The northwest North Atlantic shelf and slope waters are undergoing some of the fastest temperature increases globally (Caesar et al., 2018; Forsyth et al., 2015; Johnson and Lyman, 2020; Williams et al., 2021), a trend projected to continue under future climate scenarios (Caesar et al., 2018; Williams et al., 2021). The rapid warming is frequently attributed to a northward migration of the Gulf Stream (Gonçalves Neto et al., 2021; Seidov et al., 2021; Todd and Ren, 2023; Wu and He, 2024). The Gulf Stream (GS) is the western boundary current of the North Atlantic subtropical gyre and transports heat from the tropics to the mid-latitudes (Palter, 2015). Changes in its strength and position significantly impact weather and climate of the eastern North Atlantic and western Europe (Ma et al., 2024; Wenta et al., 2024). However, with respect to the rapid warming of the northwest North Atlantic, observations reveal that poleward shifts in the ocean gyres, including the North Atlantic subtropical gyre, are often not statistically significant (Chi et al., 2021; Yang et al., 2016, 2020). Bisagni et al. (2017) found that between 1993 and 2013, the GS actually moved southward while Todd and Ren (2023) documented a modest northward shift using observations collected predominantly between 2015 and 2023. The lack of consistent evidence challenges the narrative that a poleward shift of the GS explains the rapid warming of the northwest North Atlantic shelves.

A widely used proxy for diagnosing the location of the GS is the North Wall, defined by the 15 °C isotherm at 200 m (Fuglister and Voorhis, 1965). More recent studies have instead diagnosed the GS path using different isotherms, such as the 12 °C isotherm at 450 m (Sánchez-Román et al., 2024). These proxies, henceforth referred to as T15 and T12, are thought to be close to the maximum in GS surface current velocity. The T15 has been widely applied because long-term temperature records are available making it a convenient proxy. Many studies have assumed that changes in the North Wall, as indicated by T15, represent changes in the position of the GS, often using the two concepts interchangeably, to estimate both short- and long-term trends (Caesar et al., 2018; Hansen, 1970; Joyce et al., 2000; Seidov et al., 2019; Wu and He, 2024). However, there is a fundamental concern with using a fixed isotherm for diagnosing the GS path in a warming ocean: while the GS is likely to always be characterized by a strong north–south temperature gradient, it may not be the 15 or 12 °C isotherms, especially when background temperatures are rising (Garcia-Suarez and Fennel, 2024). Here we hypothesize that the T15 and T12 criteria overestimate a potential northward shift of the GS in a warming climate. Several previous studies have described limitations of temperature-based proxies for diagnosing the GS location. For example, air-sea heat fluxes can distort temperature gradients (Chen and Tung, 2018), and mesoscale eddies may bias the proxy northward (Chi et al., 2019). To the best of our knowledge, our hypothesis that background warming renders the temperature-based criteria unreliable has not be posed or systematically analyzed by others.

While the T15 and T12 proxies are still used, other commonly used proxies for the position of the GS are based on sea surface height (SSH) from satellite altimetry (Andres, 2016; Chi et al., 2019; Rossby et al., 2014). The SSH-based proxies build on the concept that large-scale surface circulation follows SSH contours with the strongest currents flowing along the most tightly packed SSH contours. Although SSH data are available since 1993, these methods may not yet offer a long enough record to fully capture trends in the position of the GS (Chi et al., 2021). Chi et al. (2021) estimate that an additional 22–23 years of SSH observations are required to detect statistically significant trends. Todd and Ren (2023) used a dense observational data set consisting of Argo float measurements (since 2001) and underwater glider measurements (since 2015) to reconstruct the three-dimensional water properties of the Gulf Stream. They decomposed temperature and salinity signals as they contribute to changes in density, isolated the warming trend from a shift in the GS axis, and documented a modest northward shift of 6±3 km per decade during their analysis period. Their method will become increasing useful as the record of autonomous observations grows.

The goals of this study are to (1) assess the reliability of the widely used temperature- and SSH-based proxies for tracking the position of the GS while the ocean is warming, and (2) test the hypothesis that the T15 proxy of the GS North Wall is inaccurate and overestimates a possible northward shift in a warming world. To this end, we analyze long-term preindustrial control and warming simulations from two widely used, high-resolution climate models, the CM2.6 model described in Delworth et al. (2012) and used in Saba et al. (2016), Caesar et al. (2021) and Claret et al.  2018) and the FOCI model described in Matthes et al. (2020) and subsequently used in Martin et al. (2022), Martin and Biastoch (2023) and Huo et al. (2024). We analyze a control and a warming simulation for each of the two climate models to explore the performance of different proxies over multiple decades into the future. Since the climate models are based on well-established dynamical principles, we believe they can be used to make inferences about the real ocean.

Our main finding is that the North Wall T15 criterion suggests a strong northward shift of the GS in the two widely used climate models that is not representative of the actual GS trajectory in those models. Since both models simulate pronounced shelf and slope warming, the results imply that such warming is not necessarily causally related to a northward GS shift. We thus caution against using temperature-based criteria in diagnosing changes in the GS path in warming scenarios.

The FOCI and CM2.6 climate models show a pronounced warming trend along the continental slope and over the adjacent shelf in response to increasing atmospheric CO₂ in their warming simulations (Fig. 1). Both models reasonably simulate the key features of the northwest North Atlantic circulation, including the GS and Labrador Current (Fig. 2). These circulation patterns are consistent with previous studies demonstrating the skill of both models in simulating Gulf Stream dynamics and mesoscale variability, including studies that focussed on shifts in GS position using CM2.6 (Saba et al., 2016; Caesar et al., 2021) and applications of FOCI to mesoscale and frontal dynamics in the region (Matthes et al., 2020; Martin and Biastoch, 2023; Martin et al., 2022; Huo et al., 2024). The models also capture regional currents such as the shelfbreak jet and Nova Scotia current in both the control and warming runs (Fig. 2b–d). These currents, however, are not resolved in satellite SSH observations due to the insufficient horizontal resolution (0.25°; Fig. 2a). In CM2.6, the GS velocity axis becomes diffuse near and beyond the New England Seamounts (Fig. 2d and e). Still, both models reproduce GS paths similar to those observed in satellite data (Fig. 2).

We tested T15, T12, and four SSH-based criteria for diagnosing the true GS path, defined as the maximum velocity or GS core. Among the SSH methods, the maximum SSH variance criterion performs poorly when applied to monthly averaged fields (Fig. S1 in the Supplement) and the skewness criterion underperforms in the CM2.6 control scenario, likely due to sensitivity to small-scale turbulence and eddies (Fig. S2). Despite its high resolution, CM2.6 does not appear to capture the sharp gradients needed for accurate skewness detection (Fig. S3). The other two SSH-based criteria appear more robust, with the criterion based on the location of a specific SSH isoline being closest to the axis of maximum velocity in both simulations (Fig. S4). Thus, criteria based on skewness, variance, and maximum gradient will not be used further in this study.

First, we consider the performance of the T15, T12 and SSH-based criteria in the year 2015 (Figs. 2 and 3). Close to Cape Hatteras (west of 74 °W), the GS is attached to the shelf break, and all criteria follow the axis of maximum surface velocity in both models and satellite observations (Fig. 2) to within 0.3° of latitude (33 km; Fig. 3). In observations, once the GS detaches from the shelf at ∼74° W, T15 stays close to the maximum velocity axis until reaching 67° W. Downstream (east) of this longitude, T15 diverges and is located to the north of the GS core by ∼1.4° (150 km; Fig. 2a), while the T12 and the SSH-based proxies track the GS core more closely (within 0.7° for T12 and 0.5° for SSH). The model simulations allow us to distinguish between preindustrial and warming scenarios. In the preindustrial simulations of both models, T15 is located northward of the GS core (by up to 1° in FOCI and 1.3° in CM2.6) while T12 and SSH are most faithful to the core of the GS in FOCI (within 0.2°) and biased northward by up to 0.5° in CM2.6 (dotted lines in Fig. 3c). In the warming simulations in 2015, both temperature proxies (T15 and T12) are markedly biased northward with T15 exceeding distances from the GS core of 2° in FOCI and up to 2.3°, or ∼250 km, in CM2.6 (solid lines in Fig. 3b and c) while the SSH proxy is equally close to the GS core in both pre-industrial and warming simulations. When comparing the northward bias of the two temperature criteria between preindustrial and warming simulations it appears that the northward bias roughly doubles in both models because of ocean warming. In summary, already in 2015 it appears that the T15 criterion is biased northward in observations and in the two warming simulations. A qualitatively similar bias applies to the T12 criterion although it is smaller in magnitude. The SSH criterion follows the GS core more faithfully.

Next, we consider the performance of these criteria in the warming scenarios up to 2100. Figure 4 shows the annual mean location of the GS path diagnosed by the T12, T15 and SSH criteria for the years 2015 to 2100. In FOCI it appears that the T12 criterion remains in place latitudinally, with some interannual fluctuations, while the T15 criterion moves northward, ultimately attaching to the shelf break along the Scotian Shelf and the Grand Banks, and the SSH criterion moves southward. In CM2.6, all criteria appear to suggest a northward movement that is most pronounced for T15 and smallest for the SSH.

A more quantitative assessment is presented in Fig. 5, where the evolution of the mean latitude of the GS core is presented for sections A1 (75 to 70° W) and A2 (71 to 50° W) from 2015 to 2100 (black line) as well as its diagnosed location using the T12, T15 and SSH criteria (orange, red and blue lines). Close to Cape Hatteras (section A1) the GS moves north slightly in FOCI and more pronounced in CM2.6. The T12 and SSH criteria track its location closely in both models. However, already in 2015, the T15 criterion is biased northward in both models, and the size of the bias grows until 2100. In both models, the rate of the northward movement of T15 is more than twice that of the GS core. In section A2 (east of 71° W) the GS core moves southward slightly in FOCI (by 0.02° per decade). This southward move is captured qualitatively by the SSH criterion (which moves southward by 0.06° per decade) but not by the T12 and T15 criteria which move northward by 0.02 and 0.13° per decade, respectively. In CM2.6, the GS core moves northward by 0.08° per decade. This movement is represented well by the SSH criterion which moves northward by 0.11° per decade. The latitude of the GS core is overestimated by the T12 and T15 criteria already in 2015 and the size of the bias grows until 2100. T15 suggests a northward migration of the GS of 0.2° per decade which is more than twice the actual northward migration of the core in CM2.6 of 0.08° per decade. In summary, in the warming simulations neither of the temperature criteria accurately represents the location of the GS core east of 71° W and both significantly overestimate its possible northward movement. Notably, both temperature criteria suggest a northward movement in FOCI east of 71° W even though the GS core moves southward in this simulation.

The growing discrepancy between temperature-based trajectories and the actual GS path in the warming scenarios is due to progressive warming of the slope region. In Fig. 6, we show the evolution of annual average temperature from 2025 to 2100 for the slope and subtropical regions at the surface and at 200 and 450 m depth for both the control (dotted lines) and warming (solid lines) simulations. Both models project warming in the slope and subtropical regions that exceed natural variability at all depths shown (Fig. 6). Warming in the slope region occurs at twice the rate of the subtropical region, with linear trends ranging from 0.12 to 0.29 °C per decade in the slope region, compared to 0.06 to 0.11 °C per decade in the subtropical region (Table S2). Within the slope region, the temperature rise is more pronounced at 200 m depth than at the surface and at 450 m in both models. FOCI projects a rise of 0.12, 0.22, and 0.17 °C per decade at the surface, 200, and 450 m depths, respectively. In CM2.6, the warming trends are 0.25, 0.29, and 0.18 °C per decade at the surface, 200 m, and 450 depth, respectively. The warming of the slope region results in an apparent northward shift of the GS path when inferred from a fixed isotherm as is done with both temperature-based criteria (especially with T15 at 200 m; to a lesser extent with T12 at 450 m depth), but the magnitude of this shift (and its direction in the case of A2 in FOCI) does not reflect actual shifts in latitude of the GS core.

Our main goal in this study was to assess whether temperature- and SSH-based criteria for diagnosing the path of the GS are reliable in a warming ocean. To address this, we analyzed preindustrial control and warming simulations from two high-resolution climate models, FOCI and CM2.6. Both models project a rapid warming of bottom waters on the northwest North Atlantic shelves (0.14–0.46 °C per decade) and in the slope region of 0.12–0.38 °C per decade in the upper 200 m, a rate that is two to three times faster than in the adjacent subtropical region. These projections are consistent with in situ observations reporting similar warming in the slope waters (Gonçalves Neto et al., 2021; Seidov et al., 2019; Todd and Ren, 2023).

In both models, the temperature-based criteria (T12 and T15) suggest pronounced northward shifts in the GS path in the warming scenarios that do not match the actual shifts of the GS core and do not occur in the pre-industrial simulations. In the pre-industrial control runs of both models, the SSH and T12 criteria closely predict the location of the GS core while T15 is biased northward (see Fig. 3b and c). In the warming simulations already in 2015, the T12 and T15 are located well northward of their pre-industrial analogs, while the SSH criterion remains closely aligned with the GS core (Fig. 3b and c). The same appears to be the case for observations (Fig. 3a), at least qualitatively, although we cannot separate the warming contribution from the pre-industrial, because observations for the latter are not available. The evolution of the mean latitude of the GS from 2025 to 2100 differs between FOCI and CM2.6. In FOCI, the GS core east of 71° W moves slightly southward by 0.02° per decade while T12 and T15 move northward by 0.02 and 0.13° per decade, respectively (Fig. 5). In CM2.6, the GS core east of 71° W moves northward by northward by 0.08° per decade while T12 and T15, both already located significantly further north than the GS core in 2015, move northward by 0.13 and 0.20° per decade, respectively (Fig. 5). In other words, while the North Wall closely follows the axis of maximum velocity near Cape Hatteras (west of 74° W), east of 71° W the GS location diagnosed by T15 is located significantly to the north of the GS core and moves further northward until 2100 at a much faster rate than the GS core (at more than twice the rate in the case of CM2.6; in the opposite direction in FOCI). The SSH criterion follows the location of the GS core well in all cases and thus appears to be a reliable proxy for diagnosing the GS location under warming conditions.

The climatological location of the North Wall has previously been reported to poorly match the GS high-velocity core. Chi et al. (2019) showed that the T15-based North Wall frequently lies north of the velocity maximum due to mesoscale eddy activity and frontal meandering. Frankignoul et al. (2001) demonstrated that the thermal front responds strongly to atmospheric forcing and does not consistently coincide with the dynamical jet. Pérez-Hernández and Joyce (2014) similarly found a poor agreement between a temperature-based index derived from T15 and altimeter-based Gulf Stream path indices between 2000 and 2011. These studies highlight present-day limitations of temperature-based indices. Our work goes further by showing that, in a warming ocean, background temperature increases systematically push T15 and T12 northward even when the GS core does not move, amplifying their bias under anthropogenic warming and making temperature-based criteria fundamentally inadequate for tracking the full extent of the GS path in such cases.

Our finding that the T15 criterion cannot reliably diagnose a northward shift of the GS is significant because a widely accepted hypothesis links the rapid ocean warming observed in the northwest North Atlantic to a northward migration of the GS (Caesar et al., 2018; Gonçalves Neto et al., 2021; Saba et al., 2016; Todd and Ren, 2023). This conclusion is largely based on the apparent shift of the North Wall, defined by the 15 °C isotherm at 200 m depth, i.e. the T15 criterion (Caesar et al., 2018; Seidov et al., 2019; Wu and He, 2024). These findings align with results from Alexander et al. (2020), who showed that enhanced warming on the northwest Atlantic shelf, particularly in the Gulf of Maine and Gulf of St. Lawrence, can result from multiple processes not necessarily linked to a northward migration of the Gulf Stream. These include local atmospheric warming, increased heat uptake, and the advection of warmer Arctic and Labrador Sea waters by the Labrador Current, as well as mesoscale eddy activity in the Gulf Stream region. Such mechanisms offer plausible alternatives for explaining regional warming trends without invoking large-scale latitudinal shifts in the GS.

Our analysis indicates that rising upper-slope temperatures bias the temperature-based estimates, making the North Wall an unreliable indicator of long-term circulation changes. While temperature signals are widely used in paleoclimate reconstructions due to their apparent links to ocean dynamics (Caesar et al., 2021; Palter, 2015), our findings suggest that care is needed when inferring past or future changes in northwest Atlantic circulation solely from temperature-based methods. Reconstruction of surface or subsurface temperature patterns in the Atlantic Ocean is a common approach to constraining the strength of the AMOC (Caesar et al., 2021; Rahmstorf et al., 2015; Thornalley et al., 2018). However, our results indicate that temperature patterns alone may provide a limited or biased representation of ocean currents dynamics. We caution against using the North Wall to infer future or present changes in the GS trajectory. Instead, we recommend using an SSH-based criterion.

The T15 and T12 criteria for the GS North Wall, both temperature-based proxies for the GS location, overestimate the northward shift of the GS in a warming ocean and thus create the misleading appearance of the GS rapidly migrating north. In contrast, the SSH-based criterion remains more closely aligned with the GS core, making it a more reliable tool for path estimations. The discrepancies between the temperature-based proxies and true GS core location arise because the slope region is rapidly warming. Our results call into question the notion that warming in the northwest North Atlantic is causally related to a northward migration of the GS. Future research should focus on elucidating alternative hypotheses for the observed warming in the slope region and on determining the underlying drivers.