Figure 9: (a) shows the bias in centimetres between the dynamic heights estimated from the independent T/S profiles and the ADT calculated from the CNES-CLS2 MDT. And (b) shows the reduction/increase in variance of the differences between independent dynamic heights and ADT (from CNES-CLS18 and CNES-CLS22) in percent. In blue (in red), the variance of differences is reduced (increased) using CNES-CLS22 compared with CNES-CLS18. All these statistics are calculated in 5°X5° boxes, and only boxes with at least 20 data are kept.

4.2.3 Quantitative validation with independent drifters

The CNES-CLS18 and CNES-CLS22 solutions are compared with independent velocity data. 10% of the AOML drifters dataset are randomly selected and kept for validation (independent data). These drifters are not evenly distributed across the oceans. There are few drifters close to Antarctica (south of about 50°S), at the equator and in Artic. In addition, the Atlantic is slightly better sampled than the other basins, and in particular than the North Indian.
To compare to the MDTs, we remove the wind-driven current (Ekman current and wind slippage for undrogued drifters) from total drifters current and then data are filtered at inertia frequency (if inertia frequency is between 1 and 5 days, otherwise take a minimum of 1 day and a maximum of 5 days) to remove the tide and inertial waves. The objective is to keep only the geostrophic signal. As the geostrophy hypothesis is no longer valid at the equator, we exclude drifters between 5°S and 5°N at the equator.
Absolute dynamic topography values were calculated by adding the CMEMS gridded SLA to the new CNES-CLS22 MDT. Associated geostrophic current were then derived and interpolated along the drifter trajectories. Bias and Root Mean Square differences (RMSD) between the obtained geostrophic velocities with CNES-CLS22 and CNES-CLS18 and the drifter derived geostrophic velocities were calculated spatially by 5°X5° boxes. All these statistics are calculated with at least two different drifters and with at least 100 measurement points.
Comparisons of bias (a) in current modulus (in m/s) and (b) in direction (in degrees) are shown on Figure 10. In blue (in red), there is a decrease (increase) in bias for geostrophic currents derived from the ADT estimated with the CNES-CLS22 MDT. Globally, there is a decrease in bias in current modulus (Figure 10 a) using the new solution, with a greater decrease in bias in areas of strong currents: Gulf Stream, Kuroshio, Agulhas Current and Antarctic Circumpolar Current. The areas of degradation are South Greenland and South Kerguelen. For the directional bias (Figure 10 b), the contribution of the new solution is more mixed, as the direction of geostrophic currents derived from the ADT using the CNES-CLS18 is better in the South between the Kerguelens and southern Australia. These remain areas with fewer validation drifters.