Colombia
Protecting Colombia’s Rainforests Through Acoustic Monitoring
Overview
Rainforest Connection (RFCx) worked with the Humboldt Institute, FBC, the Colombian Ministry of Environment, and local scientists to better protect and understand the ecosystems and biodiversity of the middle Magdalena Valley.
The main goal was to protect critical ecosystems in Colombia from illegal activities and empower researchers, local agencies, and communities with the insights and tools to study species of interest through ecoacoustics, including the blue-billed curassow, jaguar, South American tapir, and spider monkeys.
Above: Annotated spectrogram with species calls from a 1-min recording
Partners
Humboldt Institute
Fundación Biodiversa Colombia (FBC)
HUAWEI Colombia
Objectives
Deploy 10 real-time, long-term monitoring Guardian devices across the region to detect illegal activities and ship 100 AudioMoth Edge devices to the local team.
Implementation of real-time threat detection and biodiversity monitoring capabilities.
Protect Colombia's critical ecosystems from illegal activities and empower researchers, local agencies, and communities with the insights and tools to study species of interest using ecoacoustics.
Alert for illegal activities associated with deforestation while simultaneously gathering data that can be used to track and monitor species over time.
Train the local team on deploying AudioMoth devices/Guardian installations, and on using RFCx tools (such as the Ranger App and Dashboard).
Train local partners and scientists on using the system to extract insights through a workshop and support calls.
Create the project on the RFCx platform, add the sites, and upload the recordings to each site (guided by RFCx)
Create species-specific identification models for 80 species of interest, determined in collaboration with the Humboldt Institute, and create species distribution models for the selected species.
Create a Regional CNN capable of identifying at least 40 species for continuous biodiversity monitoring.
The goal is to establish mechanisms for smooth data transfer to enhance data availability and conservation efforts, with several alternatives and adjustments considered to align practices with existing data formats.
Implementation
Several meetings took place between the RFCx team of ecoacoustics biologists and the Humboldt Institute to map out optimal locations for AudioMoth offline recording devices.
The RFCx team trained the local team in deploying AudioMoth devices. This does not require technical expertise from the on-the-ground partners.
RFCx shipped 100 AudioMoth Edge devices to the local team.
100 AudioMoth devices were strategically deployed across the Magdalenav Valley by the local team to capture the region's biodiversity and gather data for analysis.
RFCx supported the deployment of devices across the chosen sites a total of 3 times.
Additional surveys can be done using AudioMoth devices.
The local team created the project on the RFCx platform, added the sites, and uploaded the recordings to each site (guided by RFCx).
The RFCx science team analyzed the data to determine species presence and map 80 species of interest.
RFCx and the Humboldt Institute collaborated to determine biodiversity indicators and soundscape occupancy analyses.
The RFCx team of expert conservation biologists trained local partners and scientists in using the system to extract insights through a workshop and support calls.
The RFCx data science team created a Regional CNN capable of identifying at least 40 species for continuous biodiversity monitoring. Depending on the number and quality of the training data, more species could be included.
The results from the RFCx science team and the Humboldt Institute analysis were to be included in a detailed report submitted at the end of the 1.5-year project.
All data will be continuously available to the Ministry of Environment and Humboldt Institute via the Biodiversity Insights platform.
The study utilized an AI-powered automated species detection pipeline, trained using a Convolutional Neural Network (CNN), to analyze the acoustic data.
Integration of environmental variables into the occupancy models provided insights into how factors such as precipitation, forest structure, and vegetation health influence species detectability and occupancy.
Outcomes / Challenges
Outcomes
We detected a total of 151 species across 75 sites. This total consisted of 143 bird species, 4 amphibians, and 4 mammals (all primates. We detected 6 species threatened with extinction (IUCN Red List). Five of the detected species are endemic to Colombia.
Our analyses show that the forest structural condition index (FSCI) was an important predictor of occupancy for both frogs and birds. We also found that vegetation senescence influenced bird occupancy; sites with lower vegetation senescence (i.e., lower water stress) had a higher probability of bird occurrence. Results suggest that as vegetation senescence (i.e., stress or death) increases at a site, the probability of birds occupying that site decreases. This emphasizes the importance of healthy vegetation in supporting diverse bird communities.
(From the June Pilot study)
Analysis revealed moderate correlations between acoustic indices and remote-sensing variables depicting forest integrity, such as canopy height and wetness.
We identified the remote-sensing variable that correlates better with acoustic variables and the scale at which that correlation is strongest.
Sampling areas with wider gradients showed consistently weaker slopes in the relationship, implying a less sensitive response to ecosystem changes, although the models explained more variability.
When examining the relationship between acoustic and remote-sensing variables, we found significant trends with variables that depict forest integrity - such as the structural condition index, which includes canopy height, disturbance, and forest cover - as well as radar variables that capture forest height and a wetness index.
Our results indicate that property-level comparisons are more effective, as the correspondence between acoustic and satellite variables varies slightly from property to property and more precisely follows the habitat quality gradient within each property.
Our findings are built on similar studies that integrate acoustic and satellite data to better understand ecosystem heterogeneity and recovery during restoration.
Studies collectively suggest that acoustic data can complement satellite remote sensing by providing fine-scale, temporal ecological insights that align with broader spatial data from satellites.
Challenges:
Technical constraints include differentiating between chainsaw and grass-cutter noises, the need for canopy access for maintenance, and the lack of precise location data, which complicates validation in difficult terrains.
Need for significant technological refinements and better integration with existing frameworks to ensure effective forest monitoring and conservation efforts.
The model tended to misclassify audio clips as ‘species present’ when the species was actually absent.
It is important to acknowledge that the insights derived from the MSOMs are inherently limited by the subset of species for which we obtained PM detections.
