A workflow of open-source tools for drone-based photogrammetry of marine megafauna

Abstract

Drones have revolutionized researchers' ability to obtain morphological data on megafauna, particularly cetaceans. The last decade has seen a surge in studies using drones to distinguish morphological differences among populations, calculate energetic reserves and body condition, and identify decreasing body sizes over generations. However, standardized workflows are needed to guide data collection, post-processing, and incorporation of measurement uncertainty, thereby ensuring that measurements are comparable within and across studies. Workflows containing free, open-source tools and methods that are accommodating to various research budgets and types of drones (consumer vs. professional) are more inclusive and equitable, which will foster increased knowledge in ecology and wildlife science. Here we present a workflow for collecting, processing, and analyzing morphological measurements of megafauna using drone-based photogrammetry. Our workflow connects several published open-source hardware and software tools (including automated tools) to maximize processing efficiency, data quality, and measurement accuracy. We also introduce Xcertainty, a novel R package for quantifying and incorporating photogrammetric uncertainty associated with different drones based on Bayesian statistical models. Stepping through this workflow, we discuss pre-flight setup and in-flight data collection, imagery post-processing (image selection, measuring, linking metadata with measurements, and incorporating uncertainty), and methods for including measurement uncertainty into analyses. We coalesce examples from these previously published tools and provide three detailed vignettes with code to demonstrate the ease and flexibility of using Xcertainty to estimate growth curves and body lengths, widths, and several body condition metrics with uncertainty. We also include three examples using published datasets to demonstrate how to include measurement uncertainty into analyses and provide code for researchers to adapt to their own datasets. Our workflow focuses on measuring the morphology of cetaceans but is adaptable to other taxa. Our goal is for this open-source workflow to be accessible and accommodating to research projects across a range of budgets and to facilitate collaborations and longitudinal data comparisons. This workflow serves as a guide that is easily adoptable and adaptable by researchers to fit various data and analysis needs, and emergent technology and tools.

Keywords: Bayesian; Cetaceans; Drones; Marine mammals; Marine megafauna; Morphology; Multiple imputation; Photogrammetry; R package; Uncertainty.

Figure 1. A workflow for drone-based photogrammetry, from data collection to post-processing to analysis.

Details of each step are provided throughout this article.

Figure 2. Components of the drone-based photogrammetry workflow.

Examples of (A) LidarBoX installed on a DJI Inspire 2, (B) an image of GPS (UTC) time displayed from a BadElf GPS unit from an iPhone to sync LiDAR altitude with imagery, (C) recorded launch height to be added to barometer altitude, (D) a calibration object, here a 1 m board, and (E) body lengths and widths of a pygmy blue whale (Balaenoptera musculus brevicauda) measured in MorphoMetriX v2. Note, the A and B in the image refer to the “mirror side” function in MorphoMetriX v2, see Supplementary Material for more information.

Figure 3. Schematic of Xcertainty.

The red drone has a LiDAR altimeter, while the blue drone only has a barometer. All flights by both drones include measurements from images of calibration objects (length) and whales (body length and widths). Xcertainty can incorporate multiple measurements (from different images) of an individual to produce posterior distributions for body length and each width, as indicated by the variable number of images available for Whales 1-6. The point and bars (in body length and width outputs) represent the mean and the uncertainty (here as 95% highest posterior density intervals) of each measurement’s distribution.

Figure 4. Outputs from the body_condition function of Xcertainty, which automatically calculates standardized widths, body volume, projected area (orthogonal to the dorsoventral), and body area index.

Instead of the full distribution for each individual, the mean and uncertainty (here as 95% Highest Posterior Density Intervals) are shown, represented by the dot and vertical bars, respectively. Drones 1 and 2 correspond to the red and blue drones in Fig. 3.

Figure 5. Reproduced results from Pirotta et al. (2024) using the growth_curve_sampler function in Xcertainty.

(A) Outputs of modeling growth for male and female gray whales (n = 130) and (B) comparison of the uncorrected measurements (gray) and Xcertainty outputs (green) for two individuals over the study period using five different drones (represented by shape).

Figure 6. Comparing single-stage (full Bayesian) analysis and multiple imputation.

Estimating the linear relationship between total body length (TL) and rostrum-blowhole (RB) of Antarctic minke whales (n = 27). Each point with bars represents the mean and uncertainty (here as the 95% highest posterior density interval, HPDI) of the posterior distribution for each estimate (TL, RB). The blue solid line and shading represents the mean and HPDI, respectively, of the linear relationship estimated via multiple imputations, while the red solid line represents single-stage results from Bierlich et al. (2021b).