Modeling change in ecosystem structure by combining lidar and ecoacoustics

Expanding biodiversity monitoring capabilities by integrating Earth observations with high-throughput sound recordings.

There is global interest in quantifying changing biodiversity in human-modified landscapes. Operational capacity to monitor known biodiversity is extremely limited, resulting in incomplete species inventories and sparse data coverage. Monitoring protocols that integrate sensors to capture a diversity of taxa and minimize the need for expert knowledge during data collection offer a promising pathway for routine monitoring of biodiversity.

Published in: Acoustic space occupancy: Combining ecoacoustics and lidar to model biodiversity variation and detection bias across heterogeneous landscapes. Ecological Indicators, 113, 106172.

Rappaport, D. I., Royle, J. A., & Morton, D. C. (2020).

By capturing aspects of the 3D physical interactions between forest structure and animal-generated sounds, we designed a novel modeling framework to predict biodiversity responses to variation in habitat structure, while correcting for imperfect detection of sound signals due to variation in habitat structure.

Ecoacoustics may offer a promising pathway for supporting multi-taxa monitoring, but its scalability has been hampered by the sonic complexity of biodiverse ecosystems and the imperfect detectability of animal-generated sounds. The acoustic signature of a habitat, or soundscape, contains information about multiple taxa and may circumvent species identification, but robust statistical technology for characterizing community-level attributes is lacking. Here, we present the Acoustic Space Occupancy Model, a flexible hierarchical framework designed to account for detection artifacts from acoustic surveys in order to model biologically relevant variation in acoustic space use among community assemblages. We illustrate its utility in a biologically and structurally diverse Amazon frontier forest landscape, a valuable test case for modeling biodiversity variation and acoustic attenuation from vegetation density. We use complementary airborne lidar data to capture aspects of 3D forest structure hypothesized to influence community composition and acoustic signal detection. Our novel analytic framework permitted us to model both the assembly and detectability of soundscapes using lidar-derived estimates of forest structure. Our empirical predictions were consistent with physical models of frequency-dependent attenuation, and we estimated that the probability of observing animal activity in the frequency channel most vulnerable to acoustic attenuation varied by over 60%, depending on vegetation density. There were also large differences in the biotic use of acoustic space predicted for intact and degraded forest habitats, with notable differences in the soundscape channels predominantly occupied by insects. This study advances the utility of ecoacoustics by providing a robust modeling framework for addressing detection bias from remote audio surveys while pre-serving the rich dimensionality of soundscape data, which may be critical for inferring biological patterns pertinent to multiple taxonomic groups in the tropics. Our methodology paves the way for greater integration of remotely sensed observations with high-throughput biodiversity data to help bring routine, multi-taxa monitoring to scale in dynamic and diverse landscapes.