Regional surface chlorophyll trends and uncertainties in the global ocean

Matthew L Hammond,Claudie Beaulieu,Stephanie A Henson,Sujit K Sahu

doi:10.1038/s41598-020-72073-9

Abstract

Changes in marine primary productivity are key to determine how climate change might impact marine ecosystems and fisheries. Satellite ocean color sensors provide coverage of global ocean chlorophyll with a combined record length of ~ 20 years. Coupled physical–biogeochemical models can inform on expected changes and are used here to constrain observational trend estimates and their uncertainty. We produce estimates of ocean surface chlorophyll trends, by using Coupled Model Intercomparison Project (CMIP5) models to form priors as a “first guess”, which are then updated using satellite observations in a Bayesian spatio-temporal model. Regional chlorophyll trends are found to be significantly different from zero in 18/23 regions, in the range ± 1.8% year−1. A global average of these regional trends shows a net positive trend of 0.08 ± 0.35% year−1, highlighting the importance of considering chlorophyll changes at a regional level. We compare these results with estimates obtained with the commonly used “vague” prior, representing no independent knowledge; coupled model priors are shown to slightly reduce trend magnitude and uncertainties in most regions. The statistical model used here provides a robust framework for making best use of all available information and can be applied to improve understanding of global change.

Highlights

In order to assess trends in global ocean color data (September 1997–June 2018), a Bayesian spatio-temporal model is used here
Note that the uncertainties here are defined based on the width of the 95% Highest Density Interval (HDI)
Prior information is provided by chl trends in CMIP5 model output covering the period 1997–2039

Summary

Introduction

In order to assess trends in global ocean color data (September 1997–June 2018), a Bayesian spatio-temporal model is used here. We demonstrated that this model produces a more accurate fit to chl observations than statistical models that do not account for spatial relationships within the d ata[24,25] This approach relies on ‘borrowing strength’, which takes advantage of the fact that trends in chl are likely to be similar at neighbouring grid p oints[26]. This approach provides both a full assessment of u ncertainty[24] and a framework for incorporating information from other sources through a prior distribution. We assess the sensitivity of the CMIP5 priors on the resulting trends and uncertainties

Methods

Results

Conclusion