Quantifying Free-Roaming Horse Body Condition: A Methodological Investigation Using Game Cameras and Observer Bias Corrections

Large populations of free-roaming horses (Equus caballus ) inhabit the western United States, Australia, and other arid and semi-arid rangelands globally. In the United States, horse body condition has been the impetus for gathering and removing horses under an emergency designation and has been noted as a common chronic mortality-causing condition. The Henneke body condition scoring (BCS) system is a qualitative index that includes visual appraisal and physical palpation at multiple anatomical locations on individual horses. This method has been shown to be repeatable between different observers with varying degrees of experience when assessing domestic horses at close proximity. Repeatability using the Henneke BCS system is unknown for free-roaming equids in vast environments, and adequately sampling populations in a minimally biased manner can be logistically challenging. However, estimating body condition using photographs has been shown to be a highly accurate alternative to proximal scoring for multiple scoring systems and species. Our study investigates if an automated, remote camera trap system can be used to estimate free-roaming equid body condition. We compiled existing camera trap data sets from 65 traps into a single data set including more than 3.7 million photos containing images of free-roaming horses from California, Nevada, Utah, and Wyoming. We employed veterinarian-trained animal science students to score horses within randomly sampled, spatially/temporally congruous sequences of animals in our data sets. We compared student scores with veterinarian scores to evaluate whether scorer-specific correction factors are necessary for estimating BCS parameters. Calibration to a fixed reference set reduced between-scorer variance to zero in our population-level BCS estimates. Analysis of our between-scorer variance in our empirical data implies this calibration step can be beneficial when different observers are scoring distinct horse populations across space and/or time.