BATScan: A radar classification tool reveals large‐scale bat migration patterns

Abstract

Abstract Bat movement and behaviour are still mostly understudied over large scales. High‐altitude, nocturnal activity makes visual identification of bats from the ground virtually impossible, dramatically hindering our ability to study their movement ecology. Despite the wide use of radar in aeroecology, its application to study specific taxa is limited due to incomplete target classification abilities. BATScan is a bat classifier for vertical‐looking radar data, which enables identifying bats and characterizing their unique aeroecology. We constructed the classifier using data from 10 radar deployments, covering a wide range of habitats on a central bird migration flyway over a 7‐year period, comprising ~18 million observations. We analysed animal migration above the Hula Valley, home to over 30 species of bats spanning a range of 5–150 g in size and exhibiting a variety of ecological characteristics. We distinguished bat‐labelled radar echoes for training according to phenology, morphology and movement ecology of bats, birds and insects. Several non‐bat datasets were constructed and joined to train classifiers under increasing levels of difficulty. Class imbalance in the resulting training data was handled using a generative adversarial network for up‐sampling the much smaller bat dataset. The resulting classification tool reached a high level of accuracy and precision, and was further scrutinized with an extensive set of ecological validations. Bats perform seasonal migrations over long distances, but little is known about the spatial and temporal characteristics of this movement, and the ability to study it at a large scale has so far been limited. We present the Israeli BATScan dataset, containing over 60,000 bat observations spanning the entire country and representing multiple habitats. Using this data, we produce an unprecedented large scale, highly detailed documentation of the yearly movements of bats on a major migration flyway, and distinguish this pattern from bird migration over space and time. So far, radar aeroecology dealt primarily with birds, increasingly with insects, and only rarely with bats. We present BATScan, a classification tool that can incorporate bats into the framework of radar aeroecology to finally enable a comprehensive description of animal aeroecology.

Keywords

• aeroecology
• bat migration
• classification
• flying animals
• generative adversarial network
• machine learning