Data Processing
We downloaded 7742 observations for Galax urceolata in the example above. Of these observations, only those with locality information will be helpful when investigating this species distribution.
Identify Records Missing Locality Information
Locality information can be redacted or skewed due to protect threatened taxa, often locality information will be provided upon request or can be identified through georeferencing. We created functions to aid in this process.
Redacted Records
Locality information can be redacted or skewed due to protect threatened taxa; often locality information will be provided to aid research upon request.
To find data that needs to be manually received by an institution via a permit (or removed from the data set), use needed_records(). After receiving the data from herbaria, manually merge the obtained records with your original data set.
Example:
redacted_info <- needed_records(galaxdf)Records to Georeference
Some records may be missing latitude and longitude values, however locality information can be used to assign coordinates to the record through georeferencing.
To find data lacking coordinates but containing locality information, use need_to_georeference(). You should georeference these records and then manually merge the obtain records with your original data set.
Example:
to_georeference <- need_to_georeference(galaxdf)Merging Retained Records
After receiving the data from herbaria or through georeferencing, you will want to merge the obtained records with your original data set.
Note, the following steps will only work if columns are equivalent between the two data sets you hope to merge. If the columns are not equivalent, we recommend bcd::bcd_standardize_datasets()
Example:
# Avoid duplicates by removing the records you are about to merge
dfsub <- remove_missing(galaxdf)
# Use gator_merge to merge the main data set with retrieved records
dfnew <- gators_merge(dfsub, retrieved_records)
# Version control - save a copy prior to any additional processing!!
write.csv(dfnew, "data/merged_data_Galax_urceolata_YYYYMMDD.csv") ## Version control!
# Set gatordf equal to the newly merged data set
gatordf <- dfnewOccurrence Data Cleaning
Here we walk through each cleaning function, however we also created a simple one-step option full_clean(), see below.
Resolve Taxon Names
To find data containing scientific names corresponding to your desired species, use taxa_clean(). Use your downloaded data from the first step as input, as well as a synonyms list, the accepted name, and the filter option (exact, fuzzy, or interactive).
Example:
galaxdf <- taxa_clean(df = galaxdf,
synonyms.list = c("Galax urceolata", "Galax aphylla"),
taxa.filter = "fuzzy",
accepted.name = "Galax urceolata") # creates a new column with accepted name for easy comparisonClean Locality
Basic Locality Clean
Here we remove any records with missing coordinates, impossible coordinates, coordinates at (0,0), and any that are flagged as skewed. The skewed records can be identified with the remove_skewed() function and row value for the ‘InformationWitheld’ column. We also provide the option to round the provided latitude and longitude values to a specified number of decimal places.
galaxdf <- basic_locality_clean(df = galaxdf,
remove.zero = TRUE, # Records at (0,0) are removed
precision = TRUE, # latitude and longitude are rounded
digits = 2, # round to 2 decimal places
remove.skewed = TRUE)Find and Remove Flagged Points
To find records that may have problematic coordinates, use process_flagged(). This function can either automate the process of finding and removing problematic points (interactive = FALSE) or allow for manual inspection. The latter will let you manually remove points deemed improper by viewing the points on a graph.
This function utilizes CoordinateCleaner::clean_coordinates().
Example:
galaxdf <- process_flagged(galaxdf, interactive = TRUE)
Figure 1: Interactive map that appears when process_flagged() is run.
Figure 2: Interactive map that appears after we remove points interactively with process_flagged(). Points 1, 3, 5, 6, and 7 were removed.
Remove Duplicate Records
Here we identify and remove both (1) specimen duplicates and (2) aggregator duplicates based on each specimens coordinates, occurrenceID, and eventDate. To leverage all date information available, set remove.unparseable = FALSE to manually populate the year, month, and day columns. Here, we also confirm all ID (UUID and key) are unique to remove any within-aggregator duplicates that may accumulate due to processing errors.
Example:
galaxdf <- remove_duplicates(galaxdf, remove.unparseable = TRUE)Remove Particular Record Bases
Sometimes, certain bases of records may want to be removed from the data set. To do this, we provide basis_clean(). This function can be used interactively by simply supplying a df, but no basis.list value.If a list of basis is provided, the function will select only records where the basisOfRecord value fuzzy match values in the list provided.
Example:
galaxdf <- basis_clean(galaxdf,
basis.list = c("HUMAN_OBSERVATION", "PRESERVED_SPECIMEN",
"MATERIAL_SAMPLE", "LIVING_SPECIMEN",
"PreservedSpecimen", "Preserved Specimen"))Spatial Correction
The last processing step is spatial correction. Collection efforts can lead to the clustering of points and filtering can help reduce this clustering. Here we provide functions to reduce the effects of sampling bias using randomization approach and retain only one point per pixel.
One Point Per Pixel
Maxent will only retain one point per pixel. To make the ecological niche analysis comparable, we will retain only one point per pixel.
Example:
galaxdf <- one_point_per_pixel(galaxdf,
resolution = 0.5, # for 30 arc sec raster layers
precision = TRUE,
digits = 2) Spatial thining
We thin points by utilizing spThin::thin(). This step reduces the effects of sampling bias using a randomization approach.
Step 1: What should your minimum distance be?
Here we first calculate minimum nearest neighbor distance in km:
Example:
library(fields)
nnDm <- rdist.earth(as.matrix(data.frame(lon = df$longitude, lat = df$latitude)), miles = FALSE, R = NULL)
nnDmin <- do.call(rbind, lapply(1:5, function(i) sort(nnDm[,i])[2]))
min(nnDmin)Here the current minimum distance is 2.22 km. Based on literature, we find a 2 meters (or 0.002 km) distance was enough to collect unique genets, so we do not need to thin our points.
Step 2: Thin occurrence records using spThin through gatoRs.
When you do need to thin your records, here is a great function to do so!
Example:
df <- thin_points(df,
distance = 0.002, # in km
reps = 100)Full Clean
Finally, instead of step-by-step cleaning, we created a single function to streamline this process.
Example:
## Read raw csv
rawdf <- read.csv("base_folder/my_file.csv")
## Set your full clean preferences equal to the above
df_quick_clean <- full_clean(rawdf,
synonyms.list = c("Galax urceolata", "Galax aphylla"),
remove.NA.occ.id = FALSE,
remove.NA.date = FALSE,
accepted.name = "Galax urceolata",
remove.zero = TRUE,
precision = TRUE,
digits = 2,
remove.skewed = TRUE,
basis.list = c("HUMAN_OBSERVATION", "PRESERVED_SPECIMEN",
"MATERIAL_SAMPLE", "LIVING_SPECIMEN",
"PreservedSpecimen", "Preserved Specimen"),
remove.flagged = TRUE,
thin.points = TRUE,
distance = 0.002,
reps = 100,
one.point.per.pixel = TRUE)