Impact of Atmospheric and Model Physics Perturbations on a High‐Resolution Ensemble Data Assimilation System of the Red Sea

Abstract

The Ensemble Adjustment Kalman Filter of the Data Assimilation Research Testbed is implemented to assimilate observations of satellite sea surface temperature, altimeter sea surface height, and in situ ocean temperature and salinity profiles into an eddy‐resolving 4 km Massachusetts Institute of Technology general circulation model of the Red Sea. We investigate the impact of three different ensemble generation strategies: (1) Iexp—uses ensemble of ocean states to initialize the model on 1 January 2011 and inflates filter error covariance by 10%; (2) IAexp—adds ensemble of atmospheric forcing to Iexp; and (3) IAPexp—adds perturbed model physics to IAexp. The assimilation experiments are run for 1 year, starting from the same initial ensemble and assimilating data every 3 days. Results demonstrate that the Iexp mainly improved the model outputs with respect to assimilation‐free Massachusetts Institute of Technology general circulation model run in the first few months, before showing signs of dynamical imbalances in the ocean estimates, particularly in the data‐sparse subsurface layers. The IAexp yielded substantial improvements throughout the assimilation period with almost no signs of imbalances, including the subsurface layers. It further well preserved the model mesoscale features resulting in an improved forecast for eddies, both in terms of intensity and location. Perturbing model physics in IAPexp slightly improved the forecast statistics and also the placement of basin‐scale eddies. Increasing hydrographic coverage further improved the results of IAPexp compared to IAexp in the subsurface layers. Switching off multiplicative inflation in IAexp and IAPexp leads to further improvements, especially in the subsurface layers.

A hybrid ensemble adjustment Kalman filter based high‐resolution data assimilation system for the Red Sea: Implementation and evaluation

“…EAKFexp does not show such sporadic behaviour. It further improves the subsurface temperature and salinity biases particularly to the north of 20 • N. However, as already reported in Sanikommu et al (2020), EAKFexp misses high-resolution spatial features such as the deepening of the 23 • C isotherm around 26 • N (compare Figure 5e with 5a), and the intrusion of a fresh and cold Gulf of Aden water mass around 60 m (which manifests itself as a large overestimation of subsurface temperature and salinity south of 20 • N; Figure 5e-f). EnOIexp reproduces these features, but with significant discrepancies in the location of the deeper 23 • C isotherm and in the magnitudes of the temperature/salinity of the Gulf of Aden water mass appearing at the intermediate layers.…”

“…|W(z)| max is suspiciously large (compared to Fexp; Figure 6c with 6a) in EnOIexp almost throughout the Red Sea starting from the middle of the year. As argued in Sanikommu et al (2020), such a large |W(z)| max results from spurious vertical correlations in the quasi-static ensemble of the EnOIexp. EAKFexp does not show such sporadic behaviour.…”

“…Such spurious features are likely due to dynamical imbalances, which can be assessed through vertical velocities (e.g. Sanikommu et al, 2020). The maximum vertical speed in the water column, |W(z)| max , a proxy for 2D visualization of the abnormal vertical velocities in the ocean column, is suspiciously large in EnOInoSCLexp compared to Fexp (Figure 6a,b), suggesting important dynamical imbalances in the EnOI before scaling the quasi-static ensemble covariance.…”

“…All the assimilation experiments were conducted based on the routines provided in DART. Except for the differences discussed in the next two paragraphs, the experiments were all performed using the configuration described in Sanikommu et al (2020), with the same settings in terms of assimilated observations, assimilation cycle of 3 days, localization (only in the horizontal direction) using a radius of 300 km, and no-inflation during assimilation. The system assimilates the Reynolds et al (2007) level-4 daily sea-surface temperature (SST) data available on a 0.25 • × 0.25 • grid (which was prepared by blending in situ observations with data from the advanced very-high-resolution radiometer infrared satellite; Figure 1b), along-track satellite level-3 merged altimeter filtered sea-level anomalies (SLA; corrected for dynamic atmospheric, ocean tide, and long wavelength errors; Figure 1c) from the Copernicus Marine Environment Monitoring Service (CMEMS: Mertz et al, 2017), and in situ temperature and salinity ( Figure 1d) profiles from the EN4.2.1 dataset (Good et al, 2013).…”

“…This study proposes a new Hybrid-EAKF combining the EAKF system of Sanikommu et al (2020) and EnOI system of Toye et al (2017). The aim is to further improve the state estimate and model forecasting skill in the Red Sea while alleviating the computational load of the EAKF.…”