by Meetkumar Patel

For my EUCURA research project, I studied predator disturbance events at a ground-nesting double-crested cormorant colony at Tommy Thompson Park with Professor Gail Fraser. It is important to understand predation because it can cause birds to move to other areas to nest. For example, some cormorants have recently moved to nest in trees on the Toronto Islands.

In this study, I paired trail cameras with acoustic monitoring to assess predator disturbances. Trail camera still images permitted identification of the predator species. The acoustic monitor recorded a cormorant "eh-hr" threat call and allowed me to measure the intensity (e.g., peak frequency and loudness) and length of each predator event. The image dataset was managed with a Python script according to the dates, time, and hierarchy. I also used a machine learning algorithms (Convolutional Neural Network) model to identify and classify predators. Raven sonogram software (Cornell) was used to analyze the spectrograms.

Prior to filtering for predators, the trail cameras captured a total of 12,482 still images. Still images revealed raccoons (n=807 images) as the main predator followed by coyotes (n=3 images) and bald eagles (n=8 images) (Figure 1), but not all images were independent events since the camera would take 3 consecutive images when triggered. A total of 590 audio files (approximately 34,588 minutes) over 27 days were recorded. Over 27 days, I measured a total of 110 disturbances. After correlating both data sets, I found 87 incidents (n=85 raccoons; n=1 coyote; n =1 bald eagle) where a disturbance occurred. In some cases (n=13), predators were captured on camera without vocal disturbances, while in others (n=12), audio disturbances occurred without corresponding predator sightings.

Predator-prey interactions within the cormorant colony were complex, and visual cues alone did not always correlate with recorded audio evidence and vice versa. The response of cormorants against predators was subtle, thus analyzing their behavior through both acoustic and visual monitoring was needed for a comprehensive insight into the impacts of predators on nesting colonies. I concluded that integrating acoustic recordings with trail camera imagery provides a highly effective method for assessing predator disturbances within the double-crested cormorant colony.

One of the important aims of the project was the use of artificial intelligence (AI) and machine learning in managing larger datasets to enhance the efficiency and accuracy of conservation monitoring efforts. The research project allowed me to train a Convolutional Neural Network model to identify and classify predators from the images captured by the trail cameras. The model architecture was designed to extract and process the visual patterns (key features) of different predators. The model was able to successfully distinguish between predator species, however, significant refinement and larger datasets were required to improve the accuracy of this application. The integration of artificial intelligence tools in conservation monitoring highlights the potential of AI in streamlining data management for conservation managers, however, currently available open-access tools often require extensive programming knowledge. I believe that future research to create user-friendly AI models and clear documentation to support non-technical users to leverage AI in ecological monitoring is required.

_________________________________

Meetkumar Patel is a recipient of EUC's Undergraduate Research Award (EUCURA) in Summer 2024. He is currently pursuing an Honours degree in Sustainable Environmental Management (SEM) at York University and also working as a Communications & Marketing Assistant at EUC. He has an engineering background and his program curriculum has a solution-based approach to problem-solving and decision making. The combination of science, management, and social approach in SEM is the key factor why he chose the program at EUC which will provide him the opportunity and potential to work in the field of Spatial Data Analysis and Geoinformatics in the future.