The Nesher Bari project has come together thanks to the work of the following rangers: Sonia Moushaev and myself, Raul Bermejo, (Data Rangers), Ayan Mukhopadhyay and Victor Anton (Project Rangers), and Ofer Steinitz, Ohad Hatzofe and Kaija Gahm (Biology rangers).
Given that neither Sonia nor I have a background in ecology or conservation, the first hurdle has been wrapping our heads around the topic. By talking to experts, we’ve gained a bird’s-eye view of the technical and non-technical challenges before we start coding up the solution.
The next hurdle is understanding the sources of data: how they are related and whether they can be combined to create more enriched datasets. This is a more Data Engineering endeavor that’s crucial to constrain what ML approaches we can take. For example, with a supervised learning approach, we could feed a model with examples of tracking data from vultures at their time of death. By detecting patterns in the data that go unnoticed by humans, the model could infer what inputs are best at predicting vulture mortality. However, this assumes that we have access to enough of these examples.
At the time of writing, we have three main sources of data. Firstly, a time-series dataset provided by INPA, corresponding to 10 years of tracking data for 162 vultures. Secondly, a time series dataset provided by UCLA and Tel Aviv University, corresponding to 2 years of tracking data for 110 vultures. Finally, a look up table with more detailed information about individual vulture information (status, age, release date, … , etc) and most importantly, whether they are alive and the date of their death. As this is a living project, it’s likely that we will incorporate further data to enhance our AI/ML solution.