Variability-based constraint on ocean primary production models

Abstract

Primary production (PP) is fundamental to ocean biogeochemistry, but challengingly variable. Satellite models are unique tools for investigating PP, but are difficult to compare and validate because of the scale separation between in situ and remote measurements, which also are rarely coincident. Here, I argue that satellite estimates should be log-skew-normally distributed, because of this scale separation and because PP measurements are log-normally distributed. Whether they conform to this distributional shape is therefore a powerful variability-based constraint on such models. Satellite models that do follow a log-skew-normal may then also be concisely characterized by three parameters (log-mean, log-standard deviation, and log-skewness). I show that the output from a recent satellite model (CAFE) over 2019 agrees excellently with the log-skew-normal, globally and for most spatiotemporal subsets investigated here. The exception is the Northern Hemisphere winter, which may suggest future model improvements. PP by plankton is essential to ocean ecology and biogeochemistry, so satellite models that estimate PP from remote sensing data are indispensable in numerous scientific applications. However, because the corresponding in situ data are rarely measured when a satellite passes overhead, are measured on a much smaller spatial scale, and are highly variable, it is very difficult to compare different satellite models, evaluate how accurate they are, or to constrain their parameters. Here, a different approach to evaluating, comparing, and constraining these models is described, which accounts for or avoids all of these issues with the standard approach. This approach finds excellent agreement overall with a recent satellite model, while also identifying room for improvement.