We use in-situ observations from river gauges in Scotland from the Scottish Environment Protection Agency, and significant wave height and wind data from the WAVENET network, oil rigs and buoys maintained by other European countries. Surge data came from the UK National Tide Gauge Network owned and maintained by the Environment Agency. We use the SST data from Operational Sea Surface Temperature and Ice Analysis (OSTIA; ) for the calculation of MHW and National Climate Information Centre (NCIC; ) daily gridded data for precipitation.

To determine the origin of air parcels, we use global operational Met Office forecasts starting on the day before the air parcels are started from. The global model allows us to use the Lagrangian Analysis Tool (LAGRANTO, ; ), to compute forward and backward parcel trajectories from gridded winds. LAGRANTO integrates parcel positions with a three-time iterative forward-Euler scheme and an internal step equal to 1/12 of the input interval; we use 3 h inputs (15 min interval), which are standard for synoptic-scale trajectory analyses . Winds and scalars are linearly interpolated in space and time between forecast fields. The model outputs parcel positions and interpolated diagnostics (e.g. pressure, temperature, specific humidity, potential temperature) at a user-defined (hourly in our case) sampling interval.

This study utilises the fourth generation of km-scale regional atmosphere-ocean-wave coupled system, as documented in a parallel paper , originally described by . A detailed description of the coupled model can be found in these papers; we summarize the main characteristics here. The domain is centered on the United Kingdom and covers all the Northwest European continental shelf seas, spanning approximately 46–63° N and 19° W–13° E. It operates on a rotated pole coordinate system, with the pole origin positioned at 177.5° longitude and 37.5° latitude. The system is fully coupled, facilitating two-way feedback between its components, which include the Met Office Unified Model (UM) for the atmosphere , the Joint UK Land Environment Simulator (JULES) for land surface with additional river routing provided by the River Flow Model , the Nucleus for European Modelling of the Ocean (NEMO) for the ocean , and WAVEWATCH III^® for the wave component .

In this article, we use UKC4 in two modes: (i) 2 month hindcasts started at 00:00 UTC 21 August 2023 and 21 August 2020, with atmosphere and ocean Lateral Boundaries (LBCs) reset to day 1 global forecasts every day to diagnose the drivers of the 2023 North Sea MHW by following the observed synoptic evolution and contrasting with a near climatological year (2020); (ii) 5 d ensemble forecasts started 00:00 UTC 17 October 2023, with atmospheric LBCs coming from MOGREPS-G global forecasts, and deterministic ocean LBCs from the global Marine Forecasting Centre (GLO-MFC) global ocean model PSY4 for the Atlantic and BAL-MFC for the Baltic Sea boundaries. The ensemble forecasts are used to quantify how the pre-existing MHW modified Storm Babet and its multi-hazards under identical large-scale forcing via a factual (ocean initialised 17 October 2023) and counterfactual (ocean initialised 17 October 2020) pair.

The ensemble setup consists of the 18-member Ensemble-Regional Coupled Suite framework, which integrates the Met Office Global and Regional Ensemble Prediction System over the United Kingdom (MOGREPS-UK, ) atmosphere-land ensemble forecast with the regional ocean-wave system. It includes one unperturbed reference simulation and 17 perturbed members, where atmospheric initial and lateral boundary conditions are generated by downscaling perturbations from the global MOGREPS-G system. While the atmospheric component is ensemble-based, the ocean and wave components use deterministic initial and boundary conditions, with coupling handled consistently across all members. The ensemble spread in ocean and waves therefore only grows through coupling. SST perturbations are applied from MOGREPS-G SST perturbations through the OASIS coupler, and only seen by the atmospheric system. They are kept fixed for the whole forecast and they are applied so that their average across all ensemble members is 0 and their maximum amplitude is 2 °C . These member-indexed SST perturbations are used operationally to correct under-dispersion of near-surface air temperature; they act as spread inflation rather than representing observational SST uncertainty. They are fixed in time and zero-mean across members so that reliability of 2 m temperature is improved without biasing the ensemble mean. It was chosen to keep these perturbations in the coupled system to stay as close as possible to the operational set-up, and to provide a test of robustness. Further technical detail on the ensemble set up is available in .

Two ensemble 5 d forecast simulations were triggered for Storm Babet, both starting on 17 October 2023. In the MHW experiment, the ocean initial conditions were taken from 17 October 2023, two days before the storm. In the no_MHW experiment, the ocean initial conditions were instead taken from 17 October 2020, chosen as a non-MHW reference state for the same calendar date, while the atmospheric initial conditions and simulation start date remained 17 October 2023. 2020 is chosen as the near-climatological baseline since its SST is closest to the long-term and 2003–2022 climatology across the North Sea region, providing a representative, dynamically consistent ocean state for the no_MHW run (Fig. a), and is a year for which an operational forecast is available (2018–now). A prescribed daily climatological SST baseline would represent a different experimental design and was not used here because the coupled forecast requires a physically coherent three-dimensional ocean initial state, which is not available from a climatological reference. The MHW is slightly underestimated compared to the observed anomaly in the North Sea (Fig. b, c). We can therefore consider the coupled experiment as a lower estimate of the MHW signal on the storm.

Although SST perturbations are applied to these runs in order to stay as close as possible to the operational set-up, individual members are perturbed in the exact same way between MHW and no_MHW ensembles, so that the SST difference seen by the atmosphere and shown in Fig. c is the same for each member pair, as the inter-ensemble differences reflect only the ocean initialisation.

We analyse two regions: the Central North Sea region (black polygon outlined in Fig. b) for the ocean response, and a fixed rectangular precipitation box in northeast Scotland (red rectangle in Fig. ). Hereafter we refer to this as the Eastern Scotland box [4.0° W, 55.5° N, 2.0° W, 57.5° N].

MHWs were detected following . We use OSTIA daily SST with a fixed climatology over 1982–2011 to compute the seasonally varying 90th-percentile threshold using a ± 11 d moving window. A MHW is defined as ≥ 5 consecutive days above this threshold, with gaps of ≤ 2 d merged. We use the scheme to assign categories which reflect event intensity. For category maps (Fig. b) we use a rolling 7 d majority (5-of-7) rule: for each grid cell, check the most recent seven days. If at least five of those days show a MHW (category 1 or higher), we plot the cell as MHW and use the category from the most recent day; otherwise we plot category 0.

To quantify how atmospheric forcing translates into ocean warming, we compute a mixed-layer heat budget on hourly output from the 2023 and 2020 hindcast regional coupled simulations over the Central North Sea (Fig. ). This region was chosen as a region of homogeneous hydrodynamic regime . For every grid cell in the box, we formed the budget terms and then averaged them over the region. The mixed layer temperature tendency dTdt can be expressed as a balance between surface heat fluxes, shortwave radiation penetrating below the surface, entrainment at the base of the mixed layer, and residual processes (advection, diffusion, and unresolved terms) as in . The governing equation is: dTtdt=QSW1-(1-0.58)e-htξswρCpht+QLH+QSH+QLWρCpht-ϵTt-1-Tdh(t-1)htdhtdt+R here, the term on the left, dT(t)/dt, represents the temperature tendency of the surface mixed layer (SML) averaged over hour t. The terms on the right-hand side represent the physical processes driving this change. The first two terms represent the net surface heat flux normalized by the thermal inertia of the SML (ρCph) where ρ is the seawater density (1027 kg m⁻³), Cp is the specific heat capacity of seawater (3850 JKg⁻¹ °C⁻¹) and h(t) is the SML depth (m). The first term accounts for the depth-dependent absorption of penetrating shortwave (QSW) represented using an exponential decay with an e-folding depth ξsw (15 m) following , and the second term combines the non-penetrating surface fluxes: latent heat (QLH), sensible heat (QSH), and longwave radiation (QLW). The third term quantifies the temperature change due to entrainment at the base of the mixed layer; this term is active only when the SML deepens (dh(t)/dt>0), as governed by the Heaviside function, ϵ. It depends on the temperature difference between the SML at the previous timestep, T(t-1), and the temperature of the water being entrained from below, Tdh(t-1). R is a residual term that includes all other processes, such as advection, diffusion and other non-resolved processes.

To identify the physical drivers responsible for the formation of the 2023 MHW, we contrast the atmospheric and oceanic conditions of that year with the same period in 2020, a year that did not experience a MHW and serves as a near-climatological reference. The analysis is focused on the Central North Sea region, which the storm air streams crossed before hitting Scotland.

The primary catalyst for the MHW was a sequence of persistent atmospheric anticyclonic weather regimes in September 2023. Weather regimes between 1 and 10 correspond to weak circulation, more typical of summer, while regimes between 20 and 30 represent strong circulation, more typical in winter, and regimes 10–20 are intermediate, most frequent in autumn and spring . As shown in Fig. a, the period leading up to and during the MHW's peak (20 August–20 September) was dominated by regimes 1, 3, 6, 12, 16 which are all characterised by weak winds, and most with anticyclonic conditions over the North Sea. High-pressure systems (anticyclonic conditions) suppress cloud formation and reduce wind speeds, favouring MHWs . The prolonged presence of these regimes created an extended window of clear skies and calm seas, maximizing the amount of solar radiation reaching the ocean surface. In contrast to 2023, the atmospheric conditions in 2020 (Fig. d) were characterised by variable weather regimes, with many occurrences of weather regimes above 10 and 20, with stronger winds. The frequent shifts between different patterns resulted in more unsettled weather with greater cloud cover and higher wind speeds, which intermittently limited the solar energy input to the surface mixed layer of the ocean.

The heat budget analysis of the two periods shows that the 2023 MHW was driven by a strong and sustained net heat gain in a shallow mixed layer. The event's evolution can be broken down into four distinct phases which are labeled P1–P4 in Fig. .

The initial phase of rapid formation (P1), occurring from approximately late August to September 8th, was initiated by persistent high-pressure weather regimes (3, 6, 16) or a weak circulation weather regime (1). This delivered an intense accumulation of shortwave radiation (orange line), creating a potential warming of +7 °C over the event's full duration. This forcing was critically amplified by a pre-conditioned shallow mixed layer depth (MLD) of only 5–20 m (Fig. c). The MLD, shown in panels (c) and (f), represents the upper ocean layer where temperature, salinity and density are relatively uniform due to turbulent mixing. It acts as the ocean's immediate heat reservoir, determining the volume of water over which the net surface heat gain or loss is distributed. This shallow layer concentrated the heat gain near the surface, and with minimal entrainment cooling during this period, the cumulative temperature tendency rose sharply (+1.8 °C).

The second phase (P2), lasting until 21 September, marked a transition where the rate of warming decelerated, as the anomaly deepened. Although strong solar input continued, significant cooling fluxes began to more effectively counteract it. The most critical change was the onset of substantial entrainment cooling (magenta line) as the mixed layer deepened from 8 to 30 m, which worked in concert with slow ongoing latent heat and longwave radiation losses. The mixed layer deepening was due to stormy conditions (weather regime 30) on 19–20 September. This mixing phase culminated around 21 September, when the heat budget was momentarily balanced (Cumulative dT/dt=0). This caused the reduction in intensity of the surface MHW in mid-September when the SST anomaly was approximately +1.5 °C above the climatological mean (Fig. a), but more importantly, it meant that the excess-heat anomaly deepened, storing heat below the surface in the North Sea and increasing the upper-ocean heat content available later in the season. This behaviour is consistent with glider observations (Figs. S1–S2 in the Supplement), supporting confidence in the model's vertical temperature structure.

Subsequently, the MHW transitioned into a prolonged phase of persistence (P3), lasting from late September through mid-October. This stage was characterized by a slow but steady net heat loss, as indicated by the gentle downward slope of the cumulative temperature tendency (Fig. b). This stability occurred as the deep mixed layer prevented entrainment cooling from below, while changes in weather pattern simultaneously suppressed the primary heating from solar radiation. With both major heating and cooling dynamics neutralized, the MHW's very slow decay was governed primarily by the persistent surface cooling from latent and longwave fluxes, allowing it to maintain a high intensity for nearly a month.

The prolonged period of persistence in P3 was brought to an abrupt end by Storm Babet around 20 October. It was triggered by a rapid cooling dominated by surface heat fluxes. In the shallow (40–50 m) Central North Sea, the water column was already well-mixed in P2, making entrainment a negligible factor. Instead, the storm’s intense winds drove a large increase in latent heat flux, forcing a rapid loss of heat from the sea surface to the atmosphere. This efficient heat extraction quickly dissipated the remaining thermal anomaly throughout the water column, bringing the MHW to a conclusion.

The 2020 period provides a climatological example. The weaker shortwave radiation input provided a more modest potential warming of only +4.5 °C for the same period. This energy gain was more than compensated for by the combined cooling effects of latent heat (-2.5 °C), longwave radiation (-2 °C), and entrainment (-1.5 °C) resulting in a net cooling of approximately -2 °C.

In summary, the 2023 MHW was the result of a synergistic combination of factors: persistent high-pressure weather regimes delivered an anomalous amount of solar radiation to the ocean, and anomalously weak winds favoured a shallow mixed layer, which trapped this heat at the surface, amplifying the increased solar radiation effect. The absence of this atmospheric forcing and oceanic susceptibility in 2020 prevented MHW formation and instead led to a net cooling of the surface ocean, more typical of September and October in this region.

The synoptic development of Storm Babet (Fig. ) was characterized by the interaction of two low-pressure systems and a blocking high-pressure area over Scandinavia. This configuration prevented the storm from progressing eastwards, resulting in a strong pressure gradient and sustained southeasterly flow over the North Sea lasting over 48 h.

The storm brought exceptional rainfall to parts of eastern Scotland, with 150 to 200 mm falling in the wettest areas (Fig. a). This was favoured by the persistent southeasterly flow (Fig. ), which transported moist air across the North Sea towards eastern Scotland and its topography. The Met Office issued two red warnings for rain . Heavy, persistent and widespread rain also affected much of England, Wales and Northern Ireland from 18–20 October, with 100 mm falling fairly widely across the UK (Fig. a). The storm also brought some very strong winds, with gusts over 50 Knots (58 mph) across north-east England and much of Scotland.

Storm Babet brought a combination of in-land and coastal damage; the river Tay and South Esk burst their banks; beaches in Musselburgh saw the equivalent of a 5-year erosion rate (3 m retreat in Montrose, near the mouth of river South Esk); harbour walls in North Berwick collapsed in sections; a North Sea drilling platform lost anchors; and the storm resulted in 7 deaths. Despite the storm occurring during neap tide, the persistence of the wind field led to a prolonged surge lasting for several tidal cycles.

Figure  shows that the ensemble forecast performs well over eastern Scotland for 5 d accumulated precipitation, the region where the red warning was in place. As this is a single case, formal quantitative verification is not meaningful, so we restrict ourselves to a qualitative assessment. In this region the forecast behaved near-deterministically. Elsewhere, higher observed totals fall within the ensemble envelope (seen in Fig. c), indicating the system can produce such amounts, but those regions are outside our scope. We therefore focus on the Eastern Scotland box in Fig. , the primary impact area with concurrent inland and coastal hazards; the forecast there is sufficiently accurate for the process analysis that follows.

Figure  presents a comparison between observations (black) and the coupled ensemble forecast (orange) for multiple hazards associated with Storm Babet. The panels show river discharge for the river Tay, surge and tide at Leith tide gauge, river discharge for the South Esk, and significant wave height at the Firth of Forth WaveNet buoy. In both the Tay and South Esk catchments (panels a and c), observed river discharge shows a rapid increase during the storm, consistent with the heavy rainfall. While the ensemble forecast reproduces the overall rising trend in river discharge, it lags 18 h behind the observations in capturing the peak. For each river, a few members capture the correct amplitude of river discharge increase generated by the storm, which shows good skill two days before the storm starts. Upstream observations show better model/observation agreement, which suggests the river speed is too slow in the coupled system (not shown).

At Leith (panel b), the tide gauge provides measurements of total water level, which can be decomposed into tidal and surge components. Although the event's occurrence during neap tide limited the overall hazard compared to a spring tide scenario, both high and low tides were higher relative to climatology by 0.3–0.6 m, reflecting the influence of storm surge. This storm surge was likely caused by the combined effect of low atmospheric pressure, which raises sea level, and strong winds that drove water toward the coast. Overall, some ensemble members of the coupled ensemble capture the surge amplitude of 0.3–0.6 m, although the ensemble mean shows a 0.1–0.2 m underestimation.

On the East Coast of Scotland, in a southeasterly wind regime, the significant wave height (panel d) is directly linked to wind forcing in the North Sea. Observations show significant wave heights exceeding 8 m, whereas the coupled ensemble mean peaks at around 6–7 m. Nevertheless, the timing of the wave peak is well captured. Figure S6 confirms this wave underestimation also against Central North Sea oil rig observations. Figure S5 indicates that the strongest winds (in the range of 20–25 m s⁻¹) off the coasts of Denmark are underestimated by 3–5 m s⁻¹ in all model members. Sensitivity studies conducted by showed the wave amplitude could be increased to some extent by changing the coupling strength between wind and waves. However, this deteriorated the winds. Therefore, the origin of the problem is thought to be a lack of free-tropospheric momentum entrainment into the unstable boundary layer. Further work is needed to improve this in the next atmosphere and land configuration of this model.

The highlighted red areas in Fig.  represent times when elevated river discharge, high tide, strong waves, and storm surge occurred simultaneously. It is this temporal overlap that produced the severe multi-hazard coastal impacts of Storm Babet. The coupled ensemble is more successful at capturing multi-hazard on the afternoon of 20/10 than in the morning, given its delay in river peak.

To explain the origins of the multi-hazard in Storm Babet, we performed air mass Lagrangian trajectories (Sect. 2.2) using the Met Office global deterministic forecasts to provide insight into the pathways and moisture history of air parcels feeding precipitation during Storm Babet. Backward trajectory analyses were conducted for two trajectory release times on 19 October at 04:00 UTC (Fig. ) and at 22:00 UTC (Fig. ), chosen to represent two different stages of the precipitation event and to examine whether the pathways and thermodynamic evolution of the air parcels varied through time. Air parcels were released from near-surface levels (950–970 hPa) over the target region (56.0–57.0° N, 1.0° W–0.0° E), with a spatial resolution of 0.2° in both latitude and longitude. Each parcel was tracked 16 h backward and 2 h forward in time, providing a continuous reconstruction of their pathways and thermodynamic history.

At 04:00 UTC on 19 October (Fig. a), air parcels originate from the southeast and cross Denmark before entering the North Sea. They accelerate over the sea: the longer marine segment is covered in comparable time to the land crossing (Fig. b, c). The specific-humidity time series shows the parcels are already moist on entering the marine domain. The small air potential temperature increase (+2.5 K) reflects limited turbulent heat uptake during the sea crossing, due to limited air/sea temperature difference (Figs. S3–S4).

By 22:00 UTC on 19 October (Fig. ), the air parcels originated from the east, from cooler and drier land areas before arriving over the North Sea. Initially, these parcels contained lower moisture content, but as they passed over the MHW region, the specific humidity roughly doubles over the sea and potential temperature rises by 5–6 K, indicating respectively intense latent heat and sensible heat flux. This behaviour points to strong air-sea interaction later on 19 October: the North Sea provided an abundant moisture source that primed the atmosphere for heavy rainfall.

These trajectory analyses show distinct pathways and thermodynamic histories, pointing to the North Sea as a source of moisture for precipitation in Eastern Scotland on the evening of 19 and 20 October. Section 4 examines how the MHW preconditioning led to precipitation extremes and multi-hazard amplification.

To assess the atmospheric response to the MHW, a counterfactual analysis was performed between two 18-member ensemble simulations over the 5 d forecast window from 17–21 October. As detailed in Sect. 2.4, these experiments represent the observed 2023 event with the marine heatwave (MHW) and a counterfactual scenario where the observed 2023 event occurred over near-climatological sea conditions from 2020 (no_MHW). This approach isolates the impact of the anomalously warm Northwest European Shelf sea on the stor's development. Figure  presents the temporal evolution of key meteorological variables, and clearly shows that the presence of the MHW led to a systematic intensification of the storm system.

The MHW simulation maintained consistently higher SST throughout the 17–22 October period (Figure a), by about 0.8–1 °C. This anomaly was spread throughout the 40–50 m of the Central North Sea.

A consistent atmospheric response to this anomalous warmth is evident in mean sea-level pressure (MSLP) (panel b). The MHW ensemble shows a slightly deeper low-pressure centre, with reductions of 1–2 hPa around 19 October. Although small, this difference is statistically significant and reflects a modest but systematic enhancement of storm intensity. These results align with the synoptic-scale pressure patterns discussed earlier (Sect. 3.1), where a blocking anticyclone over Scandinavia confined Babet over the North Sea and allowed the low-pressure system to intensify locally. This intensification is likely linked with increased latent heat release from increased precipitation (panel f).

The MHW run consistently produced slightly stronger 10 m wind speeds, with an increase of about 1 %, especially during the wind decay phase. The model evaluation (Fig. S5) showed a realistic temporal evolution of winds, supporting confidence in these results. The trajectory analyses (Sect. 3.2) and the model evaluation of sea and air temperature (Figs. S3 and S4 respectively) showed the air parcels over the North Sea were always cooler than the sea, and in particular on 20 October, meaning the boundary layer was both thermally and (given high wind speed) dynamically unstable. In these conditions, convective downdrafts bring free-tropospheric momentum near the surface, enhancing wind speed. We hypothesise that the warmer sea in the MHW experiment contributed to a higher boundary layer (panel e): an increase of 30–50 m relative to the no_MHW case, which means more momentum likely to be carried to the surface. Storm MSLP deepening, and the associated strengthening of the cross-basin pressure gradient (Fig. S7), may also have contributed to stronger wind speeds.

The most direct atmospheric signal of the MHW is seen in latent heat flux (panel d). Fluxes in the MHW simulation were 15 %–20 % higher than in the no_MHW case, especially during 20 October when the storm reached peak intensity. The ensemble mean shows a statistically significant enhancement across much of the storm’s evolution. This amplification reflects the greater air-sea humidity contrast provided by the warm MHW waters, but its strength depended strongly on air-mass origin. As shown in the trajectory analysis (Sect. 3.1), early on 19 October the air parcels were already moisture-laden, winds were weaker, and little additional moisture was taken up. By contrast, later on 19 October, the air arrived drier and under stronger winds, enabling vigorous evaporation from the MHW region. This led to a sharp increase in latent heat flux and set the stage for enhanced precipitation over Eastern Scotland.

Figure 8f shows the accumulated precipitation in Eastern Scotland. Both ensembles reproduce the extreme rainfall totals associated with Babet, but the MHW run yields a consistent enhancement of 5 %. While small relative to the storm total, the increase is statistically significant and occurs mainly on the afternoon of 19 October and on 20 October, coinciding with the period when dry air parcels crossed the warm North Sea and acquired substantial additional moisture. In contrast, the MHW exerted little influence on 19 October, when incoming air masses were already humid. Importantly, the timing of the enhancement corresponds to the period of strongest latent heat flux and boundary layer deepening, reinforcing the link between the ocean anomaly and atmospheric intensification. The magnitude of the moisture increase aligns with Clausius-Clapeyron expectations for the observed SST anomaly (7 % more moisture per degree of warming), but its manifestation depended on the interplay between the MHW and the evolving air-mass origin. This is consistent with who emphasise that air-mass pathways and origin humidity, rather than background SST alone, are key controls on precipitation over the UK. For Storm Babet, amplification of the precipitation was confined to periods when dry continental air masses crossed the warm basin under strong winds.

The increases in precipitation, near-surface wind speed, and cyclone deepening associated with the North Sea MHW extended into the hydrological and coastal hazards, amplifying river discharge, surge, and wave power. To quantify these impacts, we analyse the same pair of 18-member ensemble forecasts described in Sect. 2.4, spanning 17–21 October. Figure  presents the temporal evolution of these hazard indicators: river discharge at river mouths, wave power and surge averaged over the red box of Fig.  highlighting the role of anomalously warm SSTs in amplifying multi-hazard risk.

Figure b, c show that both the river South Esk and Tay experienced a systematic increase in flow in the MHW ensemble. By 20–21 October, flows were enhanced by around 12 %–18 % (∼ 30 m³ s⁻¹). The South-Esk peaked earlier, on 20 October, whereas the Tay catchment peaked later, on 21 October, reflecting different basin response times. This response is not expected to scale linearly with rainfall, because runoff also depends on antecedent soil moisture, rainfall intensity, and catchment saturation. Importantly, both rivers burst their banks during Storm-Babet, and the model indicates that the MHW amplified the peak flows, thereby raising the probability of fluvial flooding.

The coastal storm surge Fig. d was amplified by about 20 % in the MHW ensemble on 19 October (∼ 3 cm averaged on the Eastern Scotland box), consistent with slightly stronger near-surface winds (+1%) and a deeper storm core (widespread ∼ 2 hPa mean sea-level pressure anomaly) on this day. The persistence of the anomaly nonetheless illustrates how MHW-driven intensification can increase the frequency of hazardous water levels, even when tidal phasing mitigates the outcome.

Finally, Fig. e shows that wave power, defined as the energy flux carried by waves (proportional to wave height squared times wave period), was elevated in the MHW case by about 9 % on 19–20 October. This enhancement is consistent with the ∼ 1 % increase in wind speeds, arising from greater boundary-layer depth and enhanced thermal instability over warmer waters. Indeed, waves are an integrator of wind speed, and a 1 % increase in wind speed in our case leads to 3 % wave height increase, and therefore 9 % wave power increase. Stronger and more energetic waves increase the risk of coastal erosion and overtopping of sea defences.

At the end of 19 October, the MHW run shows that wave power was increasing while the surge had already started to go down. Even though the surge was smaller than earlier, it was still higher than in the no_MHW case. This means that the peaks of surge and waves did not fully happen at the same time, but the MHW still made both hazards larger than they would have been otherwise. If these waves happen at the same time as high water levels, they can also make it harder for rivers to empty into the sea, which can worsen flooding in estuaries. Our model does not link the rivers directly with the sea, but this kind of interaction is well known, and it shows why it is important to look at all the hazards together.

These diagnostics demonstrate how modest SST anomalies can propagate through multiple pathways to amplify storm impacts. Even a small increase in wind, precipitation and decrease in pressure can integrate over time and space and generate much larger river, surge and wave power increases. The MHW systematically raised all components, thereby increasing the overall hazard envelope and amplifying the multi-hazard events (river, surge, tide, wave) shown in the observations on 20 October (Fig. ). This highlights the potential for MHWs to act as multi-hazard amplifiers: even relatively small SST anomalies (∼ 1 °C) can trigger a chain of processes that lead to greater and longer-lasting riverine, coastal, and wave impacts.

This study has several limitations that should be considered when interpreting the results. The analysis focuses on a single case study, Storm Babet, meaning that the identified processes cannot yet be assumed to apply to all extratropical cyclones. Every storm has unique synoptic characteristics, so generalisation requires a larger event set. In addition, the experimental design relies on the Met Office regional coupled prediction system which, although state-of-the-art for UK forecasting, represents only one model configuration and is subject to biases. The difference between 2023 and 2020 sea states used in the coupled model tends to underestimate the MHW amplitude relative to observations (Fig. ), implying that the simulated anomaly represents a conservative estimate of the true MHW signal (it corresponds to an anomaly with a 2003–2022 baseline, rather than 1982–2012). Indeed, repeating this study with an atmosphere-only ensemble prescribing the observed MHW anomaly showed a 6 %–7 % increase in precipitation, slightly higher than the 5 % of the coupled system (not shown). Although the coupled system remarkably captured individual hazards, and the timing of their increase, it did underestimate most of them, so it is possible that the MHW amplification of these hazards may also be underestimated. Nevertheless, the coupled system captured very well the SST cooling (Fig. S3), giving confidence in how much energy was extracted from the North Sea by the storm in the coupled forecast. Furthermore, the regional model setup focuses on the Northwest European shelf and surrounding regions, so feedbacks on larger-scale circulation features influencing the storm may not be fully represented. Regarding the seasonal context, a previous NWES MHW also occurred in June 2023, raising the question of whether cumulative seasonal heat storage may have contributed to the autumn event. However, suggest that much of the excess heat content associated with the June event was removed during the very stormy conditions of July 2023. We therefore do not have clear evidence that the June event directly contributed to the atmospheric response examined here, although possible seasonal persistence or subsurface memory effects remain an open question for future work.

Despite these limitations, the results highlight important implications for understanding and predicting the effect of MHWs for mid-latitude storm impacts. The results show that MHWs can intensify multiple hazards associated with extratropical cyclones. In this case, the warm anomaly in the North Sea enhanced latent heat fluxes, which in turn supported greater atmospheric instability and additional precipitation over Eastern Scotland. However, the expected thermodynamic effects alone do not explain the response; the amplification also depended on a shift in the air-mass origins, showing that local and large-scale factors acted together. While show that background SST alone does not explain precipitation variability over the UK, our results illustrate that, for this event, the anomalously warm North Sea provided locally significant additional moisture when initially dry air crossed the basin. The model experiments also showed consistent increases in river discharge, storm surge, and wave power under MHW conditions, suggesting that the influence of the ocean surface extended across several components of the hazard chain. These findings highlight that even a modest SST anomaly was sufficient to produce systematic differences, underlining the sensitivity of shallow seas to air-sea coupling. In this sense, the MHW influence should be understood as a causal amplifier operating through enhanced air-sea fluxes, boundary-layer moisture and heat supply, and their interaction with the evolving storm dynamics, rather than as a uniform direct effect on all hazards.

We can formalise these pathways, in a causal network that links the pre-existing warm shelf-sea state and anomalous ocean heat content to surface fluxes, storm intensity, and downstream hazards (Fig. ). This framework provides a structure for attributing observed responses to specific drivers within a causal explanatory framework, and for testing whether similar connections appear in other cases. In the conditional framework used here, the counterfactual experiments indicate that the pre-existing MHW amplified the hazards under the observed broad storm setup. With a larger event set, the approach could establish more robust relationships between North Sea heat content and storm severity, allowing generalisation beyond single events.

These findings have broader implications for both predictability and risk assessment. In this case, the anomaly was not confined to the surface but extended through much of the shallow water column helping to sustain warm surface conditions that supported enhanced air-sea fluxes to persist during the storm. Ocean heat content is an observable quantity that can be monitored and predicted at sub-seasonal to seasonal (S2S) timescales. If storms with similar characteristics to Babet occur during periods of elevated heat content in the North Sea, they are likely to be amplified in a comparable way. This provides a physical basis for using ocean heat content as a precursor variable in forecasts of storm impacts. It also shows that the interaction between MHWs and storms is not a rare coincidence, but rather a systematic process that should be integrated into risk management frameworks. Storms tend to always end MHWs, and this example shows that MHWs can amplify the storms that end them.

The MHW that coincided with Storm Babet represents both internal variability and the influence of long-term climate change. The observed North Sea anomaly can be viewed as representative of warming levels that may become typical in future mean conditions. Indeed, the North Sea warms faster than the global warming trend, and warmed by 1 °C between 2003–2022 and 1982–2002 (Fig. ), and 2023 was 0.9 °C warmer than 2003–2022, which, if the past trend continues, would be close to a 2023–2042 average. Increases of up to 4.1 °C for the Central North Sea and 4.5 °C for the Southern North Sea are possible for the last two decades of the century compared to the first two . From this perspective, Babet is not only an example of present-day hazard amplification, but also a prototype of how storms may evolve as mean ocean temperatures rise. A key part of this amplification is the preconditioning of the shallow North Sea, with excess heat stored before the storm maintaining a warm shelf-sea state that supported enhanced air-sea fluxes during the event. When such warming becomes the baseline rather than an extreme, the amplification observed here would apply to a much larger set of storms. This interpretation assumes, to a first approximation, that the relevant internal variability and circulation patterns remain structurally similar, even as the thermodynamic background shifts in a warmer climate. At the same time, the large-scale background in which such events occur may also be influenced by changes in baroclinicity, jet variability, and North Atlantic internal variability. Assessing the role of these wider-scale drivers is beyond the scope of the present event-based study, which focuses on the conditional influence of the pre-existing MHW within the observed Storm Babet setup.

Nevertheless, to establish whether these mechanisms are general or case-specific, it is important to extend the analysis beyond Babet. Different storms are likely to show distinct balances of response; for instance, larger changes in mean sea-level pressure but smaller changes in precipitation. Analysing a broader set of events will make it possible to test whether the causal pathways identified here recur across storms, or whether they depend on particular synoptic configurations. This comparative approach will provide a more robust basis for linking MHW conditions to storm impacts.