Improving the predictability of plankton population dynamics using the portfolio effect

Individual species are often affected by complex mechanisms across multiple scales, making accurate predictions for individual taxa a challenge. Aggregating species data to reduce variability (i.e. the portfolio effect) is an approach that has previously been applied within fisheries restoration, conservation, and monitoring programs in order to reduce noise and examine long-term patterns and trends. Here, we use a 22 year phytoplankton dataset with 198 species from the Western Channel Observatory (1992 – 2014) to explore whether and how the portfolio effect can improve prediction of plankton population dynamics on monthly and longer timescales. Using a non-parametric framework to assess predictability, called empirical dynamic modeling, we show that the prediction skill is significantly affected by how species time-series are grouped together, observational error, and stochastic behaviour within species. Our study highlights the importance of the portfolio effect in ecosystem forecasting and provides monitoring programs a valuable tool that can guide the collection of data.