The relevance of a right scale for sampling when studying high-resolution behavioral dynamics

Abstract

Many species used in behavioral studies are small vertebrates with high metabolic rates and potentially enhanced temporal resolution of perception. Nevertheless, the selection of an appropriate scales to evaluate behavioral dynamics has received little attention. Herein, we studied the temporal organization of behaviors at fine-grain (i.e. sampling interval ≤1s) to gain insight into dynamics and to rethink how behavioral events are defined. We statistically explored high-resolution Japanese quail (Coturnix japonica) datasets encompassing 17 defined behaviors. We show that for the majority of these behaviors, events last predominately <300ms and can be shorter than 70ms. Insufficient sampling resolution, even in the order of 1s, of behaviors that involve spatial displacement (e.g. walking) yields distorted probability distributions of event durations and overestimation of event durations. Contrarily, behaviors without spatial displacement (e.g. vigilance) maintain non-Gaussian, power-law-type distributions indicative of long-term memory, independently of the sampling resolution evaluated. Since data probability distributions reflect underlying biological processes, our results highlight the importance of quantification of behavioral dynamics based on the temporal scale pertinent to the species, and data distribution. We propose a hierarchical model that links diverse types of behavioral definitions and distributions, and paves the way towards a statistical framework for defining behaviors.

High-resolution ethograms, accelerometer recordings, and behavioral time series of Japanese quail

“…From all behavioral datasets straightforward estimations of time spent performing a given behavior, number and duration of behavioral events (i.e. continuous time spent performing a given behavior 10 ) and transitions between behaviors can be estimated from each animal (Fig. 4 ).…”

“…High resolution data, with sampling interval <100 ms, was only considered necessary for evaluating the male-females interaction test since social interactions can potentially occur at high speeds. Prior experimentation that found that locomotor (spatial displacement) events in Japanese quail can be as short as 100 ms 10 . Thus, given this reported possible high-speed transition between behaviors, accelerometer data was obtained at 25 Hz (i.e.…”

“…First, as mentioned previously, to evaluate the effects of attaching the accelerometer to adult quail on behavior dynamics, as well as, the potential habituation process 13 . Second, in the assessment of the probability distribution of the duration of adult male quail behavioral events within social environments 10 . Three, to begin training neural networks to detect male reproductive behavior from accelerometer recordings 14 , 15 .…”

“…Due to the high speed of reproductive behavior, the values of the offsets were corroborated and fine-tuned by visually comparing behavioral time series obtained from video recording with acceleration vectors. Focus was placed on body shacks since they are events that predominantly last 70 milliseconds 10 , and are clearly evident as a high amplitude peak in the three axes of the acceleration vector. It also should be recalled that, as stated previously, behavioral recordings are provided at a frame-by-frame basis, hence a resolution of 1/15 s while accelerometer recordings were obtained at a higher resolution of 1/25 s.…”

Thanks to repetition, dustbathing detection can be automated combining accelerometry and wavelet analysis

“…The accelerometers used were small with dimensions 9.5 × 15 × 4 mm and lightweight of 0.7 g. Each was set to gather data with a sampling frequency of 25 Hz (i.e. 25 data points per second) based on previous studies that show the possibility of high-speed transitions between behaviours in this species (Barberis et al, 2023).…”