Towards operational 3D-Var assimilation of chlorophyll Biogeochemical-Argo float data into a biogeochemical model of the Mediterranean Sea

G Cossarini,L Mariotti,L Feudale,A Mignot,S Salon,V Taillandier,A Teruzzi,F D'Ortenzio

doi:10.1016/j.ocemod.2018.11.005

Abstract

The emerging availability of Biogeochemical-Argo (BGC-Argo) float data creates new opportunities to combine models and observations in investigations of the interior structures and dynamics of marine ecosystems. An existing variational data assimilation scheme (3DVarBio) has been upgraded and coupled with the Copernicus Marine Environment Monitoring Service biogeochemical model of the Mediterranean Sea to assimilate BGC-Argo chlorophyll profile observations. Our results show that the assimilation of BGC-Argo float data is feasible. Moreover, the proposed data assimilation framework provides significant corrections to the chlorophyll concentrations and is able to consistently re-adjust the shapes of chlorophyll profiles during surface blooms occurring in winter vertically mixed conditions, and in the case of the summer deep chlorophyll maxima.Sensitivity analysis and diagnostic metrics have been applied to evaluate the impact of assimilation and the relevance of different factors of the 3DVarBio method. A key factor is the observation error that has been tuned on a monthly basis to incorporate the representation error. Different frequencies of the assimilation cycle have been tested: daily or 3-day assimilation fosters the highest skill performances despite the reduced impacts and the increase of computational burden. Considering the present size of the BGC-Argo Mediterranean network (about 15 floats) and the estimated non-homogeneous correlation radius length scale (15.2 km on average), the chlorophyll assimilation can constrain the phytoplankton dynamics along the whole water column over an area up to 10% of the Mediterranean Sea.

Highlights

Operational ocean forecasting systems coordinate observations and modeling systems and have been widely recognized as important assets in monitoring the state of the ocean, addressing current challenges in increasing our understanding of the ocean and its role in society (Le Traon et al, 2017; She et al, 2016)
The observation error is a critical element of biogeochemical data assimilation (Teruzzi et al, 2014) because quantities are usually measured indirectly, leading to additional uncertainty (Dowd et al, 2014)
This result indicates that the specific biogeochemical processes taking place at depth in spring and summer might require a finer vertical model resolution to depths of at least 150 m in the Mediterranean Sea to fully resolve the vertical plankton structure observed by the BGC-Argo floats

Summary

Introduction

Operational ocean forecasting systems coordinate observations and modeling systems and have been widely recognized as important assets in monitoring the state of the ocean, addressing current challenges in increasing our understanding of the ocean and its role in society (Le Traon et al, 2017; She et al, 2016). Data assimilation is a well-developed practice in atmospheric science and in physical oceanography, whereas it remains a challenge in biogeochemical ocean modeling (Brasseur et al, 2009; Dowd et al, 2014). A not exhaustive list includes popular implementations of Kalman filter methods such as the singular evolutive extended Kalman (Brasseur and Verron, 2006; Fontana et al, 2013). The singular evolutive interpolated Kalman (Triantafyllou et al, 2013; Nerger and Gregg, 2007; Tsiaras et al, 2017) and the ensemble Kalman filters (Ciavatta et al, 2011, 2014; Cossarini et al, 2009; Hu et al, 2012). Variational approaches that minimize a least-squares-based cost function include 3DVAR (Teruzzi et al, 2014) and 4DVAR (Song et al, 2016; Verdy and Mazloff, 2017) methods

Methods

Results

Discussion

Conclusion