at first, they follow a single formula, detailed below.

#Anything within <> braces will be replaced in an actual function call.
ggplot(data = <DATA>, mapping = aes(<MAPPINGS>)) + <GEOM_FUNCTION>

In the above example, there are three important parts: <DATA>, <MAPPINGS>, and <GEOM_FUNCTION>.

<DATA> refers to the data that we’ll be plotting. In general, this will be held in a dataframe like the one we prepared in the previous lessons.

<MAPPINGS> refers to the aesthetic mappings for the data- that is, which columns in the data will be used to determine which attributes of the graph. For example, if you have columns for latitude and longitude, you may want to map these onto the X and Y axes of the graph. We’ll cover how to do exactly that in a moment.

Finally, <GEOM_FUNCTION> refers to the style of the plot: what type of plot are we going to make. GEOM is short for “geometry” and ggplot2 contains many different ‘geom’ functions that you can use. For this lesson, we’ll be using geom_point(), which produces a scatterplot, but in the future you may want to use geom_path(), geom_bar(), geom_boxplot() or any of ggplots other geom functions. Remember, since these are functions, you can use the help syntax (i.e ?geom_point) in the R console to find out more about them and what you need to pass to them.

Now that we’ve introduced ggplot2, let’s build a functional example with our data.

# Begin by importing the ggplot2 library, which you should have installed as part of setup.
library(ggplot2)

# Build the plot and assign it to a variable.
tqcs_10_11_plot <- ggplot(data = tqcs_matched_10_11,
                          mapping = aes(x = longitude, y = latitude)) #can assign a base

With a couple of lines of code, we’ve already mostly completed a simple scatter plot of our data. The ‘data’ parameter takes our dataframe, and the mapping parameter takes the output of the aes() function, which itself takes a mapping of our data onto the axes of the graph. That can be a bit confusing, so let’s briefly break this down. aes() is short for ‘aesthetics’- the function constructs the aesthetic mappings of our data, which describe how variables in the data are mapped to visual properties of the plot. For example, above, we are setting the ‘x’ attribute to ‘longitude’, and the ‘y’ attribute to latitude. This means that the X axis of our plot will represent longitude, and the Y axis will represent latitude. Depending on the type of plot you’re making, you may want different values there, and different types of geom functions can require different aesthetic mappings (colour, for example, is another common one). You can always type ?aes() at the console if you want more information.

We still have one step to add to our plotting code: the geom function. We’ll be making a scatterplot, so we want to use geom_point().

tqcs_10_11_plot +
  geom_point(alpha=0.1,
             colour = "blue")  
#This will layer our chosen geom onto our plot template.
#alpha is a transparency argument in case points overlap. Try alpha = 0.02 to see how it works!

With just the above code, we’ve added our geom to our aesthetic and made our plot ready for display. We’ve built only a very simple plot here, but ggplot2 provides many, many options for building more complex, illustrative plots.

Basic plots

As a minor syntactic note, you can build your plots iteratively, without assigning them to a variable in-between. For this, we make use of tidyverse pipes.

tqcs_matched_10_11 %>%  
  ggplot(aes(longitude, latitude)) +
  geom_point() #geom = the type of plot

tqcs_matched_10_11 %>%  
  ggplot(aes(longitude, latitude, colour = commonname)) +
  geom_point()


#anything you specify in the aes() is applied to the actual data points/whole plot,
#anything specified in geom() is applied to that layer only (colour, size...). sometimes you have >1 geom layer so this makes more sense!

You can see that all we need to do to make this work is omit the ‘data’ parameter, since that’s being passed in by the pipe. Note also that we’ve added colour = commonname to the second plot’s aesthetic, meaning that the output will be coloured based on the species of the animal (if there is more than one included).

Remembering which of the aes or the geom controls which variable can be difficult, but here’s a handy rule of thumb: anything specified in aes() will apply to the data points themselves, or the whole plot. They are broad statements about how the plot is to be displayed. Anything in the geom_ function will apply only to that geom_ layer. Keep this in mind, since it’s possible for your plot to have more than one geom_!

Plotting and dplyr Challenge

Try combining with dplyr functions in this challenge! Try making a scatterplot showing the lat/long for animal “TQCS-1049258-2008-02-14”, coloured by detection array

Solution

tqcs_matched_10_11  %>%  
 filter(catalognumber=="TQCS-1049258-2008-02-14") %>%
 ggplot(aes(longitude, latitude, colour = detectedby)) +
 geom_point()

What other geoms are there? Try typing geom_ into R to see what it suggests!

Key Points

• You can feed output from dplyr’s data manipulation functions into ggplot using pipes.

• Plotting various summaries and groupings of your data is good practice at the exploratory phase, and dplyr and ggplot make iterating different ideas straightforward.

Telemetry Reports - Imports

Overview

Teaching: 10 min
Exercises: 0 min

Questions

• What datasets do I need from the Network?

• How do I import all the datasets?

Objectives

Importing all the datasets

Now that we have an idea of what an exploratory workflow might look like with Tidyverse libraries like dplyr and ggplot2, let’s look at how we might implement a common telemetry workflow using these tools.

We are going to use OTN-style detection extracts for this lesson. If you’re unfamiliar with detection extracts formats from OTN-style database nodes, see the documentation here.

For the FACT Network you will receive Detection Extracts which include (1) Matched to Animals YYYY, (2) Detections Mapped to Other Trackers - Extended YYYY (also called Qualified Extended) and (3) Unqualified Detections YYYY. In each case, the YYYY in the filename indicates the single year of data contained in the file and “extended” refers to the extra column provided to FACT Network members: “species detected”. The types of detection extracts you receive will differ depending on the type of project you have regitered with the Network. If you have both an Array project and a Tag project you will likely need both sets of Detection Extracts.

To illustrate the many meaningful summary reports which can be created use detection extracts, we will import an example of Matched and Qualified extracts.

First, we will comfirm we have our Tag Matches stored in a dataframe.

View(tqcs_matched_10_11) #already have our Tag matches, from a previous lesson.

# if you do not have the variable created from a previous lesson, you can use the following code to re-create it:

tqcs_matched_2010 <- read_csv("tqcs_matched_detections_2010.zip", guess_max = 117172) #Import 2010 detections
tqcs_matched_2011 <- read_csv("tqcs_matched_detections_2011.zip", guess_max = 87131) #Import 2011 detections
tqcs_matched_10_11_full <- rbind(tqcs_matched_2010, tqcs_matched_2011) #Now join the two dataframes
# release records for animals often appear in >1 year, this will remove the duplicates
tqcs_matched_10_11_full <- tqcs_matched_10_11_full %>% distinct() # Use distinct to remove duplicates.
tqcs_matched_10_11 <- tqcs_matched_10_11_full %>% slice(1:100000) # subset our example data to help this workshop run smoother!

Next, we will load in and join our Array matches.

teq_qual_2010 <- read_csv("teq_qualified_detections_2010.zip")
teq_qual_2011 <- read_csv("teq_qualified_detections_2011.zip")
teq_qual_10_11_full <- rbind(teq_qual_2010, teq_qual_2011)

teq_qual_10_11 <- teq_qual_10_11_full %>% slice(1:100000) #subset our example data for ease of analysis!

To give meaning to these detections we should import our Instrument Deployment Metadata and Tagging Metadata as well. These are in the standard VEMBU/FACT-style templates which can be found here.

# Array metadata

teq_deploy <- read.csv("TEQ_Deployments_201001_201201.csv")
View(teq_deploy)

# Tag metadata

tqcs_tag <- read.csv("TQCS_metadata_tagging.csv")
View(tqcs_tag)

#remember: we learned how to switch timezone of datetime columns above, if that is something you need to do with your dataset!!

Key Points

Telemetry Reports for Array Operators

Overview

Teaching: 30 min
Exercises: 0 min

Questions

• How do I summarize and plot my deployments?

• How do I summarize and plot my detections?

Objectives

Mapping my stations - Static map

Since we have already imported and joined our datasets, we can jump in. This section will use the Deployment metadata for your array. We will make a static map of all the receiver stations in three steps, using the package ggmap.

First, we set a basemap using the aesthetics and bounding box we desire. Then, we will filter our stations dataset for those which we would like to plot on the map. Next, we add the stations onto the basemap and look at our creation! If we are happy with the product, we can export the map as a .tiff file using the ggsave function, to use outside of R. Other possible export formats include: .png, .jpeg, .pdf and more.

library(ggmap)


#first, what are our columns called?
names(teq_deploy)


#make a basemap for your stations, using the min/max deploy lat and longs as bounding box

base <- get_stamenmap(
  bbox = c(left = min(teq_deploy$DEPLOY_LONG),
           bottom = min(teq_deploy$DEPLOY_LAT),
           right = max(teq_deploy$DEPLOY_LONG),
           top = max(teq_deploy$DEPLOY_LAT)),
  maptype = "terrain-background",
  crop = FALSE,
  zoom = 8)

#filter for stations you want to plot

teq_deploy_plot <- teq_deploy %>%
  mutate(deploy_date=ymd_hms(DEPLOY_DATE_TIME....yyyy.mm.ddThh.mm.ss.)) %>% #make a datetime
  mutate(recover_date=ymd_hms(RECOVER_DATE_TIME..yyyy.mm.ddThh.mm.ss.)) %>% #make a datetime
  filter(!is.na(deploy_date)) %>% #no null deploys
  filter(deploy_date > 2010-07-03) %>% #only looking at certain deployments!
  group_by(STATION_NO) %>%
  summarise(MeanLat=mean(DEPLOY_LAT), MeanLong=mean(DEPLOY_LONG)) #get the mean location per station

# you could choose to plot stations which are within a certain bounding box!
# to do this you would add another filter to the above data, before passing to the map
# ex: add this line after the mutate() clauses:
	# filter(latitude >= 0.5 & latitude <= 24.5 & longitude >= 0.6 & longitude <= 34.9)


#add your stations onto your basemap

teq_map <-
  ggmap(base, extent='panel') +
  ylab("Latitude") +
  xlab("Longitude") +
  geom_point(data = teq_deploy_plot, #filtering for recent deployments
             aes(x = MeanLong,y = MeanLat), #specify the data
             colour = 'blue', shape = 19, size = 2) #lots of aesthetic options here!

#view your receiver map!

teq_map

#save your receiver map into your working directory

ggsave(plot = teq_map, file = "code/day1/teq_map.tiff", units="in", width=15, height=8)

Mapping my stations - Interactive map

An interactive map can contain more information than a static map. Here we will explore the package plotly to create interactive “slippy” maps. These allow you to explore your map in different ways by clicking and scrolling through the output.

First, we will set our basemap’s aesthetics and bounding box and assign this information (as a list) to a geo_styling variable.

library(plotly)

#set your basemap

geo_styling <- list(
  scope = 'usa',
  fitbounds = "locations", visible = TRUE, #fits the bounds to your data!
  showland = TRUE,
  showlakes = TRUE,
  lakecolor = toRGB("blue", alpha = 0.2), #make it transparent
  showcountries = TRUE,
  landcolor = toRGB("gray95"),
  countrycolor = toRGB("gray85")
)

Then, we choose which Deployment Metadata dataset we wish to use and identify the columns containing Latitude and Longitude, using the plot_geo function.

#decide what data you're going to use. Let's use teq_deploy_plot, which we created above for our static map.

teq_map_plotly <- plot_geo(teq_deploy_plot, lat = ~MeanLat, lon = ~MeanLong)  

Next, we use the add_markers function to write out what information we would like to have displayed when we hover our mouse over a station in our interactive map. In this case, we chose to use paste to join together the Station Name and its lat/long.

#add your markers for the interactive map

teq_map_plotly <- teq_map_plotly %>% add_markers(
  text = ~paste(STATION_NO, MeanLat, MeanLong, sep = "<br />"),
  symbol = I("square"), size = I(8), hoverinfo = "text"
)

Finally, we add all this information together, along with a title, using the layout function, and now we can explore our interactive map!

#Add layout (title + geo stying)

teq_map_plotly <- teq_map_plotly %>% layout(
  title = 'TEQ Deployments<br />(> 2010-07-03)', geo = geo_styling
)

#View map

teq_map_plotly

To save this interactive map as an .html file, you can explore the function htmlwidgets::saveWidget(), which is beyond the scope of this lesson.

Summary of Animals Detected

Let’s find out more about the animals detected by our array! These summary statistics, created using dplyr functions, could be used to help determine the how successful each of your stations has been at detecting tagged animals. We will also learn how to export our results using write_csv.

# How many of each animal did we detect from each collaborator, by species

teq_qual_summary <- teq_qual_10_11 %>%
  filter(datecollected > '2010-06-01') %>% #select timeframe, stations etc.
  group_by(trackercode, scientificname, tag_contact_pi, tag_contact_poc) %>%
  summarize(count = n()) %>%
  select(trackercode, tag_contact_pi, tag_contact_poc, scientificname, count)

#view our summary table

teq_qual_summary #remember, this is just the first 10,000 rows! We subsetted the dataset upon import!

#export our summary table

write_csv(teq_qual_summary, "code/day1/teq_detection_summary_June2010_to_Dec2011.csv", col_names = TRUE)

You may notice in your summary table above that some rows have a value of NA for ‘scientificname’. This is because this example dataset has detections of animals tagged by researchers who are not a part of the FACT Network, and therefore have not agreed to share their species information with array-operators automatically. To obtain this information you would have to reach out to the researcher directly. For more information on the FACT Data Policy and how it differs from other collaborating OTN Networks, please reach out to Data@theFACTnetwork.org.

Summary of Detections

These dplyr summaries can suggest array performance, hotspot stations, and be used as a metric for funders.

# number of detections per month/year per station

teq_det_summary  <- teq_qual_10_11  %>%
  mutate(datecollected=ymd_hms(datecollected))  %>%
  group_by(station, year = year(datecollected), month = month(datecollected)) %>%
  summarize(count =n())

teq_det_summary #remember: this is a subset!

# number of detections per month/year per station & species

teq_anim_summary  <- teq_qual_10_11  %>%
  mutate(datecollected=ymd_hms(datecollected))  %>%
  group_by(station, year = year(datecollected), month = month(datecollected), scientificname) %>%
  summarize(count =n())

teq_anim_summary # remember: this is a subset!

# Create a new data product, det_days, that give you the unique dates that an animal was seen by a station
stationsum <- teq_qual_10_11 %>%
  group_by(station) %>%
  summarise(num_detections = length(datecollected),
            start = min(datecollected),
            end = max(datecollected),
            species = length(unique(scientificname)),
            uniqueIDs = length(unique(fieldnumber)),
            det_days=length(unique(as.Date(datecollected))))
View(stationsum)

Plot of Detections

Lets make an informative plot using ggplot showing the number of matched detections, per year and month. Remember: we can combine dplyr data manipulation and plotting into one step, using pipes!

#try with teq_qual_10_11_full if you're feeling bold! takes about 1 min to run on a fast machine

teq_qual_10_11 %>%
  mutate(datecollected=ymd_hms(datecollected)) %>% #make datetime
  mutate(year_month = floor_date(datecollected, "months")) %>% #round to month
  group_by(year_month) %>% #can group by station, species etc.
  summarize(count =n()) %>% #how many dets per year_month
  ggplot(aes(x = (month(year_month) %>% as.factor()),
             y = count,
             fill = (year(year_month) %>% as.factor())
             )
         )+
  geom_bar(stat = "identity", position = "dodge2")+
  xlab("Month")+
  ylab("Total Detection Count")+
  ggtitle('TEQ Animal Detections by Month')+ #title
  labs(fill = "Year") #legend title

Key Points

Telemetry Reports for Tag Owners

Overview

Teaching: 30 min
Exercises: 0 min

Questions

• How do I summarize and plot my detections?

• How do I summarize and plot my tag metadata?

Objectives

New data frames

To aid in the creating of useful Matched Detection summaries, we should create a new dataframe where we filter out release records from the detection extracts. This will leave only “true” detections.

#optional subsetted dataset to use: detections with releases filtered out!

tqcs_matched_10_11_no_release <- tqcs_matched_10_11 %>%
  filter(receiver != "release")

#optional full dataset to use: detections with releases filtered out!

tqcs_matched_10_11_full_no_release <- tqcs_matched_10_11_full %>%
  filter(receiver != "release")

Mapping my Detections and Releases - static map

Where were my fish observed? We will make a static map of all the receiver stations where my fish was detected in two steps, using the package ggmap.

First, we set a basemap using the aesthetics and bounding box we desire. Next, we add the detection locations onto the basemap and look at our creation!

base <- get_stamenmap(
  bbox = c(left = min(tqcs_matched_10_11$longitude),
           bottom = min(tqcs_matched_10_11$latitude),
           right = max(tqcs_matched_10_11$longitude),
           top = max(tqcs_matched_10_11$latitude)),
  maptype = "terrain-background",
  crop = FALSE,
  zoom = 8)


#add your releases and detections onto your basemap

tqcs_map <-
  ggmap(base, extent='panel') +
  ylab("Latitude") +
  xlab("Longitude") +
  geom_point(data = tqcs_matched_10_11,
             aes(x = longitude,y = latitude), #specify the data
             colour = 'blue', shape = 19, size = 2) #lots of aesthetic options here!

#view your tagging map!

tqcs_map

Mapping my Detections and Releases - interactive map

An interactive map can contain more information than a static map. Here we will explore the package plotly to create interactive “slippy” maps. These allow you to explore your map in different ways by clicking and scrolling through the output.

First, we will set our basemap’s aesthetics and bounding box and assign this information (as a list) to a geo_styling variable. Then, we choose which detections we wish to use and identify the columns containing Latitude and Longitude, using the plot_geo function. Next, we use the add_markers function to write out what information we would like to have displayed when we hover our mouse over a station in our interactive map. In this case, we chose to use paste to join together the Station Name and its lat/long. Finally, we add all this information together, along with a title, using the layout function, and now we can explore our interactive map!

#set your basemap

geo_styling <- list(
  fitbounds = "locations", visible = TRUE, #fits the bounds to your data!
  showland = TRUE,
  landcolor = toRGB("gray95"),
  subunitcolor = toRGB("gray85"),
  countrycolor = toRGB("gray85")
)

#decide what data you're going to use

tqcs_map_plotly <- plot_geo(tqcs_matched_10_11, lat = ~latitude, lon = ~longitude)

#add your markers for the interactive map

tqcs_map_plotly <- tqcs_map_plotly %>% add_markers(
  text = ~paste(catalognumber, scientificname, paste("Date detected:", datecollected),
                paste("Latitude:", latitude), paste("Longitude",longitude),
                paste("Detected by:", detectedby), paste("Station:", station),
                paste("Contact:", contact_poc, contact_pi), sep = "<br />"),
  symbol = I("square"), size = I(8), hoverinfo = "text"
)

#Add layout (title + geo stying)

tqcs_map_plotly <- tqcs_map_plotly %>% layout(
  title = 'TQCS Detections<br />(2010-2011)', geo = geo_styling
)

#View map

tqcs_map_plotly

Summary of tagged animals

This section will use your Tagging Metadata to create dplyr summaries of your tagged animals.

# summary of animals you've tagged

tqcs_tag_summary <- tqcs_tag %>%
  mutate(UTC_RELEASE_DATE_TIME = ymd_hms(UTC_RELEASE_DATE_TIME)) %>%
  #filter(UTC_RELEASE_DATE_TIME > '2019-06-01') %>% #select timeframe, specific animals etc.
  group_by(year = year(UTC_RELEASE_DATE_TIME), COMMON_NAME_E) %>%
 summarize(count = n(),
            Meanlength = mean(LENGTH..m., na.rm=TRUE),
            minlength= min(LENGTH..m., na.rm=TRUE),
            maxlength = max(LENGTH..m., na.rm=TRUE),
            MeanWeight = mean(WEIGHT..kg., na.rm = TRUE))

#view our summary table

tqcs_tag_summary

Detection Attributes

Lets add some biological context to our summaries! To do this we can join our Tag Metadata with our Matched Detections. To learn more about the different types of dataframe joins and how they function, see here.

#Average location of each animal, without release records

tqcs_matched_10_11_no_release %>%
  group_by(catalognumber) %>%
  summarize(NumberOfStations = n_distinct(station),
            AvgLat = mean(latitude),
            AvgLong =mean(longitude))

Now lets try to join our metadata and detection extracts.

#First we need to make a tagname column in the tag metadata (to match the Detection Extract), and figure out the enddate of the tag battery.

tqcs_tag <- tqcs_tag %>%
  mutate(enddatetime = (ymd_hms(UTC_RELEASE_DATE_TIME) + days(EST_TAG_LIFE))) %>% #adding enddate
  mutate(tagname = paste(TAG_CODE_SPACE,TAG_ID_CODE, sep = '-')) #adding tagname column

#Now we join by tagname, to the detections without the release information

tag_joined_dets <-  left_join(x = tqcs_matched_10_11_no_release, y = tqcs_tag, by = "tagname")

#make sure the redeployed tags have matched within their deployment period only

tag_joined_dets <- tag_joined_dets %>%
  filter(datecollected >= UTC_RELEASE_DATE_TIME & datecollected <= enddatetime)

View(tag_joined_dets)

Lets use this new joined dataframe to make summaries!

# Avg length per location

tqcs_tag_det_summary <- tag_joined_dets %>%
  group_by(detectedby, station, latitude, longitude)  %>%  
  summarise(AvgSize = mean(LENGTH..m., na.rm=TRUE))

tqcs_tag_det_summary

# count detections per transmitter, per array

tqcs_matched_10_11_no_release %>%
  group_by(catalognumber, detectedby, commonname) %>%
  summarize(count = n()) %>%
  select(catalognumber, commonname, detectedby, count)

# list all receivers each fish was seen on, and a number_of_receivers column too

receivers <- tqcs_matched_10_11_no_release %>%
  group_by(catalognumber) %>%
  mutate(stations = (list(unique(station)))) %>% #create a column with a list of the stations
  dplyr::select(catalognumber, stations)  %>% #remove excess columns
  distinct_all() %>% #keep only one record of each
  mutate(number_of_stations = sapply(stations, length)) %>% #sapply: applies a function across a List - in this case we are applying length()
  as.data.frame()

View(receivers)

animal_id_summary <- tqcs_matched_10_11_no_release %>%
  group_by(catalognumber) %>%
  summarise(dets = length(catalognumber),
            stations = length(unique(station)),
            min = min(datecollected),
            max = max(datecollected),
            tracklength = max(datecollected)-min(datecollected))

View(animal_id_summary)

Summary of Detection Counts

Lets make an informative plot showing number of matched detections, per year and month.

#try with tqcs_matched_10_11_full_no_release if you're feeling bold! takes ~30 secs

tqcs_matched_10_11_no_release  %>%
  mutate(datecollected=ymd_hms(datecollected)) %>% #make datetime
  mutate(year_month = floor_date(datecollected, "months")) %>% #round to month
  group_by(year_month) %>% #can group by station, species etc.
  summarize(count =n()) %>% #how many dets per year_month
  ggplot(aes(x = (month(year_month) %>% as.factor()),
             y = count,
             fill = (year(year_month) %>% as.factor())
  )
  )+
  geom_bar(stat = "identity", position = "dodge2")+
  xlab("Month")+
  ylab("Total Detection Count")+
  ggtitle('TQCS Detections by Month (2010-2011)')+ #title
  labs(fill = "Year") #legend title

Other Example Plots

Some examples of complex plotting options. The most useful of these may include abacus plotting (an example with ‘animal’ and ‘station’ on the y-axis) as well as an example using ggmap and geom_path to create an example map showing animal movement.

#Use the color scales in this package to make plots that are pretty,
#better represent your data, easier to read by those with colorblindness, and print well in grey scale.
library(viridis)

# monthly latitudinal distribution of your animals (works best w >1 species)

tqcs_matched_10_11 %>%
  group_by(m=month(datecollected), catalognumber, scientificname) %>% #make our groups
  summarise(mean=mean(latitude)) %>% #mean lat
  ggplot(aes(m %>% factor, mean, colour=scientificname, fill=scientificname))+ #the data is supplied, but no info on how to show it!
  geom_point(size=3, position="jitter")+   # draw data as points, and use jitter to help see all points instead of superimposition
  #coord_flip()+   #flip x y, not needed here
  scale_colour_manual(values = "blue")+ #change the colour to represent the species better!
  scale_fill_manual(values = "grey")+
  geom_boxplot()+ #another layer
  geom_violin(colour="black") #and one more layer


#There are other ways to present a summary of data like this that we might have chosen.
#geom_density2d() will give us a KDE for our data points and give us some contours across our chosen plot axes.

tqcs_matched_10_11 %>% #doesnt work on the subsetted data, back to original dataset for this one
  group_by(month=month(datecollected), catalognumber, scientificname) %>%
  summarise(meanlat=mean(latitude)) %>%
  ggplot(aes(month, meanlat, colour=scientificname, fill=scientificname))+
  geom_point(size=3, position="jitter")+
  scale_colour_manual(values = "blue")+
  scale_fill_manual(values = "grey")+
  geom_density2d(size=7, lty=1) #this is the only difference from the plot above

#anything you specify in the aes() is applied to the actual data points/whole plot,
#anything specified in geom() is applied to that layer only (colour, size...)

# per-individual density contours - lots of plots: called facets!
tqcs_matched_10_11 %>%
  ggplot(aes(longitude, latitude))+
  facet_wrap(~catalognumber)+ #make one plot per individual
  geom_violin()

# an easy abacus plot!

abacus_animals <-
  ggplot(data = tqcs_matched_10_11_no_release, aes(x = datecollected, y = catalognumber, col = detectedby)) +
  geom_point() +
  ggtitle("Detections by animal") +
  theme(plot.title = element_text(face = "bold", hjust = 0.5)) +
  scale_color_viridis(discrete = TRUE)

abacus_animals

abacus_stations <-
  ggplot(data = tqcs_matched_10_11_no_release,  aes(x = datecollected, y = station, col = catalognumber)) +
  geom_point() +
  ggtitle("Detections by station") +
  theme(plot.title = element_text(face = "bold", hjust = 0.5)) +
  scale_color_viridis(discrete = TRUE)

abacus_stations

# track movement using geom_path!!

tqcs_subset <- tqcs_matched_10_11_no_release %>%
  dplyr::filter(catalognumber %in%
                  c('TQCS-1049282-2008-02-28', 'TQCS-1049281-2008-02-28'))

View(tqcs_subset)

movMap <-
  ggmap(base, extent = 'panel') + #use the BASE we set up before
  ylab("Latitude") +
  xlab("Longitude") +
  geom_path(data = tqcs_subset, aes(x = longitude, y = latitude, col = commonname)) + #connect the dots with lines
  geom_point(data = tqcs_subset, aes(x = longitude, y = latitude, col = commonname)) + #layer the stations back on
  scale_colour_manual(values = c("red", "blue"), name = "Species")+ #
  facet_wrap(~catalognumber)+
  ggtitle("Inferred Animal Paths")

#to size the dots by number of detections you could do something like: size = (log(length(animal)id))?

movMap

Key Points

Introduction to glatos Data Processing Package

Overview

Teaching: 30 min
Exercises: 0 min

Questions

• How do I load my data into glatos?

• How do I filter out false detections?

• How can I consolidate my detections into detection events?

• How do I summarize my data?

Objectives

The glatos package is a powerful toolkit that provides a wide range of functionality for loading, processing, and visualizing your data. With it, you can gain valuable insights with quick and easy commands that condense high volumes of base R into straightforward functions, with enough versatility to meet a variety of needs.

This package was originally created to meet the needs of the Great Lakes Acoustic Telemetry Observation System (GLATOS) and use their specific data formats. However, over time, the functionality has been expanded to allow operations on OTN-formatted data as well, broadening the range of possible applications for the software. As a point of clarification, “GLATOS” (all caps acronym) refers to the organization, while glatos refers to the package.

Our first step is setting our working directory and importing the relevant libraries.

## Set your working directory ####

setwd("./data/fact")
library(glatos)
library(tidyverse)
library(VTrack)
library(utils)
library(lubridate)

If you are following along with the workshop in the workshop repository, there should be a folder in ‘data/’ containing data corresponding to your node (at time of writing, FACT, ACT, or GLATOS). glatos can function with both GLATOS and OTN Node-formatted data, but the functions are different for each. Both, however, provide a marked performance boost over base R, and both ensure that the resulting data set will be compatible with the rest of the glatos framework.

We’ll start by combining our several data files into one master detection file, which glatos will be able to read.

format <- cols( # Heres a col spec to use when reading in the files
  .default = col_character(),
  datelastmodified = col_date(format = ""),
  bottom_depth = col_double(),
  receiver_depth = col_double(),
  sensorname = col_character(),
  sensorraw = col_character(),
  sensorvalue = col_character(),
  sensorunit = col_character(),
  datecollected = col_datetime(format = ""),
  longitude = col_double(),
  latitude = col_double(),
  yearcollected = col_double(),
  monthcollected = col_double(),
  daycollected = col_double(),
  julianday = col_double(),
  timeofday = col_double(),
  datereleasedtagger = col_logical(),
  datereleasedpublic = col_logical()
)

detections <- tibble()
for (detfile in list.files('.', full.names = TRUE, pattern = "tqcs.*\\.zip")) {
  print(detfile)
  tmp_dets <- read_csv(detfile, col_types = format)
  detections <- bind_rows(detections, tmp_dets)
}

write_csv(detections, 'all_dets.csv', append = FALSE)

With our new file in hand, we’ll want to use the read_otn_detections function to load our data into a dataframe. In this case, our data is formatted in the FACT (OTN) style- if it were GLATOS formatted, we would want to use read_glatos_detections() instead.

Remember: you can always check a function’s documentation by typing a question mark, followed by the name of the function.

## glatos help files are helpful!! ####
?read_otn_detections

# Save our detections file data into a dataframe called detections
detections <- read_otn_detections(det_file='all_dets.csv')

detections <- detections %>% slice(1:100000) # subset our example data to help this workshop run

Making a 100,000 row subset of our data is not a necessary step, but it will make our code run more smoothly for this workshop, since later functions can struggle with large datasets.

Remember that we can use head() to inspect a few lines of our data to ensure it was loaded properly.

# View first 2 rows of output
head(detections, 2)

With our data loaded, we next want to apply a false filtering algorithm to reduce the number of false detections in our dataset. glatos uses the Pincock algorithm to filter probable false detections based on the time lag between detections- tightly clustered detections are weighted as more likely to be true, while detections spaced out temporally will be marked as false. We can also pass the time-lag threshold as a variable to the false_detections function. This lets us fine-tune our filtering to allow for greater or lesser temporal space between detections before they’re flagged as false.

## Filtering False Detections ####
## ?glatos::false_detections

# write the filtered data to a new det_filtered object
#This doesn't delete any rows, it just adds a new column that tells you whether #or not a detection was filtered out.
detections_filtered <- false_detections(detections, tf=3600, show_plot=TRUE)
head(detections_filtered)
nrow(detections_filtered)

The false_detections function will add a new column to your dataframe, ‘passed_filter’. This contains a boolean value that will tell you whether or not that record passed the false detection filter. That information may be useful on its own merits; for now, we will just use it to filter out the false detections.

# Filter based on the column if you're happy with it.

detections_filtered <- detections_filtered[detections_filtered$passed_filter == 1,]
nrow(detections_filtered) # Smaller than before

With our data properly filtered, we can begin investigating it and developing some insights. glatos provides a range of tools for summarizing our data so that we can better see what our receivers are telling us.

We can begin with a summary by animal, which will group our data by the unique animals we’ve detected.

# Summarize Detections ####
#?summarize_detections
#summarize_detections(detections_filtered)

# By animal ====
sum_animal <- summarize_detections(detections_filtered, location_col = 'station', summ_type='animal')

sum_animal

We can also summarize by location, grouping our data by distinct locations.

# By location ====

sum_location <- summarize_detections(detections_filtered, location_col = 'station', summ_type='location')

head(sum_location)

summarize_detections will return different summaries depending on the summ_type parameter. It can take either “animal”, “location”, or “both”. More information on what these summaries return and how they are structured can be found in the help files (?summarize_detections).

If you had another column that describes the location of a detection, and you would prefer to use that, you can specify it in the function with the location_col parameter. In the example below, we will create a new column and use that as the location.

# You can make your own column and use that as the location_col
# For example we will create a uniq_station column for if you have duplicate station names across projects
detections_filtered_special <- detections_filtered %>%
  mutate(station_uniq = paste(glatos_array, station, sep=':'))

sum_location_special <- summarize_detections(detections_filtered_special, location_col = 'station_uniq', summ_type='location')

head(sum_location_special)

For the next example, we’ll summarise along both animal and location, as outlined above.

# By both dimensions
sum_animal_location <- summarize_detections(det = detections_filtered,
                                            location_col = 'station',
                                            summ_type='both')

head(sum_animal_location)

Summarising by both dimensions will create a row for each station and each animal pair. This can be a bit cluttered, so let’s use a filter to remove every row where the animal was not detected on the corresponding station.

# Filter out stations where the animal was NOT detected.
sum_animal_location <- sum_animal_location %>% filter(num_dets > 0)

sum_animal_location

One other method we can use is to summarize by a subset of our animals as well. If we only want to see summary data for a fixed set of animals, we can pass an array containing the animal_ids that we want to see summarized.

# create a custom vector of Animal IDs to pass to the summary function
# look only for these ids when doing your summary
tagged_fish <- c('TQCS-1049258-2008-02-14', 'TQCS-1049269-2008-02-28')

sum_animal_custom <- summarize_detections(det=detections_filtered,
                                          animals=tagged_fish,  # Supply the vector to the function
                                          location_col = 'station',
                                          summ_type='animal')

sum_animal_custom

Now that we have an overview of how to quickly and elegantly summarize our data, let’s make our dataset more amenable to plotting by reducing it from detections to detection events.

Detection Events differ from detections in that they condense a lot of temporally and spatially clustered detections for a single animal into a single detection event. This is a powerful and useful way to clean up the data, and makes it easier to present and clearer to read. Fortunately, this is easy to do with glatos.

# Reduce Detections to Detection Events ####

# ?glatos::detection_events
# you specify how long an animal must be absent before starting a fresh event

events <- detection_events(detections_filtered,
                           location_col = 'station',
                           time_sep=3600)

head(events)

location_col tells the function what to use as the locations by which to group the data, while time_sep tells it how much time has to elapse between sequential detections before the detection belongs to a new event (in this case, 3600 seconds, or an hour). The threshold for your data may be different depending on the purpose of your project.

We can also keep the full extent of our detections, but add a group column so that we can see how they would have been condensed.

# keep detections, but add a 'group' column for each event group
detections_w_events <- detection_events(detections_filtered,
                                        location_col = 'station',
                                        time_sep=3600, condense=FALSE)

With our filtered data in hand, let’s move on to some visualization.

Key Points

More Features of glatos

Overview

Teaching: 15 min
Exercises: 0 min

Questions

• What other features does glatos offer?

Objectives

glatos has more advanced analytic tools that let you manipulate your data further. We’ll cover a few of these features now, to show you how to take your data beyond just filtering and event creation. We’ll also show you how to move your data from glatos to VTrack, another powerful suite of data manipulation tools. By combining the glatos package’s powerful built-in functions with its interoperability across scientific R packages, we’ll show you how to derive powerful insights from your data, and format it in a way that lets you demonstrate them.

glatos can be used to get the residence index of your animals at all the different stations. In fact, glatos offers five different methods for calculating Residence Index. For this lesson, we will showcase two of them, but more information on the others can be found in the glatos documentation.

The residence_index() function requires an events object to create a residence index. We will start by creating a subset like we did in the last lesson. This will save us some time, since running the residence index on the full set is prohibitively long for the scope of this workshop.

First we will decide which animals to base our subset on. To help us with this, we can use group_by on the events object to make it easier to identify good candidates.

#Using all the events data will take too long, so we will subset to just use a couple animals
events %>% group_by(animal_id) %>% summarise(count=n()) %>% arrange(desc(count))

#In this case, we have already decided to use these three animal IDs as the basis for our subset.
subset_animals <- c('TQCS-1049274-2008-02-28', 'TQCS-1049271-2008-02-28', 'TQCS-1049258-2008-02-14')
events_subset <- events %>% filter(animal_id %in% subset_animals)

events_subset

Now that we have a subset of our events object, we can apply the residence_index functions.

# Calc residence index using the Kessel method

rik_data <- residence_index(events_subset,
                            calculation_method = 'kessel')
# "Kessel" method is a special case of "time_interval" where time_interval_size = "1 day"
rik_data

# Calc residence index using the time interval method, interval set to 6 hours
rit_data <- residence_index(events_subset,
                            calculation_method = 'time_interval',
                            time_interval_size = "6 hours")

rit_data

Although the code we’ve written for each method of calculating the residence index is similar, the different parameters and calculation methods mean that these will return different results. It is up to you to investigate which of the methods within glatos best suits your data and its intended application.

One of the development goals of the glatos package is interoperability with other scientific R packages. Currently, we can ‘crosswalk’ data from glatos data to the package VTrack. By ‘crosswalk’, we mean that we can take data that has been manipulated and formatted by glatos, and make it usable by another package, in this case, VTrack. We’ll use the same dataset as before, but we’ll also add some related metadata from a different file.

First, let’s get the tagging metadata.

?prepare_deploy_sheet

tags <- prepare_tag_sheet('TQCS_metadata_tagging.xlsx', sheet_name=1, header_line = 1)
receivers <- prepare_deploy_sheet('TEQ_Deployments_201001_201201.xlsx', sheet_name = 1, header_line = 0, combine_arr_stn = FALSE)

We’ll need this TEQ metadata because the format that VTrack is expecting requires deployment information. The convert_otn_to_att function in glatos will handle assembling the data as long as you pass it the correct deployment metadata. As always, if you have any further questions about the function, the documentation is always available.

Because of differences with how the FACT data is formatted, we need to use prepare_deploy_sheet rather than read_otn_deployments. Depending on how much data you import and from how many sources, you may see both of these functions. More information on how they work and what they do can, as always, be found in the glatos documentation and help files. For now, it is enough to know that they are two different ways of doing the same thing- preparing receiver metadata for an OTN-format to ATT-format conversion- and which one to use depends on the format of your data. For FACT data, we use prepare_deploy_sheet.

We also need to quickly add some columns that FACT extracts do not have, since VTrack will look for them. Remember, the syntax for creating and filling a blank column in your dataframe is the same square-bracket syntax we covered in the earliest parts of this workshop. We also have to rename the station names in our receivers object so that they match the format of the station names in our detections object.

#Add columns missing from FACT extracts.
detections_filtered['sensorvalue'] = NA
detections_filtered['sensorunit'] = NA

# Rename the station names in receivers to match station names in detections (No longer needed with `combine_arr_stn`)
# receivers <- receivers %>% mutate(station=substring(station, 4))

Now that we have all the pieces- detections, tags, and deployment metadata- we can run the convert_otn_to_att function, which will take the glatos data, in OTN format, and convert it into the ATT format, for VTrack.

?convert_otn_to_att

ATTdata <- convert_otn_to_att(detections_filtered, tags, deploymentSheet = receivers)

# ATT is split into 3 objects, we can view them like this
ATTdata$Tag.Detections
ATTdata$Tag.Metadata
ATTdata$Station.Information

With this done, you can use your data (now in the ATTdata object) with the VTrack package. You may notice that not all the detections made it into the ATT object. That’s because the conversion function only keeps detections which occur on receivers for which we have deployment metadata. Detections with no deployment metadata are excluded. This is to prevent issues with VTrack.

Now that our data is in a format that VTrack can understand, we can apply VTrack’s functions to it. For example, we can call VTrack’s abacusPlot function to generate an abacus plot of our data:

# Now that we have an ATT dataframe, we can use it in VTrack functions:

# Abacus plot:
VTrack::abacusPlot(ATTdata)

This may not be especially exciting. However, VTrack has its own set of unique features, just like glatos. To use the spacial features of VTrack, however, we have to give the ATT object a coordinate system to use.

# If you're going to do spatial things in ATT:
library(rgdal)
# Tell the ATT dataframe its coordinates are in decimal lat/lon
proj <- CRS("+init=epsg:4326")
attr(ATTdata, "CRS") <-proj

Once that’s done, we can use VTrack’s functions on our dataset. For example, the COA function, which calculates your dataset’s Centers of Activity, can be used like this:

#Calculate centers of activity.
?COA
coa <- VTrack::COA(ATTdata)

coa %>% group_by(Tag.ID) %>% summarize(n())
coa

To see what this brings us, let’s take a look at a plot of the COAs from VTrack. We’ll use animal ‘TQCS-1049273-2008-02-28’ for this.

We also need to get a shapefile for the Florida coastline so that we have something onto which we plot our data. At the time of this workshop, we’ve had some problems connecting to the GADM service from which we get this shapefile. We encourage you to try it as presented in the code. If that doesn’t work, however, you can still get the shapefile this way:

• Go to http://biogeo.ucdavis.edu/data/gadm3.6/Rsp/gadm36_USA_1_sp.rds. Your browser should download a file called gadm36_USA_1_sp.rds.
• Navigate to the file in your file explorer and get the path.
• In RStudio, navigate to the console and run the command USA <- readRDS('/file/path/to/gadm36_USA_1_sp.rds')

You should now be able to continue the code from the line FL <- USA[USA$NAME_1=="Florida",].

# Plot a COA
coa_single <- coa %>% filter(Tag.ID == 'TQCS-1049273-2008-02-28')

# We'll use raster to get the polygon
library(raster)

#The line below might fail. If it does, use one of the alternate methods, or refer to the steps above to get the shapefile
USA <- getData('GADM', country="USA", level=1)

#Alternative method of getting the polygon. 
# f <-  'http://biogeo.ucdavis.edu/data/gadm3.6/Rsp/gadm36_USA_1_sp.rds'
# b <- basename(f)
# download.file(f, b, mode="wb", method="curl")
# if the above doesn't return a file:
# download.file(f,b, method='wget', extra=c('--no-check-certificate'))
# USA <- readRDS('gadm36_USA_1_sp.rds')

FL <- USA[USA$NAME_1=="Florida",]

# plot the object and zoom in to the St. Lucie River and Jupiter Inlet. Set colour of ground to green Add labels to the axises
plot(FL, xlim=c(-80.75, -80), ylim=c(27, 27.5), col='green', xlab="Longitude", ylab="Latitude")

# For much more zoomed in plot
# plot(FL, xlim=c(-80.4, -80.0), ylim=c(27, 27.3), col='green', xlab="Longitude", ylab="Latitude")

# Create a palette
color <- c(colorRampPalette(c('pink', 'red'))(max(coa_single$Number.of.Detections)))

#add the points

points(coa_single$Longitude.coa, coa_single$Latitude.coa, pch=19, col=color[coa_single$Number.of.Detections],
    cex=log(coa_single$Number.of.Stations) + 0.5) # cex is for point size. natural log is for scaling purposes


# add axes and title
axis(1)
axis(2)
title("Centers of Activities for TQCS-1049273-2008-02-28")

For even more data processing functions, here’s an example of dispersalSummary, which calculates your dataset’s metrics of dispersion.

# Dispersal information
# ?dispersalSummary
dispSum<-dispersalSummary(ATTdata)

View(dispSum)

# Get only the detections when the animal just arrives at a station
dispSum %>% filter(Consecutive.Dispersal > 0) %>%  View

This is only the beginning of what you can do with VTrack and its powerful suite of analysis functions, but a full lesson on VTrack is outside the scope of this workshop. We encourage you to look at the VTrack documentation to see what potential applications it might have to your data.

We will, however, continue with glatos for one more lesson, in which we will cover some basic, but very versatile visualization functions provided by the package.

Key Points

Basic Visualization and Plotting

Overview

Teaching: 30 min
Exercises: 0 min

Questions

• How can I use glatos to plot my data?

• What kinds of plots can I make with my data?

Objectives

Now that we’ve cleaned and processed our data, we can use glatos’ built-in plotting tools to make quick and effective visualizations out of it. One of the simplest visualizations is an abacus plot to display animal detections against the appropriate stations. To this end, glatos supplies a built-in, customizable abacus_plot function.

# Visualizing Data - Abacus Plots ####
# ?glatos::abacus_plot
# customizable version of the standard VUE-derived abacus plots


abacus_plot(detections_w_events,
            location_col='station',
            main='TQCS Detections by Station') # can use plot() variables here, they get passed thru to plot()

This is good, but you can see that the plot is cluttered. Rather than plotting our entire dataset, let’s try filtering out a single animal ID and only plotting that. We can do this right in our call to abacus_plot with the filtering syntax we’ve previously covered.

# pick a single fish to plot
abacus_plot(detections_filtered[detections_filtered$animal_id=="TQCS-1049273-2008-02-28",],
            location_col='station',
            main="TQCS-1049273-2008-02-28 Detections By Station")

Other plots are available in glatos and can show different facets of our data. If we want to see the physical distribution of our stations, for example, a bubble plot will serve us better.

We’ll continue to use the Florida raster FL from last lesson. Remember that we need to provide a coastline shape to plot our data on top of.

# Bubble Plots for Spatial Distribution of Fish ####
# bubble variable gets the summary data that was created to make the plot
detections_filtered

?detection_bubble_plot

bubble_station <- detection_bubble_plot(detections_filtered,
                                out_file = '../tqcs_bubble.png',
                                location_col = 'station',
                                map = FL,
                                col_grad=c('white', 'green'),
                                background_xlim = c(-81, -80),
                                background_ylim = c(26, 28))
bubble_station

bubble_array <- detection_bubble_plot(detections_filtered,
                                      background_xlim = c(-81, -80),
                                      background_ylim = c(26, 28),
                                      map = FL,
                                      out_file = 'act_bubbles_by_array.png')
bubble_array

These examples provide just a brief introduction to some of the plotting available in glatos.

glatos FACT Challenge

Challenge 1 —- Create a bubble plot of that bay we zoomed in earlier. Set the bounding box using the provided nw + se cordinates, change the colour scale and resize the points to be smaller. As a bonus, add points for the other receivers that don’t have any detections. Hint: ?detection_bubble_plot will help a lot Here’s some code to get you started

nw <- c(26,-81) # given
se <- c(28, -80) # given

Solution

nw <- c(26,-81) # given
se <- c(28, -80) # given

bubble_challenge <- detection_bubble_plot(detections_filtered,
                                     background_ylim = c(nw[1], se[1]),
                                     background_xlim = c(nw[2], se[2]),
                                     map = FL,
                                     symbol_radius = 0.75,
                                     location_col = 'station',
                                     col_grad = c('white', 'green'),
                                     receiver_locs = receivers, # For bonus
                                     out_file = 'fact_bubbles_challenge.png')

Key Points

Other OTN Telemetry Curriculums

Overview

Teaching: 0 min
Exercises: 0 min

Questions

• How can I expand my learning?

Objectives

OTN has hosted other workshops in the past which contain different code sets that may be useful to explore after this workshop.

• General telemetry workflow

• IdeasOTN Telemetry Workshop Series 2020: code available here and videos available on our YouTube here

• GLATOS meeting workshop 2020

• GLATOS meeting intro to R 2020

• DFO workshop 2020

• SPG workshop 2020. Videos available on our YouTube here

• FACT workshop 2020

• GLATOS workshop 2021

• ACT/MATOS workshop 2021

Many of our Intro to R workshops are based upon this curriculum from The Carpentries.

Key Points