Chapter 3 Data Abstraction

Abstracting data from large data sets (or even small ones) is critical to data science. The most common first step to visualization is abstracting the data in a form that allows for the visualization goal in mind. If you’ve ever worked with data in spreadsheets, you commonly will be faced with some kind of data manipulation to create meaningful graphs, unless that spreadsheet is specifically designed for it, but then doing something else with the data is going to require some work.

FIGURE 3.1: Visualization of some abstracted data from the EPA Toxic Release Inventory

Figure 3.1 started with abstracting some data from EPA’s Toxic Release Inventory (TRI) program, which holds data reported from a large number of facilities that must report either “stack” or “fugitive” air. Some of the abstraction had already happened when I used the EPA website to download data for particular years and only in California. But there’s more we need to do, and we’ll want to use some dplyr functions to help with it.

At this point, we’ve learned the basics of working with the R language. From here we’ll want to explore how to analyze data, statistically, spatially, and temporally. One part of this is abstracting information from existing data sets by selecting variables and observations and summarizing their statistics.

In the previous chapter, we learned some abstraction methods in base R, such as selecting parts of data frames and applying some functions across the data frame. There’s a lot we can do with these methods, and they continue to be developed and improved. We’ll continue to use them, but they can employ some fairly arcane language. There are many packages that extend R’s functionality, but some of the most important for data science can be found in the various packages of “The Tidyverse” (Wickham and Grolemund 2016), which has the philosophy of making data manipulation more intuitive.

We’ll start with dplyr, which includes an array of data manipulation tools, including select for selecting variables, filter for subsetting observations, summarize for reducing variables to summary statistics, typically stratified by groups, and mutate for creating new variables from mathematical expressions from existing variables. Some dplyr tools such as data joins we’ll look at later in the data transformation chapter.

3.1 The Tidyverse

The tidyverse refers to a suite of R packages developed at RStudio (now posit) (see https://posit.co and <https://r4ds.had.co.nz>) for facilitating data processing and analysis. While R itself is designed around exploratory data analysis, the tidyverse takes it further. Some of the packages in the tidyverse that are widely used are:

dplyr : data manipulation like a database
readr : better methods for reading and writing rectangular data
tidyr : reorganization methods that extend dplyr’s database capabilities
purrr : expanded programming toolkit including enhanced “apply” methods
tibble : improved data frame
stringr : string manipulation library
ggplot2 : graphing system based on the grammar of graphics

In this chapter, we’ll be mostly exploring dplyr, with a few other things thrown in like reading data frames with readr; and we’ll also introduce a few other base R methods. For simplicity, we can just include library(tidyverse) to get everything, and of course the base R methods are already there.

3.2 Tibbles

Tibbles are an improved type of data frame

part of the tidyverse
serve the same purpose as a data frame, and all data frame operations work

Advantages

display better
can be composed of more complex objects like lists, etc.
can be grouped

There multiple ways to create a tibble:

Reading from a CSV using read_csv(). Note the underscore, a function naming convention in the tidyverse.
Using tibble() to either build from vectors or from scratch, or convert from a different type of data frame.
Using tribble() to build in code from scratch.
Using various tidyverse functions that return tibbles.

3.2.1 Building a tibble from vectors

We’ll start by looking at a couple of built-in character vectors (there are lots of things like this in R):

letters : lowercase letters
LETTERS : uppercase letters

letters

##  [1] "a" "b" "c" "d" "e" "f" "g" "h" "i" "j" "k" "l" "m" "n" "o" "p" "q" "r" "s"
## [20] "t" "u" "v" "w" "x" "y" "z"

LETTERS

##  [1] "A" "B" "C" "D" "E" "F" "G" "H" "I" "J" "K" "L" "M" "N" "O" "P" "Q" "R" "S"
## [20] "T" "U" "V" "W" "X" "Y" "Z"

… then make a tibble of letters, LETTERS, and two random sets of 26 values, one normally distributed, the other uniform:

norm <- rnorm(26)
unif <- runif(26)
library(tidyverse)
tibble26 <- tibble(letters,LETTERS,norm,unif)
tibble26

## # A tibble: 26 × 4
##    letters LETTERS    norm   unif
##    <chr>   <chr>     <dbl>  <dbl>
##  1 a       A        0.107  0.655 
##  2 b       B       -0.885  0.0745
##  3 c       C       -0.811  0.972 
##  4 d       D        2.25   0.290 
##  5 e       E        2.42   0.558 
##  6 f       F       -1.78   0.463 
##  7 g       G        0.915  0.711 
##  8 h       H       -0.0131 0.249 
##  9 i       I       -0.152  0.569 
## 10 j       J       -1.19   0.713 
## # ℹ 16 more rows

See section 9.2.3 for more on creating random (or rather pseudo-random) numbers in R.

3.2.1.1 Combining built-in data of various types

As we’re starting to see, there are a lot of data sets that are built into base R. They tend to be fairly simple structures like vectors and time series, shown as “Values” in the Environment tab of RStudio. But we might want to combine them into tibbles to make them more useful.

Have a look at the various data sets that together form data from the 50 United States: character vectors state.name and state.abb, factors state.region, and a state.x77 matrix.

Use ?state to learn more about these data

You’ll find that they each have a length of 50, and they happen to be in the same order by state. That will make it easy to combine them. Note in the following code that we provide better names for the vectors, but use as_tibble to get all of the variables, with their names, from the matrix state.x77.

st <- bind_cols(Name=state.name, Abb=state.abb, Region=state.region, as_tibble(state.x77))
head(st)

## # A tibble: 6 × 11
##   Name     Abb   Region Population Income Illiteracy `Life Exp` Murder `HS Grad`
##   <chr>    <chr> <fct>       <dbl>  <dbl>      <dbl>      <dbl>  <dbl>     <dbl>
## 1 Alabama  AL    South        3615   3624        2.1       69.0   15.1      41.3
## 2 Alaska   AK    West          365   6315        1.5       69.3   11.3      66.7
## 3 Arizona  AZ    West         2212   4530        1.8       70.6    7.8      58.1
## 4 Arkansas AR    South        2110   3378        1.9       70.7   10.1      39.9
## 5 Califor… CA    West        21198   5114        1.1       71.7   10.3      62.6
## 6 Colorado CO    West         2541   4884        0.7       72.1    6.8      63.9
## # ℹ 2 more variables: Frost <dbl>, Area <dbl>

Use the methods you’ve learned in RStudio to see what this tibble contains, and note that not all variables are displayed in the simple head() output above. Can you find Frost, the “mean number of days with minimum temperature below freezing (1931-1960) in capital or large city”? Those would probably be fewer now.

3.2.2 tribble

As long as you don’t let them multiply in your starship, tribbles are handy for creating tibbles. (Or rather the tribble function is a handy way to create tibbles in code.) You simply create the variable names with a series of entries starting with a tilde, then the data are entered one row at a time. If you line them all up in your code one row at a time, it’s easy to enter the data accurately (Table 3.1).

peaks <- tribble(
  ~peak, ~elev, ~longitude, ~latitude,
  "Mt. Whitney", 4421, -118.2, 36.5,
  "Mt. Elbert", 4401, -106.4, 39.1,
  "Mt. Hood", 3428, -121.7, 45.4,
  "Mt. Rainier", 4392, -121.8, 46.9)
knitr::kable(peaks, caption = 'Peaks tibble')

TABLE 3.1: Peaks tibble
peak	elev	longitude	latitude
Mt. Whitney	4421	-118.2	36.5
Mt. Elbert	4401	-106.4	39.1
Mt. Hood	3428	-121.7	45.4
Mt. Rainier	4392	-121.8	46.9

3.2.3 read_csv

The read_csv function does somewhat the same thing as read.csv in base R, but creates a tibble instead of a data.frame, and has some other properties we’ll look at below.

Note that the code below accesses data we’ll be using a lot, from EPA Toxic Release Inventory (TRI) data. If you want to keep this data organized in a separate project, you might consider creating a new air_quality project. This is optional, and you can get by with staying in one project since all of our data will be accessed from the igisci package. But in your own work, you will find it useful to create separate projects to keep things organized with your code and data together.

library(tidyverse); library(igisci)
TRI87 <- read_csv(ex("TRI/TRI_1987_BaySites.csv"))
TRI87df <- read.csv(ex("TRI/TRI_1987_BaySites.csv"))
TRI87b <- tibble(TRI87df)
identical(TRI87, TRI87b)

## [1] FALSE

Note that they’re not identical. So what’s the difference between read_csv and read.csv? Why would we use one over the other? Since their names are so similar, you may accidentally choose one or the other. Some things to consider:

To use read_csv, you need to load the readr or tidyverse library, or use readr::read_csv.
The read.csv function “fixes” some things and sometimes that might be desired: problematic variable names like MLY-TAVG-NORMAL become MLY.TAVG.NORMAL – but this may create problems if those original names are a standard designation.
With read.csv, numbers stored as characters are converted to numbers: “01” becomes 1, “02” becomes 2, etc.
There are other known problems that read_csv avoids.

Recommendation: Use read_csv and write_csv.

You can still just call tibbles “data frames”, since they are still data frames, and in this book we’ll follow that practice.

3.3 Summarizing variable distributions

A simple statistical summary is very easy to do in R, and we’ll use eucoak data in the igisci package from a study of comparative runoff and erosion under eucalyptus and oak canopies (Thompson, Davis, and Oliphant 2016). In this study (Figure 3.2), we looked at the amount of runoff and erosion captured in Gerlach troughs on paired eucalyptus and oak sites in the San Francisco Bay Area. You might consider creating a eucoak project, since we’ll be referencing this data set in several places.

FIGURE 3.2: Euc-Oak paired plot runoff and erosion study (Thompson, Davis, and Oliphant (2016))

library(igisci)
summary(eucoakrainfallrunoffTDR)

##      site               site #          date              month          
##  Length:90          Min.   :1.000   Length:90          Length:90         
##  Class :character   1st Qu.:2.000   Class :character   Class :character  
##  Mode  :character   Median :4.000   Mode  :character   Mode  :character  
##                     Mean   :4.422                                        
##                     3rd Qu.:6.000                                        
##                     Max.   :8.000                                        
##                                                                          
##     rain_mm         rain_oak        rain_euc      runoffL_oak    
##  Min.   : 1.00   Min.   : 1.00   Min.   : 1.00   Min.   : 0.000  
##  1st Qu.:16.00   1st Qu.:16.00   1st Qu.:14.75   1st Qu.: 0.000  
##  Median :28.50   Median :30.50   Median :30.00   Median : 0.450  
##  Mean   :37.99   Mean   :35.08   Mean   :34.60   Mean   : 2.032  
##  3rd Qu.:63.25   3rd Qu.:50.50   3rd Qu.:50.00   3rd Qu.: 2.800  
##  Max.   :99.00   Max.   :98.00   Max.   :96.00   Max.   :14.000  
##  NA's   :18      NA's   :2       NA's   :2       NA's   :5       
##   runoffL_euc      slope_oak       slope_euc       aspect_oak   
##  Min.   : 0.00   Min.   : 9.00   Min.   : 9.00   Min.   :100.0  
##  1st Qu.: 0.07   1st Qu.:12.00   1st Qu.:12.00   1st Qu.:143.0  
##  Median : 1.20   Median :24.50   Median :23.00   Median :189.0  
##  Mean   : 2.45   Mean   :21.62   Mean   :19.34   Mean   :181.9  
##  3rd Qu.: 3.30   3rd Qu.:30.50   3rd Qu.:25.00   3rd Qu.:220.0  
##  Max.   :16.00   Max.   :32.00   Max.   :31.00   Max.   :264.0  
##  NA's   :3                                                      
##    aspect_euc    surface_tension_oak surface_tension_euc
##  Min.   :106.0   Min.   :37.40       Min.   :28.51      
##  1st Qu.:175.0   1st Qu.:72.75       1st Qu.:32.79      
##  Median :196.5   Median :72.75       Median :37.40      
##  Mean   :191.2   Mean   :68.35       Mean   :43.11      
##  3rd Qu.:224.0   3rd Qu.:72.75       3rd Qu.:56.41      
##  Max.   :296.0   Max.   :72.75       Max.   :72.75      
##                  NA's   :22          NA's   :22         
##  runoff_rainfall_ratio_oak runoff_rainfall_ratio_euc
##  Min.   :0.00000           Min.   :0.000000         
##  1st Qu.:0.00000           1st Qu.:0.003027         
##  Median :0.02046           Median :0.047619         
##  Mean   :0.05357           Mean   :0.065902         
##  3rd Qu.:0.08485           3rd Qu.:0.083603         
##  Max.   :0.42000           Max.   :0.335652         
##  NA's   :5                 NA's   :3

In the summary output, how are character variables handled differently from numeric ones?

Remembering what we discussed in the previous chapter, consider the site variable (Figure ??), and in particular its Length. Looking at the table, what does that length represent?

There are a couple of ways of removing duplicates from a data frame, such as the unique() and the similar dplyr::distinct() functions. In the Transformation chapter, we’ll use this as one step in paring down big data from groundwater wells.

We could also use unique() to simply have a quick look at our list of subcanopy runoff measurement sites, which helps us consider using site as a factor for statistical analysis. Consider what these return:

unique(eucoakrainfallrunoffTDR$site)

## [1] "AB1" "AB2" "KM1" "PR1" "TP1" "TP2" "TP3" "TP4"

factor(eucoakrainfallrunoffTDR$site)

##  [1] AB1 AB1 AB1 AB1 AB1 AB1 AB1 AB1 AB1 AB1 AB1 AB1 AB2 AB2 AB2 AB2 AB2 AB2 AB2
## [20] AB2 AB2 AB2 AB2 AB2 KM1 KM1 KM1 KM1 KM1 KM1 KM1 KM1 KM1 KM1 KM1 KM1 PR1 PR1
## [39] PR1 PR1 PR1 PR1 PR1 PR1 PR1 PR1 TP1 TP1 TP1 TP1 TP1 TP1 TP1 TP1 TP1 TP1 TP1
## [58] TP2 TP2 TP2 TP2 TP2 TP2 TP2 TP2 TP2 TP2 TP2 TP3 TP3 TP3 TP3 TP3 TP3 TP3 TP3
## [77] TP3 TP3 TP3 TP4 TP4 TP4 TP4 TP4 TP4 TP4 TP4 TP4 TP4 TP4
## Levels: AB1 AB2 KM1 PR1 TP1 TP2 TP3 TP4

3.3.1 Stratifying variables by site using a Tukey box plot

A good way to look at variable distributions stratified by a sample site factor is the Tukey box plot (Figure 3.3). We’ll be looking more at this and other visualization methods in the next chapter.

ggplot(data = eucoakrainfallrunoffTDR) + 
  geom_boxplot(mapping = aes(x=site, y=runoffL_euc))

FIGURE 3.3: Tukey boxplot of runoff under eucalyptus canopy

3.4 Database operations

As part of exploring our data, we’ll typically simplify or reduce it for our purposes. The following methods are quickly discovered to be essential as part of exploring and analyzing data.

select variable columns you want to retain with select
filter observations using logic, such as population \> 10000, with filter
subset data to do both filtering and variable selection, with base R’s subset
add new variables and assign their values with mutate
sort rows based on a field with arrange
summarize by group

3.4.1 Select, mutate, and the pipe

Read the pipe operator %>% as “and then…” This is bigger than it sounds and opens up many possibilities.

You can also use |> as the pipe operator. Once you get used to where the | character is on your keyboard, you’ll find that it makes your code a bit easier to read. Every character adds up when you’re interpreting code.

See the example below, and observe how the expression becomes several lines long. In the process, we’ll see examples of new variables with mutate and selecting (and in the process ordering) variables (Table 3.2).

runoff <- eucoakrainfallrunoffTDR %>%
  mutate(Date = as.Date(date,"%m/%d/%Y"),
         rain_subcanopy = (rain_oak + rain_euc)/2) %>%
  dplyr::select(site, Date, rain_mm, rain_subcanopy, 
         runoffL_oak, runoffL_euc, slope_oak, slope_euc)

TABLE 3.2: EucOak data reorganized a bit, first 6
site	Date	rain_mm	rain_subcanopy	runoffL_oak	runoffL_euc	slope_oak	slope_euc
AB1	2006-11-08	29	29.0	4.7900	6.70000	32	31
AB1	2006-11-12	22	18.5	3.2000	4.30000	32	31
AB1	2006-11-29	85	65.0	9.7000	16.00000	32	31
AB1	2006-12-12	82	87.5	14.0000	14.20000	32	31
AB1	2006-12-28	43	54.0	9.7472	4.32532	32	31
AB1	2007-01-29	7	54.0	1.4000	0.00000	32	31

Another way of thinking of the pipe that is very useful is that whatever goes before it becomes the first parameter for any functions that follow. So in the example above:

The parameter eucoakrainfallrunoffTDR becomes the first for mutate(), then
The result of the mutate() becomes the first parameter for dplyr::select()

To just rename a variable, use rename instead of mutate. It will stay in position.

3.4.1.1 Review: creating penguins from penguins_raw

To review some of these methods, it’s useful to consider how the penguins data frame was created from the more complex penguins_raw data frame, both of which are part of the palmerpenguins package (Horst, Hill, and Gorman (2020)). First let’s look at palmerpenguins::penguins_raw:

library(palmerpenguins)
library(tidyverse)
library(lubridate)
summary(penguins_raw)

##   studyName         Sample Number      Species             Region         
##  Length:344         Min.   :  1.00   Length:344         Length:344        
##  Class :character   1st Qu.: 29.00   Class :character   Class :character  
##  Mode  :character   Median : 58.00   Mode  :character   Mode  :character  
##                     Mean   : 63.15                                        
##                     3rd Qu.: 95.25                                        
##                     Max.   :152.00                                        
##                                                                           
##     Island             Stage           Individual ID      Clutch Completion 
##  Length:344         Length:344         Length:344         Length:344        
##  Class :character   Class :character   Class :character   Class :character  
##  Mode  :character   Mode  :character   Mode  :character   Mode  :character  
##                                                                             
##                                                                             
##                                                                             
##                                                                             
##     Date Egg          Culmen Length (mm) Culmen Depth (mm) Flipper Length (mm)
##  Min.   :2007-11-09   Min.   :32.10      Min.   :13.10     Min.   :172.0      
##  1st Qu.:2007-11-28   1st Qu.:39.23      1st Qu.:15.60     1st Qu.:190.0      
##  Median :2008-11-09   Median :44.45      Median :17.30     Median :197.0      
##  Mean   :2008-11-27   Mean   :43.92      Mean   :17.15     Mean   :200.9      
##  3rd Qu.:2009-11-16   3rd Qu.:48.50      3rd Qu.:18.70     3rd Qu.:213.0      
##  Max.   :2009-12-01   Max.   :59.60      Max.   :21.50     Max.   :231.0      
##                       NA's   :2          NA's   :2         NA's   :2          
##  Body Mass (g)      Sex            Delta 15 N (o/oo) Delta 13 C (o/oo)
##  Min.   :2700   Length:344         Min.   : 7.632    Min.   :-27.02   
##  1st Qu.:3550   Class :character   1st Qu.: 8.300    1st Qu.:-26.32   
##  Median :4050   Mode  :character   Median : 8.652    Median :-25.83   
##  Mean   :4202                      Mean   : 8.733    Mean   :-25.69   
##  3rd Qu.:4750                      3rd Qu.: 9.172    3rd Qu.:-25.06   
##  Max.   :6300                      Max.   :10.025    Max.   :-23.79   
##  NA's   :2                         NA's   :14        NA's   :13       
##    Comments        
##  Length:344        
##  Class :character  
##  Mode  :character  
##                    
##                    
##                    
##

Now let’s create the simpler penguins data frame. We’ll use rename for a couple, but most variables require mutation to manipulate strings (we’ll get to that later), create factors, or convert to integers. And we’ll rename some variables to avoid using backticks (the backward single quotation mark accessed just to the left of the 1 key and below the Esc key, and what you can use in markdown to create a monospaced font as I just used for 1 and Esc).

penguins <- penguins_raw %>%
  rename(bill_length_mm = `Culmen Length (mm)`,
         bill_depth_mm = `Culmen Depth (mm)`) %>%
  mutate(species = factor(word(Species)),
         island = factor(Island),
         flipper_length_mm = as.integer(`Flipper Length (mm)`),
         body_mass_g = as.integer(`Body Mass (g)`),
         sex = factor(str_to_lower(Sex)),
         year = as.integer(year(ymd(`Date Egg`)))) %>%
  dplyr::select(species, island, bill_length_mm, bill_depth_mm, 
                flipper_length_mm, body_mass_g, sex, year)
summary(penguins)

##       species          island    bill_length_mm  bill_depth_mm  
##  Adelie   :152   Biscoe   :168   Min.   :32.10   Min.   :13.10  
##  Chinstrap: 68   Dream    :124   1st Qu.:39.23   1st Qu.:15.60  
##  Gentoo   :124   Torgersen: 52   Median :44.45   Median :17.30  
##                                  Mean   :43.92   Mean   :17.15  
##                                  3rd Qu.:48.50   3rd Qu.:18.70  
##                                  Max.   :59.60   Max.   :21.50  
##                                  NA's   :2       NA's   :2      
##  flipper_length_mm  body_mass_g       sex           year     
##  Min.   :172.0     Min.   :2700   female:165   Min.   :2007  
##  1st Qu.:190.0     1st Qu.:3550   male  :168   1st Qu.:2007  
##  Median :197.0     Median :4050   NA's  : 11   Median :2008  
##  Mean   :200.9     Mean   :4202                Mean   :2008  
##  3rd Qu.:213.0     3rd Qu.:4750                3rd Qu.:2009  
##  Max.   :231.0     Max.   :6300                Max.   :2009  
##  NA's   :2         NA's   :2

Unfortunately, they don’t end up as exactly identical, though all of the variables are identical as vectors:

identical(penguins, palmerpenguins::penguins)

## [1] FALSE

3.4.1.2 Helper functions for `dplyr::select()`

In the select() example above, we listed all of the variables, but there are a variety of helper functions for using logic to specify which variables to select:

contains("_") or any substring of interest in the variable name
starts_with("runoff")
ends_with("euc")
everything()
matches() a regular expression
num_range("x",1:5) for the common situation where a series of variable names combine a string and a number
one_of(myList) for when you have a group of variable names
range of variables: e.g. runoffL_oak:slope_euc could have followed rain_subcanopy above
all but: preface a variable or a set of variable names with - to select all others

3.4.2 Filtering observations

filter lets you select observations that meet criteria, similar to an SQL WHERE clause (Table 3.3).

runoff2007 <- runoff %>%
  filter(Date >= as.Date("04/01/2007", "%m/%d/%Y"))

TABLE 3.3: Date-filtered EucOak data
site	Date	rain_mm	rain_subcanopy	runoffL_oak	runoffL_euc	slope_oak	slope_euc
AB1	2007-04-23	NA	33.5	6.94488	9.19892	32.0	31
AB1	2007-05-05	NA	31.0	6.33568	7.43224	32.0	31
AB2	2007-04-23	23	35.5	4.32000	2.88000	24.0	25
AB2	2007-05-05	11	25.5	4.98000	3.30000	24.0	25
KM1	2007-04-23	NA	37.0	1.56000	2.04000	30.5	25
KM1	2007-05-05	28	22.0	1.32000	1.32000	30.5	25

3.4.2.1 Filtering out `NA` with `!is.na`

Here’s a really important one. There are many times you need to avoid NAs. We thus commonly see summary statistics using na.rm = TRUE in order to ignore NAs when calculating a statistic like mean.

To simply filter out NAs from a vector or a variable use a filter:

feb_filt <- sierraFeb %>% filter(!is.na(TEMPERATURE))

3.4.3 Subsetting data with base R’s `subset`

We’ve looked at selecting variables and filtering observations, both examples of subsetting data. We also learned in the previous chapter on base R methods that there are lots of ways of subsetting data, such as using accessors like [] and $. We’ll look at another base R method that can do both variable selection and observation filtering in one step, which in some cases you might prefer.

Let’s look at a problem where we’re subsetting the palmerpenguins::pengins data to just abstract the species and body mass of female penguins. We’d need to use both filter and select to accomplish this, but of course the pipe operator helps:

library(palmerpenguins); library(dplyr)
penguins |> filter(sex=="female") |> select(species,body_mass_g)

The exact same result can be achieved with base R’s subset function, which if we use a pipe operator looks pretty similar:

library(palmerpenguins)
penguins |> subset(sex=="female", select=c(species,body_mass_g))

## # A tibble: 165 × 2
##    species body_mass_g
##    <fct>         <int>
##  1 Adelie         3800
##  2 Adelie         3250
##  3 Adelie         3450
##  4 Adelie         3625
##  5 Adelie         3200
##  6 Adelie         3700
##  7 Adelie         3450
##  8 Adelie         3325
##  9 Adelie         3400
## 10 Adelie         3800
## # ℹ 155 more rows

As a review, how would you do this without the pipe?

Given the option of either using filter and select or just subset in the above example, it’s just personal preference, which is often driven by the desire to make code readable, and that can go either way. Then there’s the desire to reduce the size of what functions you want to remember, which can also go either way; you still need to know what “select” means.

Let’s take our example a bit further and use subset to pick the most massive individual female penguin in the dataset. You could also do this with filter, but the key is using the max function.

library(palmerpenguins)
penguins |> subset(sex=="female", select=c(species,body_mass_g)) |>
            subset(body_mass_g==max(body_mass_g))

## # A tibble: 2 × 2
##   species body_mass_g
##   <fct>         <int>
## 1 Gentoo         5200
## 2 Gentoo         5200

In this case, there was a tie between the two most massive individuals.

It shouldn’t surprise us that there are lots of ways of subsetting data in R, since doing so helps us satisfy a key requirement of data science: being able to wrangle big data into something that is able to be visualized and understood.

3.4.4 Writing a data frame to a csv

Let’s say you have created a data frame, maybe with read_csv

runoff20062007 <- read_csv(ex("eucoak/eucoakrainfallrunoffTDR.csv"))

Then you do some processing to change it, maybe adding variables, reorganizing, etc., and you want to write out your new eucoak, so you just need to use write_csv

write_csv(eucoak, "data/tidy_eucoak.csv")

Note the use of a data folder data: Remember that your default workspace (wd for working directory) is where your project file resides (check what it is with getwd()), so by default you’re saving things in that wd. To keep things organized the above code is placing data in a data folder within the wd.

3.4.5 Summarize by group

You’ll find that you need to use this all the time with real data. Let’s say you have a bunch of data where some categorical variable is defining a grouping, like our site field in the eucoak data. This is a form of stratifying our data. We’d like to just create average slope, rainfall, and runoff for each site. Note that it involves two steps, first defining which field defines the group, then the various summary statistics we’d like to store. In this case, all of the slopes under oak remain the same for a given site – it’s a site characteristic – and the same applies to the euc site, so we can just grab the first value (mean would have also worked of course) (Table 3.4).

eucoakSiteAvg <- runoff %>%
  group_by(site) %>%
  summarize(
    rain = mean(rain_mm, na.rm = TRUE),
    rain_subcanopy = mean(rain_subcanopy, na.rm = TRUE),
    runoffL_oak = mean(runoffL_oak, na.rm = TRUE),
    runoffL_euc = mean(runoffL_euc, na.rm = TRUE),
    slope_oak = first(slope_oak),
    slope_euc = first(slope_euc)
  )

TABLE 3.4: EucOak data summarized by site
site	rain	rain_subcanopy	runoffL_oak	runoffL_euc	slope_oak	slope_euc
AB1	48.37500	43.08333	6.8018364	6.026523	32.0	31
AB2	34.08333	35.37500	4.9113636	3.654545	24.0	25
KM1	48.00000	36.12500	1.9362500	0.592500	30.5	25
PR1	56.50000	37.56250	0.4585714	2.310000	27.0	23
TP1	38.36364	30.04545	0.8772727	1.657273	9.0	9
TP2	34.33333	32.86364	0.0954545	1.525454	12.0	10

Summarizing by group with TRI data

library(igisci)
TRI_BySite <- read_csv(ex("TRI/TRI_2017_CA.csv")) %>%
  mutate(all_air = `5.1_FUGITIVE_AIR` + `5.2_STACK_AIR`) %>%
  filter(all_air > 0) %>%
  group_by(FACILITY_NAME) %>%
  summarize(
    FACILITY_NAME = first(FACILITY_NAME),
    air_releases = sum(all_air, na.rm = TRUE),
    mean_fugitive = mean(`5.1_FUGITIVE_AIR`, na.rm = TRUE), 
    LATITUDE = first(LATITUDE), LONGITUDE = first(LONGITUDE))

Integer variables might be used to define categories or groups. Down the road, we’ll run into situations where numeric variables might be used to define groups, such as integer days of the year we’ll find useful in studying time series, but could also be zip codes, census tracts, or other spatial enumeration units. Note that these are discrete, not continuous values.

3.4.6 Count

The count function is a simple variant on summarizing by group, since the only statistic is the count of events. The factor to group by is simply provided as a parameter: count(factor).

tidy_eucoak %>% count(tree)

## # A tibble: 2 × 2
##   tree      n
##   <chr> <int>
## 1 euc      90
## 2 oak      90

Another way to create a count is to use n():

tidy_eucoak %>%
  group_by(tree) %>%
  summarize(n = n())

## # A tibble: 2 × 2
##   tree      n
##   <chr> <int>
## 1 euc      90
## 2 oak      90

3.4.7 Sorting after summarizing

Using the marine debris data from the Marine Debris Monitoring and Assessment Project (Marine Debris Program, n.d.), we can use arrange to sort by latitude, so we can see the beaches from south to north along the Pacific coast.

shorelineLatLong <- ConcentrationReport %>%
  group_by(`Shoreline Name`) %>%
  summarize(
    latitude = mean((`Latitude Start`+`Latitude End`)/2),
    longitude = mean((`Longitude Start`+`Longitude End`)/2)
  ) %>%
  arrange(latitude)
shorelineLatLong

## # A tibble: 38 × 3
##    `Shoreline Name`   latitude longitude
##    <chr>                 <dbl>     <dbl>
##  1 Aimee Arvidson         33.6     -118.
##  2 Balboa Pier #2         33.6     -118.
##  3 Bolsa Chica            33.7     -118.
##  4 Junipero Beach         33.8     -118.
##  5 Malaga Cove            33.8     -118.
##  6 Zuma Beach, Malibu     34.0     -119.
##  7 Zuma Beach             34.0     -119.
##  8 Will Rodgers           34.0     -119.
##  9 Carbon Beach           34.0     -119.
## 10 Nicholas Canyon        34.0     -119.
## # ℹ 28 more rows

3.4.8 The dot operator

The dot (.) operator refers to here, similar to UNIX syntax for accessing files in the current folder where the path is "./filename". In a piped sequence, you might need to reference the object (like the data frame) the pipe is using, as you can see in the following code.

The advantage of the pipe is you don’t have to keep referencing the data frame.
The dot is then used to connect to items inside the data frame

TRI87 <- read_csv(ex("TRI/TRI_1987_BaySites.csv"))
stackrate <- TRI87 %>%
  mutate(stackrate = stack_air/air_releases) %>%
  .$stackrate
head(stackrate)

## [1] 0.0000000 0.0000000 0.0000000 0.0000000 0.6666667 1.0000000

3.5 String abstraction

Character string manipulation is surprisingly critical to data analysis, and so the stringr package was developed to provide a wider array of string processing tools than what is in base R, including functions for detecting matches, subsetting strings, managing lengths, replacing substrings with other text, and joining, splitting, and sorting strings.

We’ll look at some of the stringr functions, but a good way to learn about the wide array of functions is through the cheat sheet that can be downloaded from https://www.rstudio.com/resources/cheatsheets/

3.5.1 Detecting matches

These functions look for patterns within existing strings, which can then be used subset observations based on those patterns.

str_detect detects patterns in a string, returns true or false if detected
str_locate detects patterns in a string, returns start and end position if detected, or NA if not
str_which returns the indices of strings that match a pattern
str_count counts the number of matches in each string

str_detect(fruit,"qu")
str_which(fruit, "qu")
fruit[str_which(fruit,"qu")]

##  [1] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [13] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [25] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [37] FALSE FALSE FALSE FALSE FALSE FALSE  TRUE FALSE FALSE  TRUE FALSE FALSE
## [49] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [61] FALSE FALSE FALSE FALSE FALSE FALSE  TRUE FALSE FALSE FALSE FALSE FALSE
## [73] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [1] 43 46 67
## [1] "kumquat" "loquat"  "quince"

Note that str_locate returns a matrix with rows for each item where the substring is located with both start and end locations.

fruit_a <- fruit[str_count(fruit,"a")>1]
fruit_a
str_locate(fruit_a,"a")

##  [1] "avocado"      "banana"       "blackcurrant" "canary melon" "cantaloupe"  
##  [6] "guava"        "mandarine"    "papaya"       "pomegranate"  "rambutan"    
## [11] "salal berry"  "satsuma"      "tamarillo"   
##       start end
##  [1,]     1   1
##  [2,]     2   2
##  [3,]     3   3
##  [4,]     2   2
##  [5,]     2   2
##  [6,]     3   3
##  [7,]     2   2
##  [8,]     2   2
##  [9,]     7   7
## [10,]     2   2
## [11,]     2   2
## [12,]     2   2
## [13,]     2   2

The str_locate_all function carries this further and finds all locations of the substrings within the string.

fruit3a <- fruit[str_count(fruit,"a")==3]
fruit3a
str_locate_all(fruit3a,"a")

## [1] "banana" "papaya"
## [[1]]
##      start end
## [1,]     2   2
## [2,]     4   4
## [3,]     6   6
## 
## [[2]]
##      start end
## [1,]     2   2
## [2,]     4   4
## [3,]     6   6

qufruit <- fruit[str_which(fruit,"qu")]
qufruit
str_locate(qufruit, "qu")

## [1] "kumquat" "loquat"  "quince" 
##      start end
## [1,]     4   5
## [2,]     3   4
## [3,]     1   2

3.5.2 Subsetting strings

Subsetting in this case includes its normal use of abstracting the observations specified by a match (similar to a filter for data frames), or just a specified part of a string specified by start and end character positions, or the part of the string that matches an expression.

str_sub extracts a part of a string from a start to an end character position
str_subset returns the strings that contain a pattern match
str_extract returns the first (or if str_extract_all then all matches) pattern matches
str_match returns the first (or _all) pattern match as a matrix
str_remove removes a specific substring (not on the cheat sheet, but handy)

qfruit <- str_subset(fruit, "q")
qfruit
str_sub(qfruit,1,2)

## [1] "kumquat" "loquat"  "quince" 
## [1] "ku" "lo" "qu"

str_sub("94132",1,2)
str_extract(qfruit,"[aeiou]")
str_remove(str_subset(fruit,"fruit"), "fruit")

## [1] "94"
## [1] "u" "o" "u"
## [1] "bread"   "dragon"  "grape"   "jack"    "kiwi "   "passion" "star "  
## [8] "ugli "

3.5.3 String length

The length of strings is often useful in an analysis process, either just knowing the length as an integer, or purposefully increasing or reducing it.

str_length simply returns the length of the string as an integer

qfruit <- str_subset(fruit,"q")
qfruit
str_length(qfruit)

## [1] "kumquat" "loquat"  "quince" 
## [1] 7 6 6

Don’t confuse str_length with length (from base R), the length of an object like a vector. Both are highly useful.

length(fruit)

## [1] 80

length(qfruit)

## [1] 3

Using str_locate with str_length:

name <- "Inigo Montoya"
str_length(name)
firstname <- str_sub(name,1,str_locate(name," ")[1]-1)
firstname
lastname <- str_sub(name,str_locate(name," ")[1]+1,str_length(name))
lastname

## [1] 13
## [1] "Inigo"
## [1] "Montoya"

Padding and trimming lengths:

str_pad adds a specified character (typically a space " ") to either end of a string
str_trim removes white space from either end of a string

Here’s an example str_pad and str_trim (in this case, reversing the former):

str_pad(qfruit,10,"both")
str_trim(str_pad(qfruit,10,"both"),"both")

## [1] " kumquat  " "  loquat  " "  quince  "
## [1] "kumquat" "loquat"  "quince"

3.5.4 Replacing substrings with other text (“mutating” strings)

These methods range from converting case to replacing substrings.

str_to_lower converts strings to lowercase
str_to_upper converts strings to uppercase
str_to_title capitalizes strings (makes the first character of each word uppercase)

str_to_lower(name)
str_to_upper(name)
str_to_title("for whom the bell tolls")

## [1] "inigo montoya"
## [1] "INIGO MONTOYA"
## [1] "For Whom The Bell Tolls"

str_replace replaces the first matched pattern (or all with str_replace_all) with a specified string
str_sub a special use of this function to replace substrings with a specified string

str_replace(qfruit,"q","z")
str_sub(name,1,str_locate(name," ")[1]-1) <- "Diego"
name

## [1] "kumzuat" "lozuat"  "zuince" 
## [1] "Diego Montoya"

The replacement mode of str_sub used above may seem pretty tortuous and confusing. Why did we have to access [1] in str_locate(name," ")[1]? Experiment with what you get with str_locate(name," ") itself, and maybe consult the help system with ?str_locate (and run the examples), but you’ll find this is one that returns a matrix, and even if there’s only one item located will still have a start and end location. See what happens when you use it without accessing [1]. You’re seeing one effect of vectorization – reuse. It can be very confusing, but reuse of inputs is one of the (very useful but still confusing) properties of vectorization. In this case it creates duplication, creating a vector of names.

3.5.5 Concatenating and splitting

One very common string function is to concatenate strings, and somewhat less common is splitting them using a key separator like space, comma, or line end. One use of using str_c in the example below is to create a comparable join field based on a numeric character string that might need a zero or something at the left or right.

str_c The paste() function in base R will work, but you might want the default separator setting to be “” (as does paste0()) instead of " ", so str_c is just paste with a default “” separator, but you can also use " ".
str_split splits a string into parts based upon the detection of a specified separator like space, comma, or line end. However, this ends up creating a list, which can be confusing to use. If you’re wanting to access a particular member of a list of words, the word function is easier.
word lets you pick the first, second, or nth word in a string of words.

library(stringr)
phrase <- str_c("for","whom","the","bell","tolls",sep=" ")
phrase

## [1] "for whom the bell tolls"

str_split(phrase, " ")

## [[1]]
## [1] "for"   "whom"  "the"   "bell"  "tolls"

word(phrase,4)

## [1] "bell"

… and of course, these methods can all be vectorized:

sentences[1:4]

## [1] "The birch canoe slid on the smooth planks." 
## [2] "Glue the sheet to the dark blue background."
## [3] "It's easy to tell the depth of a well."     
## [4] "These days a chicken leg is a rare dish."

word(sentences[1:4],2)

## [1] "birch" "the"   "easy"  "days"

Example of str_c use to modify a variable needed for a join:

csvPath <- system.file("extdata","CA/CA_MdInc.csv",package="igisci")
CA_MdInc <- read_csv(csvPath)
join_id <- str_c("0",CA_MdInc$NAME)
# could also use str_pad(CA_MdInc$NAME,1,side="left",pad="0")
head(CA_MdInc)

## # A tibble: 6 × 3
##         trID     NAME HHinc2016
##        <dbl>    <dbl>     <dbl>
## 1 6001400100 60014001    177417
## 2 6001400200 60014002    153125
## 3 6001400300 60014003     85313
## 4 6001400400 60014004     99539
## 5 6001400500 60014005     83650
## 6 6001400600 60014006     61597

head(join_id)

## [1] "060014001" "060014002" "060014003" "060014004" "060014005" "060014006"

3.6 Dates and times with `lubridate`

Simply stated, the lubridate package just makes it easier to work with dates and times. Dates and times are more difficult to work with than numbers and character strings, which is why lubridate was developed. The lubridate package is excellent at detecting and reading in dates and times – it can “parse” many forms. Spend some time to make sure you know how to use it. We’ll be using dates and times a lot, since many data sets include time stamps. See the cheat sheet for more information but the following examples may demonstrate that it’s pretty easy to use, and does a good job of making your job easier.

library(lubridate)
dmy("20 September 2020")
dmy_hm("20 September 2020 10:45")
mdy_hms("September 20, 2020 10:48")
mdy_hm("9/20/20 10:50")
mdy("9.20.20")

## [1] "2020-09-20"
## [1] "2020-09-20 10:45:00 UTC"
## [1] "2020-09-20 20:10:48 UTC"
## [1] "2020-09-20 10:50:00 UTC"
## [1] "2020-09-20"

start704 <- dmy_hm("24 August 2020 16:00")
end704 <- mdy_hm("12/18/2020 4:45 pm")
year(start704)
month(start704)
day(end704)
hour(end704)

## [1] 2020
## [1] 8
## [1] 18
## [1] 16

A bit of date math:

end704-start704
as_date(end704)
hms::as_hms(end704)

## Time difference of 116.0312 days
## [1] "2020-12-18"
## 16:45:00

You will find that dates are typically displayed as yyyy-mm-dd, e.g. “2020-09-20” above. You can display them other ways of course, but it’s useful to write dates this way since they sort better, with the highest order coming first.

Calling functions explicitly with ::: Sometimes you need to specify the package and function name this way, for instance, if more than one package has a function of the same name. You can also use this method to call a function without having loaded its library. Due to multiple packages having certain common names (like select), it’s common to use this syntax, and you’ll find that we’ll use dplyr::select(...) throughout this book.

3.6.1 Dates used for annual time series

For obvious reasons, we’ll use lubridate a lot when we get to time series. One date abstraction that will be particularly useful is day of the year, which can be abstracted from a date object as yday, which returns the day of the year starting with 1 for January 1 in any given year. Note that as with all of the date objects, you need to work with actual date objects, which we do in the following example, which creates a simple data frame table of month, day and day of year, then displays a subset that illustrates the utility:

library(lubridate)
library(tidyverse)
Jan1 <- ymd("2022-01-01")
Dec31 <- ymd("2022-12-31")
year2022 <- as_date(Jan1:Dec31)
ydayTable <- bind_cols(month=month(year2022),day=day(year2022),doy=yday(year2022))
ydayTable[27:35,]

## # A tibble: 9 × 3
##   month   day   doy
##   <dbl> <int> <dbl>
## 1     1    27    27
## 2     1    28    28
## 3     1    29    29
## 4     1    30    30
## 5     1    31    31
## 6     2     1    32
## 7     2     2    33
## 8     2     3    34
## 9     2     4    35

3.7 Extending abstractions with functions

We’ve explored ways of extending base R to facilitate data abstraction, mostly focusing on the tidyverse in this chapter. There are many other packages that provide special capabilities, but we can also write our own, as we explored in the last chapter. Even simple functions of a few lines of code can make our main code more readable.

We’ll look at a simple extension of the readxl::read_excel function. A common practice in data logger spreadsheets is to have the variable names in the first row and the units (like °C or mm in the second row.) There’s no good way to make use of those units in a data frame, so we typically will want to skip it. The read_excel function has a skip parameter that lets us skip the first row, and also the col_names parameter that allows us to either use the first row as variable names (the default, TRUE), get default names (FALSE), or provide a vector of variable names.

We’ll write a function that creates variable names out of the first two rows, so includes the units with the original variable name by vectorizing a paste0.

library(igisci); library(tidyverse); library(readxl)

read_excel2 <- function(xl){
  vnames <- read_excel(xl, n_max=0) |> names()
  vunits <- read_excel(xl, n_max=0, skip=1) |> names()
  read_excel(xl, skip=2, col_names=paste0(vnames,"_",vunits))
}

Then this is a new function we can just use in our code later on. We’ll be reading this kind of data a lot, so even a short function can make our main code easier to read.

read_excel2(ex("meadows/LoneyMeadow_30minCO2fluxes_Geog604.xls"))

## # A tibble: 5,397 × 14
##    `Decimal time_mgC m-2 s-1` `Day of Year_mmols m-2 s-1` `Hour of Day_W m-2`
##                         <dbl>                       <dbl>               <dbl>
##  1                       138                          138                24  
##  2                       138.                         138                 0.5
##  3                       138.                         138                 1  
##  4                       138.                         138                 1.5
##  5                       138.                         138                 2  
##  6                       138.                         138                 2.5
##  7                       138.                         138                 3  
##  8                       138.                         138                 3.5
##  9                       138.                         138                 4  
## 10                       138.                         138                 4.5
## # ℹ 5,387 more rows
## # ℹ 11 more variables: `CO2 Flux_degC` <dbl>, `PAR_%` <dbl>, Qnet_degC <dbl>,
## #   Tair_degC <dbl>, RH_deg <dbl>, `Tsoil 2cm_m s-1` <dbl>,
## #   `Tsoil 10cm_mm` <dbl>, `wdir_m3 m-3` <dbl>, `wspd_mgC m-2 s-1` <dbl>,
## #   `Rain_mmols m-2 s-1` <dbl>, `VWC_W m-2` <dbl>

Note that this function accessed the igisci::ex function we’re using a lot. That function simply returns the path to the file, and that’s all we need as input to the read_excel function. It’s not always a good idea to write functions because we either need to make sure we run it, or include it in a package that we need to load. This example seemed worth it. There are many other functions you could build to do operations you need repeatedly, and some of them may get pretty complex, but you’ll want to decide which to build.

3.8 Exercises: Data Abstraction

Create an R Markdown document (.Rmd) to complete the exercise. For each question, start by copying in the question text below, then create two code chunks. The first will be used to set up libraries, read and produce data, and try methods. The code in the first chunk will probably have outputs in the form of data frame tables and graphics, but the final outputs are reserved for the second code chunk. When the .Rmd file is knitted, it should show the code for both chunks, but only the outputs from the second, which should produce a clean knitted result.

To do this use the following options for the first chunk: {r results="hide", warning=F, message=F, echo=T}. You can run the first code chunk to learn from what it produces, but only the second chunk (without those options) will produce your final results when you knit the document.

For a simple explanation of code chunk options, see https://yihui.org/knitr/options/.

There may also be questions to answer, which should go in the markdown section, just below the question.

Exercise 3.1 Create a tibble with 20 rows of two variables norm and unif with norm created with rnorm() and unif created with runif().

Consider how you might use similar methods to 3.2.2, and explain whether this might (or might not) be a good approach.

:

Code chunk output: your resulting tibble:

Exercise 3.2 Read in “TRI/TRI_2017_CA.csv” in two ways, as a normal data frame assigned to df and as a tibble assigned to tbl. Do not create an output from the code chunk; instead use the Environment tab to answer the question.

What field names result in the data frame and in the tibble for what’s listed in the CSV as 5.1_FUGITIVE_AIR? > df: tibble:

Exercise 3.3 Use the summary function or other methods to investigate the variables in either the data.frame or tibble you just created. While you can temporarily create outputs from your code chunk to investigate, after you have your the information needed to answer the question, comment out the code producing output: don’t display any output; just answer the question below.

What type of field and what values are assigned to BIA_CODE? > :

Exercise 3.4 Create a boxplot of body_mass_g by species from the penguins data frame in the palmerpenguins package (Horst, Hill, and Gorman 2020), using methods in 3.3.1.

Exercise 3.5 Use select, mutate, and the pipe to create a penguinMass tibble where the only original variable retained is species, but with body_mass_kg created as $\frac{1}{1000}$ the body_mass_g. The statement should start with penguinMass <- penguins and use a pipe plus the other functions after that.

The output should be the tibble created.

Exercise 3.6 Now, also with penguins, create FemaleChinstaps to include only the female Chinstrap penguins. Start with FemaleChinstraps <- penguins %>%.

Similarly, the output should be the resulting FemaleChinstaps.

Exercise 3.7 Now, summarize by species groups to create mean and standard deviation variables from bill_length_mm, bill_depth_mm, flipper_length_mm, and body_mass_g. Preface the variable names with either avg. or sd. Include na.rm=T with all statistics function calls.

Similarly the output should be the resulting data frame.

Exercise 3.8 Sort the penguins by body_mass_g, and provide that output.

Exercise 3.9 Using stringr methods, detect and print out a vector of sites in the ConcentrationReport that are parks (include the substring "Park" in the variable Shoreline Name), then detect and print out the longest shoreline name (make that the only output in the end). See the method in 3.2 for identifying just a single result among duplicates.

Exercise 3.10 The sierraStations data frame has weather stations only in California, but includes “CA US” at the end of the name. Use dplyr and stringr methods to create a new STATION_NAME that truncates the name, removing that part at the end, and provide the resulting data frame as the only output.

Exercise 3.11 Use dplyr and lubridate methods to create a new date-type variable sampleDate from the existing DATE character variable in soilCO2_97, then use this to plot a graph of soil CO2 over time, with sampleDate as x, and CO2pct as y. Then use code to answer the question: What’s the length of time (time difference) from beginning to end for this data set [Hint: two approaches might use sorting or max() and min()]?

References

Horst, Allison Marie, Alison Presmanes Hill, and Kristen B Gorman. 2020. Palmerpenguins: Palmer Archipelago (Antarctica) Penguin Data. https://allisonhorst.github.io/palmerpenguins/.

Marine Debris Program. n.d. NOAA Office of Response; Restoration. https://marinedebris.noaa.gov/.

Thompson, A, JD Davis, and AJ Oliphant. 2016. “Surface Runoff and Soil Erosion Under Eucalyptus and Oak Canopy.” Earth Surface Processes and Landforms. https://doi.org/10.1002/esp.3881.

Wickham, Hadley, and Garrett Grolemund. 2016. R for Data Science: Visualize, Model, Transform, Tidy, and Import Data. O’Reilly Media, Inc. https://www.tidyverse.org/learn/.