Lead supervisor: Julie Deslippe
I am working on the cusp of machine learning and ecology. I am interested in building predictive models and identifying climate change induced patterns for the state of New Zealand’s forests. My raw materials are the data from the National Vegetation Survey, which consists of tens of thousands of botanical surveys in plots nationwide over the last 50 years. I am looking for patterns of range shifts in widespread species in these plots using climate and topographical data.
However, rather than going with the classical ecological approach of fitting regression models through the data, I analyze it through a machine learning lens. Increases or decreases of abundance over time are classified using various machine learning models (SVMs, Random Forests, or neural networks, to name a few), which pick up on general patterns. These trained models are then probed with explainable AI techniques (XAI) to explain the model’s predictions rather than the data. Explainable AI has many advantages over classical statics, such as predicting where and why certain covariates are important. Additionally, the non-linearity of machine learning models allows the capture of non-linear patterns and species-specific threshold effects.
Knowing how the composition of forest species changes (and even being able to predict it) would mean that we can identify species at risk of extinction or large-range contractions. This, in turn, allows us to dedicate conservation resources more efficiently.
My PhD came late in life, and my previous careers always revolved around entrepreneurial pursuits in the IT sector. In my free time, I continued this ambition and founded the Effortless Academic. On this platform, I offer courses and webinars for academics who want to become more efficient through the use of tools, AI, and note-taking techniques.
Back to top