OM34A-3158: On the Use of Atmospheric Ensembles to Generate Physical-biogeochemical Ocean Model Uncertainties J. Karagiorgos 1 , V. Vervatis 1 , B. Lemieux-Dudon 2 , P. De Mey-Frémaux 2 , N. Ayoub 2 , S. Sofanos 1 http://www.oc.phys.uoa.gr/ – jkaragiorgos@oc.phys.uoa.gr 1 National & Kapodistrian University of Athens, Ocean Physics and Modelling Group, Athens, Greece 2 LEGOS, Laboratoire d’Etudes en Géophysique et Océanographie Spatiales, Toulouse, France Introduction In this study, we assess ocean model uncertaintes in a high-resoluton Bay of Biscay confguraton of NEMO [5], using atmospheric ensembles and stochastc ocean simulatons. Questons:  → Regional/coastal ocean-biogeochemical model uncertaintes?  → The response of a coupled ocean-biogeochemical model to diferent ensemble generaton methods?  → Is there a link between ocean processes and the ensemble spread?  → Potental beneft in real ensemble-based DA systems? Identcal subgrid confguraton of IBI36 [6] – 1/36 ◦ 2.1 km at 47 ◦ N; 50 geopotental levels – OB & inital conditons: PSY2v3 (1/12 ◦ ) – Atm. forcing: ECMWF oper. products (3h) – Period of study: 03-Dec-2016 to 30-Jun-2017 on-line coupled with PISCES-v2 [1] – biogeochemistry has no feedback on the physics The ensemble generation approaches We produce ocean ensembles over repeated periods to mirror the data assimilaton cycles used in most operatonal forecastng systems. The length of each ensemble ‘tme-chunks‘ is adjusted between 10 days and up to a month. This method gives access to performance categorizaton based on the age of errors within a given forecast lead tme. Compare diferent periods of the year with identcal age of errors. We compare ocean ensembles of 50 members using diferent generaton strategies: 1. Control run with the deterministc ECMWF-HRES (9 km). 2. Ocean ensemble simulatons incorporatng the ECMWF-EPS atm. ensemble product (18 km). 3. Ocean ensemble simulatons using stochastc perturbatons on the CR. 4. Merging the two above approaches. Figure 1. Schematc of the ensemble generaton approaches. The stochastc perturbatons based on SPPT and SPUF schemes as pro- posed by [3] and [2]. We have focused on uncertaintes from: Atm. Forcing: U,V wind components, Tair, and SLP. Model improper parametrizaton: momentum drag, latent and sensible heat, and botom drag coefcients. State variables: “Sources Minus Sink” (SMS) term of the ecosystem. Note that the SPPT scheme is similar to that implemented by [4] to sim- ulate uncertaintes in the ECMWF ensemble predicton system. This work was carried out as part of the Copernicus Marine Environment Monitoring Ser- vice (CMEMS) “Stochastc Coastal/Regional Uncertainty Modelling 2 (SCRUM2)” project and aims to contribute to ensemble-based ocean data assimilaton systems. All the simu- latons were performed at ECMWF’s High-Performance Computng Facilites using com- putatonal resources of ECMWF’s special project: (htps://www.ecmwf.int/en/research/special-projects/spgrver2-2018. Wind forcing: a primary source of ocean model errors Figure 2. Example of Wind speed (m/s) on May 01, 2017: (a-c) mean and (d-f) spread for the ensemble experiments. The black line represent the 200 m isobath. The ensemble means are similar to each other for all experiments (Fig. 2a-c) and to the CR deterministc run (not shown) The ECMWF-EPS forcing gives a small model spread for wind paterns that cover broad-domain scales. (Fig. 2a,d). The model wind spread presented in areas: – of robust wind speed regimes → EPS-STO & CR-STO. – with spatal wind speed gradients → EPS. RESULTS: Sea surface temperature The EPS ensemble has a smaller spread in the SST compared with the stochastc ensembles EPS-STO (& CR-STO). Ens. spread for EPS-STO: Winter ∼0.1 ◦ C/ Summer ∼0.5 ◦ C. The stochastc ens. spread seem identcal during winter, but their diferences lie in the spread peaks during summer. Figure 3. Temporal evoluton of ens. spread for the SST in model domain. A spin-up period of 10 days (overlapped in each tme-chunk) and a Usable Period with forecast lead tme of 20 days is presented. SST model errors: → Winter: flament-like paterns around eddies; associated with frontal actvity on the shelf (Fig. 4a). → Summer: broad paterns of uncertainty. Figure 4. (b-c): Snapshots of SST ensemble spread (in ◦ C) on 08 Feb. 2017 (b-c) and 19 Jun. 2017 (e-f). (a,d) corresponds to the SST for the CR on the same dates. RESULTS: Surface chlorophyll concentration The biogeochemical variables are considered as tracers with their evolu- ton determined by the advectve-difusive Eq.: ∂C ∂t = − A    ∇(uC ) − D h    K h ∇ 2 h C + D v    ∂ ∂z (K z ∂C ∂z )+ SMS (C ) C : PISCES 24 prognostc variables. A, D h and D v : advecton, hor. and vert. difusion respectvely. SMS : ”Source Minus Sink” term. Figure 5. (a) Snapshot of surface relatve vortcity ζ (in s −1 ) for the CR on 22 Mar. 2017. (b-c) correspond to ensemble spread of surface chlorophyll concentraton (in mg/m 3 ) for EPS and EPS-STO experiments respectvely. The chlorophyll spread follows the eddy mesoscale actvity (Fig. 5a).  → strong dependence of the biogeochemical feld on physical processes. High spread values above the antcyclonic eddies (e.g. 46 ◦ N 4.5 ◦ W ). Over the Armorican shelf, the progressively difuse processes of river plumes lead also to high chlorophyll spread values. Messages and future steps On the atmospheric ensembles and stochastc modelling optons  The stochastc perturbatons of ocean model can be considered as an efectve approach to generate upper-ocean uncertaintes in regional ocean models.  The EPS atm. forcing augments in a moderate way the ensemble spread in comparison to the stochastc modelling.  → it should have more impact in global confguratons. On the use of tme-chunk initalizaton:  Model uncertaintes with a forecast lead tme of a few days and up to two/three weeks appear to be comparable in magnitude to expected data errors for most observatonal networks.  → there is no need for long runs to augment model spread. Future steps include: Comparison of ensembles with observatons using probabilistc scores such as Brier and CRPS (work is ongoing). The use of our model ensembles in a physical-biogeochemical multvariate data assimilaton scheme. References [1] O. Aumont, C. Ethé, A. Tagliabue, L. Bopp, and M. Gehlen. PISCES-v2: an ocean biogeochemical model for carbon and ecosystem studies. Geoscientfc Model Development, 8(8):2465–2513, 2015. [2] J.-M. Brankart. Impact of uncertaintes in the horizontal density gradient upon low resoluton global ocean modelling. Ocean Modelling, 66:64–76, 2013. [3] J.-M. Brankart, G. Candille, F. Garnier, C. Calone, A. Melet, P. A. Bouter, P. Brasseur, and J. Verron. A generic approach to explicit simulaton of uncertainty in the NEMO ocean model. Geoscientfc Model Development, 8(5):1285–1297, 2015. [4] R. Buizza, M. Milleer, and T. N. Palmer. Stochastc representaton of model uncertaintes in the ECMWF ensemble predicton system. Quarterly Journal of the Royal Meteorological Society, 125(560):2887–2908, 1999. [5] G. Madec. NEMO ocean engine. Note du Pôle de modélisaton, Insttut Pierre-Simon Laplace (IPSL), (27):357pp., 2012. [6] C. Maraldi, J. Chanut, B. Levier, N. Ayoub, P. De Mey, G. Refray, F. Lyard, S. Cailleau, M. Drévillon, E. A. Fanjul, M. G. Sotllo, P Marsaleix, and the Mercator Research and Development Team. NEMO on the shelf: assessment of the Iberia–Biscay–Ireland confguraton. Ocean Sci, 9:745–771, 2013. OM34A. Advances in Ocean Data Assimilation, Forecasting, and Reanalysis VI Posters AGU Ocean Sciences Meeting | 16-21 February 2020, San Diego, CA, USA View publication stats View publication stats