Predicting global oceanic net primary productivity with reduced-dimension, linear dynamical spatiotemporal models

Oceanic net primary production (NPP) accounts for roughly half of biological carbon fixation at the global scale, determining upper bounds for fisheries and export production. Assessing our ability to predict changes in NPP has thus major implications for the analysis of climate change impacts and for the management of living marine resources. Here, we fitted a series of reduced-dimension, linear dynamical spatiotemporal models to estimates of NPP derived from 18 years of remote sensing data and from simulations of a fully coupled, ocean-atmosphere Earth System Model (ESM). The method projects the evolution of NPP anomalies at the global scale from the time decay and interactions among a reduced set of major NPP modes of variability. This approach allowed us to assess potential limits to the predictability of NPP at the seasonal scale, and to assess the agreement in regional patterns of predictability based on remote sensing and ESM NPP estimates at large scales. The models are able to anticipate changes in NPP at lead times up to 24 months, especially in subtropical latitudes. Predictability was dominated by the decay of major modes at short time scales, with a prevalence of slow moving modes related to El Niño-Southern Oscillation and the Atlantic Multidecadal Oscillation. This decay dominated short-term predictions in both remote sensing and ESM NPP, although the modes prevailed for a longer time in ESM simulations. The dominance of short time persistence declined gradually at longer lead times as interactions between modes became more important for predicting NPP changes. These interactions were related to the propagation of major climate modes. Our approach provides a novel set of diagnostics to assess the performance of ESMs, and allowed us to identify potential regions where the prediction of NPP might lead to an improved management of living marine resources.