Phytoplankton are responsible for nearly half of global primary production. In other words, they take nearly as much carbon out of the atmosphere as all land plants together. Consequently, understanding how they will respond to environmental change is an important piece of understanding how global carbon concentrations and temperatures will change in the coming decades and centuries. However, we presently have a weak understanding of how important environmental factors - temperature, light, nutrients - interact in complex ways to influence phytoplankton growth. A few experiments with one species have shown that these interactions could be extremely important in determining where different phytoplankton will be able to live and whether they will be able to grow. This will shape not just global carbon and temperature levels, but also the food available to aquatic food webs, and the probability and frequency of harmful algal blooms.
The experiments needed to accurately quantify these interactions are large and not feasible at present. Therefore, my work proposed applying machine learning methods to existing time series datasets of phytoplankton community composition and environmental factors, as an alternative to reach the same understanding. In effect, we would use natural variation in multiple environmental factors to draw inferences and learn, instead of experimental manipulation in the lab. We use machine learning instead of standard statistical approaches to capture the high-dimensional interactions that we presently do not understand well enough to write as equations for statistical fitting.
With this approach, our objectives are to (1) understand the shape of high-dimensional interactions and develop equations to describe them, that can then be used to improve Earth Systems Model predictions of environmental change, (2) quantify the traits of entire natural phytoplankton communities simultaneously, to enable the development of lake ecosystem models that can be parameterized accurately at a species level instead of the functional group or community level, (3) quantify the trade-offs experienced by phytoplankton species that govern patterns of population dynamics and coexistence, in order to better understand how changes in the environment will affect the composition of communities in the future.
