Mixed-layer salinity budget in the tropical Indian Ocean: seasonal cycle based only on observations

Abstract

The mixed-layer salinity (MLS) budget in the tropical Indian Ocean is estimated from a combination of satellite products and in situ observations over the 2004–2012 period, to investigate the mechanisms controlling the seasonal MLS variability. In contrast with previous studies in the tropical Indian Ocean, our results reveal that the coverage, resolution, and quality of available observations are now sufficient to approach a closed monthly climatology seasonal salt budget. In the South-central Arabian Sea and South-western Tropical Indian Ocean (SCAS and STIO, respectively), where seasonal variability of the MLS is pronounced, the monthly MLS tendency terms are well captured by the diagnostic. In the SCAS region, in agreement with previous results, the seasonal cycle of the MLS is mainly due to meridional advection driven by the monsoon winds. In the STIO, contrasting previous results indicating the control of the meridional advection over the seasonal MLS budget, our results reveal the leading role of the freshwater flux due to precipitation.

Appendix: error estimates

In this study, we have performed sensitivity tests before choosing the MLD criterion, the E-P dataset, the choice of the depth of the salinity vertical gradient at the base of the mixed layer, and the surface current products. We tested two density criteria (0.03 and 0.125 kg m⁻³) of the MLD which are used in several studies (e.g., de Boyer Montégut et al. 2004; Dong et al. 2009; Yu 2011), and we also used the MLD product of de Boyer Montégut et al. (2004) based on density criteria (0.03 kg m⁻³). The reference depth for the vertical density gradient is set to 10 m because of better data sampling than 0 or 5 m. For the E-P, we tested the product described in the data section with two reanalysis products: the monthly evaporation and precipitation dataset from the ERA-Interim reanalysis (Dee et al. 2011) of the European Centre for Medium-Range Weather Forecasts (ECMWF) available at 0.5° resolution and the monthly evaporation and precipitation dataset from the National Center for Environmental Prediction (NCEP) reanalysis 1 which are available at 2° resolution (Kalnay et al. 1996). The surface currents presented in the data section are tested with the near-surface velocity average at 15 m depth deduced from satellite-tracked drifting buoy observations. This product is available on a monthly mean climatology on a 0.5° × 0.5° grid (Lumpkin and Johnson 2013). The subsurface salinity at the base of the mixed layer is tested with three different values: salinity just at the mixed-layer base (S_h0) and salinity at 5 m (S_h5) and at 15 m (S_h15) below the mixed-layer base. Using different combinations of MLD, E-P, salinity at the mixed-layer base, and surface current products, we diagnosed 25 MLS tendencies time series for each box. For each of these combinations, observed and diagnosed MLS tendencies are compared and we quantify the similarity between the two estimates by computing the correlation coefficient and root mean square difference (RMSD) which are presented in the Taylor diagrams (Fig. 10). In the SCAS region, the Taylor diagram shows the sensitivity of the region to current products, and we found that the correlation coefficient between observed and diagnosed MLS tendencies is better with OSCAR than drifter currents. We find a small sensitivity when this current (OSCAR) is used with the three different E-P and also the three salinity values at the mixed-layer base. The RMSD between the MLS tendencies (using OSCAR currents) appears more important using different density criteria to compute the MLD than E-P. We noted that OAFLUX-GPCP and the salinity at 15 m below the mixed-layer base using OSCAR current and density criteria of 0.125 kg m⁻³ for the MLD give the best correlation (0.97), the smallest RMSD (0.25), and the best (closest to 1) standard deviation ratio (0.95).

Fig. 10

Taylor diagram in the SCAS and the STIO regions. Observed and diagnosed MLS tendencies are represented by points on a diagram where the correlation coefficient (R) between the observed and diagnosed time series is given by the azimuthal position, standard deviation of the observed or diagnosed time series is given by the radial distance from the origin, and the centered root mean square difference (RMSD) is given by the distance between the observed point and diagnosed point. REF is the reference experiment

In the STIO, the values in the Taylor diagram are more dispersed than in the SCAS. OSCAR and drifter currents give roughly similar results although drifter current slightly improves the correlation between the diagnosed and observed tendencies. This region is more sensitive to E-P, density criteria for MLD, and the choice of the depth of salinity gradient at the mixed layer. We selected the OSCAR current with OAFLUX-GPCP, salinity at 15 m below the mixed layer, and the density criteria 0.125 kg m⁻³ for the reference experiment in the two boxes. To complete sensitivity tests, we replace in the reference experiment, the MLD by the product of de Boyer Montégut et al. (2004) which is based on density criteria 0.03 kg m⁻³ (Fig. 10). In the SCAS, we have obtained RMSD = 0.32 instead of 0.24 with the reference MLD, while in the STIO, RMSD = 0.38 instead of 0.25 with the reference. These results suggest that the reference MLD used in this study is more appropriate to approach the salinity balance than the product of de Boyer Montégut et al. (2004).

All the sensitivity tests are used to estimate the error bar on the diagnosed MLS tendency. Following Da-Allada et al. (2013), standard error is estimated from all the diagnosed tendencies described above, for each month of the seasonal cycle.

For the observed MLS tendency, we first estimate monthly error (ε _S) in MLS as the standard error of all available observations for each month over the 2004–2012 period. Then, errors in MLS (ε _obs) are obtained following the formula of Foltz and McPhaden (2008): ${\varepsilon}_{\mathrm{obs}}=\left(\sqrt{\varepsilon_{{\mathrm{S}}_{t+1}}^2+{\varepsilon}_{{\mathrm{S}}_{t-1}}^2}\right)/\varDelta t$, with ∆t = 2 months.