About This Project
Due to the recent decline in Florida manatee populations, accurate population assessments are crucial. Current aerial surveys overlook undetectable manatees and don't track movements between wintering sites. Prior studies suggest manatees might have unique vocalizations, so we'll use data from rehabilitation centers to determine if we can reliably identify and count manatees based on their calls.
Ask the Scientists
Join The DiscussionWhat is the context of this research?
In past years, the Florida manatee population has declined, and it is expected to continue as manatees face a number of anthropogenic and environmental challenges.Manatee survival depends on wintering sites, so it's important to understand both the number of manatees using these sites and the manner in which they use them. Currently, manatee populations are estimated based on aerial surveys that tend to undercount manatees and do not provide information about movement between the sites. Since previous research indicated that manatees might have individually distinct calls, we propose to use these calls as evidence of the presence of a particular manatee at the site. We plan to train a deep learning model to identify and match these distinct calls both at the site and between sites.
What is the significance of this project?
This project seeks to establish an alternative way to identify individual manatee presence at the wintering site and to learn more about the potential utilization of different wintering sites by individual manatees. This is a non-invasive way to collect data and could be useful as an alternative (or complementary) method of counting manatees because it doesn't rely on visual observations, which can be challenging in several wintering sites with very turbid waters.
Moving between wintering sites might also mean that manatees are particularly vulnerable during transit.Recognizing this could help establish additional conservation measures to protect them, as manatees rely on wintering sites for their survival.
What are the goals of the project?
The project's goal is to develop a probabilistic model capable of counting manatees based on their distinct calls. This is grounded in the hypothesis that distinct calls can reliably indicate individual manatees, thus aiding in their counting and tracking. The model will have two components: a deep learning model to match calls within our dataset, and another model that links distinct call types to the number of manatees using Bayes' Theorem. We will validate the developed model on previously unseen acoustic data, asking it to estimate the number of manatees present in the pool at the time of recording (this value is known and can be compared to the model's estimations).
Budget
We have a multi-year, unannotated dataset from the manatee rehabilitation center, which includes behavioral data and manatee identity data/pool presence. This data must be prepared, cleaned, and annotated to make it ready for deep model training (1500 USD). Once the data is prepared, we will set up a pipeline for data storage, retrieval, and validation (500 USD). The next step will be model training, where the model will search for matching calls in our data pools (1500 USD). Once the model identifies the matches, we will analyze and validate the results based on our metadata and manatee presence data ((500 USD). Project management will also be necessary (600 USD), and the Experiment platform fee is 400 USD.
Endorsed by
Project Timeline
The unannotated data from the rehabilitation centers has already been collected. The estimation for project completion is around 12 months. The first step would be to annotate/prepare the data, and the pipeline for data retrieval and validation. Next, the model will be trained, and finally, the analysis will validate the obtained results, by using the ground truth for data collected in the rehabilitation center where the manatee's identity and presence in the pool are known.
Aug 24, 2023
Project Launched
Dec 28, 2023
Data Preparation and Annotations
Mar 20, 2024
Model training
Aug 29, 2024
Data analysis and validation
Meet the Team
Natalija Lace
Natalija Lace is a Principal investigator for Cetalingua Project. Natalija received her PhD from the University of Southern Mississippi and her main interests include form and function of marine mammal acoustic communication. You can read more about Natalija's research here.
Project Backers
- 4Backers
- 100%Funded
- $5,003Total Donations
- $1,250.75Average Donation