Storm 
Aila
: An unusually strong autumn storm in Finland

The weather during January 1974 was mild, wet and stormy. There were numerous days when the winds reached storm force but two storms stand out. The storm on the evening of the 11th and morning of the 12th, and the storm on the evening of the 27th, which continued through much of the following day in Ulster. Both of these storms were associated with deep occluding depressions as they swept northwards off the west coast of Ireland and five lives were lost as a result of the two storms, along with extensive damage and coastal flooding (Aylott et al., 2019).


Introduction
Extratropical cyclones (ETCs) are responsible for most of the day-to-day variability of weather in the mid-latitudes. ETCs occur most frequently in the winter season, and in favourable atmospheric conditions they can strengthen into powerful windstorms and cause significant damage to society due to their associated extreme winds. Preparing for the extreme winds caused by windstorms is crucial for many domains of society, such as forestry, insurance companies, the energy sec-tor and power grid operators. Thus, accurately predicting the track and intensity of ETCs remains a crucial task for weather forecasters.
Although located at the tail end of the North Atlantic storm track, Finland experiences several high-impact windstorms each year (Gregow et al., 2020). They can occur in all seasons but typically the strongest windstorms in Finland are observed in late autumn and winter. For instance, storm Aapeli in January 2019 was the strongest windstorm on record in Finland with observed maximum 10min wind speed of 32.5ms −1 (63 kn) and a maximum wind gust of 41.6ms −1 (81 kn;Tollman et al., 2019). Storm Tapani (also named as Dagmar in other Nordic countries) and the subsequent storm Hannu on the next day in December 2011 belong to the category of windstorms of greatest impact in Finland and left about 570 000 customers without electricity (Kufeoglu & Lehtonen, 2014). Windstorms can also lead to a rise of sea level and can cause damage via coastal flooding, such as during storm Gudrun in January 2005.
Windstorms with at least storm-force winds (>24.5ms −1 on the Beaufort scale) in northern Europe in the month of September are usually rare, and proportionally many of them are post-tropical cyclones (Sainsbury et al., 2020). One example of these is storm Mauri, which developed from the remnants of Hurricane Debby in September 1982 and caused significant damage in northern Finland (Laurila et al., 2020). In contrast, storm Aila, which is the topic of this paper, was a classic baroclinic storm with no tropical origins.
Storm Aila was an exceptionally strong autumn storm. Aila traversed central Finland approximately at 63°N towards the east and thus followed the track of earlier notable windstorms in Finland (see Figure 1 from Valta et al., 2019). The Finnish Meteorological Institute (FMI) issued the highest level warning (red) of wind gusts and rough seas for western Finland and Bothnian Sea. The first red warnings were given with a three day lead time, which indicated strong confidence and a high predictability of the storm. At the time of the strongest winds, FMI encouraged people to stay indoors in the coastal areas. The highest observed 10-minute wind speed was 29.4ms −1 (57 kn) and the highest wind gust was 35.3ms −1 (69 kn; Figure 2). Furthermore, the strong-gale-force winds (>20.8ms −1 on the Beaufort scale) lasted over 15 hours in western Finland. The maximum wind gust at an inland station was 26.8ms −1 (52kn; FMI, 2020), and the Finnish Forest Centre (FFC) estimated the volume of forest damage to be 0.4-0.7 million m 3 (FFC, 2020). Based on preliminary estimations, Aila caused 160 000 households to be without electricity and altogether 2950 emergency call outs (Láng et al., submitted). Additional impacts arose from unusually high precipitation totals, which in central Finland amounted to 66mmday −1 (FMI, 2020).
The high wind speeds observed during storm Aila, and the fact that the warnings could be given relatively early, is the main motivation of this study. The first objective is to give an overview of Aila's synoptic development. This is performed by analysing reanalysis data. The second objective is to quantify the predictability of storm Aila: how early was the signal for strong winds visible in the medium-range forecasts? Finally, the third objective is to put the observed winds into a climatological larger context: how unusual was storm Aila? The third aim is conducted by analysing longterm weather observations from Finland and reanalysis data.

Data
We used ERA5 reanalysis (Hersbach et al., 2020) from European Centre for Medium-Range Weather Forecasts (ECMWF) to describe the synoptic evolution of storm Aila and to verify the forecasts. ERA5 was used at 6-hourly temporal and 0.25° horizontal resolution. We used both surface and pressure level fields from ERA5. The fields were downloaded from Copernicus Climate Data Store (cds.climate.copernicus.eu).
The predictability of storm Aila was studied using high-resolution (HRES) deterministic forecasts from the Integrated Forecast System (IFS). IFS is the operational weather forecast model used by ECMWF. IFS forecasts were retrieved from ECMWF via the Meteorological Archival and Retrieval System (MARS). We used the forecasts which were initialised at 0000 utc and 1200 utc on 7-18 September 2020. The original resolution of IFS HRES forecasts is 9km, which corresponds approximately to 0.1° resolution. The forecast data were regridded to 0.25° spatial resolution (31 km) which is the same resolution as the ERA5 fields that were also used.
The winds caused by storm Aila were analysed using wind speed and wind gust observations from FMI's weather stations located across Finland. For the climatological investigation, we analyse wind distributions from two stations where the highest winds were observed. The observational dataset includes 1-hour maximum wind speed and wind gust values during September from 2004 to 2020. Before 2004, most of the stations only recorded instantaneous values which likely missed the highest winds and hence, we use here only the 1-hour maximums which are available from 2004 onwards.

Synoptic evolution of storm Aila
This section gives the synoptic overview of storm Aila's evolution based on ERA5 reanalysis. We describe the synoptic situation in northern Europe during 15-17 September 2020 using both low-level and upper-level meteorological variables from ERA5. Figure 3(a) shows the initial synoptic situation in northern Europe at the time when the storm started to develop. There was a strong upper-level ridge present over central Europe (Figure 3(a)). Due to the ridge, the jet stream was shifted northward and was oriented northwest to southeast over Scandinavia. Owing to the southerly flow on western side of the ridge, the air mass in western and northwestern Europe was very warm. A weak frontal boundary north of British Isles, evident in the low-level vorticity shown in Figure 3(a), separated the warm air mass from the colder, polar air mass in Norwegian Sea (Figure 3(c)). Because of the temperature difference, the synoptic situation in the northern edge of the ridge was highly baroclinic and thus vulnerable for cyclogenesis.
The cyclogenesis started approximately on 15 September 1200 utc when a weak upper-level trough approached the preexisting, low-level frontal zone from the west. In Figure 3(a), the trough axis was located between Iceland and northern Scotland. Vorticity advection ahead of the upper-level trough forced ascent in the vicinity of the surface front (not shown). This mid-level ascent and associated low-level convergence caused the low-level vorticity associated with the pre-existing front to rapidly increase causing cyclogenesis to occur. This rapid spin up process can be understood both physically and mathematically by considering the 'stretching' term of the full vorticity equation (see Lackmann, 2011, section 5.3.1 for more details). Furthermore, the cyclogenesis took place in the right entrance region of the jet streak (Figure 3(a)) which enhanced the cyclone development as this region is known to be a favourable location for cyclogenesis due to enhanced upper-level divergence. As a result, a surface low pressure area formed on 15 September 1800 utc with minimum pressure of 1013hPa (between the times in Figures 3(c) and (d); not shown).
On 16 September 2020, 0000 utc, the lowlevel frontal vorticity was no longer oriented linearly, but featured as a clearly identifiable frontal wave over southern Norway (Figure 3(b)). Compared with the situation 12 hours before (Figure 3(a)), the low-level vorticity maximum had also increased. The minimum pressure of the system had decreased to 1008hPa (Figure 3(d)).
The deepening of the low pressure system continued due to the favourable phasing with the upper-level trough. On 16 September at 1200 utc, storm Aila had moved to the western coast of Finland (Figure 4(a)). At this point, the strongest winds were still on the Swedish side of the Gulf of Bothnia. The intensification of the system was still in progress as indicated by the trough at 500hPa situated west from the surface low (Figure 4(a)). Thus, the structure of the system was vertically tilted which is generally a characteristic of a strengthening low pressure system. The cold-air advection west of the surface low (see northeasterly oriented isobars below the 500-hPa trough in Figure 4(a)) further contributed to the strengthening of the system by cooling the air below the trough. A decrease of thickness due to cooling causes geopotential heights to fall (decrease) at altitudes above where the maximum amount of cold advection took place, and hence deepening of the upper trough.
During the night between 16 and 17 September, storm Aila reached its maximum intensity in Finland. There was an upstream high pressure system to the west of Aila with mean sea level pressure (MSLP) greater than 1032hPa (Figure 4(b)). Thus, the pressure gradient between storm Aila and the high was very strong and resulted in powerful northerly airflow along the Gulf of Bothnia (Figure 4(b)). In sea areas, the wind speeds reached their maximum during the night. The flow was parallel with the Gulf of Bothnia, which presumably helped the formation of such high wind speeds because of the large fetch of open water over which the wind blew without obstruction. On 17 September 1200 utc, the storm had already reached its mature phase. The upper-level trough was detached from the main flow and formed a closed circulation (Figure 4(c)). At this point, the upper-and lower-level lows were almost vertically aligned, implying that the strengthening of the system had ceased. This is also seen by the minimum surface pressure, which did not decrease between the times shown in Figures 4(b) and (c). During the day on 17 September (Figure 4(c)), gusty winds were still blowing from the north and northeast, and caused damage especially over land areas in central Finland.
According to ERA5, the minimum surface pressure of storm Aila in Finland was 995hPa on 17 September 0600 utc (between the times in Figures 4(b) and (c), not shown). The maximum 24-hour deepening rate was 16hPa and took place from 15 September 1800 utc to 16 September 1800 utc.

Comparison of IFS forecasts to ERA5
The first warnings of strong gale-force winds (21ms −1 ) for western sea areas for 17 September were issued by FMI in the morning of 13 September, 4 days in advance. The highest red level warnings were issued with a lead time of three days. Thus, it was evident that storm Aila was quite well captured by numerical weather prediction models, as the warnings could be given so early. The medium-range forecasts by FMI are mostly based on the IFS model, which is why we next compare the IFS forecasts at different initialisation times to the ERA5 reanalysis.
In Figure 5, MSLP forecasts initialised every 12 hours between 11 September 0000 utc and 16 September 0000 utc by IFS are shown. The valid time of all forecasts is 17 September 0000 utc (see ERA5 analysis in Figure 5(l)), which was approximately the time when the winds were the strongest over Finnish sea areas. Thus, the forecasts have lead times ranging from 144 hours ( Figure 5(a)) to 24 hours ( Figure 5(i)).
The first impression from the MSLP fields in Figure 5 is that the low pressure system is clearly visible in all forecasts. The centre of the storm varied by several hundred kilometres in consecutive forecasts at T+108 to T+144 hour lead times (Figures 5(a-e)), but after 96 hours lead time increased consistency in the location of the centre emerged ( Figures 5(f-k)). However, regardless of the variability in the location of the centre, in all the forecasts a relatively strong pressure gradient in western Finland was present.
Most of the forecasts predicted too low MSLP values over southwestern Finland and too high MSLP values over eastern and northeastern Finland. This is seen as a dipole type of structure in the difference fields which appear to be present in the majority of the forecasts ( Figure 5). Consequently, in these forecasts, the low pressure centre was predicted to be too far west than where it was in reality (Figure 5(l)).
The magnitude of the MSLP errors are more than 15 hPa in the T+96 to T+ 144 hours forecasts (Figures 5(a-e)), but after 84 hours lead time the error generally decreased and was less than 10 hPa (Figures 5(f )-(k)). The central pressure of the system was consistently predicted to be lower than the actual value (996hPa, Figure 5(l)). Only two forecasts predicted a weaker storm (Figure 5(a, d)). Figure 6 shows the predictability of 10-metre maximum wind gust speeds for two specific areas. The x-axis of the panels describes the valid time of the forecasts, and the y-axis the lead time of the fore-casts. In the case of an accurate forecast, the wind gust values (red shading) would agree with ERA5 (shown at the bottom rows in Figure 6). Figure 6 also allows us to determine the consistency of the forecasts. A high degree of consistency means the same values are predicted in multiple, consecutive forecasts which appear in Figure 6 as vertical lines of the same colour. The first area (Figure 6(a)) is located in the western coast of Finland, including quite a big fraction of the Bothnian Sea. This was the area where the strongest 10-metre average wind speed was observed (Figure 2). The second area (Figure 6(b)) is located in southern Finland, and was considered here because this was the area over land where the strongest winds were forecast. Some early indications of the storm-force wind gusts were visible already in the T+216 to T+240 forecasts for both areas. However, the signal was not yet consistent, and partly disappeared at 192-204 hour lead time. After that, starting from 180 hours lead time, the forecast signal for stormy winds began to strengthen and became more consistent. Nevertheless, the forecasts with lead times longer than 144 hours had small timing errors (Figure 6). At 0 to 144 hours lead time, the forecasts form almost invariant vertical lines, which means that the valid times of the strongest gusts remain fixed and thus the forecasts were good. After 144 hours, the forecasts tend to drift rightward towards later valid times, meaning that the strongest winds were forecast to occur 12-24 hours later than when they did in reality.
For western Finland (Figure 6a), compared with ERA5, IFS slightly overestimated the maximum wind gusts for 17 September 0000 utc at short (12-72 hour) lead times, which is a critical time frame for preparations and communication. In southern Finland (Figure 6(b)) on 17 September 1200 utc, the magnitude of the wind gusts were forecast quite well, with some modest underestimation at 48-60 hour lead times.

Figure 5. Mean sea level pressure (contours) from IFS forecasts and their difference from ERA5 reanalysis (colours) at different forecast lead times (a-k). The initialisation times and lead times of the forecasts are shown as titles of the panel figures. The valid time of all panels is 17
September 2020, 0000 utc, which is shown in the last panel (l).
which shows that the centre of the storm in the IFS forecast on 15 September 0000 utc was predicted to be further west, indicating a stronger pressure gradient and thus stronger winds near Rauma on 17 September 0000 utc. In reality, the centre of the storm on 17 September 0000 utc was located in eastern Finland ( Figure 5l) and hence most of the short-term forecasts, including the one initialised on 15 September 0000 utc (Figure 7b), overestimated the windiness along the western coast (Figure 6a). Nevertheless, although having some errors in magnitude, the IFS forecast on 15 September 0000 utc captured fairly well the temporal evolution of wind gusts at both stations.
The observed values are not fully comparable to ERA5 and IFS for two reasons. First, the observations are point values, while both ERA5 and IFS represent spatial averages from a 0.25° grid cell which is nearest to the weather stations. Secondly, the observations are made on isolated islands, on top of lighthouses, at 30 metres altitude (Pietarsaari Kallan, Figure 7(a)) and 38m altitude (Rauma Kylmäpihlaja, Figure 7(b)) while ERA5 and IFS represent 10-metres wind gust values. The ERA5 land-sea mask for Pietarsaari grid point is 0.43 and for Rauma grid point 0.23. This means that the model interprets the Pietarsaari grid point as almost half land while the observations represent pure marine conditions. For these reasons, the observed wind gusts are 3-7ms −1 higher than the modelled wind gusts, especially in Pietarsaari where the fraction of land in the model is higher (Figure 7(a)).

Wind speed comparison to climatology
During storm Aila, the highest wind speeds were observed along the western coast of Finland (Figure 2). Out of all stations, the maxima were recorded in Rauma Kylmäpihlaja with 29.4ms −1 (57kn) 10-minute average wind speed and 34.8ms −1 (68kn) wind gust (Figure 7(b)) and Pietarsaari Kallan with 28.1ms −1 (55kn) 10-minute average wind speed and 35.3ms −1 (69kn) wind gust (Figure 7(a)). We investigated the historical distributions of the wind speeds at these two stations in more detail in order to put the observed values into a climatological context. The comparison is made against September climatology, because Aila occurred in the middle of September. Figure 8 shows histograms of the 1-hour maximum wind speed and wind gust for all observations from all Septembers 2004-2020 from both stations. In addition, the strongest ever observed wind speed and wind gust in all Septembers from 2004-2020 are shown as blue vertical lines in Figure 8. At both stations, the strongest observed winds and gusts during the entire 17-year period (2004-2020) occurred during storm Aila. Similarly, at the closest grid points to these stations in ERA5, the highest winds and gusts are associated with storm Aila (grey vertical lines in Figure 8). We also investigated a longer time period of 1979-2020 from ERA5 (not shown) and a similar result was found. Therefore, we can conclude that storm Aila had the strongest wind speeds  Furthermore, when the maximum wind speeds in all marine and land stations in Finland (excluding mountainous stations which are located over 200 m above sea level) during the whole observational history are considered, no higher September wind speeds are found (not shown). Thus, the observed wind speeds during Aila were records not only at Pietarsaari and Rauma stations (Figures 2, 7 and 8), but for the whole country. However, the caveat in this conclusion is that the observational network of wind speeds in Finland in the twentieth century was more sparse and only instantaneous values were recorded at certain times of the day. Thus, historic observations are likely biased towards lower wind speeds and the comparison must be interpreted with caution.
When comparing the wind distributions between the observations and ERA5 (Figure 8), the wind speeds are weaker and the distributions are less skewed to the right (i.e. narrower distribution) in ERA5 than in the observations. This may be due to the coarser resolution of ERA5, the difference in the wind speed level (10m in ERA5, 30-38m at the stations) and the local features which are not fully resolved in ERA5. Moreover, ERA5 captures the wind gust distribution better than the mean wind speed distribution. In the IFS, the wind gust parameter is calculated by summing up three terms: 10-metre wind speed, a term which represents surface roughness and boundary layer stability, and a convection term (ECMWF, 2013). Therefore, we suggest that while the 10-metre wind speeds are underestimated, some other term in the gust parametrisation, likely the roughness term, is overestimated, meaning that the two errors (of opposite sign) compensate each other. These are, however, values from only two stations and two grid points near the coast and may not represent a larger area over sea or land. This issue needs further research which is out of scope of this study. Furthermore, our finding that both mean and gust wind speeds in ERA5 are negatively biased in the right tail of the distributions (i.e. at high wind speeds) is in agreement with ERA5 wind distributions found in Sweden (Minola et al., 2020).

Conclusions
Storm Aila was a severe autumn windstorm in Finland, affecting mainly the western and southern part of the country. According to preliminary estimations, Aila destroyed about half a million cubic metres of forestry, left 160 000 households without electricity and caused 2950 call outs for the emergency services.
The development of storm Aila was typical for baroclinic cyclones. The storm formed from a pre-existing frontal boundary under upper-level forcing and in a right entrance region of the jet stream. The formation of the surface low occurred only 18 hours before Aila hit Finland, which means that Aila's deepening was still in progress as it arrived in Finland. Although storm Aila was exceptional in terms of its associated winds with substantial impacts, meteorologically Aila was not a deep storm and its deepening cannot be classified as an explosive.
The medium-range forecasts by ECMWF predicted the formation of storm Aila very well and meteorologists at FMI were able to issue warnings with moderately long lead times. For example, the first red warning (the highest level) was announced three days in advance, which indicates both high predictability and high impacts of the event. What is remarkable is that the storm itself formed only 30 hours before hitting Finland on 17 September 0000 utc at its full strength, which means that the IFS model captured the potential development of the storm several days in advance. We speculate that the relatively large spatial scale of the developing storm and presumably the dominance of the adiabatic contributions (e.g., vorticity advection) over the diabatic processes were largely the reasons for the high level of predictability.
The observed wind speeds during storm Aila were exceptional for the time of the year. The maximum mean wind speed and gust wind speed were not only new September records for the weather stations where the readings were observed, but also for Finland as a whole for September. Although slightly higher wind speeds have been observed later in the year when windstorms are usually stronger, storm Aila belongs to the category of most notable windstorms in Finland. Luckily, Aila was well forecast, communicated and adequate preparations were made in time.