Exploring long-term trends in marine ecosystems: machine-learning approaches to global change biology

The observation of marine systems and the acquisition of physical, chemical, biological, and ecological data have increasingly grown in the last century. Marine systems are undergoing multiple and overlapping impacts that menace ecological communities, but global change impacts on the ecosystem functioning are not easily assessable, for several reasons. For instance, because time-series datasets can include a sizeable amount of sparse and missing values, and the physical-biological interrelationships establishing within marine ecosystems can determine ample fluctuations, incoherent peaks and, overall, a high variance in the variables under study. Herein, we propose coupling linear statistics and different categories of Machine Learning techniques as a good compromise to construct and test suitable mechanistic models for hindcasting and forecasting the dynamics of key ecological processes, such as primary productivity in the open ocean.