deep|tropics Project (ecoAI)

Deep Tropics

Cooling fan on one of three deep-learning rigs in the Behavioral Complexity Lab

Funded by University of Wyoming (Faculty seed grant) and Microsoft, the DeepTropics Project leverages a long-standing interest (and work) in bioacoustics (and large amount of amassed data) to understand more about [1] behavioral complexity in hyperspeciose tropical environments, [2] the structuring of tropical communities, and [3] the effects of habitat disturbance on individuals, populations, and communities. The DeepTropics Project (a component of ecoAI) combines remote acoustic monitoring and deep-learning to document avian mixed-species foraging aggregations (MSFA)–specifically, antwren foraging flocks–along environmental gradients in central, lowland Panama. The project has four phases:

  1. Test the efficacy of artificial intelligence (AI) algorithms to detect avian sound in noisy, hyperspeciose environments (tropical forests). Former University of Wyoming undergraduate Thaddeus Lipke did some preliminary work on this front, using ~65k songs and calls from birds in Panama. 
  2. Test the efficacy of AI to recognize species, count individuals, and construct interaction networks. Michael CastaƱo (University of Antioquia, Colombia) has produced an amazing dataset on antwren flocks that will be used to test this project component. 
  3. Evaluate the ecological factors impacting efficacy of deep-learning models, a novel approach made possible only by the large number of study plots established by our previous work in Panama
  4. Use AI algorithms to estimate variation in avian networks. 
One of 43 AudioMoth recorders to be deployed in Spring 2022 in one of 21 forest plots in central Panama.

This funded work helps establish an infrastructure in the Behavioral Complexity Lab for rapidly collecting and analyzing a wide variety of ecologically relevant data (from environmental noise to biodiversity to animal behavior) using bioacoustic tools. Future work includes:

  • Developing AI-based bioacoustic tools to rapidly detect rare or transient species in tropical environments, potentially aiding in conservation efforts. The ultimate goal is to create a wireless mesh network of acoustic sensors (with on-board AI processing ability) that will collect data in inaccessible habitats.
  • Determining the factors influencing breeding phenology of birds. To date, little is known (except from a few species) about what factors drive breeding phenology in the tropics. Deployment of recorders at 24 study plots for testing of broad-scale hypotheses about breeding phenology.
  • Quantify variation in anthropogenic noise and its impact on animal behavior. Do animals vocalize or move differently in noisy (i.e. more disturbed) environments?