Many bat species live in densely vegetated habitats which pose challenges to their biosonar systems. Among these challenges, the problem of identifying prey in clutter has received the most attention. Here, the far less well-studied problem of landmark identification in forest environments has been investigated. To this end, a large data set of about 220 000 foliage echoes has been collected along different tracks located in forested area. The echoes were recorded using a biomimetic sonarhead with flexible noseleaf and pinnae modeled on the periphery of horseshoe bats. Low-dimensional representations of these foliage echoes were created with the encoder portion of a variational autoencoder deep neural network architecture. The feature vectors obtained in this manner were subjected to clustering to determine whether the echo recordings exhibited continuous variability or fell into discernible clusters. The data silhouettes indicate the presence of a small number of distinguishable clusters in the echo data. Furthermore, mapping the assigned cluster labels to the geographical coordinates of the respective echo recordings revealed that different tracks were characterized by a different “fingerprints” of echo classes. Hence, these fingerprints could be a hypothetical basis for biosonar-based navigation in forest environments.Many bat species live in densely vegetated habitats which pose challenges to their biosonar systems. Among these challenges, the problem of identifying prey in clutter has received the most attention. Here, the far less well-studied problem of landmark identification in forest environments has been investigated. To this end, a large data set of about 220 000 foliage echoes has been collected along different tracks located in forested area. The echoes were recorded using a biomimetic sonarhead with flexible noseleaf and pinnae modeled on the periphery of horseshoe bats. Low-dimensional representations of these foliage echoes were created with the encoder portion of a variational autoencoder deep neural network architecture. The feature vectors obtained in this manner were subjected to clustering to determine whether the echo recordings exhibited continuous variability or fell into discernible clusters. The data silhouettes indicate the presence of a small number of distinguishable clusters in the echo data...