Eisenhauer, E, Hanks, E. A lattice and random intermediate point sampling design for animal movement. Environmetrics. 2020;e2618. doi: 10.1002/env.2618

Researchers study animal movement for many reasons, including to understand how animals respond to human development, how disease spreads in an animal population, and where to focus conservation efforts. In these studies, researchers collect animal movement data, consisting of a list of animal locations at different times. Ideally, one would record the animal’s location as often as possible, but in most cases, there are limitations to the amount of data that can be collected. For instance, battery life limits the number of times tracking devices can record and transmit an animal’s location. Many animal movement studies collect data at regular time intervals at a scale that is within their limitations, e.g. every hour or every 24 hours. But while regular sampling is a convenient choice, is it really the best way to design a sample? Eisenhauer and Hanks (2019) propose an alternative sampling design for animal movement data which offers improvements over sampling at regular time intervals.
To understand the potential drawbacks of regular sampling, suppose one observes the position of an animal every 24 hours. If the animal is asleep in its den every 24 hours, the data would result in a poor estimate of the animal’s home range, for instance. While most examples will not be this extreme, sampling at regular intervals will always result in a loss of information about movement behavior at finer time scales. As an alternative, Eisenhauer and Hanks (2019) proposed a sampling design which combines samples at regular time intervals with additional samples at random times within each regular time interval. For example, instead of sampling every 5 seconds, their lattice and random intermediate point (LARI) sampling design would consist of samples every 10 seconds and one random time point within each 10-second interval.
Eisenhauer and Hanks (2019) illustrated the LARI design using simulation data, guppy data, and carpenter ant data. In all three cases, they started with very fine scale samples (e.g., the ant data was recorded every second), which they subsampled using both regular and LARI designs. They compared LARI and regular samples of the same sample size. In all three cases, they found that the LARI sample resulted in a better understanding of animal behavior (i.e., better estimation of model parameters) than the regular sample. Additionally, as shown in their simulation example, the LARI sample resulted in better estimation of space use (i.e., estimation of missing location data) than the regular sample. These results suggest that researchers might gain greater insight into underlying animal movement processes by choosing LARI sampling over regular sampling.