• Europe's windstorm climate has varied by over a factor two at decadal scales, in recent times
• It hit a trough about 10-15 years ago and has since been rising
• Pricing can adapt to these long timescales, but is hampered by little understanding of the variations
• Here, it is shown how the equator-to-pole heat gradient is aligned with decadal storm activity over the past 70 years - as expected from basic principles of the general circulation
• Re-framing storm variations as changes in polewards heat gradient is a very promising path towards better pricing of windstorm risk, more to follow in Part 2 of this series

 

Introduction

More experienced people may recall some very windy autumns and winters in the 1980s and 1990s, and aware of how it’s quietened down since then. The first section of this blog presents a quantitative view of windstorm variations from 1950 to the present day, to give the bigger picture. Notably, there is a new upwards trend in storminess over the past ten years.

It is entirely practical for pricing to adapt to large changes in storm climate risk over these long timescales, however, this potential is not yet realised due to little understanding of the drivers. The main part of this blog is focussed on developing insights into the drivers. Specifically, it is found that the myriad forcings at interannual and shorter timescales, and chaotic weather noise, are largely removed by decadal-scale averaging, and what remains is the fundamental driver of windstorms: the equator-to-pole heat gradients. This framework gives new insights into drivers of past windstorm activity, including what’s caused the recent upturn, and further, it lays the foundation for the second part of this blog series, on the drivers and their indications for European storminess over the next decade.

European storminess from 1950 to 2022

ERA-5 hourly winds at 10m above short grass were used to construct a record of storm damage from 1950 to the present day. In brief, the maximum winds were found per storm event, which was defined as up to a 3-day window for every grid cell in Europe. This source of winds was validated for loss modelling purposes by applying the loss model from Klawa and Ulbrich (2003), which essentially assumes damage amounts are proportional to (a) the cube of the wind excess above the local 98th percentile, and (b) the exposure, which is represented by the population density.

Figure 1 shows the timeseries of annual losses when the model's constant of proportionality is chosen to produce a €25 bn loss for 1990. The relativity of loss between years is in-line with expectations from sources of industry losses such as PERILS, Munich Re, Swiss Re and other national organisations. We expect to see 1990 and 1999 as the two stand-out years, followed by 1972, 1976 and 2007 with their extreme events (Lower Saxony, Capella and Kyrill, respectively). There are signs that ERA-5 over-estimates losses from spatially large storms (e.g. Jeanette in 2002, or the sequence in mid-February 2022) versus storms with damage caused by very active fronts (e.g. 87J), but the overall impression from Figure 1, and inspection of individual event losses, is that ERA-5 winds (not gusts) provide very reasonable storm hazard. 

Figure 1: annual timeseries of Europe-wide windstorm losses based on ERA-5 winds input into the Klawa and Ulbrich (2003) loss model. The red line shows the 9-yr running mean (centred).

The long timescale variations of the jet are manifested as changes in its strength rather than its latitude (Woollings et al., 2015). However, decadal storm losses can be sensitive to jet latitude due to much higher exposure to the south of the mean storm track, and greater vulnerability stemming from adaptation to a weaker wind climate. The more feasible first step towards understanding storminess is to explain changes in the total strength of storms. Therefore, to reflect changes in strength of all European storms, the following analysis will use a modified storm index based on a constant wind threshold at every location (rather than their local 98^th percentile). Figure 2 shows European storm index values from 1950 to the present-day. These values represent the 9-yr running mean of the log of the annual index, then standardised. The log of the storm index is more suitable for the linear analysis in the next section.

Figure 2: centred 9-yr running mean of the European storm index, standardised.

The storm index has peak values in the 1980s and '90s and the subsequent decline mentioned earlier. The recovery in storm activity over the past ten years is very intriguing. It fits with recent significant losses from storms in mid-February 2020 and 2022, and January 2018. This recent upwards trend could be analogous to the 1970s and a sign of stormier times ahead. Knowing more about these decadal variations would help manage the billions of euros of exposure to this risk.

Mid-latitude storminess and polewards heat gradients

A basic principle of the general circulation dictates how mid-latitude wind strength is a function of the poleward gradient of heat within the troposphere. A diagnostic has been devised to reflect this fact. It is defined as the difference in geopotential heights at the 70 hPa level in the tropics (20°S to 20°N), and 300 hPa level in the Arctic (north of 60°N). The two different pressure levels were chosen to reflect the different depths of the troposphere at the equator and pole. The 9-yr running means were computed, and they are compared to 9-yr running means of the European storm index in Figure 3. Evidently, the poleward heat gradient has a very close relationship to European storminess (correlation of 0.86). The decadal averaging is key here: various processes will interact to produce different storminess per mid-latitude region in different months or seasons, with chaotic weather also adding noise. However, the large-scale coupling emerges more clearly over longer timescales simply by averaging out these faster timescale, shorter spatial scale variations.

Figure 3: 9-yr running means (standardised) of the measure of tropics-to-pole heat gradient from reanalyses, and the European storm index. (The reanalyses data are a blend of ERA-5 and NCEP, to reduce uncertainty in tropical upper troposphere temperature trends in the 1980s and '90s.)

With a basis in the fundamentals of atmosphere dynamics and strongly supported by observations over the past 70+ years, we can conclude that understanding decadal variations in storminess is equivalent to understanding the changes in the polewards heat gradient. This different perspective on storminess is the focus of the remainder of this blog.

Figure 4 shows the breakdown of this polewards gradient into its two components. It can be seen that the sharp peak in the 1980s and '90s was caused by a steepening of the polewards gradient both by warming in the tropics and cooling in the Arctic. The subsequent dip to the trough around 2010 was caused by reversals in both regions: a cooling in the tropics and warming of the Arctic. The uplift in storminess over the past ten years has been caused by relative warming in the tropics versus the Arctic.

Figure 4: change in running 9-yr mean tropopause geopotential heights in the tropics and Arctic.

The total change in geopotential height in each region was broken down into component parts consisting of (a) change in column mass (surface pressure), (b) change in heating of the 1000 hPa to 300 hPa layer, and (c) change in heating in the 300 hPa to 70 hPa layer (directly relevant for tropics only). To assist with intercomparison, all anomalies are now plotted in units of height (in metres).

Figure 5 shows the component drivers for the tropics, and three main messages appear:

1. upper troposphere warming from the mid-1970s to mid-1990s was the major source of the tropical part of the forcing of this very stormy period in Europe
2. gradual warming of the bulk of the tropical troposphere, most likely due to anthropogenic activity, is material to European storminess over longer timescales
3. decadal-scale changes to column mass in the tropics are relatively small

 

Figure 5: change in tropical-mean 70 hPa height, and its three component parts (see text).

Figure 6 shows the changes in Arctic troposphere heights broken down into its two components, both of which make considerable contributions to multidecadal variability and longer timescale trends. The troposphere thickness from 1000 to 300 hPa has a warming trend which exceeds the same in the tropics, and most likely a consequence of the Arctic Amplification of global-mean warming. Note though that the mean surface pressure has been trending downwards over the Arctic in the past 70 years, and offsets some of the warming below. In terms of decadal-scale variations, the pronounced dip in surface pressure in the 1980s and '90s, causing the 300 hPa height to fall by about 20m, is perhaps the most notable feature. This was compounded by a cooling of the Arctic troposphere to produce a total reduction of 30m in geopotential heights at the pole.

Figure 6: change in Arctic 300 hPa height, and its component drivers.

In contrast to the tropics, the change in surface pressure has exerted significant influence on Arctic upper troposphere geopotential heights. In turn, surface pressure is known to be connected to the polar stratosphere: when it’s colder in the stratosphere, the polar vortex intensifies and surface pressure drops. Figure 7 shows this relationship from reanalyses: a cooler stratosphere is accompanied by lower surface pressures over the Arctic. Anthropogenic activities – both increased greenhouse gases and ozone depletion – are considered to be driving the lower stratosphere cooling over the poles in the past several decades. This is discussed more in the second blog of this series.

Figure 7: change in Arctic surface pressure, together with a measure of lower stratosphere temperatures, namely the thickness of the layer from 300 to 70 hPa.

In conclusion, changes in the gradient of heat between the tropics and the Arctic give a fresh perspective on the large decadal variations of European storminess. For example, the recent uptick in storminess is largely due to differential heating between the tropics and the Arctic: marked warming of the upper troposphere has lifted tropical troposphere heights, whereas the Arctic troposphere has warmed by a smaller amount in the past 10 or so years, and largely offset by cooling in its lower stratosphere causing surface pressures to fall. We will use this understanding of storm activity towards a medium-term outlook for European windstorms in the second blog of this series.

 

 

Summary

In addition to the change from the late 20^th to early 21^st centuries in which storminess was more than halved, a reconstruction of European windstorms over the past 70 years using ERA-5 winds reveals an upturn in activity over the past 10 years.

While the slow and substantial variations in European storminess have great potential for improved risk management, we have had to sit and suffer this climate variability due to a lack of understanding of its drivers. This blog has shown how storm variations at decadal timescales are closely tied to the equator-to-pole heat gradient, hence storm variations can be alternatively expressed in terms of changes in this slope. For example, the rising storminess over the past decade was caused by differential heating: a significant warming of the tropical upper troposphere, versus little change in Arctic heat (weaker warming of the polar troposphere largely offset by a cooling of its lower stratosphere causing a deeper polar vortex).

The re-expression in terms of a polewards heat gradient offers a new way to understand how Europe's storm climate varies. The second part of this blog series will analyse drivers of this heat gradient, towards reducing the surprise when winters such as 1990 or 1999 strike again.