Climate change increases the “perfect storm” coastal flood potential

By Emanuele Bevacqua

Some of the European low-lying coastal areas and river estuaries may see a future increase in flooding caused not only by sea-level rise but also by more frequent concurrent storm surge and heavy precipitation, we show in a new study published in Science Advances [1].

Figure 1: Compound flooding caused by the interaction between storm surge and heavy rainfall. A moderate storm surge can obstruct the gravity-feed drainage of pluvial or fluvial water into the sea. Or, rainfall on top of a storm surge inundated area can worsen the overall flooding. Image originally published on the BBC.

Coastal flood risk is usually calculated based on storm surge and extreme rainfall hazards separately. However, ignoring compound flooding arising from the local interaction between these two hazards may cause severe underestimation of the actual flooding risk in some areas.

The importance of considering compound flooding for planning coastal defences is shown, for example, by the past flooding management in Lymington [3], a port town in the southern UK. In December 1989, a storm surge led to widespread flooding in the town, causing damages to houses and the railway line. As a result, an extensive upgrade of the coastal defences took place; however, the potential for flooding resulting from the interaction of storm surge and pluvial flooding was not considered at that time. Ten years later in 1999, the Christmas Eve in Lymington was stormy: a cyclone not only caused a storm surge but, simultaneously, also very heavy rainfall leading to high discharge in the Lymington River. The new raised defences were activated and protected the town from the storm surge well; however, the prolonged storm surge did not allow the sea defences to be deactivated, thus trapping the water discharge of the Lymington River on the upstream side of the sea defences. This combination of events resulted in very deep flooding over the railway line and into residential and commercial development in Lymington.

Figure 2: Flooding around Lymington (UK) on December 24, 1999. Photo from The New Forest LIFE Projects.

Similar compound flooding happened, for example, in 2012 in The Netherlands [4], and in 2015 in Ravenna (Italy) [2] due to the cyclone Norbert causing concurring heavy rainfall and storm surge. In fact, coastal heavy rain often coincides with a storm surge because these two hazards frequently result from the passage of a cyclone. Thus, considering only one of the two hazards at a time, or assuming that they are independent, can result in a substantial underestimation of the flooding risk.

It is clear that in low-lying coastal areas and river estuaries, compound flooding should be considered to avoid underestimating the flooding risk. However, the need for considering the multiple flooding drivers simultaneously makes compound flooding risk assessment a rather complicated task. In addition, as climate changes, we need to understand how the compound flooding risk will change in the future, which is vital for adapting to flooding changes through – when necessary – building or upgrading flood defences. Thus, also future changes in sea level and precipitation need to be taken into account when assessing the flooding risk, making a compound flooding assessment extremely challenging.

In a new study, colleagues and I investigated, for the first time, how compound flooding may change in the future due to sea-level rise and changes in precipitation, storm surge, and their interaction.

Focussing on the European and North African coasts, we found that not only sea-level rise will cause an increase in the potential for compound flooding everywhere, but additional changes in compound flooding will be caused by changes in storms. In particular, the potential for compound flooding is projected to increase in northern Europe due to increasing precipitation intensities, mainly caused by a future warmer atmosphere that allows storms to carry more moisture.

The results of the study cannot be interpreted as a precise assessment of the local flooding risk, though it is evident that compound flooding will become more likely in the future if we do not take any action against climate change. The study aims at offering a large-scale view of the present and future potential for the compound flooding hazard. Identifying European regions potentially facing compound flooding in a warmer climate, it provides a basis for follow-up local compound flooding risk assessments and adaptation planning.

In locations prone to compound flooding, adapting to sea-level rise only might be insufficient, and additional changes in precipitation, storm surges, and their interaction should be considered to avoid too optimistic adaptation planning.


 [1] Bevacqua, E., D. Maraun, M. I. Vousdoukas, E. Voukouvalas, M. Vrac, L. Mentaschi, and M. Widmann, 2019: Higher probability of compound flooding from precipitation and storm surge in Europe under anthropogenic climate change. Science Advances, 5, no. 9, eaaw5531,

[2] Bevacqua, E., D. Maraun, I. Hobæk Haff, M. Widmann, and M. Vrac, 2017: Multivariate statistical modelling of compound events via pair-copula constructions: analysis of floods in Ravenna (Italy). Hydrology and Earth System Sciences, 21, no. 6, 2701-2723,

 [3] Hendry, A., I. Haigh, R. Nicholls, H. Winter, R. Neal, T. Wahl, A. Joly-Laugel, and S. Darby, 2019: Assessing the characteristics and drivers of compound flooding events around the UK coast. Hydrology and Earth System Sciences, 23, 3117-3139,

[4] van den Hurk, Bart, Erik van Meijgaard, Paul de Valk, Klaas-Jan van Heeringen, and Jan Gooijer, 2015: Analysis of a compounding surge and precipitation event in the Netherlands. Environmental Research Letters, 10, no. 3, 035001,

Posted in Climate, Climate change, Flooding | Leave a comment

The giant space plasma waves that can destroy our satellites

By: Sarah Bentley

Everyday life is becoming more and more dependent on satellite services. From critical communications to forecasting and GPS, we would feel the impact of these lost services quickly. The location and accurate time provided by GPS is vital for navigation, the power grid, computers, phones, financial transactions… even the food we eat in the UK relies on timely transport.  (The BBC recently published an article on the impact of GPS which you can read here.) However, satellites are susceptible to several types of damage through space weather.

The region of space near Earth that is dominated by our planet’s magnetic field is known as the magnetosphere. This provides us and our satellites significant protection from the fast-moving and highly charged solar wind streaming out from the Sun. However, this barrier is highly dynamic as the solar wind constantly buffets us on its way through the solar system. This buffeting causes giant large-scale waves that can bounce around inside the magnetosphere, energising and transporting trapped electrons and posing a hazard to satellites that reside in the radiation belt region. The radiation belts are doughnut-shaped regions that contain many charged particles, trapped by the magnetic field and continually travelling around the Earth.

Figure 1: The Van Allen probes spent seven years observing the radiation belts, the region of energetic trapped particles that are hazardous to satellites. Credits: NASA Goddard’s Scientific Visualization Studio

Actually, there are multiple ways in which space weather can threaten satellite services. High energy particles such as “killer electrons” can penetrate and ionise individual components. Electrostatic charging and the sudden discharges can develop on internal or external surfaces, and communications through the ionosphere can be severely disrupted. However, only some of these are directly related to the giant plasma waves I study – ultra-low frequency or ULF waves. These waves affect the energisation and transport of high-energy electrons. Since satellite operators don’t like to advertise when they have spacecraft failures or the details of those failures, it’s difficult to pinpoint many occasions where these waves were the dominant factor. One well-known example is the failure of the two Telesat spacecraft in 1994, Anik-E1 and E2 [Lam et al, 2012]. We’re aware of this because of the sheer amount of disruption caused; these satellites carried virtually all Canadian television and a significant amount of communications capability. Energetic electrons accelerated by ULF waves damaged the satellites such that it took over six months to regain full services and a hundred thousand customers had to manually re-orient their satellite dishes towards the recovered spacecraft.

Figure 2: Perturbations at the magnetopause can drive waves that propagate inwards, disturbing Earth’s magnetic field. These ULF waves can reflect, bounce or form standing waves that can be measured at the ground. Credit: Sarah Bentley

Technically, we classify “ultra-low frequency” to be below 30 Hz, but I’m mostly interested in the 2-10 mHz range of waves. ULF waves are plasma waves, which means that as well as involving density compressions like a sound (or fluid) wave, they can also incorporate oscillations in the electric and magnetic field. All these oscillations are so large that we measure their period in minutes or hours and their wavelengths in thousands of kilometres – comparable to the radius of the Earth. These waves have their strongest effects when we see sustained wave activity. In this case, repeated electric field pulses can coincide with the motion of the multitude of electrons zipping around the Earth. The electric field exerts a force on the trapped electrons, accelerating them or moving them to different areas of the magnetosphere. So, predicting the occurrence, amplitude and extent of these, is an important aspect of forecasting the radiation belt environment. This would enable satellite operators to take steps to protect their spacecraft, for example by moving the satellite or shutting down vulnerable components.

Unfortunately, predicting the extent of these waves isn’t always that easy. Because they are so big, it’s more computationally feasible to simulate them numerically than other aspects of the magnetosphere. But it’s still very time-consuming, and we can’t run simulations that correspond to a given time in real life because we would need to know what the boundary of the magnetosphere looks like and what the driving solar wind is doing. Typically, we only have a single point measurement in the solar wind near the Earth, which is just not enough to fully describe an environment hundreds of thousands of kilometres wide. An easier approximation we have been making is to simply use an empirical, statistical model which gives median ULF wave power under different solar wind conditions [Bentley et al, 2019]. To our surprise, just using solar wind properties predicted the ULF wave power better than if we assume that we would see the same power from one hour to the next – our model was much more successful than anticipated! We found that the amount of energy seen in these waves changes with different solar wind properties to those we expected. This suggests that the driving of these waves is slightly more complicated than previously realised and that the processing of the magnetosphere may be more important. 

Eventually, we expect that a model like this will improve radiation belt forecasting. This is becoming more important than ever – while the last decade or so has been particularly quiet for the radiation belts, there’s no guarantee this will last. In this time, we have become ever more dependent on satellites, and the way that we use them has made them more susceptible to radiation damage. Spacecraft now often use off-the-shelf rather than custom radiation hard components, and cheaper methods of getting into orbit means that they spend even longer in the radiation belts. So, we hope to understand and predict these giant waves better and discover more about the complex and weird behaviour in the area dominated by the Earth’s magnetic field.


Bentley, S. N., Watt, C. E. J., Rae, I. J., Owens, M. J., Murphy, K., Lockwood, M., & Sandhu, J. K. ( 2019). Capturing uncertainty in magnetospheric ultralow frequency wave models. Space Weather, 17, 599– 618.

Lam, H.‐L., Boteler, D. H., Burlton, B., and Evans, J. ( 2012), Anik‐E1 and E2 satellite failures of January 1994 revisited, Space Weather, 10, S10003, doi:10.1029/2012SW000811.

Horne, R.B., Glauert, S.A., Meredith, N.P., Koskinen, H., Vainio, R., Afanasiev, A., Ganushkina, N.Y., Amariutei, O.A., Boscher, D., Sicard, A., Maget, V., Poedts, S., Jacobs, C., Sanahuja, B., Aran, A., Heynderickx D., and Pitchford, D., (2013), Forecasting the Earth’s radiation belts and modelling solar energetic particle events: Recent results from SPACECAST, J. Space Weather Space Clim., 3 (2013) A20, doi: 10.1051/swsc/2013042

Posted in Climate, space weather | Leave a comment

The latest on aerosol radiative forcing

By: Nicolas Bellouin

Aerosols are tiny liquid or solid particles suspended in the Earth’s atmosphere. Some aerosols form naturally, like the sea spray emitted by breaking waves, the mineral dust that form sandstorms, or smoke from wildfires. But human activities, from combustion of fossil fuels, cement manufacturing, fertilisers, and agricultural and forest clearing fires also emit aerosols into the atmosphere.

Once in the atmosphere, aerosols affect the energy budget of the Earth by reflecting and absorbing sunlight. They also play an important role in the formation of liquid clouds, and aerosols from human activities lead to the formation of more reflective clouds, which may also have a different liquid water content or be slower to rain out. The extent to which aerosols from human activities modify the energy budget of the Earth is measured by a quantity called aerosol radiative forcing.

Scientists have studied aerosols for a long time. Their role in the formation of fog was identified by the Scottish meteorologist John Aitken in 1880. In the 1960s and 1970s, scientists observed ships making clouds thicker (the so-called ship tracks) and paper mills affecting precipitation downwind of their location. But the global extent of aerosol perturbations was only revealed during the 1980s and 1990s when ground-based observations networks, aircraft campaigns, and satellite instruments tremendously increased our ability to observe aerosol properties.

Figure 1: Distributions of aerosol optical depth, a measure of aerosol loading of the atmosphere, as estimated in (left) 1984 based on knowledge of the time; and (right) 2019 based on a numerical model of atmospheric composition that ingests satellite observations of aerosols.

Figure 1 shows how understanding of global aerosol patterns has dramatically improved thanks to better observations and better models of aerosol sources, transport, and removal. The research done over the past 40 years identified aerosol sources and their variability, characterised the wide spectrum of aerosol properties as they “age” in the atmosphere, and discovered the complexity of the natural atmosphere.

Figure 2: Uncertainty ranges for aerosol radiative forcing from interactions with radiation (ari, left in blue columns) and interactions with clouds (aci, right in orange columns) in past Assessment Reports (AR) of the Intergovernmental Panel on Climate Change. The first report did not quantify aerosol radiative forcing but noted that it was potentially substantial.

That improved understanding quickly translated into the first estimates of aerosol radiative forcing, but progress then seemed to stall. Figure 2 shows estimates of the uncertainty ranges of aerosol radiative forcing as assessed by successive reports of the Intergovernmental Panel on Climate Change. Despite the large, international intellectual effort dedicated to observing, understanding, and modelling the impact of aerosols on climate, uncertainty in aerosol radiative forcing has not changed much over the past decades.

Over the past year, I have led more than 30 colleagues, each bringing complementary expertise in the many ways we observe and model aerosols, to review the scientific literature on aerosol radiative forcing. Our aim was to take a fresh and comprehensive look at present understanding of aerosol radiative forcing and identify prospects for progress on some of the most pressing open questions. On 1 November 2019, the scientific journal Reviews of Geophysics published a preliminary version of our review article .

We found that important aerosol radiative forcing mechanisms are getting better constrained. The balance between aerosol reflection and absorption of sunlight, and the degree to which aerosols from human activities increase the number of liquid cloud droplets, are now well understood. In addition, a series of recent observations, including from volcanic eruptions, found that cloud liquid water content was less sensitive to aerosol perturbations than previously thought.

But important gaps in our understanding remain and those gaps explain why uncertainty in aerosol radiative forcing remains large. Some of the most important research questions are:

  • What were the aerosols levels before industrialisation dramatically increased human emissions? How polluted was the natural atmosphere by wildfires and emissions from plants, deserts, and oceans? Those questions are difficult because preindustrial aerosols have not been observed, and numerical models incompletely represent the sources and properties of natural aerosols.
  • Do aerosol perturbations increase cloud fraction, as predicted by large-scale models of the atmosphere? Or is the picture more nuanced, as suggested by satellite observations or predicted by smaller-scale, more detailed models? Those questions are difficult because cloud fraction is a poorly-defined quantity and depends on the scale at which clouds are observed and simulated.
  • Do ice clouds respond to perturbations of aerosol amounts by human emissions? And if so, what is the associated radiative forcing? Those questions are difficult because ice crystal number is poorly observed, and aerosol-ice interactions are not well understood theoretically.

To quantify the remaining uncertainties, my colleagues and I worked out that aerosol radiative forcing offset at least a fifth and up to half of the radiative forcing by greenhouse gases, with 2 chances out of 3 that the right answer lies between those two bounds. We also identified new approaches that hold great promise to make further progress:

  • Statistical methods to compare multiple model configurations to better, more numerous observations improves our understanding of the sources of uncertainty in models.
  • Increases in computing power allow simulations of the whole atmosphere with detailed models of aerosol-cloud interactions, which better represent subtle processes.
  • Natural laboratories like volcanic eruptions and ship tracks provide opportunities to understand aerosol perturbations of ice clouds.
  • Analyses of sediments to find traces of ancient charcoal give unprecedented insights into past fire activity.
  • Observed changes in surface temperature, ocean heat content, or sunlight levels at the surface can be used to constrain aerosol radiative forcing. The use of differences between the two hemispheres, or between time periods where aerosol pollution increases or decreases, may make those inferences stronger.

Attempting to quantify aerosol radiative forcing has given us fascinating insights into the atmosphere of our planet and its climate system. Many questions remain wide open and provide much to excite the curiosity of the best physicists, chemists, and mathematicians. If a new group of aerosol scientists repeats our review in 10 years’ time, I hope they will be able to marvel at the progress accomplished.


 Bellouin, N., Quaas, J., Gryspeerdt, E., Kinne, S., Stier, P., Watson‐Parris, D., et al (2019). Bounding global aerosol radiative forcing of climate change. Reviews of Geophysics, 57.

Tanre, D., J.-F. Geleyn, and J. Slingo (1984), First results of the introduction of an advanced aerosol-radiation interaction in the ECMWF low resolution global model, in Aerosols and Their Climatic Effects, edited by H. Gerber and A. Deepak, pp. 133–177, A. Deepak, Hampton,Va.

Inness, A., Ades, M., Agustí-Panareda, A., Barré, J., Benedictow, A., Blechschmidt, A.-M., Dominguez, J. J., Engelen, R., Eskes, H., Flemming, J., Huijnen, V., Jones, L., Kipling, Z., Massart, S., Parrington, M., Peuch, V.-H., Razinger, M., Remy, S., Schulz, M., and Suttie, M.: The CAMS reanalysis of atmospheric composition, Atmos. Chem. Phys., 19, 3515–3556,, 2019.


Posted in Aerosols, Climate, Climate modelling | Leave a comment

Howling Space Gales and why we should photograph them.

By: Luke Barnard

Most people are familiar with the fact the Sun emits a range of electromagnetic radiation (e.g. sunlight), and that this radiation is necessary to sustain life on Earth as we know it. What is less well known is that alongside the Sun’s electromagnetic radiation, it also generates a wind of plasma that continuously blows out through the Solar System, with speeds of 250 km/s to 750 km/s.

This solar wind impacts our everyday lives through its effects on the technology we increasingly depend on; particularly spacecraft in orbit around Earth. We rely on satellites for critical services such as communications, GPS, and weather forecasting. When services like these are disrupted, it can have both expensive and dangerous consequences [1].

During periods of intense solar wind activity, it squeezes and shakes Earth’s magnetic field. This produces energetic charged particles which are harmful to satellite electronics and can also make it difficult to maintain radio communications with them. Depending on how intense the solar wind is, satellites can be temporarily or permanently damaged, with knock on impacts to the services they provide.

Space Weather Forecasting grew out of the need to understand and predict when situations like this would occur. A key challenge in space weather forecasting is to be able to forecast the solar wind flow throughout the Solar System. This is difficult because there are only a handful of spacecraft able to measure the solar wind, and these only measure it at single points which are vastly separated. By way of analogy, it is like trying to forecast the weather at Reading, with only weather observations at a few other far away cities, like Exeter, Manchester and Brighton. The limited information is still useful, but there is a lot that can happen in between and a lot of uncertainty.

Our research [2] aims to help solve this problem by using images of the solar wind plasma to characterise the solar wind flow near the Sun. This would be an extra source of information on the solar wind flow, which we could use to help improve computer models that forecast the solar wind.

Figure 1: This shows the relative locations of Earth, STEREO-A and STEREO-B. The purple shaded regions show the field-of-view of the inner Heliospheric Imager camera on STEREO-A.

NASA’s STEREO mission consists of two spacecraft which are in Earth-like orbits, but they drift relative to Earth [Figure 1], so that they can observe the space between the Sun and Earth. On each spacecraft are a pair of cameras called the Heliospheric Imagers, which produce images of the solar wind plasma [Figure 2]. The cameras record visible sunlight that has scattered off of electrons in the solar wind. Interpreting the images is tricky because there are electrons and visible light everywhere in space, and so we don’t actually produce an image of a specific feature or object. But, because we understand the physics of sunlight well, and of how sunlight scatters off of electrons, we are able to use these images to identify regions where there are relatively more electrons, and a denser solar wind.

Figure 2: A movie of heliospheric imager images from July 2008. Movie obtained from the UK Solar System Data Centre

Our aim was to show that variability in the images could be statistically related to the direct single point measurements of solar wind flow observed by other spacecraft. This would be the first step in creating and calibrating a technique to estimate the solar wind flow directly from the images.

We compared the solar wind point measurements and images directly, computing the correlation between variability in the images recorded by STEREO-A with the solar wind speed measured directly at Earth, STEREO-A, and STEREO-B. We found that there is a strong correlation between variability in the images and the solar wind speed observations at the three spacecraft, but that the correlation was largest when a delay was applied between the image and solar wind observations. This delay was different for each pair of spacecraft, and changed in time in a way that can only be explained by the orbits of the spacecraft. Based on this statistical analysis we have concluded that we probably can trace the flow of the solar wind in the Heliospheric Imager data. Our next step is to investigate how to best compute a reliable estimate of the solar wind speed directly from the images.


[1] Cannon, P., et al. (2013), Extreme space weather: Impacts on engineered systems and infrastructure, Tech. Rep., Royal Acad. of Eng., London. ISBN:1903496950 Link to PDF here

[2] Barnard, L.A., Owens, M.J., Scott, C.J., Jones, S.R.: 2019, Extracting inner-heliosphere solar wind speed information from heliospheric imager observations. Space Weather 17.

Posted in Climate, space weather | Leave a comment

Trading Evil lasers for MAGIC Doppler lidars

By: Janet Barlow 

Lasers may have an evil reputation in Hollywood, but they are very good for observing urban meteorology. We recently took part in the MAGIC project field campaign in London, deploying a Doppler lidar to measure wind-speed around tall buildings.

Just like a duck in water, a tall building causes a wake behind it. The wake can be 100s of metres long downstream, causing reduced wind speeds and increased turbulence. Wakes can thus affect air quality, so it is important to represent them in pollutant dispersion models.

Recently we reported on wind tunnel experiments where we measured flow around a model tall building at the MAGIC project experimental site. One question was whether the wake affected natural ventilation of the test building at the centre of the site. Measuring flow around actual tall buildings is impossible using traditional meteorological instruments like cup anemometers: they are simply too small to measure the whole wake. Instead, we used a Doppler lidar which can measure wind-speed remotely over a wide area (Drew et al. 2011).

Figure 1: Principle of infra-red Doppler lidar operation. Image taken from

The principle behind radar observations of rainfall used for a weather forecast is that a pulse of electromagnetic radiation of a certain wavelength is beamed out into the atmosphere (Figure 1). A lidar uses infra-red light that interacts with particles of a similar size to the light wavelength. Some light is scattered back to the instrument and measured. But the backscattered waves are shifted in frequency by an amount proportional to the wind-speed blowing the particles around. This is the same “Doppler effect” that we hear when an ambulance goes by and its siren seems to change pitch: the soundwaves change wavelength in proportion to its speed. One advantage of using infra-red frequencies is that lidars are eyesafe. Not evil at all!

Figure 2: Photo showing MAGIC experimental site. The tall building (height: 81 m) and the lidar (white box) are highlighted with a red circle. The London Eye is on the far left and the Shard is on the far right.

We placed our Doppler lidar on the roof of a building at the MAGIC experimental site in London (Figure 2). At a height of 27 m we had a good view above most rooftops. We scanned the laser beam horizontally in a circle, meaning that laser light was reflected from tall buildings, allowing us to locate them.  

Figure 3: Lidar horizontal scan of local wind-speeds minus the average wind-speed across the whole scan. The wind direction was north-westerly. The building is shown as a red square and its wake is the yellow area to the south-east of it.

Figure 3 shows a horizontal scan of wind-speed measured by the lidar. The velocity measurement at each pixel has been subtracted from the average velocity across the whole scan (NB: as velocity is negative towards the lidar, a wake appears as a positive difference). The wake is approximately 150 m long, which means the test building is definitely affected by the wake – it is 85 m away from the tall building. Flow around it is weaker and more turbulent, affecting pollutant levels and the ability to ventilate rooms through open windows.

So, does a wake measured around a real building resemble wind tunnel measurements? We also found that the building wake in the wind tunnel was long enough to affect the test building – but how much was wind-speed reduced, compared to if the tall building was not there? The wind tunnel experiments suggested around 40% reduction at the location of the MAGIC test building (Hertwig et al. 2019); the lidar measurements for our case study suggest around 25%. With 6 months of data, we have many more cases to analyse to quantify wake behaviour under different weather conditions.

This amazing instrument allows us to “see” urban winds and provides invaluable data to improve forecasting and building design. But we definitely don’t need to attach our lidar to a shark’s head. That would just be evil.

Thanks to Eric Mathieu, Elsa Aristodemou, Jess Brown, Ian Read and Selena Zito for technical assistance.


Drew, D.R., Barlow, J.F. and Lane, S.E. (2013) Observations of wind speed profiles over Greater London, UK using a Doppler lidar, Journal of Wind Engineering and Industrial Aerodynamics, 121, 98-105, DOI: 10.1016/j.jweia.2013.07.019

Hertwig, D, Gough, H., Grimmond, C.S.B., Barlow, J.F., Kent, C.W., Lin, W., Robins, A.R. and Hayden, P. (2019) Wake characteristics of tall buildings in a realistic urban canopy, Boundary-Layer Meteorology, 172, 239-270, doi: 10.1007/s10546-019-00450-7

Posted in Boundary layer, Climate, Urban meteorology | Leave a comment

Don’t (always) blame the weather forecaster

By: Ross Bannister

There are (I am sure) numerous metaphors that suggest that a small, almost immeasurable event, can have a catastrophic outcome – that adding the proverbial straw to the load of the camel will break its back. In 1972, the mathematical meteorologist Ed Lorenz famously gave the presentation, “Does the flap of a butterfly’s wings in Brazil set off a tornado in Texas?” Unlike for folks who do keep a domestic camel, this title was not intended to be interpreted literally, but instead to ask how a system like the Earth’s atmosphere is affected by vanishingly small perturbations. But is it possible for a butterfly’s flap to really have consequences? Without the ability to experiment on two or more otherwise identical Earths, demonstrating this is impossible.

Learning from computer simulations

Atmospheric scientists are acutely aware that computer-derived forecasts are sensitive to the ‘initial conditions’ provided to them. Modern weather forecasting is done by representing the atmosphere at an initial time with vast sets of numbers stored inside a computer (this set is called the initial conditions of the model). The computer marches this state forward in steps into the computer’s version of the future. The rules that the computer uses to do this task boil down to Newton’s laws of motion (i.e. how forces acting on air masses change their motion), and other processes that affect the behaviour of the atmosphere, like heating and cooling by radiation and by condensation/evaporation of water. Unlike in the real world, it is possible in the computer to create two identical sets of initial conditions apart from small differences, and then to let the computer calculate the two possible future states.

Sensitivity to initial conditions

So, what do scientists find from these experiments? At first the forecasts are virtually indistinguishable, but at some time they start to show noticeable differences. These appear typically on small scales and then start to affect larger scales (known as the inverse energy cascade). Lorenz discovered this serendipitously in the 1950s when he ran simplified weather simulations on a research computer (a valve/diode-based Royal-McBee LGP-30 with the equivalent of 16 kilobytes of memory). He found that if he stopped the simulation, and restarted it with similar, but rounded, sets of numbers representing the weather, the computer simulated weather patterns that became very different from those that are forecast had he not stopped and restarted the simulation. Lorenz had discovered sensitive dependence to initial conditions (or colloquially, the “butterfly effect” in connection with the title of his presentation). Faced with two such different outcomes, which one, if any, is the better forecast? Hmm …

Figure 1:

Numerical solutions of two x, y, z trajectories obeying the (non-linear) Lorenz-63 equations to demonstrate sensitive dependence to initial conditions (red and yellow lines/points). At t = 0 the initial conditions are indistinguishably close and at t = 3 the two trajectories virtually overlap. At t = 6 small differences appear, which become more obvious at t = 9. By t = 12 and t = 15 the two trajectories are so different that they occupy separate branches. The beauty of the structure that emerges by solving the Lorenz-63 equations is quite amazing. For the record, the Lorenz-63 equations are: dx/dt = σ(yx), dy/dt = –xz + rx y, and dz/dt = xy bz, with σ = 10, r = 28, and b = 8/3. The multiplication of one of x, y, z with another such variable gives these equations their non-linear property.

Try this at home

This effect is also seen in simple non-linear equations. In 1963, Lorenz published a seminal work, “Deterministic non-periodic flow”, where he introduced some equations that describe how variables, x, y, and z change in time. These equations may be regarded as representing a highly simplified version of the atmosphere. It is only possible to solve these equations approximately with the help of a computer (note to reader – try this, it’s fun!). One can visualise the solution by taking particular x, y, and z values at a given time as the co-ordinates of a point in space. Joining the points up in time shows the forecasts as trajectories, and one may think of different positions as representing different kinds of weather. Figure 1 shows two such trajectories (red and yellow), whose initial conditions are nearly identical at time t = 0. As time progresses, they diverge, slowly at first, until by t = 12 they represent completely different states (note the resemblance to a butterfly).

Ensemble weather prediction

Scientists routinely run large models from many initial conditions, each subject to a slight variation – a technique called ensemble forecasting. The initial conditions differ by amounts believed to be around the level of uncertainty that the weather is known using observations and previous forecasts. These are combined in a physically consistent way, using data assimilation (which is my area of research). As a rule of thumb, differences seen in the small-scale weather forecast patterns emerge first. Indeed if the forecast grid is small enough to resolve cloud systems then the ensemble members will likely first disagree in the forecast of convective events, like showers and thunderstorms. This is why patterns of convective precipitation are so hard to predict beyond a few hours. One forecast may predict heavy rain at a particular location between 4.00 and 4.10pm, another between 4.30 and 4.35pm, and another may predict no heavy rain. Ensemble forecasting allows forecasters to understand the range of likely outcomes (usually all ensemble members will predict heavy showers, but with slightly different locations), and to give probabilistic forecasts for individual locations. While small-scale features will differ, large-scale weather patterns (such as high and low pressure systems) are usually predicted accurately at these early stages. As forecast time progresses the uncertainty develops in larger scales and eventually the forecast of the large-scale systems become unpredictable.

Fundamental limits

As a rule of thumb, km-scale motion is predictable to no more than about half-hour, 10 km-scales to about one hour, 100 km-scales to about 10 hours, 1000 km-scales to about one day, 10000 km-scales to about four or five days, and the largest scales no more than about a week or two. In extra-tropical regions, for example, there is a particular kind of atmospheric instability (baroclinic instability) between scales of around one to three thousand km which can lead to a lowering of predictability on those scales, although observing the weather at these scales is given special attention so that the uncertainty at these scales is reduced in the initial conditions.

[We should note that climate models make projections many years, decades, or centuries into the future and use the same building blocks as weather models. Climate models though predict different things: long-time averaged conditions rather than the weather at particular times, which is thought to be very useful as long as realistic forcings (e.g. the radiative forcing associated with changes in greenhouse gas concentrations in the atmosphere) are known.]

Room for improvement?

So what hope is there of improving weather prediction given these fundamental limits? There are other factors that can be improved. The spread in the ensemble’s initial conditions can be reduced with more observations and better assimilation. Model error can also be reduced. No model is perfect, but there is room for improvement by decreasing the grid size and time step (severely restricted by cost and available computer power), and by improving the representation of physical processes (also restricted by computing and on research activity).  While scientific and technological barriers can be broken, the fundamental limits of nature cannot. As the air motion of the butterfly’s flap mixes with all the other fluctuations, it is impossible to say exactly how it will change the course of the atmosphere, just that it will.


Lorenz E.N., The Essence of Chaos, UCL Press Ltd., London (1993), ISBN-13: 978-0295975146. A readable a thought provoking popular account of chaos theory.

Lorenz E.N., Deterministic nonperiodic flow, Journal of the Atmospheric Sciences 20 (1963), 130–141, DOI:10.1175/1520-0469(1963)020<0130:DNF>2.0.CO;2. An exploration of the derivation and interpretation of the Lorenz-63 equations.

Lorenz E.N., The predictability of a flow which possesses many scales of motion, Tellus 21 (1969), 289–307, DOI:10.1111/j.2153-3490.1969.tb00444.x. This paper explores different kinds of predictability and how predictability depends on scale.

Tribbia J.J. and Baumhefner D.P., Scale interactions and atmospheric predictability: An updated perspective, Monthly Weather Review 132 (2004), 703–713, DOI:10.1175/1520-0493(2004)132<0703:SIAAPA>2.0.CO;2. An update on earlier work of Lorenz with more modern weather prediction models.

Palmer T.N., Dring A., and Seregin G., The real butterfly effect, Nonlinearity 27 (2014), R123–R141, doi:10.1088/0951-7715/27/9/R123. A discussion of the “butterfly effect” term to necessarily refer to a finite time limit to predictability in fluids with many scales of motion.

Data Assimilation Research Centre, What is data assimilation?, A brief introduction to data assimilation.

Met Office, The Met Office ensemble system, An introduction to the Met Office’s ensemble prediction system.

University of Hamburg, Forecasts diagrams for Europe, Choose a European city for ensemble forecasts of temperature and precipitation. A graphic illustration of the growth of uncertainty with forecast time from weather forecast models.

Posted in Climate, Climate modelling, data assimilation, Numerical modelling, Predictability | Leave a comment

High-resolution insights into future European winters

By: Alexander Baker

Figure 1: Observed UK rainfall anomaly as a percentage of 1981-2010 monthly average for (a) December 2013, (b) January 2014, and (c) February 2014. Figure from Huntingford et al. (2014).

Most – roughly 70% – of Europe’s winter rainfall is brought by extratropical storms, which are steered our way by the westerly North Atlantic jet stream. The wettest winters, such as 2013/14 (Figure 1), are often when Europe is at the receiving end of a veritable convoy of storm systems, the human and economic impacts of which are felt far and wide.

An analogy: the sharpness of a digital photograph depends on how many pixels comprise it – in other words, on the camera’s resolution. The more pixels; the higher the resolution; the sharper the image. Climate models break down Earth’s atmosphere into three-dimensional pixels called grid cells. With a relatively low number of large grid cells (each typically 100-200 km wide), simulated weather systems not only appear pixelated when visualised, but are actually not all that realistic; their flows of air, heat and moisture don’t properly resemble those in the real world. This limits our confidence in using these models to make predictions. However, developments in high-performance computing have enabled the size of a climate model’s grid cells to be shrunk (to roughly 25 km in our case), thereby increasing their number and enabling the simulation of air flows over mountainous terrain, weather processes, and other aspects of atmospheric variability in more detail.

In a recent paper published in Journal of Climate, we address two questions. Does increasing a model’s atmospheric resolution improve the fidelity of simulated European winter hydroclimate? How do high-resolution future projections differ, if at all, from those at low-resolution? We compared simulations with low- and high-resolution versions of the same climate model – Met Office’s Hadley Centre Global Environmental Model (version 3) Global Atmosphere 3.0 (hereafter HadGEM3-GA3; Walters et al. 2011) – to establish exactly what the impact of resolution is on the North Atlantic jet and on downstream storm activity and precipitation. At the lowest resolution (‘N96’), the latitude-longitude grid is made up of grid cells each 135 km. At mid- (‘N216’) and high-resolution (‘N512’), grid cells are each 60 and 25 km, respectively. We’ll focus here on how the North Atlantic jet behaves at different model resolutions.

Figure 2: Frequency of North Atlantic eddy-driven jet latitude in reanalyses and HadGEM3-GA3 under historical climate (upper panel) and the projected future change (lower panel). We use ‘N’ notation to describe resolutions: ‘N96’ (135 km), ‘N216’ (60 km) and ‘N512’ (25 km). Figure adapted from Baker et al. (2019).

What about future projections? Under climate change (RCP 8.5), at all model resolutions, southern jet occurrences decrease but northern jet occurrences increase (Figure 2, lower panel). The upshot of this is fewer storms making landfall over southern Europe and more across northern Europe towards the end of the twenty-first century. These climate change consequences are significantly enhanced by increased resolution. Crucially, this reveals the extent to which lower-resolution models may have previously underestimated aspects of the jet’s response to climate change, and thereby changes in winter storms and precipitation, and their associated hazards. There is much more work to do: further studies investigating other models and climate change scenarios are needed, but our study offers insight into how high-resolution might bring the picture of Europe’s future winters into sharper focus.


Baker, A. J. et al., 2019: Enhanced climate change response of wintertime North Atlantic circulation, cyclonic activity and precipitation in a 25 km-resolution global atmospheric model. Journal of Climate

Huntingford, C. et al., 2014: Potential influences on the United Kingdom’s floods of winter 2013/14. Nature Climate Change 4, 769. 10.1038/nclimate2314

Walters, D. N. et al., 2011: The Met Office Unified Model Global Atmosphere 3.0/3.1 and JULES Global Land 3.0/3.1 configurations. Geoscientific Model Development 4, 919-941. 10.5194/gmd-4-919-2011 

Posted in Climate change, Climate modelling, extratropical cyclones, Numerical modelling | Leave a comment

Turbulence Matters

By: Torsten Auerswald

Most people are only consciously aware of the existence of turbulence when the pilot announces it. But apart from the discomfort of a bumpy flight, turbulence affects us in many other important aspects of daily life. The fact that turbulent mixing is much more efficient than molecular diffusion does not only come in handy when stirring the tea after adding sugar and milk (whether it is better to first put the milk or the tea into the cup can be discussed in another blog entry), but also has impacts on important problems in medicine, engineering, biology and physics, to name just a few. Turbulence is responsible for noise created at the blades of wind turbines and knowledge about it can help engineers to design quieter wings. It also affects the delivery of drugs to the lung, and therapeutic aerosols can be designed to optimise their effect in the body by modifying their aerodynamic properties. Turbulent effects are important for designing efficient filters for power plants or improving the efficiency of fuel engines. For our weather, turbulence plays a critical role. For example, it controls the exchange of heat and moisture between the soil and the atmosphere and is one of the factors which influences the development and characteristics of clouds. This is the reason why it is of so much importance for weather forecast models to describe turbulent processes accurately.

Unfortunately, the importance of turbulence is directly proportional to the difficulty to study its properties. The underlying set of equations which describe all fluid flows are the Navier-Stokes equations. This set of equations is extremely difficult (and most of the times impossible) to solve analytically. This is why for most real-world applications computer models are used which are able to find numerical solutions of the Navier-Stokes equations with good precision. Especially for turbulent flows, these computer models are numerically very expensive and the direct numerical simulation of turbulent flows remains restricted to relatively simple cases. In most applications, a certain level of approximation for the smaller scales, or even the whole turbulent part of the flow, is necessary to be able to simulate turbulent flows. This is necessary despite the fact that computer performance has rapidly increased over the last decades.

When the first numerical weather forecast was computed by hand by Lewis F. Richardson in 1922, it took him 6 weeks to calculate a 6-hour forecast for Europe. This forecast, apart from coming approximately 5.96 weeks too late, was also wrong. Nevertheless, his work was revolutionary and kicked off the era of numerical weather forecasts. In the publication of his results he estimated that it would take 64000 human computers (people who solve the numerical equations by hand) to simulate the global weather in real-time (meaning it would take one hour to compute a one hour forecast). He envisioned a weather centre in which hundreds of human computers would work together solving the equations for their respective parts of the forecast domain, and coordinators would make sure that everybody stays in sync, collect the results for each time step from the human computers and unify them to one big data set. He basically described the parallelisation of numerical flow models, including the communication between the “nodes”, which decades later would be used to compute global weather forecasts in only a few hours on modern supercomputers.

One of the first supercomputers was the CDC 6600. In 1964 it was considered to be the most powerful computer in the world and could compute an astonishing three million floating point operations per second (3 megaflops). Today’s fastest supercomputer is the IBM Summit which is able to perform 122×1015 floating point operations per second, 40 billion times more than the CDC 6600. Despite this impressive increase in computational power, current supercomputers are still not fast enough to allow the direct simulation of turbulent flows for most real-life applications, leaving plenty of interesting research topics for current and future scientists to investigate.

Richardson, L. F., 1922: Weather prediction by numerical process. Cambridge university press, 236 pp.


Posted in Numerical modelling, Turbulence | Leave a comment

The consequences of climate change: how bad could it get?

By: Nigel Arnell

The United Nations Climate Action Summit held in New York on 23rd September was meant to be the occasion where countries and industry organisations made stronger commitments to reduce the emissions of the greenhouse gases that are causing global warming. Whilst 65 countries pledged to achieve net-zero emissions by 2050, the overall commitment – according to some commentators – fell “woefully short” of expectations. Most news headlines after the event focused on Greta Thunberg’s speech where she passionately challenged world leaders to do more: “the eyes of all future generations are upon you”.  Meanwhile, campaign groups such as Extinction Rebellion warn of ‘unprecedented global emergency’, ‘climate breakdown’ and ‘mass extinction’. The best-selling author David Wallace-Wells writes of ‘The Uninhabitable Earth’.

So what are the consequences of climate change, and how bad could it get? The impacts of climate change in the future depend not only on how climate – temperature, precipitation and so on – changes, but also on how societies and economies change. Our estimates of changes in climate depend on two things. First is a projection of how emissions will change. This depends on how economies and energy use change, and we cannot predict this: it will depend on policy choices, such as the actions to reduce emissions announced in New York last week. We, therefore, use ‘scenarios’ to describe plausible changes in emissions, but these should not be seen as predictions. The second part is where climate science fits in. Projections of the effect of an emissions scenario on changes in local weather are made using climate models. Whilst all models produce broadly similar changes in climate (temperatures increase in high latitudes more than in tropical regions, wet areas tend to get wetter and dry areas tend to get drier), the detailed projections can vary considerably between climate models. We can estimate the consequences of these changes in local weather for local climate hazards and resources – such as floods and droughts – using separate ‘impacts’ models.

Figure 1: Change in global heatwave, drought and flood hazard through the 21st century under three plausible emissions scenarios, and numbers of people affected in 2100. The five estimates of people affected for each emissions scenario are from five different socio-economic scenarios. See Arnell et al. (2019) for specific definitions of the indicators.

In practice, policymakers and others want to know the human impacts of climate change. We can estimate the direct impacts – such as the number of people affected by flooding, droughts or heatwaves – by combining our estimates of the physical changes in climate with socio-economic scenarios describing plausible changes in population and the economy. Figure 1 shows the global direct impacts of climate change on exposure to heatwaves, river flooding and drought (from Arnell et al., 2019). The left panels show changes in the physical hazard through the 21st century, under different emissions. The range in estimates for each emissions scenario shows the uncertainty due to different model projections of local temperature and precipitation. By 2100, the chance of experiencing a major heatwave has risen from 8% to 100%, the average proportion of time in drought has gone up from 6 to 30%, and the chance of a flood has increased from 2% to 7% (in some regions the changes are greater). The right-hand panels show the human impacts in 2100, under three of the emissions scenarios and for five socio-economic scenarios. The biggest impacts – in terms of numbers of people affected – are obviously with the highest emissions, but the differences between the socio-economic scenarios can be very large. When presenting such results we often focus on the central estimates (the solid lines in the figures), but we could instead look at the ‘worst-case’ impacts at the top end of the distribution. These can be a lot higher than the central estimates.

These direct impacts have knock-on consequences: changes in the frequency of droughts, for example, could plausibly lead to loss of livelihoods, food insecurity, political instability and displacement of people. It is these potential ‘systemic’ risks that are of greatest significance for policymakers and indeed are behind many of the most extreme warnings about climate emergencies and crises. However, these knock-on consequences depend not only on the physical changes in climate and future socio-economic scenarios that we can model but also – and largely – on how societies and governments react and behave. In order to estimate the likelihood of the real ‘worst-case scenarios’ that are ringing the loudest alarm bells we, therefore, need to link the work of climate scientists, impact modellers and experts in institutions, governance and human behaviour.


Arnell, N.W. et al. (2019) The global and regional impacts of climate change under Representative Concentration Pathway forcings and Shared Socioeconomic Pathway socioeconomic scenarios. Environmental Research Letters 14 084046.  doi:10.1088/1748-9326/ab35a6

Wallace-Wells, D. (2019) The Uninhabitable Earth: a story of the future. Penguin: London

Posted in Climate, Climate change, Climate modelling | Leave a comment

Probing the atmosphere with sound waves

By: Javier Amezcua

Summer is a quiet time for both the University of Reading and the town itself. The buzzing that fills campus during term time is gone, the population decreases and activities are reduced. Some people find it relaxing – I find it boring and lethargic.  There is an exception to this quietness which occurs during the August Bank Holiday weekend. Any music aficionado knows this is the time when the annual Reading and Leeds Festival takes place. Thousands of people from all around the UK descend into our town to inhabit a patch of land next to the Thames for three days and enjoy some of their favourite bands, as well as some other excesses…

During my seven years in Reading, I have indulged myself with attending the festival four times. In each of these occasions, I have had a similar conversation with my colleagues when we were back at work the day after the Bank Holiday Monday. First, they hypothesize that I was the oldest person in the whole festival (not true, but it is accurate to say that I am too old to camp and instead I go home each day). Second, they state that at night they could hear the music all the way to their houses. I found the latter comment interesting given what I had experienced. Let me explain: there have been days in which I was not interested in the headliner act and hence I left the festival while the music was still playing. In each occasion, I remember walking away from the festival and noticing the music fading progressively until it was gone. So how could other people hear it at their homes when these were further away from the stage? The answer, not obvious to me at first, is that there were some sound waves that were ‘jumping’ over me.

Figure 1: Simple representation of the reflection of a vertically propagating wave in the atmosphere. The sound wave (yellow line) departs from a source (S) at the surface, reaches a maximum height (Zmax) and it is reflected back towards the surface where it can be detected in a receiver (R). The cross-winds that the wave-front encounters make it seem to come from a false source in a slightly different direction than the real source.

To understand my explanation, we need to think about the way sound travels from the loud-speakers around and above the stage. Without going into the specifics of the type of loud-speaker (there are lots), the sound waves can be transmitted in both the horizontal and vertical directions (you can see a classical illustration of how spherical sound waves work in the youtube link in the references). At the end of the night, as I walked away from the source, the sound waves coming horizontally in my direction were attenuated by the medium (air) and obstacles, and hence I stopped hearing the music. What about those waves with a vertical component? Figure 1 answers this: the yellow line represents the path of these waves. They travel up to some maximum height until they are reflected back towards the surface. The attenuation over that path is different from the attenuation of the horizontal propagating wave (in the case I am discussing it is less). So, the people in town were receiving a sound-wave that had been reflected in an upper level and still had enough intensity. It also helped that is was night-time and not day-time (at night the conditions are more prone for reflection/refraction, but that is another issue).

Figure 2: Vertical sensitivities for the infra-sound waves generated by the detonation of old ammunition in Finland and detected in Norway. The horizontal axis is the time; we indicate the year but not the exact time of detonation. The vertical axis is height. Notice that most infra-sound waves reach about 40km in height.

Why am I telling this story? Because lately, I have been using the behaviour I just described as a tool to probe the winds in the stratosphere (roughly between 12 to 50 km in the mid-latitudes). Finland has a lot of old ammunition, mainly from the Cold War, that it is trying to get rid of. Therefore, every summer the Finnish army performs a series of controlled detonations at a remote location during the course of several days. These explosions produce infra-sound waves (waves below 20 Hz which cannot be detected by the human ear). Some of them follow a path with a vertical component, they reach a maximum level and they are reflected back to the surface where they are detected, about 10 minutes after the explosion, in a station in Norway. This station is about 178 km due north from the explosion site and it has quite powerful micro-barometers which are able to measure precisely the pressure variations caused by the infrasound waves. I have some enthusiastic colleagues in the Norway Seism Array (NORSAR) who have shared these observations with me. Figure 2 shows a model-based reconstruction of the maximum height the waves reach for explosion events from 2001 to 2018. There are different numbers of detonations per year, which is why the horizontal axis looks irregular. Notice that most of the waves reach about 40 km in height, and some up to 60 km.

So how do I probe the atmosphere with these waves? As the waves travel through the atmosphere they are affected by several atmospheric conditions: winds, humidity, etc. In particular, the presence of cross-winds (i.e. winds perpendicular to the direction of the wavefront) can shift the detection angle of the waves when they reach the ground. Hence the waves appear to have come from a false source in the direction of the blue line in Figure 1. Since I know the exact location of the source, the time it took for the waves to be detected, and the shift angle towards the apparent source, I can deduce some values for the cross-winds each infrasound wave encountered, including those in upper levels of the atmosphere. In order to solve this estimation problem, I use techniques from inverse problems and data assimilation which I will not discuss in this post; I only mention that I use an implementation of the ensemble Kalman filter.

It is quite difficult to measure winds in the stratosphere, hence any source of information is valuable and this includes the strategy discussed here. There are other sources of infrasound waves that we can exploit around the world, and some of them are natural! For instance, the ocean swell in a spot near Iceland is a natural producer of infra-sound waves. So, we could use this hot-spot to probe the winds between that location and the same receiver in Norway, or many other receiving stations at the moment. At the moment I am participating in a collaborative project with some people from different institutions in Europe to solve this problem.


Amezcua J., S. P. Naeholm, E. M. Blixt, and A. J. Charlton-Perez, 2019: Assimilation of atmospheric infrasound data to constrain tropospheric and stratospheric winds. QJRMS, submitted.

Extract of Sound Waves And Their Sources (1933), you can see a classical illustration of how spherical sound waves work in this animation.


Posted in Climate, data assimilation, Stratosphere, Wind | Leave a comment