In jniedballa/camtrapR: Camera Trap Data Management and Preparation of Occupancy and Spatial Capture-Recapture Analyses
exiftool_present <- Sys.which("exiftool") != ""
print("WARNING: THIS VIGNETTE WAS CREATED WITHOUT EXIFTOOL. OUTPUT IS INCOMPLETE SINCE ESSENTIAL FUNCTIONS DID NOT RUN!")
library(camtrapR)
library(secr)
Tabulating species and individual records: the recordTable functions
There are 2 function to tabulate species records after identification of species from images and videos.
recordTable tabulates records of all images after species-level identification and recordTableIndividual tabulates individuals of one species.
Nevertheless, the underlying idea is the same. For each image, the date and time it was taken are read from the image's Exif metadata using ExifTool. Species or individual ID are read from the directory structure or image metadata (see vignette "Species and Individual Identification"). Video data are extracted analogously.
recordTable: tabulating species records
recordTable is typically run after identifying species from images. It reads species IDs from the directory structure the images are placed in or from image metadata tags.
First we define the directory containing our renamed, identified images
# find the directory with sample images contained in the package
wd_images_ID <- system.file("pictures/sample_images_species_dir", package = "camtrapR", lib.loc = .libPaths())
and see how many JPG images we have (this is not necessary, but informative here).
length(list.files(wd_images_ID, pattern = "JPG", recursive = TRUE))
Now we can run recordTable. Here is a minimal example:
rec.db.species0 <- recordTable(inDir = wd_images_ID,
IDfrom = "directory")
head(rec.db.species0)
Argument IDfrom tells the function to look for species directories within the station directories and to take species IDs from these. IDfrom must be set to "metadata" if metadata tagging was used for species identification.
By default, the function returns all records. Nevertheless, we get a data frame containing r ifelse(exiftool_present, nrow(rec.db.species0), 56) records, less than the number of images in the image directory (r length(list.files(wd_images_ID, pattern = "JPG", recursive = TRUE))). This is because a number of images were taken at the same time (measured with 1 minute precision by these cameras) and the function removes duplicate records.
It may suffice to illustrate this with the Pig-tailed macaque images from stationB:
list.files(file.path(wd_images_ID, "StationB", "MNE"))
There are, amongst others, 3 images taken at 17:48:00 on the 28th of April 2009. Of these, only 1 will be returned.
Here is what the columns of the record table contain:
| column | content |
|-------|-------------------------------------------------------------|
| Station | the station the image is from |
| Species | species name |
| DateTimeOriginal | Date and time of record in R-readable format |
| Date | record date |
| Time | record time of day |
| delta.time.secs | time difference between record and last (independent) record of same species at same station / camera (in seconds) |
| delta.time.mins | time difference between record and last (independent) record of same species at same station / camera (in minutes) |
| delta.time.hours | time difference between record and last (independent) record of same species at same station / camera (in hours) |
| delta.time.days | time difference between record and last (independent) record of same species at same station / camera (in days) |
| Directory | directory the image is in |
| FileName | image file name |
*see below: Independence between cameras within stations
Temporal independence between records
Imagine a species that loves to hang out in front of your cameras. You will end up with hundreds of shots of the same species, maybe even the same animal. Therefore, images can be filtered using an adjustable criterion for temporal independence between subsequent records of the same species in an attempt to remove non-independent records. This is achieved via argument minDeltaTime. It is the minimum time difference (in minutes) between two records of the same species at the same station which are to be considered independent. The default is 0, causing the function to return all records. Setting it to a higher number, e.g. 60 (i.e., 1 hour), is commonly done to thin the number of records. Note that you will not lose records of different species, even if they fall within the specified time interval from a record.
The argument deltaTimeComparedTo further controls how independence between records is assessed. Setting it to "lastRecord" returns only records taken minDeltaTime minutes after the last record. Setting it to "lastIndependentRecord" returns only records taken minDeltaTime minutes after the last independent record.
Let's now apply an 1-hour independence criterion and define a time zone.
rec.db.species60 <- recordTable(inDir = wd_images_ID,
IDfrom = "directory",
minDeltaTime = 60,
deltaTimeComparedTo = "lastRecord",
timeZone = "Asia/Kuala_Lumpur")
nrow(rec.db.species60)
Now r ifelse(exiftool_present, nrow(rec.db.species60), 40) records were returned instead of r ifelse(exiftool_present, nrow(rec.db.species0), 56). The missing records were taken less than 1 hour after the prior record of the same species at the same station and therefore omitted.
Time zones and daylight saving time
Note the warning about the missing time zone in the minimal example above. By default, camtrapR will assume UTC time zone. This should work well in most situations (even though the time zone may not be correct, strictly speaking), because UTC does not use daylight saving time (DST, aka 'summer time'), and camera traps normally don't use time zones (hence, camera traps will normally not respect DST of the area you work in and will not adjust image timestamps accordingly).
Now, if you work in an area that uses DST, and your camera does not know about DST, things can go wrong in various ways and you may end up with a systematic 1-hour offset. If you set your cameras up during DST (in summer) and set the internal clock accordingly, all records taken in winter (non-DST period) will be recorded as 1 hour later than according to actual clock time. If you set your cameras up during winter (not DST) and set the internal clock accordingly, all records taken in summer (DST period) will be recorded as 1 hour earlier than according to actual clock time.
In addition if you specify a time zone with DST and your cameras don't respect it, records may fall into the non-existing hour when clocks are advanced in spring, leading to an error in interpreting the date/time.
So, the question whether or not your cameras record the time zone becomes very important. Here is some recommendations on how to use the argument timeZone.
If your cameras don't save the time zone, and your time zone does not have DST, set argument timeZone to your study area time zone
If your cameras don't save the time zone, and your time zone has DST, leave argument timeZone at the default UTC.
If your cameras save the time zone and adjust image timestamps accordingly, set argument timeZone to your study area time zone (both if your area has DST or not).
In other words, it is advised to set argument timeZone to your study area's time zone (one of OlsonNames()), unless the time zone of your study area has DST, but your cameras don't record it.
Independence between cameras within stations
The issue of temporal independence between records becomes slightly more complex if more than one camera was operated at stations. That information can and should be included in the output of recordTable (and recordTableIndividual). Users can then decide whether temporal independence is to be assessed within stations or within cameras at each station. In the first case, argument camerasIndependent must be set to FALSE. 2 images taken at different 2 cameras at the same station within minDeltaTime minutes will be reported as 1 record in the record table (suitable if cameras were places in pairs). In the second case camerasIndependent must be set to TRUE and 2 images taken at different 2 cameras at the same station within minDeltaTime minutes will be reported as 2 record in the record table (suitable e.g., if cameras were located at some distance to one another and faced different trails).
The cameraID argument controls where camtrapR will look for camera IDs: in the file names (after renaming using imageRename, e.g. "renamedImages/StationA/StationACamera12015-12-3123-59-59(1).JPG") or in the directory structure (e.g. renamedImages/StationA/Camera1/StationACamera12015-12-3123-59-59(1).JPG"). If missing, it will be assumed there was only 1 camera per station.
Ignoring species
Argument exclude can be used to ignore certain species. This is useful for omitting images in directories like "team" or "unidentified". Here is an example:
# see what species we recorded
table(rec.db.species60$Species)
# remove "UNID" by setting argument exclude = "UNID"
rec.db.species60.exclude <- recordTable(inDir = wd_images_ID,
IDfrom = "directory",
minDeltaTime = 60,
deltaTimeComparedTo = "lastIndependentRecord",
timeZone = "Asia/Kuala_Lumpur",
exclude = "UNID")
# note that "UNID" is gone now
table(rec.db.species60.exclude$Species)
Extracting image metadata
recordTable and recordTableIndividual can both extract additional metadata from images (apart from date and time). For example, some camera models record ambient temperature or moon phase, which may be of interest. Metadata tags are stored in the images at the time they are taken and can be accessed and extracted if their tag names are known. Some tag names are standardised (e.g. "DateTimeOriginal") while others are manufacturer-specific. Therefore, function exifTagNames returns all Exif metadata it finds in a sample image. Users can then choose which to extract with recordTable and recordTableIndividual.
wd_images_ID <- system.file("pictures/sample_images_species_dir", package = "camtrapR")
exifTagNames(inDir = wd_images_ID)
The output table contains tag groups, tag descriptions, tag names and tag values. The values in tag_group help to unambiguously identify specific metadata tags.
Use the entries in tag_name (not tag_description) when providing values to the argument additionalMetadataTags in therecordTable functions.
Of the tags shown here, "DateTimeOriginal" contains the date and time that camtrapR reads out. Apart from that, there is little information of ecological interest in the example data. However, for demonstration purposes, let's extract information about the camera model and make. To be as precise as possible, we provide tags in the format: tag_group:tag_name (e.g. EXIF:Model)
rec.db.species.metadata1 <- recordTable(inDir = wd_images_ID,
IDfrom = "directory",
timeZone = "Asia/Kuala_Lumpur",
additionalMetadataTags = c("EXIF:Model", "EXIF:Make"))
head(rec.db.species.metadata1)
There are 2 additional columns containing the information from the metadata we requested.
recordTableIndividual: tabulating individuals of a species
Individual identification of species is a prerequiste for running (spatial) capture-recapture models. recordTableIndividual and the subsequent spatialDetectionHistory function prepare data for these models.
# find the directory with tagged sample images contained in the package
wd_images_individual_ID <- system.file("pictures/sample_images_indiv_tag/LeopardCat", package = "camtrapR")
# missing space in species = "LeopardCat" is because of CRAN package policies
rec.db.pbe <- recordTableIndividual(inDir = wd_images_individual_ID,
IDfrom = "metadata",
minDeltaTime = 60,
deltaTimeComparedTo = "lastIndependentRecord",
hasStationFolders = FALSE, # images are not in station directories
metadataIDTag = "individual", # the name of the metadata tag containing individual IDs
timeZone = "Asia/Kuala_Lumpur"
)
Extracting custom image metadata
In addition to the metadata that were saved when the image was taken (see above), custom metadata tags assigned in image management software can also be extracted. These may contain species ID tags if species were identified using metadata tags (instead of moving images into species directories), but also individual IDs, animal counts, sex of animals recorded, etc..
head(rec.db.pbe)
Camera Operation
The camera operation matrix is a day-by-station matrix that states how many cameras were active at a station on a given day. Rows represent stations and columns days, beginning with the day the first camera was set up and ending the day the last camera was retrieved.
It is calculated from the camera trap table mention in vignette 1 (see ?camtraps for an example). Users can give the setup / retrieval / problem columns in the input table as dates or date-times (since version 2.1.0).
Important: Prior to version 2.1.0, the values were integer numbers, with 1 indicating the camera was active on a day (or possibly higher values if multiple cameras at a station are combined).
From version 2.1.0, the values are the fraction of the day a camera was active. If only dates are provided in the camera trap table, the function will assume 12 noon as setup/retrieval time Problem periods will be assumed to last the entire day.
If imput is provided as date-times, output will be the exact fraction of day a station or camera was active (taking into account setup / retrieval time as well as time of Problem periods).
Dates as input
# first load the camera trap station table
data(camtraps)
dateFormat <- "dmy" # requires lubridate package
# alternatively, use "%d/%m/%Y" (from base R)
camop_problem <- cameraOperation(CTtable = camtraps,
stationCol = "Station",
setupCol = "Setup_date",
retrievalCol = "Retrieval_date",
writecsv = FALSE,
hasProblems = TRUE,
dateFormat = dateFormat
)
# as a reminder, these are the dates in our station information table
camtraps[,-which(colnames(camtraps) %in% c("utm_y", "utm_x"))]
# now let's have a look at the first few columns of the camera operation matrix
camop_problem[, 1:5]
# and the last few
camop_problem[, (ncol(camop_problem)-6):ncol(camop_problem)]
If stations were not set up, values are NA. Operational stations get value 1. If cameras were set up but malfunctioning, it is 0 (if hasProblems = TRUE).
Date-times as input
camtraps_hrs <- camtraps
# assign hours for setup and retrieval
camtraps_hrs$Setup_date <- paste(camtraps_hrs$Setup_date, c("12", "15", "18"))
camtraps_hrs$Retrieval_date <- paste(camtraps_hrs$Retrieval_date, c("18", "15", "12"))
# assign a random hours for begin of Problem at station 3
camtraps_hrs$Problem1_from[3] <- paste(camtraps_hrs$Problem1_from[3], "20")
# Problem ends with retrieval (i.e., malfunction until camera retrieved)
camtraps_hrs$Problem1_to[3] <- camtraps_hrs$Retrieval_date[3]
# create camera operation matrix
camop_hours <- cameraOperation(CTtable = camtraps_hrs,
stationCol = "Station",
setupCol = "Setup_date",
retrievalCol = "Retrieval_date",
hasProblems = TRUE,
dateFormat = "dmy H"
)
# as a reminder, these are the date-times in our station information table
camtraps_hrs
# now let's have a look at the first few columns of the camera operation matrix
camop_hours[, 1:5]
# and the last few
camop_hours[, (ncol(camop_hours)-6):ncol(camop_hours)]
Differences between version 2.1 and previous versions
Version 2.1 introduced support of cameraOperation() for date-times in the setup / retrieval / problem columns. By giving the setup / retrieval / problem columns as date-times (and adjusting argument dateFormat accordingly), one can calculate precise daily effort.
If the times are unknown and users provide dates only to argument dateFormat, camtrapR will assume that setup and retrieval were at 12 noon. Malfunction periods (indicated by the problem columns) will be interpreted as the cameras having malfunctioned the whole day.
This is how the handling of setup/retrieval days differs between dates and date-times in dateFormat:
| version | dateFormat is date | dateFormat is date-time |
|-|-|-|
| prior to 2.1 | 1 | not implemented |
| 2.1 and later | 0.5 | calculated precisely |
This is how the handling of problems / malfunction periods differs between dates and date-times in dateFormat:
| version | dateFormat is date | dateFormat is date-time |
|-|-|-|
| prior to 2.1 | 0 | not implemented |
| 2.1 and later | 0 | calculated precisely |
Camera days vs. camera nights
By default, the camera operation matrix refers to days, centered on noon and lasting from midnight to midnight the next day. In some situations it can make sense to shift the time frame for analyses, so the camera operation matrix describes trap nights (centered on midnight, lasting from noon to noon the next day).
To that end, the argument occasionStartTime can be set. Prior to version 2.1.0, it was an argument of detectionHistory and spatialDetectionHistory, but was moved to cameraOperation in v2.1.0. It can be set to an hour between 0 and 23, describing the hour the occasions begin.
# create camera operation matrix with occasions / trap days starting on noon (until noon the next day)
camop_hours_12 <- cameraOperation(CTtable = camtraps_hrs,
stationCol = "Station",
setupCol = "Setup_date",
retrievalCol = "Retrieval_date",
hasProblems = TRUE,
dateFormat = "dmy H",
occasionStartTime = 12
)
# now let's have a look at the first few columns of the camera operation matrix
camop_hours_12[, 1:5]
# and the last few
camop_hours_12[, (ncol(camop_hours_12)-6):ncol(camop_hours_12)]
The column names now indicate the shift, "+12h" in the example above. That means that the first column shows the effort from 12 noon on 2009-04-02 until 12 noon 2009-04-03.
When using this camera operation matrix in detectionHistory or spatialDetectionHistory, the shift will be extracted and trapping effort will be calculated accordingly. Therefore, the argument occasionStartTime in these two functions is deprecated from version 2.1.0.
Combining multiple cameras per station
Often multiple cameras are set at a station, but data are to be analysed on a station-level, not at the level of individual cameras. cameraOperation can combine these effort of individual cameras according to how they were set up.
First, let's create a simple artifial data set. Note how the columns contain an hour (e.g. 12 noon in this case: "2020-01-01 12"). Other formats for date-time are possible and can be specified with argument dateFormat in cameraOperation.
camtraps_by_camera <- data.frame(Station = c("A", "A"),
camera = c("cam1", "cam2"),
setup = c("2020-01-01 12"),
retrieval = c("2020-01-08 18", "2020-01-07 18"),
Problem1_from = c("2020-01-03 12", NA),
Problem1_to = c("2020-01-05 18", NA),
Problem2_from = c("2020-01-06 12", NA),
Problem2_to = c("2020-01-07 12", NA))
camtraps_by_camera
To aggregate data from cameras to station, you need a camera column ("camera" in the example above). Furthermore, set argument byCamera = FALSE. Then, the combinations of the arguments camerasIndependent and allCamsOn define how effort from individual cameras is combined. allCamsOn means all cameras need to be active for a station to be considered active. camerasIndepndent specifies whether the cameras accumulate effort independently or now. See ?cameraoperation for details.
Below we'll run cameraOperation of the little sample table camtraps_by_camera, for all possible combinations of allCamsOn, cameraOperation and hasProblems.
# list of common arguments for the different function runs
args.common <- list(CTtable = camtraps_by_camera,
stationCol = "Station",
cameraCol = "camera",
setupCol = "setup",
retrievalCol = "retrieval",
dateFormat = "ymd H",
byCamera = FALSE,
writecsv = FALSE
)
# combinations of the arguments of interest
fun.args0 <- expand.grid(allCamsOn = c(T,F),
camerasIndependent = c(T,F),
hasProblems = c(T,F))
# create a list of function arguments with the varying arguments
fun.args1 <- apply(fun.args0, MARGIN = 1, FUN = function(x) modifyList(args.common, as.list(x)))
# run cameraOperation on each combination of arguments, and combine into an output table with the function arguments
camOp_station_aggregation <- cbind(fun.args0, do.call(rbind, lapply(fun.args1, FUN = function(x)
do.call(cameraOperation, x))), row.names = NULL)
Now let's see how the daily effort was combined for the different combinations of allCamsOn and camerasIndependent when hasProblems is FALSE.
camOp_station_aggregation [camOp_station_aggregation$hasProblems == FALSE, -3]
sS you can see, allCamsOn has no effect if all cameras were active (due to hasProblems = FALSE). The differences on the setup/retrieval day are due to camerasIndependent.
Now with hasProblems = TRUE:
camOp_station_aggregation [camOp_station_aggregation$hasProblems == TRUE, -3]
Here's how the daily aggregates are calculate for the different combinations of allCamsOn and camerasIndependent, and depending on whether input columns are date or date-time (from v2.1.0). In the table below, x is a vector of values at individual cameras for one day (e.g. c(1,0.5) when one camera was active the whole day, the other cameras half a day).
| camerasIndependent | allCamsOn | input columns | aggregate with | example c(1, 0.5) | example c(1, 1) |
|---|---|---|---|---|---|
| TRUE | TRUE | date-time | ifelse(all(x == 1), sum(x), min(x) * length(x)) | 1 | 2 |
| TRUE | TRUE | date | ifelse(all(x == 1), sum(x), 0) | 0 | 2 |
| TRUE | FALSE | date-time | sum | 1.5 | 2 |
| TRUE | FALSE | date| sum | 1.5 | 2 |
| FALSE |TRUE | date-time | min | 0.5 | 1 |
| FALSE| TRUE | date | ifelse(all(x == 1), 1, 0) | 0 | 1 |
| FALSE| FALSE | date-time | mean | 0.75 | 2 |
| FALSE |FALSE | date| ifelse(sum(x, na.rm = TRUE) >= 1, 1, sum(x, na.rm = TRUE)) | 1 | 1 |
Plotting camera operation matrices
There is a little function for plotting the camera operation matrix included in the package, but not exported in the Namespace. It is accessible with camtrapR:::camopPlot(). Imagine there is a typo in one of your date fields and the setup or retrieval year is wrong. You will easily be able to spot it this way.
Here is the camera operation matrix calculated above (in two different color palettes, the default being the viridis palette - requires R 3.6.0 or higher).
par(mfrow = c(1,2))
camtrapR:::camopPlot(camOp = camop_problem)
camtrapR:::camopPlot(camOp = camop_problem, palette = "Heat")
One can also plot with the levelplot function from the lattice package:
camtrapR:::camopPlot(camOp = camop_problem, lattice = TRUE)
camtrapR:::camopPlot(camOp = camop_problem, palette = "Heat", lattice = TRUE)
Here is what it would look like if the year is wrong for one station.
camtraps_typo <- camtraps
# replace 2009 with 2008 in setup date of station A
camtraps_typo$Setup_date[1] <- gsub("2009", "2008", camtraps_typo$Setup_date[1])
camop_typo <- cameraOperation(CTtable = camtraps_typo,
stationCol = "Station",
setupCol = "Setup_date",
retrievalCol = "Retrieval_date",
writecsv = FALSE,
hasProblems = TRUE,
dateFormat = dateFormat
)
camtrapR:::camopPlot(camop_typo)
Note that the colors are stretched between the lowest and highest value. There is no dedicated color for 0, designating non-operational cameras.
For a list of color palettes, see here.
Saving and loading camera operation matrices
The camera operation matrix can easily be saved as a csv file (by setting argument writecsv = TRUE, check.names = FALSE and defining outdir). In order to load the csv into R again, it is necessary to tell R to use the station IDs (the first column) as row names:
camOp <- read.csv(file = ..., row.names = 1, check.names = FALSE)
check.names = FALSE ensures that column names (the dates) are read back into R as they are (e.g. "2015-12-01"). Otherwise one may end up with unreadable column names (at least for camtrapR) such as "X2015.12.01".
Input for subsequent analyses (single-season)
It is very easy to prepare input for single-season occupancy and spatial capture-recapture (SCR) analyses in camtrapR. All one needs it a record table and the camera operation matrix. Making input for SCR analyses further requires the camera trap station table.
Occupancy analyses
Occupancy models use detection/non-detection matrices in which for every station and every occasion "1"" signifies a detection of a given species, "0"" signifies non-detecion, and NA missing data.
Here is how to obtain a detection/non-detection matrix using function detectionHistory. Because the function builds on prior functions (recordTable and cameraOperation) we also show these function here to provide the context.
# create camera operation matrix
camop_no_problem <- cameraOperation(CTtable = camtraps,
stationCol = "Station",
setupCol = "Setup_date",
retrievalCol = "Retrieval_date",
hasProblems = FALSE,
dateFormat = dateFormat
)
# define image directory
wd_images_ID <- system.file("pictures/sample_images_species_dir", package = "camtrapR")
# make record table
recordTableSample <- recordTable(inDir = wd_images_ID,
IDfrom = "directory",
minDeltaTime = 60,
deltaTimeComparedTo = "lastIndependentRecord",
timeZone = "Asia/Kuala_Lumpur"
)
# make detection history (without trapping effort)
DetHist1 <- detectionHistory(recordTable = recordTableSample,
camOp = camop_no_problem,
stationCol = "Station",
speciesCol = "Species",
recordDateTimeCol = "DateTimeOriginal",
species = "VTA",
occasionLength = 7,
day1 = "station",
includeEffort = FALSE
)
DetHist1
Note the warning about the missing time zone (as in the functions recordTable and recordTable). Normally, it should be fine, but to be on the safe side, better set it to your study area's time zone.
If trapping effort is thought to influence detection probability, it can be returned by setting includeEffort = TRUE. This way the number of active trapping days per occasion and station is returned.
# make detection history (with trapping effort)
DetHist2 <- detectionHistory(recordTable = recordTableSample,
camOp = camop_no_problem,
stationCol = "Station",
speciesCol = "Species",
recordDateTimeCol = "DateTimeOriginal",
species = "VTA",
timeZone = "Asia/Kuala_Lumpur",
occasionLength = 7,
day1 = "station",
includeEffort = TRUE,
scaleEffort = FALSE
)
DetHist2[[1]] # detection history
DetHist2[[2]] # effort (in days per occasion)
To help with convergence of models, the effort matrix can be scaled to mean = 0 and sd = 1 by setting scaleEffort = TRUE. If writecsv = TRUE, the scaling parameters will also be saved in a separate csv file.
DetHist3 <- detectionHistory(recordTable = recordTableSample,
camOp = camop_no_problem,
stationCol = "Station",
speciesCol = "Species",
recordDateTimeCol = "DateTimeOriginal",
species = "VTA",
timeZone = "Asia/Kuala_Lumpur",
occasionLength = 7,
day1 = "station",
includeEffort = TRUE,
scaleEffort = TRUE
)
DetHist3[[1]] # detection history (same as above)
DetHist3[[2]] # effort (scaled)
DetHist3[[3]] # scaling parameters for back-transformation
# backtransform scaled effort like this if needed
(DetHist3[[2]] * DetHist3[[3]]$effort.scaled.scale) + DetHist3[[3]]$effort.scaled.center
Handling of incomplete occasions
The following table shows the behaviour of the detectionHistory function for different combinations of the function arguments includeEffort and minActiveDaysPerOccasion and different occasion-level camera operation values (table head, column 3-7). Depending on these arguments, incomplete occasions will either NA or have values of 0/1 (depending on whether there was a detection) in the output detection matrix .
| includeEffort | minActiveDaysPerOccasion | camera operation all 1 | at least one 1 | all 0 | 0 and NA | all NA |
|---|---|---|---|---|---|---|---|
| TRUE | not defined | 0/1 | 0/1 | NA | NA | NA |
| TRUE | defined | 0/1 | 0/1/NA | NA | NA | NA |
| FALSE | not defined | 0/1 | NA | NA | NA | NA |
| FALSE | defined | 0/1 | 0/1/NA | NA | NA | NA |
*: NA if there were less active days in an occasion than minActiveDaysPerOccasion
The same applies to generation of input for spatial capture-recapture analyses using spatialDetectionHistory as described below.
Saving and loading detection histories
The detection history and effort matrices can easily be saved as csv files (by setting argument writecsv = TRUE and defining outdir). In order to load the csv into R again, it is necessary to tell R to use the station IDs as row names:
detHist <- read.csv(file = ..., row.names = 1)
effort <- read.csv(file = ..., row.names = 1)
Spatial Capture-Recapture analyses
Input for spatial capture-recapture analyses can be generated in the form of capthist-objects as defined in the secr package with the function spatialDetectionHistory. Output can be in the form of counts (number of individual detections per occasion, argument, argument output = "count") or binary (was an individual detected during an occasion, argument output = "binary"). note that the detector type will change accordingly: "proximity" if output = "binary" and "count" if output = "count".
data(recordTableIndividualSample)
# create camera operation matrix (with problems/malfunction)
camop_problem <- cameraOperation(CTtable = camtraps,
stationCol = "Station",
setupCol = "Setup_date",
retrievalCol = "Retrieval_date",
writecsv = FALSE,
hasProblems = TRUE,
dateFormat = dateFormat
)
sdh <- spatialDetectionHistory(recordTableIndividual = recordTableIndividualSample,
species = "LeopardCat",
output = "binary",
camOp = camop_problem,
CTtable = camtraps,
stationCol = "Station",
speciesCol = "Species",
Xcol = "utm_x",
Ycol = "utm_y",
individualCol = "Individual",
recordDateTimeCol = "DateTimeOriginal",
recordDateTimeFormat = "%Y-%m-%d %H:%M:%S",
occasionLength = 10,
day1 = "survey",
includeEffort = TRUE,
timeZone = "Asia/Kuala_Lumpur"
)
# missing space in species = "LeopardCat" was introduced by recordTableIndividual
# (because of CRAN package policies).
# In your own data you can have spaces in your directory names.
summary(sdh)
plot(sdh, tracks = TRUE)
Input for subsequent analyses (multi-season)
Since version 1.2, camtrapR can prepare input for multi-season occupancy and spatial capture-recapture analyses (in unmarked and secr). The process is very similar to the one for single-season models.
Multi-season occupancy analyses
camtrapR prepares detection histories that can be used as input for argument y in function unmarkedMultFrame. unmarkedMultFrame creates the input for function colext which fits multi-season occupancy models.
For multi-season occupancy, we need a session column in the camera trap table. Records will be assigned to session automatically based on their dates.
First we simulate data for 2 seasons by duplicating the existing sample data and adjusting the dates (add 1 year)
# load multi-season data
data(camtrapsMultiSeason)
data(recordTableSampleMultiSeason)
# also, for clarity, lets remove all unnecessary columns from the record table
recordTableSampleMultiSeason <- recordTableSampleMultiSeason[, c("Station", "Species", "DateTimeOriginal")]
# create camera operation matrix
camop_season <- cameraOperation(CTtable = camtrapsMultiSeason,
stationCol = "Station",
setupCol = "Setup_date",
sessionCol = "session",
retrievalCol = "Retrieval_date",
hasProblems = TRUE,
dateFormat = dateFormat
)
# plot camera operation matrix
par(oma = c(0,7,0,0))
camtrapR:::camopPlot(camop_season)
# make multi-season detection history
DetHist_multi <- detectionHistory(recordTable = recordTableSampleMultiSeason,
camOp = camop_season,
stationCol = "Station",
speciesCol = "Species",
species = "VTA",
occasionLength = 10,
day1 = "station",
recordDateTimeCol = "DateTimeOriginal",
includeEffort = TRUE,
scaleEffort = FALSE,
timeZone = "UTC",
unmarkedMultFrameInput = TRUE
)
DetHist_multi
Note that the function makes sure that all seasons have identical number of occasions, and that all stations are represented, even if they were not sampled in all seasons.
From here, the resulting object can be used in unmarked.
year_matrix <- matrix(unique(as.character(camtrapsMultiSeason$session)),
ncol = length(unique(as.character(camtrapsMultiSeason$session))),
nrow = length(unique(camtrapsMultiSeason$Station)),
byrow = TRUE)
# this is a made up example table with station covariates for demonstration
site_covariates <- data.frame(Station = rownames(DetHist_multi$detection_history),
elevation = c(100, 200, 500, 300),
treecover = c(80, 100, 50, 10))
umf <- unmarked::unmarkedMultFrame(y = DetHist_multi$detection_history,
siteCovs = site_covariates,
yearlySiteCovs = list(year = year_matrix),
obsCovs = list(effort = DetHist_multi$effort),
numPrimary = 2)
colext_example <- unmarked::colext(psiformula = ~ treecover, # First-year occupancy
gammaformula = ~ 1, # Colonization
epsilonformula = ~ 1, # Extinction
pformula = ~ effort, # Detection
data = umf,
method="BFGS")
summary(colext_example)
Multi-season spatial capture-recapture
For secr, session IDs must be a gapless sequence of numbers beginning with 1. So 2009 becomes season 1 and 2010 season 2
camtrapsMultiSeason$session[camtrapsMultiSeason$session == 2009] <- 1
camtrapsMultiSeason$session[camtrapsMultiSeason$session == 2010] <- 2
# we also want a few records in season 2
recordTableIndividualSample_season2 <- recordTableIndividualSample[1:10,]
recordTableIndividualSample_season2$DateTimeOriginal <- gsub("2009", "2010", recordTableIndividualSample_season2$DateTimeOriginal)
recordTableIndividualSample_season <- rbind(recordTableIndividualSample, recordTableIndividualSample_season2)
# for clarity, lets remove all unnecessary columns
recordTableIndividualSample_season <- recordTableIndividualSample_season[, c("Station", "Species", "Individual", "DateTimeOriginal")]
# create camera operation matrix (with problems/malfunction), same as above for multi-season occupancy
camop_season <- cameraOperation(CTtable = camtrapsMultiSeason,
stationCol = "Station",
setupCol = "Setup_date",
sessionCol = "session",
retrievalCol = "Retrieval_date",
hasProblems = TRUE,
dateFormat = dateFormat
)
# create capthist object
sdh_multi <- spatialDetectionHistory(recordTableIndividual = recordTableIndividualSample_season,
species = "LeopardCat",
output = "binary",
camOp = camop_season,
CTtable = camtrapsMultiSeason,
stationCol = "Station",
speciesCol = "Species",
sessionCol = "session",
Xcol = "utm_x",
Ycol = "utm_y",
individualCol = "Individual",
recordDateTimeCol = "DateTimeOriginal",
recordDateTimeFormat = "%Y-%m-%d %H:%M:%S",
occasionLength = 10,
day1 = "survey",
includeEffort = TRUE,
timeZone = "Asia/Kuala_Lumpur",
stationCovariateCols = "utm_y", # made up, a potential site covariate from camtrapsMultiSeason
individualCovariateCols = "Individual" # made up, a potential individual covariate from recordTable
)
summary(sdh_multi)
par(mfrow = c(1,2))
plot(sdh_multi)
secr.fit.example <- secr.fit(capthist = sdh_multi,
start = c(-1, -2, 10)) # with starting values, since its only very little data
summary(secr.fit.example)
jniedballa/camtrapR documentation built on Dec. 11, 2024, 10:25 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
exiftool_present <- Sys.which("exiftool") != ""
print("WARNING: THIS VIGNETTE WAS CREATED WITHOUT EXIFTOOL. OUTPUT IS INCOMPLETE SINCE ESSENTIAL FUNCTIONS DID NOT RUN!")
library(camtrapR) library(secr)
Tabulating species and individual records: the recordTable functions
There are 2 function to tabulate species records after identification of species from images and videos.
recordTabletabulates records of all images after species-level identification andrecordTableIndividualtabulates individuals of one species.
Nevertheless, the underlying idea is the same. For each image, the date and time it was taken are read from the image's Exif metadata using ExifTool. Species or individual ID are read from the directory structure or image metadata (see vignette "Species and Individual Identification"). Video data are extracted analogously.
recordTable: tabulating species records
recordTable is typically run after identifying species from images. It reads species IDs from the directory structure the images are placed in or from image metadata tags.
First we define the directory containing our renamed, identified images
# find the directory with sample images contained in the package wd_images_ID <- system.file("pictures/sample_images_species_dir", package = "camtrapR", lib.loc = .libPaths())
and see how many JPG images we have (this is not necessary, but informative here).
length(list.files(wd_images_ID, pattern = "JPG", recursive = TRUE))
Now we can run recordTable. Here is a minimal example:
rec.db.species0 <- recordTable(inDir = wd_images_ID, IDfrom = "directory") head(rec.db.species0)
Argument IDfrom tells the function to look for species directories within the station directories and to take species IDs from these. IDfrom must be set to "metadata" if metadata tagging was used for species identification.
By default, the function returns all records. Nevertheless, we get a data frame containing r ifelse(exiftool_present, nrow(rec.db.species0), 56) records, less than the number of images in the image directory (r length(list.files(wd_images_ID, pattern = "JPG", recursive = TRUE))). This is because a number of images were taken at the same time (measured with 1 minute precision by these cameras) and the function removes duplicate records.
It may suffice to illustrate this with the Pig-tailed macaque images from stationB:
list.files(file.path(wd_images_ID, "StationB", "MNE"))
There are, amongst others, 3 images taken at 17:48:00 on the 28th of April 2009. Of these, only 1 will be returned.
Here is what the columns of the record table contain:
| column | content |
|-------|-------------------------------------------------------------|
| Station | the station the image is from |
| Species | species name |
| DateTimeOriginal | Date and time of record in R-readable format |
| Date | record date |
| Time | record time of day |
| delta.time.secs | time difference between record and last (independent) record of same species at same station / camera (in seconds) |
| delta.time.mins | time difference between record and last (independent) record of same species at same station / camera (in minutes) |
| delta.time.hours | time difference between record and last (independent) record of same species at same station / camera (in hours) |
| delta.time.days | time difference between record and last (independent) record of same species at same station / camera (in days) |
| Directory | directory the image is in |
| FileName | image file name |
*see below: Independence between cameras within stations
Temporal independence between records
Imagine a species that loves to hang out in front of your cameras. You will end up with hundreds of shots of the same species, maybe even the same animal. Therefore, images can be filtered using an adjustable criterion for temporal independence between subsequent records of the same species in an attempt to remove non-independent records. This is achieved via argument minDeltaTime. It is the minimum time difference (in minutes) between two records of the same species at the same station which are to be considered independent. The default is 0, causing the function to return all records. Setting it to a higher number, e.g. 60 (i.e., 1 hour), is commonly done to thin the number of records. Note that you will not lose records of different species, even if they fall within the specified time interval from a record.
The argument deltaTimeComparedTo further controls how independence between records is assessed. Setting it to "lastRecord" returns only records taken minDeltaTime minutes after the last record. Setting it to "lastIndependentRecord" returns only records taken minDeltaTime minutes after the last independent record.
Let's now apply an 1-hour independence criterion and define a time zone.
rec.db.species60 <- recordTable(inDir = wd_images_ID, IDfrom = "directory", minDeltaTime = 60, deltaTimeComparedTo = "lastRecord", timeZone = "Asia/Kuala_Lumpur") nrow(rec.db.species60)
Now r ifelse(exiftool_present, nrow(rec.db.species60), 40) records were returned instead of r ifelse(exiftool_present, nrow(rec.db.species0), 56). The missing records were taken less than 1 hour after the prior record of the same species at the same station and therefore omitted.
Time zones and daylight saving time
Note the warning about the missing time zone in the minimal example above. By default, camtrapR will assume UTC time zone. This should work well in most situations (even though the time zone may not be correct, strictly speaking), because UTC does not use daylight saving time (DST, aka 'summer time'), and camera traps normally don't use time zones (hence, camera traps will normally not respect DST of the area you work in and will not adjust image timestamps accordingly).
Now, if you work in an area that uses DST, and your camera does not know about DST, things can go wrong in various ways and you may end up with a systematic 1-hour offset. If you set your cameras up during DST (in summer) and set the internal clock accordingly, all records taken in winter (non-DST period) will be recorded as 1 hour later than according to actual clock time. If you set your cameras up during winter (not DST) and set the internal clock accordingly, all records taken in summer (DST period) will be recorded as 1 hour earlier than according to actual clock time.
In addition if you specify a time zone with DST and your cameras don't respect it, records may fall into the non-existing hour when clocks are advanced in spring, leading to an error in interpreting the date/time.
So, the question whether or not your cameras record the time zone becomes very important. Here is some recommendations on how to use the argument timeZone.
If your cameras don't save the time zone, and your time zone does not have DST, set argument timeZone to your study area time zone
If your cameras don't save the time zone, and your time zone has DST, leave argument timeZone at the default UTC.
If your cameras save the time zone and adjust image timestamps accordingly, set argument timeZone to your study area time zone (both if your area has DST or not).
In other words, it is advised to set argument timeZone to your study area's time zone (one of OlsonNames()), unless the time zone of your study area has DST, but your cameras don't record it.
Independence between cameras within stations
The issue of temporal independence between records becomes slightly more complex if more than one camera was operated at stations. That information can and should be included in the output of recordTable (and recordTableIndividual). Users can then decide whether temporal independence is to be assessed within stations or within cameras at each station. In the first case, argument camerasIndependent must be set to FALSE. 2 images taken at different 2 cameras at the same station within minDeltaTime minutes will be reported as 1 record in the record table (suitable if cameras were places in pairs). In the second case camerasIndependent must be set to TRUE and 2 images taken at different 2 cameras at the same station within minDeltaTime minutes will be reported as 2 record in the record table (suitable e.g., if cameras were located at some distance to one another and faced different trails).
The cameraID argument controls where camtrapR will look for camera IDs: in the file names (after renaming using imageRename, e.g. "renamedImages/StationA/StationACamera12015-12-3123-59-59(1).JPG") or in the directory structure (e.g. renamedImages/StationA/Camera1/StationACamera12015-12-3123-59-59(1).JPG"). If missing, it will be assumed there was only 1 camera per station.
Ignoring species
Argument exclude can be used to ignore certain species. This is useful for omitting images in directories like "team" or "unidentified". Here is an example:
# see what species we recorded table(rec.db.species60$Species) # remove "UNID" by setting argument exclude = "UNID" rec.db.species60.exclude <- recordTable(inDir = wd_images_ID, IDfrom = "directory", minDeltaTime = 60, deltaTimeComparedTo = "lastIndependentRecord", timeZone = "Asia/Kuala_Lumpur", exclude = "UNID") # note that "UNID" is gone now table(rec.db.species60.exclude$Species)
Extracting image metadata
recordTable and recordTableIndividual can both extract additional metadata from images (apart from date and time). For example, some camera models record ambient temperature or moon phase, which may be of interest. Metadata tags are stored in the images at the time they are taken and can be accessed and extracted if their tag names are known. Some tag names are standardised (e.g. "DateTimeOriginal") while others are manufacturer-specific. Therefore, function exifTagNames returns all Exif metadata it finds in a sample image. Users can then choose which to extract with recordTable and recordTableIndividual.
wd_images_ID <- system.file("pictures/sample_images_species_dir", package = "camtrapR") exifTagNames(inDir = wd_images_ID)
The output table contains tag groups, tag descriptions, tag names and tag values. The values in tag_group help to unambiguously identify specific metadata tags.
Use the entries in tag_name (not tag_description) when providing values to the argument additionalMetadataTags in therecordTable functions.
Of the tags shown here, "DateTimeOriginal" contains the date and time that camtrapR reads out. Apart from that, there is little information of ecological interest in the example data. However, for demonstration purposes, let's extract information about the camera model and make. To be as precise as possible, we provide tags in the format: tag_group:tag_name (e.g. EXIF:Model)
rec.db.species.metadata1 <- recordTable(inDir = wd_images_ID, IDfrom = "directory", timeZone = "Asia/Kuala_Lumpur", additionalMetadataTags = c("EXIF:Model", "EXIF:Make")) head(rec.db.species.metadata1)
There are 2 additional columns containing the information from the metadata we requested.
recordTableIndividual: tabulating individuals of a species
Individual identification of species is a prerequiste for running (spatial) capture-recapture models. recordTableIndividual and the subsequent spatialDetectionHistory function prepare data for these models.
# find the directory with tagged sample images contained in the package wd_images_individual_ID <- system.file("pictures/sample_images_indiv_tag/LeopardCat", package = "camtrapR") # missing space in species = "LeopardCat" is because of CRAN package policies rec.db.pbe <- recordTableIndividual(inDir = wd_images_individual_ID, IDfrom = "metadata", minDeltaTime = 60, deltaTimeComparedTo = "lastIndependentRecord", hasStationFolders = FALSE, # images are not in station directories metadataIDTag = "individual", # the name of the metadata tag containing individual IDs timeZone = "Asia/Kuala_Lumpur" )
Extracting custom image metadata
In addition to the metadata that were saved when the image was taken (see above), custom metadata tags assigned in image management software can also be extracted. These may contain species ID tags if species were identified using metadata tags (instead of moving images into species directories), but also individual IDs, animal counts, sex of animals recorded, etc..
head(rec.db.pbe)
Camera Operation
The camera operation matrix is a day-by-station matrix that states how many cameras were active at a station on a given day. Rows represent stations and columns days, beginning with the day the first camera was set up and ending the day the last camera was retrieved. It is calculated from the camera trap table mention in vignette 1 (see ?camtraps for an example). Users can give the setup / retrieval / problem columns in the input table as dates or date-times (since version 2.1.0).
Important: Prior to version 2.1.0, the values were integer numbers, with 1 indicating the camera was active on a day (or possibly higher values if multiple cameras at a station are combined).
From version 2.1.0, the values are the fraction of the day a camera was active. If only dates are provided in the camera trap table, the function will assume 12 noon as setup/retrieval time Problem periods will be assumed to last the entire day.
If imput is provided as date-times, output will be the exact fraction of day a station or camera was active (taking into account setup / retrieval time as well as time of Problem periods).
Dates as input
# first load the camera trap station table data(camtraps) dateFormat <- "dmy" # requires lubridate package # alternatively, use "%d/%m/%Y" (from base R) camop_problem <- cameraOperation(CTtable = camtraps, stationCol = "Station", setupCol = "Setup_date", retrievalCol = "Retrieval_date", writecsv = FALSE, hasProblems = TRUE, dateFormat = dateFormat ) # as a reminder, these are the dates in our station information table camtraps[,-which(colnames(camtraps) %in% c("utm_y", "utm_x"))] # now let's have a look at the first few columns of the camera operation matrix camop_problem[, 1:5] # and the last few camop_problem[, (ncol(camop_problem)-6):ncol(camop_problem)]
If stations were not set up, values are NA. Operational stations get value 1. If cameras were set up but malfunctioning, it is 0 (if hasProblems = TRUE).
Date-times as input
camtraps_hrs <- camtraps # assign hours for setup and retrieval camtraps_hrs$Setup_date <- paste(camtraps_hrs$Setup_date, c("12", "15", "18")) camtraps_hrs$Retrieval_date <- paste(camtraps_hrs$Retrieval_date, c("18", "15", "12")) # assign a random hours for begin of Problem at station 3 camtraps_hrs$Problem1_from[3] <- paste(camtraps_hrs$Problem1_from[3], "20") # Problem ends with retrieval (i.e., malfunction until camera retrieved) camtraps_hrs$Problem1_to[3] <- camtraps_hrs$Retrieval_date[3] # create camera operation matrix camop_hours <- cameraOperation(CTtable = camtraps_hrs, stationCol = "Station", setupCol = "Setup_date", retrievalCol = "Retrieval_date", hasProblems = TRUE, dateFormat = "dmy H" ) # as a reminder, these are the date-times in our station information table camtraps_hrs # now let's have a look at the first few columns of the camera operation matrix camop_hours[, 1:5] # and the last few camop_hours[, (ncol(camop_hours)-6):ncol(camop_hours)]
Differences between version 2.1 and previous versions
Version 2.1 introduced support of cameraOperation() for date-times in the setup / retrieval / problem columns. By giving the setup / retrieval / problem columns as date-times (and adjusting argument dateFormat accordingly), one can calculate precise daily effort.
If the times are unknown and users provide dates only to argument dateFormat, camtrapR will assume that setup and retrieval were at 12 noon. Malfunction periods (indicated by the problem columns) will be interpreted as the cameras having malfunctioned the whole day.
This is how the handling of setup/retrieval days differs between dates and date-times in dateFormat:
| version | dateFormat is date | dateFormat is date-time | |-|-|-| | prior to 2.1 | 1 | not implemented | | 2.1 and later | 0.5 | calculated precisely |
This is how the handling of problems / malfunction periods differs between dates and date-times in dateFormat:
| version | dateFormat is date | dateFormat is date-time | |-|-|-| | prior to 2.1 | 0 | not implemented | | 2.1 and later | 0 | calculated precisely |
Camera days vs. camera nights
By default, the camera operation matrix refers to days, centered on noon and lasting from midnight to midnight the next day. In some situations it can make sense to shift the time frame for analyses, so the camera operation matrix describes trap nights (centered on midnight, lasting from noon to noon the next day).
To that end, the argument occasionStartTime can be set. Prior to version 2.1.0, it was an argument of detectionHistory and spatialDetectionHistory, but was moved to cameraOperation in v2.1.0. It can be set to an hour between 0 and 23, describing the hour the occasions begin.
# create camera operation matrix with occasions / trap days starting on noon (until noon the next day) camop_hours_12 <- cameraOperation(CTtable = camtraps_hrs, stationCol = "Station", setupCol = "Setup_date", retrievalCol = "Retrieval_date", hasProblems = TRUE, dateFormat = "dmy H", occasionStartTime = 12 ) # now let's have a look at the first few columns of the camera operation matrix camop_hours_12[, 1:5] # and the last few camop_hours_12[, (ncol(camop_hours_12)-6):ncol(camop_hours_12)]
The column names now indicate the shift, "+12h" in the example above. That means that the first column shows the effort from 12 noon on 2009-04-02 until 12 noon 2009-04-03.
When using this camera operation matrix in detectionHistory or spatialDetectionHistory, the shift will be extracted and trapping effort will be calculated accordingly. Therefore, the argument occasionStartTime in these two functions is deprecated from version 2.1.0.
Combining multiple cameras per station
Often multiple cameras are set at a station, but data are to be analysed on a station-level, not at the level of individual cameras. cameraOperation can combine these effort of individual cameras according to how they were set up.
First, let's create a simple artifial data set. Note how the columns contain an hour (e.g. 12 noon in this case: "2020-01-01 12"). Other formats for date-time are possible and can be specified with argument dateFormat in cameraOperation.
camtraps_by_camera <- data.frame(Station = c("A", "A"), camera = c("cam1", "cam2"), setup = c("2020-01-01 12"), retrieval = c("2020-01-08 18", "2020-01-07 18"), Problem1_from = c("2020-01-03 12", NA), Problem1_to = c("2020-01-05 18", NA), Problem2_from = c("2020-01-06 12", NA), Problem2_to = c("2020-01-07 12", NA)) camtraps_by_camera
To aggregate data from cameras to station, you need a camera column ("camera" in the example above). Furthermore, set argument byCamera = FALSE. Then, the combinations of the arguments camerasIndependent and allCamsOn define how effort from individual cameras is combined. allCamsOn means all cameras need to be active for a station to be considered active. camerasIndepndent specifies whether the cameras accumulate effort independently or now. See ?cameraoperation for details.
Below we'll run cameraOperation of the little sample table camtraps_by_camera, for all possible combinations of allCamsOn, cameraOperation and hasProblems.
# list of common arguments for the different function runs args.common <- list(CTtable = camtraps_by_camera, stationCol = "Station", cameraCol = "camera", setupCol = "setup", retrievalCol = "retrieval", dateFormat = "ymd H", byCamera = FALSE, writecsv = FALSE ) # combinations of the arguments of interest fun.args0 <- expand.grid(allCamsOn = c(T,F), camerasIndependent = c(T,F), hasProblems = c(T,F)) # create a list of function arguments with the varying arguments fun.args1 <- apply(fun.args0, MARGIN = 1, FUN = function(x) modifyList(args.common, as.list(x))) # run cameraOperation on each combination of arguments, and combine into an output table with the function arguments camOp_station_aggregation <- cbind(fun.args0, do.call(rbind, lapply(fun.args1, FUN = function(x) do.call(cameraOperation, x))), row.names = NULL)
Now let's see how the daily effort was combined for the different combinations of allCamsOn and camerasIndependent when hasProblems is FALSE.
camOp_station_aggregation [camOp_station_aggregation$hasProblems == FALSE, -3]
sS you can see, allCamsOn has no effect if all cameras were active (due to hasProblems = FALSE). The differences on the setup/retrieval day are due to camerasIndependent.
Now with hasProblems = TRUE:
camOp_station_aggregation [camOp_station_aggregation$hasProblems == TRUE, -3]
Here's how the daily aggregates are calculate for the different combinations of allCamsOn and camerasIndependent, and depending on whether input columns are date or date-time (from v2.1.0). In the table below, x is a vector of values at individual cameras for one day (e.g. c(1,0.5) when one camera was active the whole day, the other cameras half a day).
| camerasIndependent | allCamsOn | input columns | aggregate with | example c(1, 0.5) | example c(1, 1) | |---|---|---|---|---|---| | TRUE | TRUE | date-time | ifelse(all(x == 1), sum(x), min(x) * length(x)) | 1 | 2 | | TRUE | TRUE | date | ifelse(all(x == 1), sum(x), 0) | 0 | 2 | | TRUE | FALSE | date-time | sum | 1.5 | 2 | | TRUE | FALSE | date| sum | 1.5 | 2 | | FALSE |TRUE | date-time | min | 0.5 | 1 | | FALSE| TRUE | date | ifelse(all(x == 1), 1, 0) | 0 | 1 | | FALSE| FALSE | date-time | mean | 0.75 | 2 | | FALSE |FALSE | date| ifelse(sum(x, na.rm = TRUE) >= 1, 1, sum(x, na.rm = TRUE)) | 1 | 1 |
Plotting camera operation matrices
There is a little function for plotting the camera operation matrix included in the package, but not exported in the Namespace. It is accessible with camtrapR:::camopPlot(). Imagine there is a typo in one of your date fields and the setup or retrieval year is wrong. You will easily be able to spot it this way.
Here is the camera operation matrix calculated above (in two different color palettes, the default being the viridis palette - requires R 3.6.0 or higher).
par(mfrow = c(1,2)) camtrapR:::camopPlot(camOp = camop_problem) camtrapR:::camopPlot(camOp = camop_problem, palette = "Heat")
One can also plot with the levelplot function from the lattice package:
camtrapR:::camopPlot(camOp = camop_problem, lattice = TRUE) camtrapR:::camopPlot(camOp = camop_problem, palette = "Heat", lattice = TRUE)
Here is what it would look like if the year is wrong for one station.
camtraps_typo <- camtraps # replace 2009 with 2008 in setup date of station A camtraps_typo$Setup_date[1] <- gsub("2009", "2008", camtraps_typo$Setup_date[1]) camop_typo <- cameraOperation(CTtable = camtraps_typo, stationCol = "Station", setupCol = "Setup_date", retrievalCol = "Retrieval_date", writecsv = FALSE, hasProblems = TRUE, dateFormat = dateFormat ) camtrapR:::camopPlot(camop_typo)
Note that the colors are stretched between the lowest and highest value. There is no dedicated color for 0, designating non-operational cameras.
For a list of color palettes, see here.
Saving and loading camera operation matrices
The camera operation matrix can easily be saved as a csv file (by setting argument writecsv = TRUE, check.names = FALSE and defining outdir). In order to load the csv into R again, it is necessary to tell R to use the station IDs (the first column) as row names:
camOp <- read.csv(file = ..., row.names = 1, check.names = FALSE)
check.names = FALSE ensures that column names (the dates) are read back into R as they are (e.g. "2015-12-01"). Otherwise one may end up with unreadable column names (at least for camtrapR) such as "X2015.12.01".
Input for subsequent analyses (single-season)
It is very easy to prepare input for single-season occupancy and spatial capture-recapture (SCR) analyses in camtrapR. All one needs it a record table and the camera operation matrix. Making input for SCR analyses further requires the camera trap station table.
Occupancy analyses
Occupancy models use detection/non-detection matrices in which for every station and every occasion "1"" signifies a detection of a given species, "0"" signifies non-detecion, and NA missing data.
Here is how to obtain a detection/non-detection matrix using function detectionHistory. Because the function builds on prior functions (recordTable and cameraOperation) we also show these function here to provide the context.
# create camera operation matrix camop_no_problem <- cameraOperation(CTtable = camtraps, stationCol = "Station", setupCol = "Setup_date", retrievalCol = "Retrieval_date", hasProblems = FALSE, dateFormat = dateFormat ) # define image directory wd_images_ID <- system.file("pictures/sample_images_species_dir", package = "camtrapR") # make record table recordTableSample <- recordTable(inDir = wd_images_ID, IDfrom = "directory", minDeltaTime = 60, deltaTimeComparedTo = "lastIndependentRecord", timeZone = "Asia/Kuala_Lumpur" ) # make detection history (without trapping effort) DetHist1 <- detectionHistory(recordTable = recordTableSample, camOp = camop_no_problem, stationCol = "Station", speciesCol = "Species", recordDateTimeCol = "DateTimeOriginal", species = "VTA", occasionLength = 7, day1 = "station", includeEffort = FALSE ) DetHist1
Note the warning about the missing time zone (as in the functions recordTable and recordTable). Normally, it should be fine, but to be on the safe side, better set it to your study area's time zone.
If trapping effort is thought to influence detection probability, it can be returned by setting includeEffort = TRUE. This way the number of active trapping days per occasion and station is returned.
# make detection history (with trapping effort) DetHist2 <- detectionHistory(recordTable = recordTableSample, camOp = camop_no_problem, stationCol = "Station", speciesCol = "Species", recordDateTimeCol = "DateTimeOriginal", species = "VTA", timeZone = "Asia/Kuala_Lumpur", occasionLength = 7, day1 = "station", includeEffort = TRUE, scaleEffort = FALSE ) DetHist2[[1]] # detection history DetHist2[[2]] # effort (in days per occasion)
To help with convergence of models, the effort matrix can be scaled to mean = 0 and sd = 1 by setting scaleEffort = TRUE. If writecsv = TRUE, the scaling parameters will also be saved in a separate csv file.
DetHist3 <- detectionHistory(recordTable = recordTableSample, camOp = camop_no_problem, stationCol = "Station", speciesCol = "Species", recordDateTimeCol = "DateTimeOriginal", species = "VTA", timeZone = "Asia/Kuala_Lumpur", occasionLength = 7, day1 = "station", includeEffort = TRUE, scaleEffort = TRUE ) DetHist3[[1]] # detection history (same as above) DetHist3[[2]] # effort (scaled) DetHist3[[3]] # scaling parameters for back-transformation # backtransform scaled effort like this if needed (DetHist3[[2]] * DetHist3[[3]]$effort.scaled.scale) + DetHist3[[3]]$effort.scaled.center
Handling of incomplete occasions
The following table shows the behaviour of the detectionHistory function for different combinations of the function arguments includeEffort and minActiveDaysPerOccasion and different occasion-level camera operation values (table head, column 3-7). Depending on these arguments, incomplete occasions will either NA or have values of 0/1 (depending on whether there was a detection) in the output detection matrix .
| includeEffort | minActiveDaysPerOccasion | camera operation all 1 | at least one 1 | all 0 | 0 and NA | all NA |
|---|---|---|---|---|---|---|---|
| TRUE | not defined | 0/1 | 0/1 | NA | NA | NA |
| TRUE | defined | 0/1 | 0/1/NA | NA | NA | NA |
| FALSE | not defined | 0/1 | NA | NA | NA | NA |
| FALSE | defined | 0/1 | 0/1/NA | NA | NA | NA |
*: NA if there were less active days in an occasion than minActiveDaysPerOccasion
The same applies to generation of input for spatial capture-recapture analyses using spatialDetectionHistory as described below.
Saving and loading detection histories
The detection history and effort matrices can easily be saved as csv files (by setting argument writecsv = TRUE and defining outdir). In order to load the csv into R again, it is necessary to tell R to use the station IDs as row names:
detHist <- read.csv(file = ..., row.names = 1) effort <- read.csv(file = ..., row.names = 1)
Spatial Capture-Recapture analyses
Input for spatial capture-recapture analyses can be generated in the form of capthist-objects as defined in the secr package with the function spatialDetectionHistory. Output can be in the form of counts (number of individual detections per occasion, argument, argument output = "count") or binary (was an individual detected during an occasion, argument output = "binary"). note that the detector type will change accordingly: "proximity" if output = "binary" and "count" if output = "count".
data(recordTableIndividualSample) # create camera operation matrix (with problems/malfunction) camop_problem <- cameraOperation(CTtable = camtraps, stationCol = "Station", setupCol = "Setup_date", retrievalCol = "Retrieval_date", writecsv = FALSE, hasProblems = TRUE, dateFormat = dateFormat ) sdh <- spatialDetectionHistory(recordTableIndividual = recordTableIndividualSample, species = "LeopardCat", output = "binary", camOp = camop_problem, CTtable = camtraps, stationCol = "Station", speciesCol = "Species", Xcol = "utm_x", Ycol = "utm_y", individualCol = "Individual", recordDateTimeCol = "DateTimeOriginal", recordDateTimeFormat = "%Y-%m-%d %H:%M:%S", occasionLength = 10, day1 = "survey", includeEffort = TRUE, timeZone = "Asia/Kuala_Lumpur" ) # missing space in species = "LeopardCat" was introduced by recordTableIndividual # (because of CRAN package policies). # In your own data you can have spaces in your directory names. summary(sdh) plot(sdh, tracks = TRUE)
Input for subsequent analyses (multi-season)
Since version 1.2, camtrapR can prepare input for multi-season occupancy and spatial capture-recapture analyses (in unmarked and secr). The process is very similar to the one for single-season models.
Multi-season occupancy analyses
camtrapR prepares detection histories that can be used as input for argument y in function unmarkedMultFrame. unmarkedMultFrame creates the input for function colext which fits multi-season occupancy models.
For multi-season occupancy, we need a session column in the camera trap table. Records will be assigned to session automatically based on their dates.
First we simulate data for 2 seasons by duplicating the existing sample data and adjusting the dates (add 1 year)
# load multi-season data data(camtrapsMultiSeason) data(recordTableSampleMultiSeason) # also, for clarity, lets remove all unnecessary columns from the record table recordTableSampleMultiSeason <- recordTableSampleMultiSeason[, c("Station", "Species", "DateTimeOriginal")] # create camera operation matrix camop_season <- cameraOperation(CTtable = camtrapsMultiSeason, stationCol = "Station", setupCol = "Setup_date", sessionCol = "session", retrievalCol = "Retrieval_date", hasProblems = TRUE, dateFormat = dateFormat ) # plot camera operation matrix par(oma = c(0,7,0,0)) camtrapR:::camopPlot(camop_season) # make multi-season detection history DetHist_multi <- detectionHistory(recordTable = recordTableSampleMultiSeason, camOp = camop_season, stationCol = "Station", speciesCol = "Species", species = "VTA", occasionLength = 10, day1 = "station", recordDateTimeCol = "DateTimeOriginal", includeEffort = TRUE, scaleEffort = FALSE, timeZone = "UTC", unmarkedMultFrameInput = TRUE ) DetHist_multi
Note that the function makes sure that all seasons have identical number of occasions, and that all stations are represented, even if they were not sampled in all seasons.
From here, the resulting object can be used in unmarked.
year_matrix <- matrix(unique(as.character(camtrapsMultiSeason$session)), ncol = length(unique(as.character(camtrapsMultiSeason$session))), nrow = length(unique(camtrapsMultiSeason$Station)), byrow = TRUE) # this is a made up example table with station covariates for demonstration site_covariates <- data.frame(Station = rownames(DetHist_multi$detection_history), elevation = c(100, 200, 500, 300), treecover = c(80, 100, 50, 10)) umf <- unmarked::unmarkedMultFrame(y = DetHist_multi$detection_history, siteCovs = site_covariates, yearlySiteCovs = list(year = year_matrix), obsCovs = list(effort = DetHist_multi$effort), numPrimary = 2) colext_example <- unmarked::colext(psiformula = ~ treecover, # First-year occupancy gammaformula = ~ 1, # Colonization epsilonformula = ~ 1, # Extinction pformula = ~ effort, # Detection data = umf, method="BFGS") summary(colext_example)
Multi-season spatial capture-recapture
For secr, session IDs must be a gapless sequence of numbers beginning with 1. So 2009 becomes season 1 and 2010 season 2
camtrapsMultiSeason$session[camtrapsMultiSeason$session == 2009] <- 1 camtrapsMultiSeason$session[camtrapsMultiSeason$session == 2010] <- 2
# we also want a few records in season 2 recordTableIndividualSample_season2 <- recordTableIndividualSample[1:10,] recordTableIndividualSample_season2$DateTimeOriginal <- gsub("2009", "2010", recordTableIndividualSample_season2$DateTimeOriginal) recordTableIndividualSample_season <- rbind(recordTableIndividualSample, recordTableIndividualSample_season2) # for clarity, lets remove all unnecessary columns recordTableIndividualSample_season <- recordTableIndividualSample_season[, c("Station", "Species", "Individual", "DateTimeOriginal")]
# create camera operation matrix (with problems/malfunction), same as above for multi-season occupancy camop_season <- cameraOperation(CTtable = camtrapsMultiSeason, stationCol = "Station", setupCol = "Setup_date", sessionCol = "session", retrievalCol = "Retrieval_date", hasProblems = TRUE, dateFormat = dateFormat ) # create capthist object sdh_multi <- spatialDetectionHistory(recordTableIndividual = recordTableIndividualSample_season, species = "LeopardCat", output = "binary", camOp = camop_season, CTtable = camtrapsMultiSeason, stationCol = "Station", speciesCol = "Species", sessionCol = "session", Xcol = "utm_x", Ycol = "utm_y", individualCol = "Individual", recordDateTimeCol = "DateTimeOriginal", recordDateTimeFormat = "%Y-%m-%d %H:%M:%S", occasionLength = 10, day1 = "survey", includeEffort = TRUE, timeZone = "Asia/Kuala_Lumpur", stationCovariateCols = "utm_y", # made up, a potential site covariate from camtrapsMultiSeason individualCovariateCols = "Individual" # made up, a potential individual covariate from recordTable ) summary(sdh_multi) par(mfrow = c(1,2)) plot(sdh_multi) secr.fit.example <- secr.fit(capthist = sdh_multi, start = c(-1, -2, 10)) # with starting values, since its only very little data summary(secr.fit.example)
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.