Assessing trends and associated uncertainties in global ocean chlorophyll using Bayesian spatio-temporal techniques

Abstract

Climate change is predicted to affect oceanic phytoplankton abundance with impacts on fisheries and feedbacks on climate. The presence, magnitude, and even direction of long-term trends in phytoplankton abundance over the past few decades is still debated in the literature. The challenges affecting these studies include the low signal-to-noise ratio, the large degree of natural variability, and the shortness of the satellite ocean colour record, which is itself a composite of multiple shorter records. Previous work, however, has typically focused on using linear temporal models to determine the presence of trends in chlorophyll, where each grid cell is considered independently. To improve the assessment of trends a statistical model that explicitly models the relationship between neighbouring grid cells is used. A hierarchical Bayesian spatio-temporal model is fitted to global ocean colour data (1997 – 2013). This results in a notable improvement in accuracy in model fit, an order of magnitude smaller global trend, and larger uncertainty when compared to a model without spatial correlation. To help separate long-term trends from natural variability, trends from coupled physical-biogeochemical models are incorporated in to the model as Bayesian priors. The introduction of priors tends to reduce the magnitude and uncertainty of trend estimates, although the amount is deemed to be not statistically different from zero in any of the regions considered. Finally, the model is used to analyse the effect of taking into account discontinuities on estimated chlorophyll trends. The discontinuities considered are those relating to the launch and termination of individual ocean colour sensors. Considering discontinuities leads to statistically different trends in most regions, which can have a reversed sign as well as increased uncertainty. The improvement in trend estimate accuracy, and the more realistic representation of their uncertainty, emphasizes the solution that spatio-temporal modelling offers for studying global long-term change.

More information

Identifiers

ORCID for Matthew Lee Hammond: orcid.org/0000-0002-8918-2351

Catalogue record

Export record