Figure 8: (a) ROMS model MDT
contours, (b) Mid Atlantic Bight circulation, (c) CNES-CLS18 MDT
contours and (d) CNES-CLS22 MDT contours.
CNES-CLS18 MDT (Fig. 8 (c)) shows a slightly more organized circulation
on the shelf, although contours on the inner shelf are noisy. Coastal
currents in the Gulf of Maine emerge: Coastal currents on the Scotian
Shelf or in the Gulf of Maine are present but weak. In addition, there
are still MDT contours cutting the coast, and sea-level slope inversions
along the shelf.
The CNES-CLS22 MDT (Fig. 8 (d)) is also noisy on the continental shelf,
and there are still contours cutting the coast (associated with low
geostrophic velocities). Coastal currents on the Scotian Shelf are more
organized, but there still flows into the coast of central New Jersey.
So CNES-CLS22 MDT is a significant improvement, but that there is still
a way to go to bring MDT to the coast on broad shelves.
4.2.2 Quantitative validation with independent T/S
profiles
The CNES-CLS22 and CNES-CLS18 MDTs are compared with independent data.
Around 5% of the T/S profiles dataset are randomly selected from 2017
and kept for validation (independent data for CNES-CLS22 and CNES-CLS18
MDTs). The dynamic height estimated from T/S profiles (from a reference
depth) characterizes a baroclinic component of the dynamic circulation.
Whereas altimetry measures a height that is also influenced by
baroclinic processes occurring from the reference depth to the bottom of
the water column, and by barotropic processes. For validation purposes,
we choose to keep only the deepest profiles (with a reference depth of
1900m), thus reducing the number of deep baroclinic processes not taken
into account; this leaves us with a validation set of 2% of the
database.
As a first step, we compare the CNES-CLS18 and CNES-CLS22 MDTs against
these independent dynamic heights, by looking at the correlation of the
ADT (SLA+MDT considered) and the independent dynamic heights (figures
not shown). Correlations are calculated in boxes of 5° by 5° (with at
least 20 data) and are high, mostly between 0.8 and 1 for both MDTs, and
it is difficult to differentiate between them.
Secondly, Figure 9 (a) shows the mean bias per 5°X5° box between the ADT
estimated from the CNES-CLS22 MDT and the independent dynamic heights.
This global mean bias is 1.30 m, with spatial variations in the
Norwegian Sea and to the south near Antarctica (equivalent for
CNES-CLS18, not shown) and mainly represents the barotropic component
not observed by the dynamic heights. This is why it is removed from the
estimate of mean synthetic heights for the calculation of the MDT and
the following validation diagnosis on Fig. 9 (b) shows a comparison
between the variances of the differences (not considering the bias)
between the different ADTs and the dynamic heights, in percent. In blue
(in red), the variance of differences is reduced (increased) using
CNES-CLS22 compared with CNES-CLS18.
Globally, we see an improvement in CNES-CLS22 MDT compared with
CNES-CLS18, but this is not true in all regions. South of the Atlantic
and the Indian Ocean (as far south as Australia), the variance of
differences is reduced for CNES-CLS22 by more than 10% for many boxes
(even if boxes of strong reduction are juxtaposed with boxes of
increased variance of differences). Areas of degradation are
concentrated in the north-western Atlantic, particularly south of
Greenland, in the very north of the Pacific (along the Gulf of Alaska to
the Fox Islands, and close to Russia) and in the south of the Pacific,
where there are degradations of over 10% (also juxtaposed with
improvement boxes).