Inconsistent strategies to spin up models in CMIP5: implications for ocean biogeochemical model performance assessment

Roland Séférian,Christoph Heinze,Anastasia Romanou,James C Orr,Tatiana Ilyina,Scott C Doney,John P Dunne,Olivier Aumont,Olivier Marti,Marion Gehlen,Laurent Bopp,James R Christian,Paul R Halloran,Keith Lindsay,Laure Resplandy,Jerry Tjiputra,Joachim Segschneider

doi:10.5194/gmd-9-1827-2016

Abstract

Abstract. During the fifth phase of the Coupled Model Intercomparison Project (CMIP5) substantial efforts were made to systematically assess the skill of Earth system models. One goal was to check how realistically representative marine biogeochemical tracer distributions could be reproduced by models. In routine assessments model historical hindcasts were compared with available modern biogeochemical observations. However, these assessments considered neither how close modeled biogeochemical reservoirs were to equilibrium nor the sensitivity of model performance to initial conditions or to the spin-up protocols. Here, we explore how the large diversity in spin-up protocols used for marine biogeochemistry in CMIP5 Earth system models (ESMs) contributes to model-to-model differences in the simulated fields. We take advantage of a 500-year spin-up simulation of IPSL-CM5A-LR to quantify the influence of the spin-up protocol on model ability to reproduce relevant data fields. Amplification of biases in selected biogeochemical fields (O2, NO3, Alk-DIC) is assessed as a function of spin-up duration. We demonstrate that a relationship between spin-up duration and assessment metrics emerges from our model results and holds when confronted with a larger ensemble of CMIP5 models. This shows that drift has implications for performance assessment in addition to possibly aliasing estimates of climate change impact. Our study suggests that differences in spin-up protocols could explain a substantial part of model disparities, constituting a source of model-to-model uncertainty. This requires more attention in future model intercomparison exercises in order to provide quantitatively more correct ESM results on marine biogeochemistry and carbon cycle feedbacks.

Highlights

ContextEarth system models (ESMs) are recognized as the current state-of-the-art global coupled models used for climate research (e.g., Hajima et al, 2014; IPCC, 2013)
Most of these approaches can be considered as “blind” given that they are routinely applied without considering models’ specific characteristics and treat models a priori as equivalently independent of observations. Since these models are typically initialized from observations, the spin-up procedure has the potential to exert a significant control over the statistical metrics calculated for each model, using those observations
It relies on a 500-year long spin-up simulation from a state-of-the-art Earth system model, IPSL-CM5A-LR to investigate the impacts of spin-up strategy on selected biogeochemical tracers and residual model drift across the various

Summary

Introduction

ContextEarth system models (ESMs) are recognized as the current state-of-the-art global coupled models used for climate research (e.g., Hajima et al, 2014; IPCC, 2013). The ESMs that contributed to CMIP5 substantially differed from each other in terms of their simulations of physical and biogeochemical components of the Earth system These differences in design translate into a significant variability between the skill with which the different models reproduce the observed biogeochemistry and carbon cycle, which in turn may impact projected climate change responses (IPCC, 2013). With the goal of constraining future projections, statistical metrics are often used for model ranking (e.g., Anav et al, 2013), weighting of model projections (e.g., Steinacher et al, 2010) or selection of the most skillful models across a wider ensemble (e.g., Cox et al, 2013; Massonnet et al, 2012; Wenzel et al, 2014) Most of these approaches can be considered as “blind” given that they are routinely applied without considering models’ specific characteristics and treat models a priori as equivalently independent of observations. Since these models are typically initialized from observations, the spin-up procedure (e.g. the length of time for which the model has been run since initialization, and the degree to which it has approached its own equilibrium) has the potential to exert a significant control over the statistical metrics calculated for each model, using those observations

Objectives

Methods

Results

Conclusion