• The recent upturn in European storminess is mainly due to multidecadal variations in the two major oceans

• The first is the rebound in tropical Pacific warming (via the IPO cycle)

• The second is cooler waters in the northern North Atlantic which subdues anthropogenic Arctic warming

• Both steepen the polewards heat gradient, and are precursors of stormier mid-latitudes

• Their current states point to rising storminess over the next decade

• An intriguing upwards trend of storminess over the past 70 years is most likely anthropogenic in origin, but is not material on decadal time horizons

 

1. Introduction

Part 1 of this blog series presented a timeseries of European storm damage with pronounced multidecadal changes. It is widely accepted that the storm damage climate varies by more than a factor of two between recent multidecadal peaks and troughs. While insurance can profit from adapting to these storm climate changes, and comfortably able to do so over such slow timescales, there is no cohesive explanation of these past changes to inspire confidence in predictions. Many drivers have been identified by researchers, though their relative roles and interactions are unclear due to the wide variety of storm metrics used by researchers (e.g. discussion in Yau and Chang, 2020), including proxies such as jet waviness or polar vortex changes which are linked to storminess but not easily relatable. The picture gets further blurred by climate models with limited ability to simulate storms (e.g. Priestley et al., 2022).

The first blog found the large-scale equator-to-pole heat gradient was strongly related to decadal-mean European storminess over the past 70 years. This fits with our understanding of the climate system whereby heat gradients control mid-latitude wind strength, and decadal timescales are accompanied by the largest spatial scales of atmosphere variability. How is this useful? Heat anomalies are measured quite robustly, and their evolution is well modelled, therefore storm climate variation becomes a more tractable problem, one which is more amenable to solution. The heat gradient model is used here to understand past variations in decadal storminess, and bolster predictions of future storm activity.

Figure 1 is from the first blog, and shows a breakdown of the polewards gradient into components. In this blog, the first step is to partition these timeseries of changes in Figure 1 into two parts, namely a long-term linear trend and its residual, multidecadal variations. It is assumed that the linear heating trends are caused by anthropogenic activity, while the remaining changes are due to internal climate variability (with the exception of ozone forcing of the Arctic, which contains a multidecadal signal due to a decline which has been reversed, albeit slowly, by the Montreal Protocol). This partition of full timeseries into a linear trend and remainder helps identify drivers and their relative roles in past change.

 

Figure 1: the upper plot shows the change in tropical-mean 70 hPa height, and its three component parts, while the lower plot contains similar information for the Arctic.

 

 

The next section will present the long-term trends in the different components of the polewards gradient, and analyse their causes. The subsequent section describes an investigation of the components of the multidecadal variations of tropical tropopause heights, while Section 4 contains the same for the Arctic tropopause. Findings are brought together into the summary in Section 5.

 

2. Linear trends in storminess over the past 70 years

The best-fitting linear trends of the quantities plotted in Figure 1 are shown in Figure 2, in units of geopotential height (metres) per decade. The headline finding is how warming in the tropics exceeded that in the Arctic by about 3m per decade, hence anthropogenic activity caused storminess in Europe to increase over the past 70 years. This estimate is confounded by the specific phases of multidecadal variability at the start and end of each 72-year timeseries, and it is seen later how the Arctic was unusually warm in the 1950s, and the tropics less so. If the linear trend is computed from the early 1960s onwards, then it is found that the tropics-to-pole gradient rose by 1.5m per decade, and if starting from the 1970/71 season then the trend reverses to a 1.4m decline per decade. To put this uncertainty into perspective, Figure 4 in the first blog shows a 60m increase in the equator-to-pole gradient from the early 1970s to about 1990, followed by the same-sized decline to the mid- 2000s. Therefore, any linear trend from anthropogenic forcing of European storms could be impactful at century timescales, but is not material over one or two decades. Finally, for loss context, the 60m multidecadal changes correspond to roughly a factor two loss change, hence this linear trend causes low single digit percent changes in loss over a decade.

 

Figure 2:  linear trends in geopotential heights (m) per decade for the tropics and the Arctic.

The sign of the linear trends in the components displayed in Figure 2 match prior expectations of the impacts of anthropogenic forcing. A significant warming of the 1000-300 hPa layer in the tropics is expected, as is the corresponding warming at the pole which is slightly larger due to Arctic Amplification. The smaller increase in the tropical upper troposphere is caused by large warming up to about 150 hPa being partly offset by cooling in a layer from 150 to 70 hPa. This is a transitional layer, from pure troposphere air below to the stratosphere which has experienced large cooling from raised greenhouse gases, and ozone destruction (e.g. Mitchell et al., 2020). The negative height trend from surface pressure changes over the Arctic is consistent with the cooling of the polar stratosphere due to both ozone loss and raised greenhouse gas amounts, driving a deeper polar vortex.

In summary, a small rising trend in storminess is observed over the past 70 years. This is presumably due to anthropogenic forcing, and more specifically, Arctic stratosphere cooling outweighing the differential warming of the Arctic versus tropical troposphere. While the direction is notable, the magnitude is quite immaterial at decadal timescales.

 

3. Multidecadal variations in the tropics

The linear trends presented in Figure 2 were removed from their respective full timeseries to leave the residual, multidecadal signals shown in Figure 3 for the tropical region. Clearly, the thickness of the upper troposphere has been the key driver of multidecadal changes in the tropics in the past 70 years. These upper troposphere heat anomalies are the subject of the remainder of this section.

 

Figure 3: the de-trended variations of tropical-mean 70 hPa height, and its three component parts.

 

Zhang et al. (1997, their Figure 11) showed the Pacific region contributes the majority of tropic-wide multidecadal variability. In turn, Power et al. (2021) show that around half of all multidecadal variability in the tropical Pacific is explained by a pattern very similar to the Interdecadal Pacific Oscillation (IPO, e.g. Power et al., 1999) in their Figure 2A, and reproduced here as Figure 4. Further, Kamae et al. (2015) showed how a spatial pattern of sea surface temperature (SST) changes very similar to the IPO was responsible for the hiatus in tropical upper troposphere warming from 1997 to 2011. In short, researchers find the IPO is the dominant driver of multidecadal variability in the tropical upper troposphere.

 

Figure 4: the SST signal associated with the dominant mode of tropical Pacific decadal variability. A copy of Figure 2A in Power et al. (2021).  

Figure 5 shows the timeseries of de-trended upper troposphere thickness, alongside the IPO (based on the Tripole Index of Henley et al., 2015, available from here) and their connection is apparent. The move to the positive phase of the IPO from the 1970s to 1980s played a significant role in raising European storminess in the 1980s and 1990s, while the reversal to a negative IPO contributed to the big storm decline in the first decade of the 21^st century. The processes which combine to produce the IPO are actively being investigated: Meehl et al. (2016) analysed IPO mechanisms simulated by a climate model, and found sub-tropical cells in the Pacific Ocean play a key role. The full picture is not established, though it seems more likely to be a predictable process given the high level of skill of Meehl et al.’s climate model in predicting the IPO and phase changes, and in a multi-model ensemble described in Meehl et al. (2014).

 

Figure 5: timeseries of 9-yr mean tropical upper troposphere thickness, and IPO, both converted to standardised anomalies for comparison purposes.

As discussed in the previous section, the upper half of the 70-300 hPa layer partly contains lower stratospheric changes too. The lower stratosphere has experienced substantial multidecadal changes, from significant cooling in the latter part of the 20^th century, to much less change over the past 20 years (Mitchell et al., 2020). The main driver is widely thought to be the decline of ozone due to anthropogenic activity and its subsequent partial recovery. The ozone changes occur alongside the continual rise of greenhouse gases driving a radiative cooling of this level. Recent observations indicate the ozone recovery offsets the greenhouse gas build-up to produce little change over the 21^st century.

In terms of the medium-term outlook, Thoma et al. (2015) describe prediction skill in their system for the PDO (Pacific Decadal Oscillation, a closely related phenomenon found when isolating the North Pacific variations), and they forecasted that the recent flip to a positive phase will be maintained out to the end of their forecast period, to 2024. In addition, the latest WMO Global Annual to Decadal Climate Update issue a forecast for near-surface air temperatures for the period 2022-2026 which contains no IPO-like signal, which suggests a continued upturn of the IPO index in Figure 5 to neutral. Ozone amounts in the stratosphere are expected to continue to recover, which suggests recent small heat trends in the lower stratosphere will persist. In summary, the more likely outcome for the 70-300 hPa layer over the next decade is a continuation of its recent enhanced heating trend.

 

4. Multidecadal variations in the Arctic

The linear trends from Figure 2 were removed from the full timeseries to reveal the multidecadal variations in the Arctic region shown in Figure 6. Following the first blog in this series, the Arctic column mass changes will be viewed as heat anomalies in the Arctic lower stratosphere in the following discussion, arising from the tight coupling between these two quantities.

 

Figure 6: the de-trended variations of Arctic-mean 300 hPa height, and its two component parts.

Figure 6 shows largely in-phase changes in the Arctic troposphere (red) and stratosphere (green), which fits with the vertical coherency of the polar vortex. As a result, both component parts will be analysed together. The deep dip in the 1980s/90s is by far the most important multidecadal feature: it has the largest amplitude, and it coincides with peak European storminess. Several candidate drivers have been identified by researchers. First, the northern North Atlantic Ocean waters had negative temperature anomalies and this cools the Arctic in two ways: (i) by cooling the Arctic lower stratosphere (e.g. Omrani et al., 2014), and (ii) raising sea-ice amounts (Årthun et al., 2017) through less heat transport to Arctic Seas, which cool the lower troposphere of the Arctic. Second, there were two explosive volcanoes in 1982 (El Chichon) and 1991 (Pinatubo) which are known to strengthen hence cool the Arctic winter vortex (e.g. Graf et al., 1993). Third, the ozone destruction from human activity cooled the lower stratosphere (e.g. Philipona et al., 2018).

There have been no recent volcanic eruptions with explosivity close to the climate-changing impacts of Pinatubo, and the radiative impacts of the ongoing recovery of ozone seem to be cancelled by the ongoing rise of greenhouse gases (e.g. Mitchell et al., 2020). Therefore, the northern Atlantic Ocean heat content is likely to be the main driver of multidecadal variations of Arctic heat in the near future. The two effects of this driver mentioned in the previous paragraph have been assessed using heat contents in the top 100m of the ocean in the regions shown in Figure 7. In brief, one effect was measured using a bipole index following the spatial pattern found to force the polar vortex in Hu et al., 2019 (their Figure 1c), while the forcing of sea-ice extents by anomalous ocean heat transport used a region in the northeast Atlantic (green box in Figure 7).

 

Figure 7: the ocean regions used to define heat contents in the North Atlantic Ocean. A bipole index was estimated as the average heat content in the two red boxes, following Figure 1c of Hu et al., 2019. The heat content of waters flowing into the Arctic Seas is tracked using the green box.

The timeseries of the Atlantic bipole index is shown in the upper panel of Figure 8, and it has recently dipped to a negative state. This forces a cooling of the polar vortex. In the absence of specific guidance on how the bipole heat content may evolve, the persistence of recent behaviour is a reasonable default, and suggests a continuation of the recent downward trend. However, less confidence can be placed in persistence-based forecasts, and in this case their uncertainties are amplified by the surprisingly volatile flips of such enormous ocean heat amounts, almost at annual timescales (e.g. the mid-1990s and around 2010), and caused by anomalous ocean currents. There is low confidence in the bipole index forecast.

 

Figure 8: heat contents for the bipole index (upper) and upstream of the Arctic Seas (lower). Heat contents of the top 100m are from EN4.2.2 ocean reanalyses (Good et al., 2013). The timeseries in each region in Figure 7 was de-trended, then 9-yr means computed, and subsequently converted into standardised anomalies. The sea-ice in the lower panel is from HadISST (Rayner et al., 2003) with a correction applied to pre-satellite era data using ORAS5 to reduce a major inhomogeneity in sea-ice area observations.

The lower panel in Figure 8 shows how waters have cooled markedly in the region upstream of the Norwegian Sea over the past ten years (black solid line). Its relationship to Arctic sea-ice areas (red dashed line, note values are reversed in plot) differs before and after 1980. Given much larger uncertainties in observed Arctic sea-ice area in the pre-satellite era before 1979, more confidence is placed in the modern period, showing how ocean heat contents lead Arctic sea-ice area by 5 to 10 years, as described in Årthun et al., 2017. The current cool state of the waters upstream of the Norwegian Sea will lead to increases in Arctic sea-ice area and cooling of this region over the next few years. Note these increases from multidecadal variations are independent of anthropogenic forcing. The latter will continue melting sea-ice at a pace which might overwhelm the multidecadal tendency and produce net reductions over the next decade. Nevertheless, the tendency from multidecadal variability is towards more sea-ice. Confidence in this forecast is boosted by its basis in process-based analysis (e.g. Zhang 2015; Årthun et al., 2017) however those researchers studied heat transport anomalies, and omission of the effects of ocean currents lead to less confidence here, compared to what was reported by the researchers.

 

5. Summary

This blog develops a storm outlook for the next ten years by considering how the drivers of the polewards heat gradient will evolve.

Long-term trends are consistent with anthropogenic forcing and reveal a slowly rising storminess in Europe over the past 70 years. The sign of change is intriguing, but the most salient point is they are quite immaterial to the windstorm loss climate in the next decade.

In contrast, multidecadal variations have substantial impacts on decadal-mean losses. Two major ocean regions are likely to have caused the recent upturn in European storminess, and both are primed to drive further increases over the next decade. First, the enhanced warming of the tropical Pacific seems likely to continue to heat the tropics and boost the polewards gradient over the next few years. Second, the northern Atlantic Ocean is now in a cool phase of multidecadal variability, which cools the Arctic both by reducing heat transport into the Arctic Seas, and by lowering temperatures in the Arctic stratosphere via atmosphere dynamics. The ongoing slow recovery of ozone levels will warm the Arctic but unlikely to fully offset the ocean drivers. The stage is set for rising storminess in Europe over the next decade, and most likely in fits and starts. We should not be lulled by quieter years: January-March 1990 and January 1993 are flanked by seasons with historically low activity. 

Model limitations include the fact that the polewards heat gradient has an imperfect correlation with windstorm damage, and it is near-certain that one or more of the model components will follow a different path from the estimates here. Further, a new driver could emerge, especially in these changing times. Though the model’s physical basis and accurate accounting of past variations in decadal storm damage means it should provide useful guidance.