Information-theoretic analysis of realistic odor plumes: What cues are useful for determining location?

Many species rely on olfaction to navigate towards food sources or mates. Olfactory navigation is a challenging task since odor environments are typically turbulent. While time-averaged odor concentration varies smoothly with the distance to the source, instaneous concentrations are intermittent and obtaining stable averages takes longer than the typical intervals between animals’ navigation decisions. How to effectively sample from the odor distribution to determine sampling location is the focus in this article. To investigate which sampling strategies are most informative about the location of an odor source, we recorded three naturalistic stimuli with planar lased-induced fluorescence and used an information-theoretic approach to quantify the information that different sampling strategies provide about sampling location. Specifically, we compared multiple sampling strategies based on a fixed number of coding bits for encoding the olfactory stimulus. When the coding bits were all allocated to representing odor concentration at a single sensor, information rapidly saturated. Using the same number of coding bits in two sensors provides more information, as does coding multiple samples at different times. When accumulating multiple samples at a fixed location, the temporal sequence does not yield a large amount of information and can be averaged with minimal loss. Furthermore, we show that histogram-equalization is not the most efficient way to use coding bits when using the olfactory sample to determine location.Author summary: Navigating towards a food source or mating partner based on an animals’ sense of smell is a difficult task due to the complex spatiotemporal distribution of odor molecules. The most basic aspect of this task is the acquisition of samples from the environment. It is clear that odor concentration does not vary smoothly across space in many natural foraging environments. Using data from three different naturalistic environments, we compare different sampling strategies and assess their efficacy in determining the sources’ location. Our findings show that coarsely encoding the concentration of samples at separate sensors and/or multiple times provides more information than encoding fewer samples with higher resolution. Furthermore, coding resources should be focused on discriminating rare high-concentration odor samples, which are very informative about the sampling location. Such a nonlinear transformation can be implemented biologically by the receptor binding kinetics that bind odorants as a first stage of the sampling process. A further implication is that animals as well as computational models of algorithms can operate efficiently with a coarse representation of the odor concentration.