Stratospheric Nudging And Predictable Surface Impacts (SNAPSI): A Protocol for Investigating the Role of the Stratospheric Polar Vortex in Subseasonal to Seasonal Forecasts

Major disruptions of the winter season, high-latitude, stratospheric polar vortices can result in stratospheric anomalies that persist for months. These sudden stratospheric warming events are recognized as an important potential source of forecast skill for surface climate on subseasonal to seasonal timescales. Realizing this skill in operational subseasonal forecast models remains a challenge, as models must capture both the evolution of the stratospheric polar vortices in addition to their coupling to the troposphere. The processes involved in this coupling remain a topic of open research. 5 We present here the Stratospheric Nudging And Predictable Surface Impacts (SNAPSI) project. SNAPSI is a new model intercomparison protocol designed to study the role of the Arctic and Antarctic stratospheric polar vortices in sub-seasonal to 1 https://doi.org/10.5194/gmd-2021-394 Preprint. Discussion started: 4 January 2022 c © Author(s) 2022. CC BY 4.0 License.

The basic experimental design proposes to focus on the evolution of specific events of interest, using the following sets of 75 forecast ensembles: free A standard forecast ensemble in which the atmosphere evolves freely after initialization. The method of initialization and of generating ensemble members is not specified and can be determined by the participating modeling groups.
nudged A nudged ensemble in which the zonally symmetric stratospheric state is nudged globally to the observed time evolution of the stratospheric event of interest. 80 control A nudged control ensemble in which the zonally symmetric stratospheric state is nudged globally to a time-evolving climatological state.
As discussed in the introduction, the free ensembles along with the zonally symmetric nudged and control ensembles are of highest priority. However, particularly for models with grids that are not aligned along fixed latitudes, the zonally symmetric nudging is difficult to realize. Thus two additional ensembles are requested at lower priority: 85 nudged-full A nudged ensemble in which the full stratospheric state (including zonally asymmetric components) is nudged globally to the observed time evolution of the stratospheric event of interest.
control-full A nudged control ensemble in which the full stratospheric state (including zonally asymmetric components) is nudged globally to a time-evolving climatological state.
The reference states for the nudged and control ensembles are computed from ERA5 reanalysis output (Hersbach et al.,90 2020) on native model levels. These coincide with isobaric surfaces at the stratospheric levels where the nudging is applied.
Details of how the climatological state is computed are given in the Methods section below.
The protocol targets forecast integrations of 45 days, and an ensemble size of 50 to 100 members. For each of the three case studies, two specific initialization dates for each type of integration are proposed; these are discussed in the context of the specific target events described in Section 3. 95 The impact of the stratosphere on the troposphere can be confounded by unrelated dynamical variability within the troposphere. Hence the choice to emphasize ensemble size over the number of initialization dates is intended to allow for statistically and dynamically meaningful comparisons of these specific events across participating models.
There are four central motivations for the proposed forecast experiments, presented in the following subsections. In addition, these experiments are expected to provide useful insights into coupling between the tropical stratosphere and troposphere. 100 These secondary motivations are discussed following the four primary goals.

Quantify stratospheric contributions to surface predictability
Through nudging the stratosphere to observations, the nudged ensemble will provide a 'perfect' forecast of the stratosphere's zonal mean state. The forecast skill attained can be compared to that attained by the control ensemble (amounting to a 'clima-These experiments will provide a multi-model assessment of the potential increase in skill associated with an improved representation of the stratospheric state, and an up to date assessment of the present skill that is achieved by each model. By including multiple case studies of interest, this suite of experiments also makes it possible to study the dependence of the tropospheric response to stratospheric anomalies on the tropospheric state itself.

Attribute extreme events to stratospheric variability 110
The proposed protocol will also provide a means of assessing or formally attributing the contribution of the stratosphere to an extreme event of interest (Domeisen and Butler, 2020). Extremes that have been associated with sudden warmings in recent years include cold air outbreaks in the Northern Hemisphere (Kolstad et al., 2010;Afargan-Gerstman et al., 2020;Huang et al., 2021;Charlton-Perez et al., 2021) and hot, dry extremes over Australia (Lim et al., 2019). This goal is closely related to the growing sub-discipline that focuses on attributing the occurrence of particular extremes to climate change and variability 115 (National Academies of Sciences, Engineering, and Medicine, 2016).
Consider some extreme event A that is thought to have been associated with a specific sudden stratospheric warming (SSW), for instance the cold air outbreak (CAO) that occurred in Europe following the sudden warming in February 2018. The probability of such an event occurring p 0 = p(A) might be estimated from the observed climatological frequency of similar events, or from a set of forecasts that sufficiently represent the variability of the climate system from a given subseasonal forecast 120 model, or from a combination of both (Sippel et al., 2015). Given the nudged ensemble, one can then estimate the probability of a similar event occurring given the weakened state of the stratospheric polar vortex p 1 = p(A|V − ). The Relative Risk (see, e.g. Paciorek et al., 2018) of this CAO might then be calculated as RR = p 1 /p 0 . Relative Risk values of RR > 1 would then imply an increased risk of a CAO under a weakened vortex state, whereas RR < 1 would imply the opposite. This can also be compared to the probability of such an event occurring in the counterfactual situation that the sudden warming did not occur, 125 p 0 = p(A|V 0 ), computed from the control ensemble, allowing further for the calculation of necessary or sufficient causation probabilities (Hannart et al., 2016). In the context used here, the RR is the most appropriate measure of risk because the data are likely to be non-Gaussian (Christiansen, 2015).
As an example, Fig. 1 shows monthly mean NAO indices (from Hitchcock and Simpson, 2014). The probability of occurrence of a strongly negative monthly mean NAO state is much more likely in the aftermath of a sudden stratospheric warming 130 than under a 'counterfactual' scenario during which the stratosphere was close to its climatological state.
The interpretation of the Relative Risk becomes more challenging in a forecast context, since the probability of an extreme event is strongly conditional on the initial conditions known at the time of the forecast. As the forecast initialization date grows closer to the event of interest, the forecast ensembles will begin to forecast the event with increasing fidelity; that is, the probability of occurrence conditional on initial conditions n days prior to an event, p(A|IC(n)) will grow.

135
One practical way to frame this question is to ask whether improving the forecast of the stratospheric state can lead to earlier accurate forecasts of the event in question. Alternately, one may ask whether degrading the forecast of the stratosphere leads to degraded forecasts of the event. Both framings are enabled by the proposed experiments. SSW with any certainty until the initializations in the 1-10 February period. There is a significant change in the March surface temperatures over Europe for initializations before and after the stratospheric event was captured in the prediction system, with forecasts initialized with the SSW information more closely capturing the observed March temperatures. But do these differences arise solely because the forecast model finally captured the SSW, or because the lead-time had decreased? With the three experiments proposed and applying this to multiple initializations before the event, it would be clear whether or 145 not having the "perfect" stratosphere (nudged ensemble) for runs initialized in mid-January would have given more accurate forecasts at longer leads.
A very similar approach has been adopted by Kautz et al. (2020) who made the distinction between 'probabilistic' and 'deterministic' forecasts of the extreme event in question. They presented evidence from the ECMWF model that a perfect forecast of the stratospheric anomalies in early 2018 would increase the predicted odds of extreme cold weather over Europe 150 from 5% to 45%. These odds then increase further as forecasts are made closer to the event.
The common and comparable set of integrations from a range of operational centers made available by this project will allow this finding to be extended to other extreme events and will allow further development of this methodology. We aim to have a large enough ensemble size to allow for the direct study of well-constrained extreme events, but if necessary may also supplement the model data with extreme value statistics (e.g. Sippel et al., 2015). Ensemble sizes of 50-100 are sufficient 155 to understand large-spatial scale and persistent (weekly) extremes, such as cold or NAO events, but the latter method will be required for 'noisier' fields such as precipitation.

Assess the mechanisms underlying stratospheric coupling in individual models
Imposing stratospheric anomalies through a nudging procedure has been shown to significantly impact the near surface flow (e.g. Douville, 2009), even if only the zonally symmetric component is imposed (e.g. Simpson et al., 2011;Hitchcock and 160 Simpson, 2014;Zhang et al., 2018;Jiménez-Esteve and Domeisen, 2020). By comparing the difference between the nudged and control ensembles, the processes that drive this downward coupling can be diagnosed in each model for a variety of events of interest. It is of particular interest to better understand why some specific stratospheric events are followed by the 'canonical' equatorward shift of the tropospheric eddy driven jets, while others are not. The two boreal and one austral case studies proposed were followed by a diversity of tropospheric responses, including two cases which exhibited the 'canonical' 165 response (the 2018 boreal and 2019 austral cases) and one which did not (the 2019 boreal case). This set of experiments will thus shed light on whether these diverse responses were determined by stratospheric causes, or whether they are determined by competing effects such as tropical tropospheric variability or independent mid-latitude dynamical processes (e.g. Knight et al., 2020). In either case, the statistical sampling afforded by a multi-model set of forecast ensembles with detailed diagnostics will allow for new and deeper insights into the mechanisms responsible for the tropospheric response. Moreover, each event 170 also coincided with specific surface extremes that produced significant societal impacts. This set of experiments will provide quantitative insight into the mechanisms responsible for these surface extremes. The data request has been designed to allow for a more detailed analysis of these processes than has been possible with existing databases of subseasonal forecasts. Ultimately this understanding will help both future operational system design and practical use of subseasonal forecasts.
2.4 Quantify the role of the stratosphere in upward wave propagation 175 The onset of a sudden stratospheric warming is marked by the reversal of the climatologically westerly zonal mean zonal winds in the mid stratosphere. Operational forecasts can, on average, successfully forecast this reversal starting about two weeks prior, but this depends strongly on the specifics of the event in question (Tripathi et al., 2015;Domeisen et al., 2020a;Rao et al., 2020a, b). A key issue is the successful forecasting of the rapid growth in planetary-scale Rossby waves that drives the breakdown of the stratospheric polar vortex. This requires capturing both tropospheric precursors for these waves (e.g. 180 Garfinkel et al., 2010), as well as their interaction with the stratospheric flow (e.g. Hitchcock and Haynes, 2016;de la Cámara et al., 2018;Lim et al., 2021;Weinberger et al., 2021).
A fourth goal for this protocol is to determine how well forecast systems capture this initial amplification of planetary waves.
In particular, the first of the initialization dates has been chosen just prior to the periods of enhanced wave driving that lead to the breakdown of the stratospheric polar vortex (as discussed in section 4 below). By comparing the evolution of the wave 185 field in the control and nudged ensembles, the role of the stratospheric state in determining the wave amplification can be isolated and compared with the importance of capturing specific tropospheric precursors. This will reveal how well forecast models can predict the evolution of the planetary waves on a given zonally symmetric background, allowing for quantitative intercomparison. Further comparison with the free ensemble will provide detailed insight into the ability of individual models to forecast the complex interactions responsible for the amplification of the wave field.

Secondary Science Questions
Although the emphasis in the design of SNAPSI has been on extratropical coupling between the stratosphere and troposphere, the experiments are expected to provide further insights into coupling between the tropical stratosphere and troposphere, and between the tropics and extratropics in both the troposphere and stratosphere. We outline in this section several potential questions that may be addressed with these experiments.

Representation of the Quasibiennial Oscillation
These experiments may be useful for examining the model representation of the QBO. Since the QBO is a nonlinear oscillation driven by wave-mean flow interactions, the waves and the mean flow are tightly coupled: the waves influence the evolution of the mean flow, and vice versa. In the nudged ensemble, upward-propagating equatorial waves that force the QBO (both resolved and parameterized) will encounter essentially identical zonal-mean zonal wind profiles in all models. This allows 200 wave forcing to be directly compared between models absent the complication of differing background zonal-mean winds. This approach has been used previously to assess the response of wave forcing to changing vertical resolution in a single model (Anstey et al., 2016). In the free experiment, equatorial winds can respond to the wave forcing. To the extent that model biases have time to develop over the 45-day hindcast period, results from the nudged ensemble may yield insight into the origin of biases in the free runs. In particular, current QBO-resolving models are typically unable to maintain realistic QBO amplitude 205 in the lowermost tropical stratosphere (Stockdale et al., 2020;Richter et al., 2020), and some S2S models lose nearly the entire QBO signal within a typical 40 day reforecast (Garfinkel et al., 2018).

Stratospheric influences on tropical convection
Recent work has highlighted a variety of potential stratospheric impacts on organized tropical convection (Haynes et al., in press). Notably, the phase of the QBO has been shown to have a significant impact on the strength and persistence of the 210 MJO (Son et al., 2017). This impact has an apparent effect on the predictive skill of the MJO, in that forecasts of the MJO remain skillful at longer lead times during the easterly phase of the QBO (Martin et al., 2021b). Sudden stratospheric warmings have also been shown to shift and enhance regions of tropical convection (Kodera, 2006;Noguchi et al., 2020). A wide range of coupling pathways and mechanisms have been proposed, but fundamental understanding remains limited, in part due to the large scale separation between the planetary scales of the stratospheric variability and the mesoscale to synoptic scale of 215 tropical convection (Haynes et al., in press).
Imposing stratospheric variability through nudging as is proposed here can isolate the importance of the stratospheric state on convection in the forecast models. Nudging techniques similar to those adopted by SNAPSI have been used to study both tropical (Martin et al., 2021a) and extratropical (Noguchi et al., 2020) pathways in single-model contexts; SNAPSI will allow for a multi-model investigation of these effects.

Stratospheric pathways for teleconnections
The stratosphere is thought to modulate the remote impacts of a variety of climate drivers, spanning from shorter time-scale blocking events and seasonal snow-cover anomalies to ENSO and sources of decadal variability such as the solar cycle (for a more complete list, see Butler et al., 2019). Thus the stratosphere may play an important role in correctly capturing the response to a broad range of subseasonal predictors. In many cases the detailed mechanisms responsible for the modulation remains an 225 open area of study. Comparisons between the nudged and control ensembles will provide a clear means of assessing the stratospheric pathway at play for those teleconnections that are active during the selected case studies. An analysis of the teleconnection in the free ensemble may then provide an assessment of the skill of each model in capturing the relevant pathway. It may also yield insight into specific model biases or deficiencies that prevent skill arising from the stratospheric pathways from being realized.

Reference states
The reference states have been prepared from the ERA5 reanalysis (Hersbach et al., 2020). The zonally symmetric reference states used for the nudged and control ensembles are the instantaneous zonal mean temperature and zonal wind output from ERA5 at the native 137 model levels at six hourly intervals, interpolated to a 1 degree horizontal grid. No relaxation is imposed  3.2 Nudging specification and reference states "Nudging" specific components of the atmospheric circulation by means of an artificially imposed relaxation to a given state has been used by many studies as a means of testing dynamical hypotheses. However, the introduction of an artificial linear relaxation into the equations of motion can produce unintended consequences (e.g. Shepherd et al., 1996;Orbe et al., 2017;Chrysanthou et al., 2019).

250
This section describes in detail the nature of the nudging relaxation to be used in this protocol, which is designed to avoid such consequences. The intent for the nudged and control ensembles is to prescribe the zonally-symmetric component of the stratospheric flow without indirectly constraining the troposphere or affecting the planetary waves that play a central role in the coupling between the two. The nudging is specified as a relaxation tendency of the form −τ −1 (X − X r ), where X is either the zonal mean temperature or zonal wind, and X r is the zonally symmetric reference state to which the flow is constrained.

255
The nudging tendency is imposed equally on all longitudes (at a given latitude and height), to avoid directly affecting the wave field. The timescale of the nudging varies with pressure, tapering gradually from infinite (i.e. no nudging) below a lower limit of p b = 90 hPa, to full strength at p t = 50 hPa, following a cubic profile p b −p p b −pt 3 . At full strength the nudging timescale is 6 hours. The nudging is to be imposed at all latitudes equally.
While the nudging profile is specified in pressure coordinates, the intent is for the nudging strength to be constant on model 260 levels and can be converted using 'typical' pressures appropriate for the details of the vertical coordinate system of a given model.
For the nudged-full and control-full ensembles, the same nudging profile is applied to the horizontal winds and temperature.
For the control ensemble, the protocol specifies nudging the zonally symmetric components of the stratosphere towards the climatology. Since the initial conditions are in some cases some ways away from the climatology, nudging at full strength to 265 the climatology will generate undesirable transients as the stratosphere adjusts towards the climatological state.
In order to reduce this initial shock, the reference state for the control ensemble is interpolated smoothly from the observed evolution to the climatology over the first 5 days of the forecast period. For instance, the temperatures is relaxed towards a state where T o is the instantaneous reference state, T c is the reference climatology computed as described in the previous section, t is the time, t i is the starting time of the forecast, and f is an interpolating function given by A similar adjustment should be adopted for the control-full ensemble.
The zonally symmetric nudging approach has been successfully applied by a number of studies to explore aspects of 275 stratosphere-troposphere interactions (Simpson et al., 2011;Hitchcock and Simpson, 2014;Simpson et al., 2018). The approach has also been used to impose a QBO in models that do not internally generate one, and to study the dynamics of the QBO and its teleconnections (Anstey et al., 2016;Martin et al., 2021a). However, these studies have been carried out either in uninitialized climate models or in idealized general circulation models, not in the context of operational forecasting, and the technique in general can produce undesirable artifacts, particularly with regards to transport effects (Chrysanthou et al., 2019;280 Hitchcock and Haynes, 2014).
The nudging specification in the nudged and control runs is intended not to directly impact the zonally asymmetric component of the flow. The statistics of the planetary waves in particular are found in test experiments to be largely unaffected by the constraint on the zonal mean. One exception is that wave amplitudes in the upper stratosphere can grow larger in the presence of nudging; this is in part because the nudging prevents the wave transience from decelerating the mean flow, allowing plane-285 tary waves to propagate higher before they encounter critical levels. In some cases this can result in unusually strong winds in the upper stratosphere and lower mesosphere; however this is not expected to influence the evolution of the lower stratosphere or its interactions with the troposphere.
Because the wave field is not directly controlled by the nudging, the zonal mean forcing produced by the internally-generated wave field can differ substantially from that consistent with the evolution of the reference state, particularly for the control 290 ensemble. Since the meridional circulation is largely determined by the forcing associated with the waves (e.g. Plumb, 1982;Haynes et al., 1991), misrepresentation of the wave field can result in spurious meridional circulations and the potential for unintended remote effects. However, Hitchcock and Haynes (2014) has shown that the spurious circulations are largely confined to within the region of nudging, while the non-local circulation below the region of the nudging associated with 'downward Table 1. Case studies and forecast initialization dates. The nudged-full and control-full ensembles are requested only for the later of the two initialization dates for a given event. control' is to a close approximation consistent with the forcings that produced the reference state. This implies that any down-295 ward influence associated with these circulations can be expected to be present in the nudged ensemble, and absent in the control ensemble. Spurious circulations within the nudging region may give rise to anomalous transport of constituents within the stratosphere, but this is not expected to be of concern on the subseasonal timescales relevant to the present protocol.
The presence of a nudging layer can also give rise to a 'sponge-layer feedback' like response (Shepherd et al., 1996), which is characterized by spurious zonal mean temperature and wind anomalies generated just below the layer of nudging in response 300 to tropospheric torques that differ from the reference state. These effects have also been shown to be negligible on these timescales (Hitchcock and Haynes, 2014).

Case Studies
The ensemble forecasts just described will be applied to three recent events: Two initialization dates are requested for each event (Table 1). One date is chosen about three weeks prior to the surface extreme of interest, in order to identify the contribution of the stratosphere to its forecast on subseasonal timescales (motivations 1, 2, and 3). A second date is chosen prior to the onset of the stratospheric warming in order to assess the representation of the onset of the event (motivations 1, 3, and 4). The former has higher priority than the latter, although they are listed 315 chronologically in Table 1. Thursdays are chosen since nearly all models that contributed to the S2S database contributed forecasts initialized on Thursdays, making it easier to compare the two datasets. Further justification for the initialization dates selected are provided in the case-by-case discussion below. The tropospheric NAM responded strongly to these stratospheric anomalies, exhibiting a shift to negative values from mid-February through mid-March, consistent with the composite mean response to sudden stratospheric warmings. The NAO index 325 was strongly negative in late February, coinciding with unusually cold weather over much of Europe and Asia during the last two weeks of Feb. (Lü et al., 2020), bringing, for example, snow to Rome and several notable winter storms to the UK.
Precipitation patterns also shifted, bringing persistent rain to the Iberian peninsula, ending an extended period of drought (Ayarzagüena et al., 2018).
Of the three proposed case studies, this first case has been the most actively studied to date. In a study of the S2S database, 330 Rao et al. (2020a) showed that those ensemble members that capture the amplitude of the lower stratospheric anomalies during this event (and the 2019 case considered next) were also more successful in forecasting the surface extremes; they also showed that this was more relevant than whether the model forecasted a split or displacement of the vortex. As discussed above, Kautz  Garfinkel and Schwartz, 2017). A week after the event the MJO entered phase 8, which is linked to a negative NAO pattern. (2020) do find that nudging the tropical evolution produces a negative NAO response in late February, suggesting that tropical circulation anomalies contributed to the anomalous European weather regimes.
The S2S prediction systems forecast the event about 11 days in advance (Karpechko et al., 2018;Rao et al., 2020a), making this event less predictable than some other sudden warmings. Proximately, this is likely due to the nature of the relevant wave 345 driving which amplified rapidly during the week prior to the stratospheric wind reversal (Fig. 4). Subseasonal forecasts that captured this wave event were more successful in forecasting the vortex breakdown. The difficulty in forecasting the pulse of wave activity has in turn been tied to both anomalous blocking over Siberia (Karpechko et al., 2018) as well as to an episode of anticyclonic Rossby wave breaking in the North Atlantic .
On longer timescales, Knight et al. (2020) further suggest a role for the large-amplitude MJO event that preceded the moderate La Niña state, and the QBO winds were persistently westerly at 50 hPa and easterly at 30 hPa throughout the winter.
Thus the state of both ENSO and the QBO may have also contributed.
The first initialization date proposed is 25 Jan 2018, just prior to the first pulse of wave activity leading to the vortex split.

355
The ensembles will thus produce some diversity in the tropospheric precursors outlined above. By considering the ensemble  In late December 2018, the Arctic vortex was displaced off of the pole, prior to splitting. The 10 hPa winds reversed on 2 Jan 2019. In contrast to the 2018 event, the stratospheric vortex anomalies developed much more gradually through late December and early January of the 2018-19 winter . The vortex remained split for several weeks. Anomalies in the lower stratosphere persisted nearly to March of 2019. The gradual weakening of the vortex was due to persistent wavenumber-365 one forcing that was well predicted even from mid-December (Rao et al., 2020a).
In strong contrast to the 2018 case, the tropospheric NAM did not respond strongly to the stratospheric anomalies, remaining near neutral or even slightly positive through much of the troposphere until early February (Fig.5). However, an extensive cold snap occurred over North America in late January (roughly 23-29 Jan) in a region vertically aligned with one of the daughter vortices generated by the split.

370
This event was also considered by Rao et al. (2020a), who found that the surface temperatures and precipitation patterns 20 days following the onset date were generally not well forecast by the S2S models. Note, however, that they did not focus specifically on the cold air outbreak over North America. Knight et al. (2020) also performed nudging experiments to explore the impacts of the stratospheric anomalies on the surface. They found that the ensemble mean again reproduced the 'canonical' tropospheric response, with an anomalously persistent negative AO pattern coinciding with NAM anomalies in the lower 375 stratosphere, implying that the lack of tropospheric signal in observations was due to some competing effect(s). One possibility is that these competing effects arise from the tropics; the tropical nudging experiments of Knight et al. (2020) gave rise to North Atlantic mean sea-level pressure anomalies that more closely resembled observations in January. For instance, the MJO also progressed through phase 6 and 7 in early January 2019, but at amplitudes considerably weaker than in 2018.
The S2S prediction systems forecast the stratospheric wind reversal more than 18 days prior to the central date in some 380 cases (Rao et al., 2020a), but did not predict the vortex would split more than a few days in advance . The longer forecast horizon in this case seems to be related to the persistent wavenumber-one forcing from mid-December 2018 that displaced the vortex off the pole, prior to its ultimate splitting (see Fig.6). Rao et al. (2020a) propose a range of contributing factors for the wave amplification, including the state of ENSO, the QBO, the solar cycle, and the MJO. In the fall of 2018, the QBO at 50 hPa was strongly easterly, below a westerly shear zone that

Austral Minor Warming of September 2019
The final event of interest is the minor warming that occurred in the Southern Hemisphere in September of 2019. Significant

395
Southern Annular Mode anomalies began to emerge in the upper stratosphere towards the end of August (Fig. 7). However, in contrast to the first two cases, the zonal mean winds at 10 hPa, 60 • S did not reverse. However, they did decelerate dramatically, reaching their minimum value on 18 Sep 2019, which can be considered as the 'central' date for the event (Fig. 8). This was slightly earlier in the spring than the major austral warming event that occurred in 2002, during which the Antarctic polar vortex westerlies did fully reverse. In late August the mid-stratospheric winds were near their climatological values, before  The event was forecast nearly 18 days prior by models with a reasonably resolved stratosphere (Rao et al., 2020b), including 410 the persistent stratospheric wavenumber-one flux anomalies (Fig. 8). A number of tropospheric precursors have been linked to this wave activity pulse, including a persistent blocking high over the Antarctic Peninsula and a region of anomalously low pressure over the Southern Indian Ocean (Rao et al., 2020b;Lim et al., 2021). The first suggested initialization date is 29 August 2019, early in the development of the wave activity pulse responsible for the stratospheric event. The second suggested initialization date is 1 October 2019, after the stratospheric anomalies are established, two to three weeks prior to the onset of 415 the tropospheric SAM response.

Data Request
In order to meet the scientific goals of this project, we request output from the forecast models that includes both surface quantities needed for identifying and quantifying high-impact surface extremes, as well as dynamical quantities needed to diagnose the processes that couple the stratosphere and troposphere. Given the relatively short integration periods and small 420 number of initialization dates, data is requested at relatively high temporal and vertical resolution to enable more detailed comparisons of the relevant processes than has been possible with existing subseasonal forecast databases. The data request is closely related to the DynVarMIP request (Gerber and Manzini, 2016), including a request for quantities required to close the zonally averaged zonal momentum and thermodynamic budgets.

430
6hr Surface quantities and fluxes averaged, maximized, or minimized over the 6 hours preceding the timestep (Table 2).
6hrZ Zonally-averaged atmospheric quantities averaged over the 6 hours preceding the timestep (Table 3). Includes quantities needed to close atmospheric momentum and thermodynamic budgets. These follow closely the DynVarMIP request (Gerber and Manzini, 2016), but include the imposed tendencies from the nudging as well.
6hrPt Instantaneous basic meteorological and surface quantities output every 6 hours (Table 4).
There are two levels of priority for the requested variables. The higher priority variables (1) are considered necessary to meet the primary science goals, and include meteorological quantities (winds, temperatures, specific humidity, and geopotential height) required to compute commonly used dynamical diagnostics, measures of precipitation, and surface quantities including 440 pressure, temperature, and horizontal winds. Variables at the lower priority level (2) include zonally averaged quantities that would permit closing the zonally averaged momentum and thermodynamic budgets, as well as surface quantities that would permit a more detailed analysis of surface processes. The zonal mean stratospheric ozone field is also requested at this lower priority level to allow for some assessment of the importance of ozone anomalies in forecasting the evolution of the polar vortex over subseasonal timescales.

6 Summary and Outlook
The SNAPSI project aims to produce a set of controlled ensemble forecasts, initialized around several recent sudden stratospheric warmings. This dataset will allow for an unprecedentedly thorough, multi-model assessment of the contribution of stratospheric extreme events to surface predictability on subseasonal timescales. The proposed forecast ensembles include standard, free-running ensembles, in addition to 'nudged' ensembles in which the evolution of the stratosphere is constrained 450 either to the observed or to climatological conditions. The use of zonally symmetric nudging will enable detailed investigations of the representation of planetary waves that play a central role in the evolution of the events.   exhibited a range of tropospheric responses, but in each case, an extreme event with significant societal impacts followed the 455 stratospheric perturbation.  The experiments have been designed with four primary scientific motivations. First, as outlined above to assess the contribution of the stratosphere to subseasonal forecast skill. Second, to develop methods of formally attributing specific surface extremes to this stratospheric variability. Third, to quantify in detail mechanisms responsible for the surface impacts across the forecast models, controlling for the magnitude and nature of the zonally symmetric stratospheric anomalies that are thought 460 to be most directly responsible for the surface impacts. Fourth, and finally, to improve understanding of the upward coupling from the troposphere to the stratosphere. The experimental design, specific case studies and forecast initialization dates have been chosen to meet these four goals.
Beyond these central goals, the experiments are further expected to shed light on a number of other aspects of dynamical coupling on subseasonal timescales between the stratosphere and troposphere, and between the tropics and extratropics.  tably, both the 2018 and 2019 boreal sudden warming case studies span periods with significant MJO activity and differing phases of the QBO, and the 2019 austral sudden warming case spans the development phase of a disruption to the QBO that occurred in early 2020.
At the time of submission, eleven modeling groups from ten modeling centers are participating in this project (Table 6).
Output from the contributing models will be stored in a central archive hosted by CEDA. Initial analysis of the output will be 470 carried out by community working groups organized through the SNAP project; following an initial embargo period intended to allow time for this analysis to be carried out, the data will be made available to the broader community.