Cleaning and Cropping Quadrat Data from CoralNet • quadcleanR

Introduction

This vignette will walk you through cleaning up quadrat data from CoralNet to produce easy to analyze data frames.

A little about the data:

The data being cleaned will be the softcoral_SQuads and this vignette will build off of the cropping vignette as well. Data was collected by the Baum Lab and Kiritimati Field Teams and is the uncleaned version of data found in Maucieri and Baum 2021. Biological Conservation. The softcoral_SQuads data are from photo quadrats (0.9m by 0.6m) which were randomly annotated with 54 random points each, while the softcoral_annotations data are from photo quadrats (1m by 1m) which were randomly annotated with 100 random points each. At each of these annotated points, the substrate was identified. Photo quadrats were collected on Kiritimati Island in the Republic of Kiribati and document coral cover over time and space. The annotations and output of the data frame were produced using CoralNet and all annotations were done manually, by trained researchers.

So this vignette will follow the following steps:

Take the small quadrats (softcoral_SQuads) and show how to clean the data
Then crop the large quadrat data (softcoral_annotations) and clean it too
Join the two together to produce a large, easy to use data frame for future analyses, with some further cleaning of this joint data set.

First lets load the package, the dplyr package, tidyr package and the data, plus a few extras used to create this vignette.

library(quadcleanR)
library(dplyr)
library(tidyr)
library(shiny)
library(knitr)
library(kableExtra)


data("softcoral_SQuads")

Now let me point out some unique aspects of this data:

tail(softcoral_SQuads)

	Image.ID	Image.name	Annotation.status	Points	Trans	Unidentified
2517	1403339	KI2011_site9_Q5.jpg	Confirmed	54	5.556	94.444
2518	1403340	KI2011_site9_Q6.jpg	Confirmed	54	5.556	94.444
2519	1403341	KI2011_site9_Q7.jpg	Confirmed	54	1.852	98.148
2520	1403342	KI2011_site9_Q8.jpg	Confirmed	54	0.000	100.000
2521	1403343	KI2011_site9_Q9.jpg	Confirmed	54	0.000	100.000
2522	ALL IMAGES			NA	0.000	100.000

This data as an Image.ID column which was arbitrarily added to this data set, so we are going to remove that as it holds no scientific value. There is also a final row which sums all quadrats but since we will be removing quadrats and points to clean these data up, we will remove that final row as well. The Image.name column is the unique ID for each photo quadrat, but it is very messy and not easy to use, so we will make this into new columns and add more information. Annotation.status is a column from CoralNet which tells if the annotations in each photo quadrat have been confirmed by human researchers or are only based on AI. The Points column tells us how many randomly annotated points there are in each quadrat, and since they are all 54, we know these data are from the smaller quadrats. The rest of the columns are the different coral and substrate IDs and how many points where annotated for each tag in each photo.

So first, lets remove unneeded columns and make sure we are only working with “Confirmed” annotations.

SQuad_confirmed <- softcoral_SQuads %>% filter(Annotation.status == "Confirmed") %>% select(-c(Image.ID, Points, Annotation.status))

Now we will separate the Image.name column into more descriptive columns.

SQuad_separated <- separate(SQuad_confirmed, Image.name, sep="_", into=c("Field.Season", "Site","Quadrat"))

## Warning: Expected 3 pieces. Additional pieces discarded in 3 rows [275, 446,
## 477].

But if you notice, there are still .jpg and .jpeg in the quadrat names, so lets remove those, and change the naming of siteT19 to site40.

SQuad_nojpg <- rm_chr(SQuad_separated, c(".jpg", ".jpeg"))
SQuad_site40 <- change_values(SQuad_nojpg, "Site", "siteT19", "site40")

Now lets look at the levels of some of my columns.

unique(SQuad_site40$Field.Season)

## [1] "KI2007" "KI2009" "KI2010" "KI2011"

unique(SQuad_site40$Site)

##  [1] "site10" "site11" "site12" "site13" "site14" "site15" "site16" "site17"
##  [9] "site18" "site19" "site1"  "site20" "site21" "site22" "site23" "site24"
## [17] "site25" "site26" "site27" "site28" "site29" "site2"  "site30" "site31"
## [25] "site32" "site33" "site34" "site35" "site36" "site37" "site3"  "site4" 
## [33] "site5"  "site6"  "site7"  "site8"  "site9"  "site38" "site39" "site40"

Sometimes you may not want every year and every site of the data, so you may want to subset some of them out. For this example I don’t want to subset any of these out, but I will demonstrate how to subset out based on values later, when we get to the large quadrats.

Now to prep for the addition of the large quadrats, I want to update the column names for this data frame. The column names are currently set as the tag shorthand used during the annotation process, but now I want them to better reflect the actual substrate names.

data("coral_labelset")
head(coral_labelset)

short_name	full_name	taxonomic_name	functional_group	life_history
Rubble	Broken_coral_rubble	Broken_coral_rubble	Abiotic_Substrate	Not_Coral
SCRO	Consolidated_hard_rock	Consolidated_hard_rock	Abiotic_Substrate	Not_Coral
Sand	Sand	Sand	Abiotic_Substrate	Not_Coral
Sediment	Sediment	Sediment	Abiotic_Substrate	Not_Coral
CCA	Crustose_coralline_algae	Crustose_coralline_algae	Crustose_Algae	Not_Coral
Peysson	Peyssonnelia	Peyssonnelia	Crustose_Algae	Not_Coral

This is what my label set document looks like, but you could also make this in R by joining a series of vectors. Now lets fix the column names.

SQuad_colnames <- change_names(SQuad_site40, coral_labelset, "short_name", "full_name")
names(SQuad_colnames)[1:16]

##  [1] "Field.Season"                  "Site"                         
##  [3] "Quadrat"                       "Acropora_corymbose"           
##  [5] "Acropora_digitate"             "Acropora_arborescent"         
##  [7] "Acropora"                      "Acropora_tabulate"            
##  [9] "Astreopora"                    "Bleached_Acropora_arborescent"
## [11] "Bleached_Acropora"             "Bleached_Astreopora"          
## [13] "Bleached_Acropora_tabulate"    "Bleached_Coscinarea"          
## [15] "Bleached_Echinophyllia"        "Bleached_Favites_halicora"

Much better.

Now because these are smaller quadrats with fewer annotated points, before we add in the large quadrat data, lets deal with the substrate that was not able to be accurately identified. For these photo quadrats, the Shadow, Transect_hardware and Unclear tags need to be removed and not used when we calculate percent cover. If I was going to use this data to do a diversity analysis with hard corals, I would also include unknown_hard_coral, and Bleached_unknown_hard_coral to this list, but as this data will be used for soft coral analyses, we will leave it.

SQuad_colnames <-  mutate_at(SQuad_colnames, c(4:134), as.numeric)

SQuad_usable <- usable_obs(SQuad_colnames, c("Shadow", "Transect_hardware", "Unclear"), 
                            max = TRUE, cutoff = 10)

SQuad_removed <- usable_obs(SQuad_colnames, c("Shadow", "Transect_hardware", "Unclear"),
                             max = TRUE, cutoff = 10, above_cutoff = TRUE)

By identifying how many usable points there are in each quadrat, and removing any quadrats that had over 10% of the identified points unusable, we have removed 0 quadrats from analysis, which you could view with the SQuad_removed data frame but as we have no quadrats to removed, it will be an empty data frame.

Now we can start working on the large quadrat data.

data("softcoral_annotations")
data("softcoral_LQuads")

I will be cropping this data as outlined in the cropping vignette, so check that out for the details, but before I crop it, I want to remove any unconfirmed quadrats just as we did for the small quads.

LQuad_confirmed <- softcoral_LQuads %>% filter(Annotation.status == "Confirmed") 

LQuad_sub <- softcoral_annotations %>% filter(Name %in% unique(LQuad_confirmed$Image.name))

Then we will crop the annotations data.

LQuad_cropped <- crop_area(data = LQuad_sub, row = "Row",
                             column = "Column", id = "Name", dim = c(0.9, 0.6), 
                             obs_range = c(36,64))

Now I will just format this to look like the small quad data so we can join the two.

LQuad_wide <- LQuad_cropped %>% select(-c(obs, Row, Column)) %>% group_by(Name) %>% pivot_wider(names_from = Label, values_from = Label, values_fn = length, values_fill = 0)
LQuad_wide <- as.data.frame(LQuad_wide)

LQuad_separated <- separate(LQuad_wide, Name, sep="_", into=c("Field.Season", "Site","Quadrat"))

LQuad_nojpg <- rm_chr(LQuad_separated, c(".jpg", ".jpeg"))

LQuad_colnames <- change_names(LQuad_nojpg, coral_labelset, "short_name", "full_name")


LQuad_colnames <-  mutate_at(LQuad_colnames, c(4:11), as.numeric)

LQuad_usable <- usable_obs(LQuad_colnames, c("Shadow", "Transect_hardware", "Unclear"),
                             max = TRUE, cutoff = 0.1*54)

LQuad_removed <- usable_obs(LQuad_colnames, c("Shadow", "Transect_hardware", "Unclear"),
                             max = TRUE, cutoff = 0.1*54, above_cutoff = TRUE)

With the large quadrats, only 4 photos needed to be removed because they had too many unusable points.

Alright, now we can join these data frames together, and continue with one large data set. As both data sets used the same label set of names, This is easy with the bind_rows() function out of the dplyr package

AllQuads <- bind_rows(SQuad_usable, LQuad_usable) %>% select(-c(unusable))

AllQuads[, 4:131][is.na(AllQuads[, 4:131])] <- 0

Now lets continue with the cleaning of these data.

There are a few more things that need to be removed including MPQs (Mega photo quadrats), site8.5 and DEEP photo quadrats.

AllQuad_noDEEP_site8.5 <- keep_rm(AllQuads, c("DEEP", "site8.5"), select = "row", exact = FALSE, colname = "Site", keep = FALSE)

AllQuad_noMPQ <- keep_rm(AllQuad_noDEEP_site8.5, c("MPQ"), select = "row", exact = FALSE, colname = "Quadrat", keep = FALSE)

Now we have removed unusable annotations, so lets rescale and recalculate cover with proportion cover.

AllQuad_cover <- cover_calc(AllQuad_noMPQ, names(AllQuad_noMPQ[,4:131]), prop = TRUE)

This data frame is now nicely formatted and could be used for many community based analyses. This might be a great stopping point for some analyses, but to further clean this up I am going to convert this into long format data.

AllQuad_long <- AllQuad_cover %>% pivot_longer(cols = names(AllQuad_cover[,4:131]), names_to = "Tag_Name", values_to = "prop_cover")

One thing you may notice by looking at the Tag_Name column, is that these species names are not unique species, but there are duplicates of the same species, categorized into bleaching and non bleaching forms. For any kind of diversity analysis, this would inflate the number of different species, so it is important to combine different forms of the same species if diversity analyses are being done.

For this clean up, we will walk through 3 ways of dealing with this based on what you want to accomplish.

Option A. Categorizing rows.

If you want to use your data in this long format, want to just categorize everything and you will use these various categories based on your different research questions, you could just add a bunch of category columns like so:

A_AllQuad_Bleach <- categorize(AllQuad_long, "Tag_Name", values = c("Bleach"), name = "Bleached", binary = TRUE, exact = FALSE)

This categorizes each Tag_Name to whether it is a bleaching or nonbleaching tag.

And you could also add other information in if you have it, like taxonomy.

A_AllQuad_Taxa <- categorize(A_AllQuad_Bleach, "Tag_Name", values = coral_labelset$full_name, name = "Taxonomic_Name", binary = FALSE, categories = coral_labelset$taxonomic_name)

Option B. Categorizing rows and then combining.

Now after you categorize your rows, perhaps you want to have all the cover values summed at a different level, like at the taxonomy level. To do this, the summarise() function from dplyr will work great.

B_AllQuad_taxonomy <- A_AllQuad_Taxa %>% group_by(Field.Season, Site, Quadrat, Taxonomic_Name) %>% summarise(prop_cover = sum(prop_cover))

## `summarise()` has grouped output by 'Field.Season', 'Site', 'Quadrat'. You can
## override using the `.groups` argument.

Option C. Wide format summing columns

If you wanted to keep the data in a wide format, and sum columns based on taxonomy, to allow for community level analyses, you could also use the sum_cols() function. To do this, we first need a vector of what to change the names too, which can be done with a simple match, unless you have a vector with the new names already in the right order.

current_names <- colnames(AllQuad_cover[,4:131])
new_names <- coral_labelset[match(current_names, coral_labelset$full_name),]$taxonomic_name

AllQuad_wide_summed <- sum_cols(AllQuad_cover, from = current_names, to = new_names)

Whichever of the options you choose, you will be able to customize the data to your analysis needs. After that your data is nearly cleaned. Some other things you may want to add would be environmental data, or more taxonomic data. The add_data() function can help with adding multiple columns from a data set at a time.

B_AllQuad_LH_FG <- add_data(B_AllQuad_taxonomy, coral_labelset, cols = c("functional_group", "life_history"), data_id = "Taxonomic_Name", add_id = "taxonomic_name", number = 5)

data("environmental_data")

B_AllQuad_enviro <- add_data(B_AllQuad_LH_FG, environmental_data, cols = c("HD_Cat", "HD_Cont", "NPP",  "WE", "Region", "WaveEnergy"), data_id = "Site", add_id = "Site", number = 4)

The final things I will add to this data to get it in shape for analysis is a final categorization of the study years based on the timing of the 2015/2016 El Niño, subset the species to only soft coral, and refactor the levels of some of the variables.

B_AllQuad_timeblock <- categorize(B_AllQuad_enviro, column = "Field.Season", values = unique(B_AllQuad_enviro$Field.Season), name = "TimeBlock", binary = FALSE, exact = TRUE, categories = c(rep("Before", times = 8), rep("During", times = 3), rep("After", times = 4)))

AllQuads_cropclean <- keep_rm(B_AllQuad_timeblock, values = "Soft_coral", select = "row", colname = "functional_group")

AllQuads_cropclean$TimeBlock <- factor(AllQuads_cropclean$TimeBlock, levels = c("Before", "During", "After"))
AllQuads_cropclean$Site <- factor(AllQuads_cropclean$Site, levels = paste("site", seq(1:40), sep = ""))
AllQuads_cropclean$HD_Cat <- factor(AllQuads_cropclean$HD_Cat, levels = c("Very Low", "Low", "Medium", "High", "Very High"))

Exploration

This data has now been sufficiently cleaned and can be used for many different analyses. Often once data has been cleaned, the first step is to start exploring the data. One thing we can look at is the sample sizes, to see how many quadrats I have over the different sites and years.

sample_size(AllQuads_cropclean, dim_1 = "Site", dim_2 = "Field.Season", count = "Quadrat")

	KI2007	KI2009	KI2010	KI2011	KI2013	KI2014	KI2015a	KI2015b	KI2015c	KI2015d	KI2016a	KI2016b	KI2017	KI2018	KI2019
site1	22	30	0	0	28	0	0	0	0	0	0	0	29	0	0
site10	20	21	32	0	0	0	0	0	30	0	0	0	25	0	0
site11	29	0	0	0	0	0	0	0	0	0	0	0	0	0	0
site12	30	25	0	0	0	0	0	0	0	0	22	30	31	0	30
site13	30	0	0	0	0	0	0	0	0	0	0	0	23	0	0
site14	29	22	22	25	13	21	0	0	30	0	16	0	30	0	0
site15	30	23	26	26	24	0	29	15	29	0	29	30	30	30	0
site16	28	26	0	0	0	0	0	0	0	0	0	0	0	0	0
site17	29	20	0	19	0	0	0	0	0	0	0	0	0	0	0
site18	29	28	0	24	0	0	0	0	0	0	0	0	20	0	0
site19	30	28	26	21	21	0	0	0	28	0	0	0	26	0	0
site2	25	29	0	0	0	0	0	0	0	0	0	0	0	0	0
site20	28	0	0	0	31	0	0	0	0	0	0	0	19	0	0
site21	30	0	0	0	0	0	0	0	0	0	0	0	0	0	0
site22	30	20	0	0	19	0	0	0	0	0	0	0	0	0	0
site23	30	28	0	0	25	0	0	0	0	0	0	0	29	0	0
site24	30	35	0	22	31	0	0	0	0	0	0	0	0	0	0
site25	30	22	21	24	26	25	0	0	31	0	0	0	17	27	0
site26	30	27	22	25	30	0	0	0	0	0	0	0	30	0	0
site27	30	26	18	24	28	30	30	30	29	30	30	30	29	29	30
site28	21	0	0	0	0	0	0	0	0	0	0	0	0	0	0
site29	21	0	0	0	0	0	0	0	0	0	0	0	0	0	0
site3	28	20	30	25	28	32	0	0	30	0	0	0	29	28	30
site30	30	28	23	0	29	25	30	30	30	0	30	30	30	21	30
site31	20	27	21	26	0	0	0	0	0	0	0	0	20	0	30
site32	10	18	18	24	30	0	0	30	30	34	30	30	30	30	30
site33	21	27	0	27	0	0	0	0	0	0	30	0	31	0	0
site34	22	27	23	19	28	31	0	30	29	0	30	30	29	27	30
site35	21	20	44	25	26	31	30	30	29	9	26	30	29	30	30
site36	21	22	0	0	0	0	0	0	0	0	0	0	27	0	0
site37	22	27	0	0	0	0	0	0	0	0	30	30	30	30	30
site4	30	0	0	0	0	0	0	0	0	0	0	0	0	0	0
site5	29	0	0	0	0	0	0	29	30	0	30	30	26	29	30
site6	30	19	0	27	29	0	0	0	0	0	0	30	30	0	30
site7	30	20	0	0	0	0	0	0	0	0	0	0	25	0	0
site8	30	0	0	0	29	27	30	30	30	27	30	30	30	30	29
site9	35	28	13	27	25	0	0	0	0	0	30	0	30	0	0
site38	0	25	0	21	0	31	0	0	30	0	29	0	0	28	30
site39	0	22	0	0	0	0	0	0	0	0	0	0	0	0	0
site40	0	21	0	0	0	0	0	0	0	0	23	30	31	0	30

Visualizing the data can be easy with a built in shiny app function. To see an example shiny app you can go here but to visualize the data we have cleaned up here, we can run the following code.

A good combination to examine this shiny with is : - y-axis: prop_cover - x-axis: Field.Season - color: TimeBlock (treat as discrete) - facet: HD_cat - group by: Field.Season, TimeBlock, Site and HD_Cat - view as a box plot

visualize_app(data = AllQuads_cropclean, xaxis = colnames(AllQuads_cropclean[,1:13]), yaxis = "prop_cover")