Project

For the past few summers, I have gone to the study site (Caples Creek Canyon) and conducted point counts. Point counts involve going to a point and counting all of the birds that I see (or hear). In addition to conducting point counts, I also moved ARUs (automated recording units) to different points in the study area. The ARUs collect audio data which captures birds vocalizing.

Right now, I am helping my mentors look through the audio data. The recordings from the ARUs were run through a machine learning algorithm that attempted to classify each 2.5-second sound snippet by the species that were present. This machine learning model was created by expert machine learnerers at Google. Due to the incredibly large number of recordings, this classification resulted in a very large csv containing all of the data. As of Spring 2022, I have created a script that is able to ingest and filter the data in order to reduce its size.

In the future, I hope to collaborate on the paper that the Caples team is going to write and then carve out my own question that I create as I am working with the researchers at the Academy of Sciences. My project will be mostly data analysis driven as all of the data collection has already been completed of will be completed during the summer as a group.

One of my preliminary thoughts on what I will be studying is to see how the "effort" affects the confidence of the models used in order to calculate the occupancy of the birds. "Effort" is basically a fancy word for amount of time that are we collecting data for. If we are able to collect less data and maintain the same amount of information, then we can design future studies for efficiently. By changing how much automated recording unit (ARU) and point count (PC) data is given to the model, I will be able to see how important each of them are for finding the occupancy of birds. This is very important as it will inform future studies on the optimal proportion of ARUs to point counts.