Exploring Marine Biogeochemical Prediction With Linear Dynamical Frameworks

Abstract Prediction of marine primary production (PP) forms the basis for forecasting marine ecosystems. Despite many prediction approaches, few have used linear dynamics to investigate PP predictability. Here, we implement Linear Inverse Models (LIM) to examine the predictability of PP in the eastern Pacific. Forecast skill substantially improves when moving from white‐noise‐only LIMs to those incorporating colored‐noise, revealing a role for temporally correlated climate forcing in extending predictability. Seasonally varying dynamics also contribute to enhancing forecast skill, particularly within the first annual cycle. Sea surface temperature and sea‐level provide the primary source of long‐term memory for extended forecasts. Building on this memory, the inclusion of the biolimiting nutrient iron further enhances predictive skill by interacting with these physical drivers, despite its inherently high short‐term variability. This insight has broader implications for marine biogeochemical prediction and data‐driven forecasting more generally, highlighting the role of memory interactions among predictors in improving forecast performance.