The in-situ measurements were obtained from the World Ocean Database (WOD) (Boyer et al., 2018). In the study area (120˚W - 100˚W) of Amundsen Sea, there are total 21,754 profiling casts, 608 of which are CTD casts (as shown in Figure 1). All the profiles contain more than a million potential temperature-salinity (θ-S) data pairs. The reanalysis product used in this study are provided by the Global Ocean Reanalysis Multi-Model Ensemble Products (GREP), which contains four sets of global reanalysis data, C-GLORS (hereinafter CGLO), FOAM-GLOSEA5v13 (hereinafter FOAM), GLORYS2V4 (hereinafter GLOR) and ORAS5 (hereinafter ORAS). The uniform horizontal resolution of them is 0.25˚ in all 75 levels over each vertical column. In order to compare to a best one that is closest to reality among the four sets, the θ-S data of reanalysis products were interpolated to the sampling positions of all CTD casts and the Taylor diagram (Taylor, 2001) was plotted for comparison (Figure 2). It is found that FOAM is the closest product to the observation measurements under any indicators.

To validate further the accuracy of FOAM’s simulation of θ-S, the horizontal distributions of FOAM reanalysis and WOD observation data are shown in Figure 3. We calculated the maximum of time-averaged θ and S in the depths deeper than 300 m, where the most CDW is located (Wåhlin et al., 2012; Walker et al., 2013), from 1993 to 2019 in FOAM. Similarly, the maximum value of θ and S for each profile (deeper than 300 m) in WOD data were selected; to minimize overlap, 10% of the sampling points were randomly shown in Figure 3. For comparison, the WOD data were scattered on the coloring maps of FOAM. Even

if the simulations of FOAM are biased in central areas, it is still accurate in troughs; and the θ-S tendency of FOAM and WOD along latitude and longitude are almost identical.

The numerical investigation in this study is performed using the Regional Ocean Modelling System (ROMS) (Haidvogel et al., 2008; Shchepetkin & McWilliams, 2005); based on which a coupled sea ice-ice shelf-ocean general circulation model was developed (Wang et al., 2024). The average horizonal resolution of this model is ~6 km of the grid meshes in all 50 layers; more detailed model description and verification can be found in Wang et al. (2024). In this study, the climatological outputs of two numerical tests were used to compare variations in CDW intrusion at different area. The tests simulated climate change by enhancing westerly winds; and the detailed description and results are shown in Section 3.3.

The cross-shelf break CDW intrusions in Amundsen Sea have been well-studied near the troughs, which were considered the main CDW intrusion pathways (Kim et al., 2021; Lars & Isabella, 2021; Wåhlin et al., 2010). However, an observed high-temperature, high-salinity signal at shelf break, away from the deep troughs, was captured in WOD data. As shown in Figure 4, there are clear CDW

intrusions at the locations (near 118˚W, 113˚W and 102˚W) near the three major troughs WT, CT and ET. Although there is no deep-water trough in the vicinity of 106˚W, CDW signals have also been observed there. Due to the lack of observed flow velocity data near 106˚W, in order to further validate the existence of this intrusion site in SA, numerical particle tracking experiments were implemented through the FOAM.

As shown in Figure 5, the numerical particle tracking experiments were implemented at depth 300 m and 400 m. 11 particles were released at intervals of 0.1˚on the slopeat 106˚W - 107˚W. Some of particles were transported along the isobath to the east, and eventually they are transported by the ET to the south. Meanwhile, it is clearly found that there are many particles across the shelf break in SA, and they could be transported in the PIT within 9 months. Due to the PIT is connect to the ice shelves, the CDW intrusions in the SA also have a chance to reach the vicinity of the ice shelves.

Both observations and reanalysis product indicate that there are CDW intrusions in SA of Amundsen Sea. In order to weigh the heat contribution of in SA and compare the differences of heat contributions between SA and deep trough area, two zonal sections (S1 and S2 in Figure 1) were selected and the onshore temperature transport (TT) across the sections was calculated. As shown in Figure 6(a), Figure 6(b), it is suggested that S1 has the higher temperature than S2 and there is a shorter duration of the cold period (θ < 1.6˚C) and a longer duration

of warm period (θ > 1.8˚C) at section S2. Shifts in cold and warm periods in 2015 may be associated with El Niño events (Azaneu et al., 2023). As Figure 6 (c) shows, in year 2010-2019, there is the same magnitude of the temperature transport (negative value represents onshore transport) in S1 and S2. And it is clearly found that the heat contribution to the shelf of SA is more than that of CT during the most time after year 2016; in this period, the mean temperature transport of SA is 3.9 times as large as which of CT. Thus, the heat contribution into Amundsen Sea continental shelf of SA cannot be ignored.

As Figure 1 shows, the zonal winds in the Amundsen Sea are westerly winds outside the continental slope; and with the movement of the westerlies and the Amundsen Low under the climate change, the westerly winds have a tendency to increase (Schneider et al., 2015; Thoma et al., 2008; Thompson et al., 2011).Since westerly winds are a potential driver of CDW intrusions (Assmann et al., 2013;

Steig et al., 2012), predicting the intrusions in SA and deep trough area under climate change is meaningful. We did two sets of numerical tests; as the control group, Test 0 provided the climatological results in one year; as an experimental group, Test 1simulated the situation under climate change by adding 3 ms⁻¹ westerly winds to the test area (67˚S-72˚S, 130˚W-100˚W). The results of two tests are shown in Figure 7.

The averaged climatological zonal winds in test area of Test 0 and Test 1 are shown in Figure 7(a). Due to the spin up of model, the results of the first two months of the outputs are not analyzed. As Figure 7(b), Figure 7(c) shows, the strength of the cross-section flow is stronger in Test 1 than in Test 0 in the most of time. To further quantify the intrusion, cross-section volume transport Q was calculated:

Q = ∫ − H η ∫ x 1 x n u ⊥ d x d z . (1)

where η is the sea surface height; H is the seabed depth; u ⊥ is the cross-section flow velocity. As shown in Figure 7(d), the monthly results suggested that Q is significantly larger for Test 1 than Test 0 in all months except for in October, when they were close to each other, suggesting that the CDW intrusion in CT tends to strengthen under the climate change. However, the intrusion in SA is not sensitive to the variations of the westerly winds. As Figure 7(e) & Figure 7(f) shows, the flow velocity remained almost unchanged between the results of Test 0 and Test 1. Only in July was there a significant increase in on shore transport, as Figure 7(g) shows, but this was due to increased outflow on the

surface, not stronger bottom intrusion. Thus, in our model results, the CDW intrusion in shallow area seem to be unaffected by the variations of westerly winds.