README

Comparing uncertainty associated with 1-, 2-, and 3D aerial photogrammetry-based body condition measurements of baleen whales
Data and model code

Steps:
1. Run uncertainty model on total length and width measurements to produce a posterior predictive distribution for each measurement of each whale.
2. Calculate the different body condition measurements (1D, 2D,and 3D) for each individual using the uncertainty model outputs (the posterior predictive distributions for each measurement).
3. Analyze and compare the different body condition measurements.

 

1. Running the uncertainty model

Uncertainty model https://github.com/KCBierlich/Uncertainty_Model.
Bierlich KC, Schick RS, Hewitt J, Dale J, Goldbogen JA, Friedlaender AS, Johnston DW (2021) Bayesian approach for predicting photogrammetric uncertainty in morphometric measurements derived from drones. Mar Ecol Prog Ser 673:193-210. https://doi.org/10.3354/meps13814

The code is designed to be run using the drake package, by running the code in the script make.R either via the command line (i.e., R CMD BATCH make.R), or from within an interactive R session. The outline of all project components (i.e., the drake plan) is available in the R/subplans directory.

The drake Github page page has a good overview of what drake does, and includes some code snippets. After running the make.R script, the drake::loadd function is the most helpful function for loading output.

The drake documentation itself is also a good resource for learning how to use the drake package.
Starting with the Walkthrough is a decent place to just jump in to some details.

Data formats

The Error measurement model is implemented in the model_joint.R script. The model implementation is abstract, to allow multiple observations of any object. The data_plan.R script creates the necessary, raw data structures for the model. The flatten_data() function munges the data structures for use with the model’s implementation in nimble.

Training object information

Each training object must be documented in the same format as the training_obj object (below). The first two columns specify the training object, and the third column records its known length. The flatten_data() function will use the information in training_obj to associate known lengths with image and pixel measurements of the training objects.

## Warning: Auto-saved .RData file detected. Remove it to enhance reproducibility.

Subject	Measurement	Length
WAP-180611	Total length	1.400
WAP-180305	Total length	1.330
CA-180906	Total length	1.270
CA-170813	Total length	1.525
NC-190626	Total length	1.480

Image data

Each image must be documented in the same format as the Mns_images object (preview below). The first column Image provides a unique identifier for the image and will be used to create an altitude variable, which will be estimated from the altitude sensor readings provided by AltitudeBarometer and AltitudeLaser in conjunction with training data. The remaining columns in Mns_images report the focal length (mm), image width (pixels), and camera sensor width (mm) associated with the image, which will be used to estimate the image’s ground sampling distance (GSD) per pixel.

Image	AltitudeBarometer	AltitudeLaser	FocalLength	ImageWidth	SensorWidth
180829_A_F3_DSC09961.JPG	60.2411	56.39	35	6000	23.5
180829_A_F3_DSC00039.JPG	60.7411	56.98	35	6000	23.5
180830_A_F3_DSC00505.JPG	61.7111	60.38	35	6000	23.5
180830_L_F4_DSC01483.JPG	103.2718	103.77	50	6000	23.5
180830_L_F3_DSC01423.JPG	72.6719	73.79	50	6000	23.5

Pixel count “measurement” data

All pixel measurements must also be documented in the same format as the Mns_pixels object (preview below). The first two columns specify the measurement. One measurement variable, which will be estimated, will be associated with each unique combination of the first two columns. The Image column links the measurement to an image and it’s estimated GSD. Lastly, the PixelCount column records the pixel-length of the object as it appears in the image.

Subject	Measurement	Image	PixelCount
BW180829-30	TL	180829_A_F3_DSC09961.JPG	3211.9149
BW180829-30	TL.05.00..Width	180829_A_F3_DSC09961.JPG	240.0000
BW180829-30	TL.10.00..Width	180829_A_F3_DSC09961.JPG	331.9149
BW180829-30	TL.15.00..Width	180829_A_F3_DSC09961.JPG	385.5319
BW180829-30	TL.20.00..Width	180829_A_F3_DSC09961.JPG	434.0426

Many-to-Many relationships

Storing Image and Pixel count data in separate structures allows multiple measurements to be estimated from a single image. For example, total length and maximum width can be measured from the same image through a pixel count table like the following:

Subject	Measurement	Image	PixelCount
Animal A	Total length	Image 1	1000
Animal A	Max. width	Image 1	100

Both PixelCount entries above relate to a different object being measured, but use the same estimated altitude and implied GSD for Image 1 when estimating lengths from pixel counts.

Similarly, the Image and Pixel count structures also allow one object to be estimated from multiple images. For example, total length can be estimated from multiple observations through a pixel count table like the following:

Subject	Measurement	Image	PixelCount
Animal A	Total length	Image 1	1000.0000
Animal A	Total length	Image 2	1005.4916
Animal A	Total length	Image 3	999.1957
Animal A	Total length	Image 4	989.1591
Animal A	Total length	Image 5	1026.1055

Data

Data located in data folder

calibration_object-measurements.csv - training/calibration data for the LemHex-44 and FreeFly Alta 6.
Data from: Bierlich, K. C., Schick, R. S., Hewitt, J., Dale, J., Goldbogen, J. A., Friedlaender, A. S., & Johnston, D. W. (2020). Data and scripts from: A Bayesian approach for predicting photogrammetric uncertainty in morphometric measurements derived from UAS. Duke Research Data Repository. V2 https://doi.org/10.7924/r4sj1jj6s

• CO.ID = calibration object ID
• CO.Length = the true length of the CO.ID
• Image = image used for measuring calibration object
• Lab = the lab that collected the data
• Cruise = research expedicition ID
• Date = date that imagery was collected of calbiration object
• Flight = the flight number/name
• Pilot = pilot during data collection
• VO = visual observer
• Aircraft = the UAS aircraft used to collect imagery of calibration object to measure
• Focal_length = focal length of camera
• Iw = image width in pixels
• Sw = sensor width in mm
• pix.dim = pixel dimensions; Sw/Iw
• Baro_raw = the raw relative altitude recorded by the barometer
• Launch.Ht = the launch height of the drone, to be added to the BarAlt to get the absolute barometric altitude above the surface of the water
• Baro+Ht = the baro_raw + Launch.Ht to get the absolute barometer altitude
• Laser_Alt = the altitude recorded by the laser altimeter. Blanks spaces/NAs indicate no/false reading
• Altitude = altitude used in measurement, either Laser_Alt or Baro + Ht
• Altimeter = which altimeter was used for altitude in measurement; either barometer or laser
• Lpix = the length in pixels of the known sized object (calibration object)
• object_position = indicates if calibration object is in center of corner of image frame
• Analyst = analyst that performed the measurement

measurement_inputs.csv - total length and body width measurements for blue, humpback, and Antarctic minke whales. Measurements performed using MorphoMetriX

• AID = unique animal ID for individual whale
• Species = the species of whale
• Image = the image ID used for measuring the whale
• TL = total length of the whale in meters, measure rostrum to fluke notch
• TL.05.00..Width - TL.95.00..Width = width measurements in % increments of TL
• Reproductive_Class = actually pretty limited here, just “calf” or “adult”
• HTRange = Head to Tail Range used to calculate 2D and 3D body condition measurements
• SHrange = Short Range, only includes the three widths with the largest standard deviation for the population. Not included in the analysis
• SW = the Single Width (1D) measurement, width with the largest standard deviation for population
• BaroAlt = the raw relative altitude recorded by the barometer
• Launch_Ht = the launch height of the drone, to be added to the BarAlt to get the absolute barometric altitude
• LaserAlt = the altitude recorded by the laser altimeter. Blanks spaces/NAs indicate no/false reading
• Altitude = altitude used in measurement, either LaserAlt or BaroAlt
  • Launch_Ht
• Focal_Length = focal length (mm) of the camera used
• Iw = image width, in pixels
• Sw = sensor width, in mm
• alt_diff = percent difference between BaroAlt + Launch_Ht and LaserAlt

Steps

run ‘make.R’

This will generate an “output” folder that contains an ‘mcmc’ folder and a ‘reports’ folder.

• The ‘mcmc’ folder contains all the .rds outputs from the model.
  • “length_samples_mns.rds” contains the posterior predictive measurement distributions for TL and widths and is used to calculate body condition in the “calculating_body_condition” folder.
• The ‘reports’ folder contains .html files to evaluate the results from the model
  • For example, see “posterior_diagnostics_mns.html” for a summary of model outputs for each measurement

After model is finished running, proceed to calculate_body condition

2. Calculating body condition

In the “calculate_body_condition” folder, run “Calculate_body_condition.Rmd”.

This uses CollatriX to calculate body condition measurements.

Creates:
* whales_merged.csv – cleaned dataframe to input into CollatriX
* collated_allMC.rds – each animal’s dataframe containing the posterior distribution (the mcmc iterations) for each measurement
* collated_MC_summarystats.csv – CollatriX summary sheet with the mean and 95% HPD intervals for the posterior predictive distribution of each body condition metric for each individual. Used in body condition analysis.

3. body condition analysis

Run “Body_condition_analysis.Rmd”

This will run the analysis comparing the different body condition metrics and re-create figures from manuscript.

Contact

KC Bierlich, kevin.bierlich@oregonstate.edu

Core contributors:
Clara N. Bird
Dr. Josh Hewitt