The pre-Argo ocean reanalyses may be seriously affected by the spatial coverage of moored buoys

Abstract

Assimilation methods, meant to constrain divergence of model trajectory from reality using observations, do not exactly satisfy the physical laws governing the model state variables. This allows mismatches in the analysis in the vicinity of observation locations where the effect of assimilation is most prominent. These mismatches are usually mitigated either by the model dynamics in between the analysis cycles and/or by assimilation at the next analysis cycle. However, if the observations coverage is limited in space, as it was in the ocean before the Argo era, these mechanisms may be insufficient to dampen the mismatches, which we call shocks, and they may remain and grow. Here we show through controlled experiments, using real and simulated observations in two different ocean models and assimilation systems, that such shocks are generated in the ocean at the lateral boundaries of the moored buoy network. They thrive and propagate westward as Rossby waves along these boundaries. However, these shocks are essentially eliminated by the assimilation of near-homogenous global Argo distribution. These findings question the fidelity of ocean reanalysis products in the pre-Argo era. For example, a reanalysis that ignores Argo floats and assimilates only moored buoys, wrongly represents 2008 as a negative Indian Ocean Dipole year.

The response of ocean parameters to tropical cyclones in the Bay of Bengal

“…Argo observations within ±10 days of analysis time are used. It is noted that the tropical mooring observations are not assimilated in the present analyses to avoid assimilation shocks (Sivareddy et al, 2017). The quality of INCOIS-GODAS analyses has been assessed in various assessments (e.g.…”

“…The quality of INCOIS-GODAS analyses has been assessed in various assessments (e.g. Ravichandran et al, 2013;Sivareddy, 2015;Sivareddy et al, 2017) including for tropical cyclone studies (Pothapakula et al, 2017). For instance, the INCOIS-GODAS compared to the in situ observations show a better SST representation with a root-mean-square error (RMSE) of 0.26 and 0.48 • C, respectively, for the pre-and post-monsoon seasons over the BoB (Pothapakula et al, 2017).…”

Impact of dynamical representational errors on an Indian Ocean ensemble data assimilation system

“…These coefficients are set as in the operational INCOIS‐GODAS, the quality of which was assessed on various occasions (e.g. Ravichandran et al ., ; Sivareddy, ; Pothapakula et al ., ; Sivareddy et al ., ). We are also motivated by the weak sensitivity of the dynamic RE experiments results to different values of this coefficient, as discussed in the Supplementary Information Appendix S1 (sections 1 and 2). Assim_DynMod – same as Assim_DynObs but uses model‐based dynamic REs as OEs after adding the uniform SE s of 0.75 °C and 0.35 psu for temperature and salinity, respectively.…”

“…These coefficients are set as in the operational INCOIS-GODAS, the quality of which was assessed on various occasions (e.g. Ravichandran et al, 2013;Sivareddy, 2015;Pothapakula et al, 2017;Sivareddy et al, 2017). We are also motivated by the weak sensitivity of the dynamic RE experiments results to different values of this coefficient, as discussed in the Supplementary Information Appendix S1 (sections 1 and 2).…”

“…While weak corrections limit the benefits of assimilated observations, large corrections may cause assimilation shocks, which can be generated by spurious corrections. These assimilation shocks propagate as planetary waves and degrade the analysis elsewhere, unless the observation coverage is sufficient enough to suppress them (Sivareddy et al, 2017). The use of spatially and temporally varying (here onwards dynamic) observational error may help mitigating these adverse effects by imposing proper weighting of the innovations, and is therefore crucial to obtain a reliable analysis state.…”