Mapping bio-UV products from space

By: Michael Taylor

Solar radiation arriving at the Earth’s surface in the UV part of the spectrum modulates photosynthetically-sensitive life on the land and in the oceans. UV radiation also drives important chemical reaction pathways in the atmosphere that impact air quality. It can cause DNA-damage in the epithelial cells of our skin and is a key factor for tuning the rate of Vitamin D production in our metabolism.

Solar UV radiation may be measured in radiometric units or spectrally-weighted to account for biologically-effective UV radiation doses. The Commission Internationale de l’Éclairage (CIE) defines the reference action spectrum for the ability of UV radiation as a function of wavelength to produce just perceptible erythema (colour change from the Greek word “ερυθρός” for red) in human skin. The standard erythemal dose (Jm-2) is equivalent to an erythemal radiant exposure of 100 Jm-2 (ISO 17166:1999). According to the Bunsen-Roscoe law of reciprocity (Bunsen & Roscoe, 1859), a given biological effect due to UV radiant exposure is directly proportional to the total energy dose given by the product of irradiance (Wm-2) and exposure time (s).

Figure 1: TEMIS erythemal UV dose products (kJ m-2) from KNMI/ESA (Van Geffen et al, 2017): daily “Clear sky” UV from SCIAMACHY/GOME-2, daily “cloud-modified” UV from SCIAMACHY/GOME-2 revealing the impact of a weather system over Sicily, and the global climatological “clear sky” June mean from GOME showing the impact of desert dust as revealed (lower right) by the global climatology of aerosol mixtures (Taylor et al, 2015).

Satellites like GOME, GOME-2 and SCIAMACHY have operational processing algorithms that retrieve erythemal UV dose (kJ m-2) once daily from top of the atmosphere irradiance measurements which are strongly affected by both cloud and atmospheric aerosol (Fig. 1).

Recent studies performed in the context of solar energy (Kosmopoulos et al., 2017; 2018), have revealed that atmospheric aerosol, and desert dust in particular, strongly attenuates solar radiation and the UV component arriving at the ground. In the context of increasing our global capacity for renewable energy with solar power as a major component, this is important. Since atmospheric aerosols reduce solar radiation by absorbing and scattering light and reduce the strength of the direct beam from which solar power generation is most efficient, they also cause forecast uncertainty. As a result, electricity supplied to national grids from solar power must balance the demand by coping with these unexpected fluctuations.

Figure 2: (a) Window functions used by KNMI/ESA to derive Bio-UV dose products from UV spectral irradiances and (b) the back-propagation neural network used to perform ground-based validation. See Zempila et al (2017) for details.

Nevertheless, it is straightforward to obtain biological ultraviolet (“Bio-UV”) products from the UV spectra retrieved at the ground. By applying weighting functions to the ultraviolet part of the irradiance spectra over the range 285-400 nm, important Bio-UV products like Vitamin D dose, DNA-damage dose and photosynthetically active radiation can be calculated. Satellite Bio-UV products from TEMIS (KNMI/ESA) have been successfully validated with high temporal resolution (1-minute) ground-based measurements by Zempila et al. (2017). Fig. 2 shows how the weighting functions vary with wavelength together with the neural network model developed to convert combinations of UV irradiances (I) and solar zenith angle (sza) to Bio-UV products for the ground-based validation.

Figure 3: Zoom sequence showing how the surface solar radiance spectra (280-2500 nm) retrieved under cloudy conditions from space with fast neural network radiative transfer solvers (Taylor et al., 2016) can be used to extract erythemal UV spectral irradiances for calculation of Bio-UV products as per Zempila et al., (2017).

While polar orbiting satellites like GOME, GOME-2, SCIAMACHY and OMI allow global maps of Bio-UV products to be generated, geostationary satellites like Meteosat Second Generation (MSG) provide high spatial resolution images of the Earth disc (3 km x 3 km) every few minutes and allow us to dramatically increase the frequency of the data. In support of this, an operational algorithm capable of retrieving the UV part of the solar spectrum at the surface was recently developed (Taylor et al., 2016). This was achieved with a synergistic model that uses both machine learning with neural networks and a look-up table of radiative transfer simulations to help unravel the complexity of the atmosphere. The model includes the effects of clouds, aerosols, ozone, elevation and surface albedo (Taylor et al; 2016; Kosmopoulos et al., 2017) and provides the surface global horizontal irradiance (GHI), direct normal irradiance (DNI) and diffuse horizontal irradiance (DHI) spectrum over the broad wavelength range 285-2700 nm. Application of the weighting functions of Fig. 2 to the UV part of the solar radiation spectrum can then provide global maps of Bio-UV products at high frequency. Fig. 3 illustrates how various UV products can be obtained from surface solar radiation spectra retrieved from space.

One of the most exci­ti­ng appl­icati­ons of being able to map UV spectral information ­from space is the potential for creating mob­ile appli­cat­ions that pull data from surface UV spectra data cloud computing resources and combi­ne them wi­th users’ GPS i­nformati­on to produce real-ti­me UV alerts to the general publi­c with unprecedented precision. By improving our capacity to map UV impact on the quality of life in the global ecosystem from space at high frequency, we will be better placed to monitor progress towards achievement of the UN’s sustainable development goals as we proceed to a more climate resilient society.


Bunsen, R., Roscoe, H. E., 1859: Photochemische untersuchungen. Annalen der Physik  184(10), 193-273, DOI: 10.1002/andp.18591841002

Kosmopoulos, P., S. Kazadzis, H. El-Askary, M. Taylor, A. Gkikas, E. Proestakis, C. Kontoes, M. M. El-Khayat, 2018: Earth-Observation-Based Estimation and Forecasting of Particulate Matter Impact on Solar Energy in Egypt. Remote Sens. 10(12), 1870, DOI:

Kosmopoulos, P. G., S. Kazadzis, M. Taylor, E. Athanasopoulou, O. Speyer, P. I. Raptis, E. Marinou, E. Proestakis, S. Solomos, E. Gerasopoulos, V. Amiridis, 2017: Dust impact on surface solar irradiance assessed with model simulations, satellite observations and ground-based measurements. Atmos. Meas. Tech., 10(7), 2435-2453, DOI: 10.5194/amt-10-2435-2017

Taylor, M., P. G. Kosmopoulos, S. Kazadzis, I. Keramitsoglou, C. T. Kiranoudis, 2016: Neural network radiative transfer solvers for the generation of high resolution solar irradiance spectra parameterized by cloud and aerosol parameters. J. Quant. Spectrosc. Radiat. Transfer, 168, 176–192, DOI: 10.1016/j.jqsrt.2015.08.018

Taylor, M., S. Kazadzis, V. Amiridis, R. A. Kahn, 2015: Global aerosol mixtures and their multiyear and seasonal characteristics. Atmos. Environ., 116, 112–129, DOI: 10.1016/j.atmosenv.2015.06.029

Van Geffen, J., Van Weele, M., Allaart, M. and Van der A, R., 2017: TEMIS UV index and UV dose operational data products, version 2, KNMI Dataset, DOI:

Zempila, M. M., J. H. van Geffen, M. Taylor, I. Fountoulakis, M. E. Koukouli, M. van Weele, R. J. van der A, A. Bais, C. Meleti, D. Balis, 2017: TEMIS UV product validation using NILU-UV ground-based measurements in Thessaloniki, Greece. Atmos. Chem. Phys., 17(11), 7157–7174, DOI: 10.5194/acp-17-7157-2017


I am very grateful to colleagues from the Tropospheric Emission Monitoring Internet Service (TEMIS) at KNMI and ESA for kindly making available plots of UV radiation monitoring products generated from the v2 processing algorithm, and to colleagues from the Greek national network for the measurement of ultraviolet solar radiation ( for permission to present results from Zempila et al., (2017) using their ground-based NILU-UV multi-filter radiometer measurement data and associated UV dose data obtained from a Brewer MKIII spectrophotometer. I would also like to acknowledge colleagues from with whom I collaborated with to develop the solar radiation neural network modeling aspects presented.

Posted in Climate, earth observation, Remote sensing, Solar radiation | Leave a comment

The sky is the limit – How tall buildings affect wind and air quality

By: Denise Hertwig

Based on current UN estimates, by 2050 over 6.6 billion people (68% of the total population) will be living in cities. Across the world, tall (> 50 m height) and super-tall (> 300 m) buildings already define the skylines of many large cities and will become increasingly more common outside of city centres to accommodate growing urban populations, especially when horizontal urban sprawl is geographically limited. For London, 2019 was declared the “year of the tall building” (NLA London Tall Buildings Survey 2019). At the moment, 541 buildings over 20 storeys (approx. 60 m in height) are planned or already under construction in the UK capital. Tall buildings are currently being built in 22 out of the 33 London boroughs and 76 of them are expected to be completed this year.

Tall buildings, in isolation or as clusters, affect the urban micro-climate of the local surroundings and the neighbouring region. The impact on aerodynamics (e.g. local flow distortions, long-range wake effects), radiation budget (e.g. shadowing, radiative trapping) and components of the surface energy balance (e.g. storage of heat in building materials, anthropogenic heat emissions) can be large compared to low-rise buildings. Such modifications challenge current modelling frameworks for urban areas. Urban land-surface models used in numerical weather prediction, for example, typically do not account for building-height variations. They also rely on the concept that the flow within the urban canopy is sufficiently decoupled from the flow aloft, which is not the case if tall buildings protrude deep into the urban boundary layer.

Figure 1: Normalised pollutant concentrations (a,b) in an idealised building array. Pollutants are released from point sources located (a) in the street canyon behind a tall building, (b) in an intersection upwind of the tall building. (c) Mean-flow streamlines near the tall building with colours showing the mean vertical velocity. The black arrow indicates the upwind flow direction. Data are results from large-eddy simulations by Fuka et al. (2018) for the DIPLOS project.

Similarly, operational urban air quality and dispersion models do not usually account for tall-building effects (Hertwig et al. 2018). Tall buildings strongly change pedestrian-level winds in the surrounding streets and the flow field above the roofs of the low-lying buildings. This affects pollutant pathways and the overall ventilation potential of cities. Pollutants released near the ground in a street canyon on the leeward side of a tall building (Fig. 1a) can be rapidly lifted out of the building canopy by updrafts (Fig. 1c). Although the pollutants are emitted at the ground, the tall building causes a large proportion of the released mass to be transported above the roofs of the low-rise neighbourhoods, thereby reducing street-level pollution. A pollutant source located in an upwind intersection leads to drastically different results (Fig. 1b). The downdrafts on the windward side of the tall building result in strong horizontal flow out of the upwind street canyon (Fig. 1c). This outflow shifts the pollutants away from their release point in the intersection, creating a virtual source location in the adjacent street canyon and deteriorating air quality in the streets downwind.

Figure 2: (a) Building heights and (b) wind-tunnel model buildings of the neighbourhood between Waterloo station and Elephant & Castle in London (MAGIC project study area). Wind-tunnel measurements of the wake behind the central tall building (81 m height) in isolation and together with the low-rise building canopy shown in terms of (c) height profiles of flow speeds at several sites downwind of the tall building, (d) velocity differences to the ambient (undisturbed) flow with downwind distance at several heights. Details in Hertwig et al. (2019).

Flow interactions between tall and low-rise buildings also change the structure of the momentum deficit region (wake) that forms behind tall buildings. Wake models used for local air-quality predictions currently do not account for such interactions as they were derived for isolated buildings. Wind-tunnel experiments in a realistic scale model of the area between the Waterloo and Elephant & Castle stations in London (Fig. 2a,b) documented the strong impact of the canopy on tall-building wakes (Hertwig et al. 2019). Compared to tall buildings in isolation, the presence of a low-rise canopy displaces the wake vertically (Fig. 2c), so that flow speeds are reduced over longer distances downwind well-above the canopy (Fig. 2d). In the case shown, the wake extends over distances larger than 5 times the height of the tall building (i.e. > 400 m). The increasing spatial resolution (of the order of 100 m) of mesoscale and microscale atmospheric models means that tall-building wakes no longer are subgrid-scale phenomena, but have an impact at the grid-scale. Understanding and quantifying tall-building impacts on the boundary layer over cities is essential to identify needs for model refinements.


Fuka, V., Z.-T. Xie, I.P. Castro, P. Hayden, M. Carpentieri, A.G. Robins, 2018: Scalar fluxes near a tall building in an aligned array of rectangular buildings. Boundary-Layer Meteorology 167, 53–76, DOI: 10.1007/s10546-017-0308-4

Hertwig, D., L. Soulhac, V. Fuka, T. Auerswald, M. Carpentieri, P. Hayden, A. Robins, Z.-T. Xie and O. Coceal, 2018: Evaluation of fast atmospheric dispersion models in a regular street network. Environmental Fluid Mechanics 18, 1007–1044, DOI: 10.1007/s10652-018-9587-7

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

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

Balloon measurements at Stromboli suggest radioactivity contributes charge in volcanic plumes

By: Martin Airey

Volcanic lightning is an awe-inspiring and humbling display of nature’s power. It results from the breakdown of large electric fields that are generated within the volcanic plume. The processes that result in the accumulation of charge are varied and complex and by no means fully understood. Current knowledge of the key established mechanisms that are known to contribute to plume charging centre around the role played by ash. These mechanisms fall broadly into the categories of fractoemission and triboelectrification (Mather and Harrison, 2006). Fractoemission is the release of neutral and charged (electrons, positive ions, and photons) particles from fracture surfaces as magma fragments upon eruption (James et al., 2000); these particles may then interact with ash and aerosols to impart a net charge. Triboelectrification is a mechanism by which charge is transferred between the ash particles as they collide.

When charge has been produced, it must then be separated in order for an electric field to develop and a discharge to occur. The plume is a dynamic and chaotic environment, where primitive constituents of the magma, such as solid particles, gases, and metal species are mixed with atmospheric material as it is entrained by the plume. Above the initial jet region, thermal buoyancy-driven dynamics enable the plume to grow to an altitude at which neutral buoyancy is attained. Within this setting, charged aerosols and charged ash grains settle differently resulting in the separation of positively and negatively charged regions in the plume (Mather and Harrison, 2006), which can ultimately cause a discharge to occur.

But what if there are other additional mechanisms that contribute to either the charging or separation processes? As it is a complex, rapidly evolving, multiphase environment, there is the potential for many other chemical and physical interactions occurring within the plume that may currently be overlooked by this simplistic view. To test this, sensors and instrumentation developed at Reading over many years for deployment on weather balloons was combined through a NERC-funded project into a disposable modular payload called VOLCLAB (VOLCano LABoratory). The range of sensors that can be incorporated into the VOLCLAB package includes an optical backscatter droplet detector, a charge sensor, a sulphur dioxide sensor, an oscillating microbalance particle collector, and a turbulence sensor.

View from Stromboli’s summit into the vent complex showing the gas-rich plumes

In September 2017, a team of scientists from the University of Reading, Ludwig Maximillians Universität (Munich), and the University of Bath set off to Stromboli on fieldwork funded by National Geographic, equipped with VOLCLAB sensors, radiosondes, balloons, a thunderstorm detector, and lots of helium. Stromboli was an ideal choice for this expedition as it erupts frequently (several times an hour) and produces a wide range of plume types ranging from ash-rich to predominately gaseous. By launching these instruments directly into the plumes, in situ measurements may be acquired from all these plume types. The two-week long campaign required a daily hike to the summit at 900 m, often with very heavy kit. Many sensor-equipped balloons were launched from the summit with a range of success in encountering a plume, and VOLCLAB packages were deployed in fixed locations around the summit to continually record passing plumes.

   Martin Airey (holding VOLCLAB package) and Corrado Cimarelli

                 Keri Nicoll, Kuang Koh, and Martin Airey

Most interesting was the discovery of significant electric charge in plumes that contained negligible or no ash. This led to the investigation of what might be causing this unexpected charging mechanism. It is widely known that volcanoes emit a broad range of chemical products (Allard et al, 2000), one of which is radon, which is produced in high concentrations from all volcanoes. Radon is routinely monitored at many volcanoes, including Stromboli, which is known to constantly emit very large quantities through the soil near the vents, and even more during eruptions (Cigolini et al, 2009). As radon radioactively decays, it increases the charge present by ionising the air. This additional source of charge, inferred for the first time with these new direct measurements inside gaseous plumes, will inevitably contribute to the overall charge structure and may affect the likelihood of lightning strikes.

The original open access article, published in Geophysical Research Letters, may be found at:

And additional press can be found in the following links:
New York Times:
Atlas Obscura:
The VOLCLAB package covered in Meteorological Technology magazine:
Some footage of the fieldwork was also included in the Arte documentary “Living with Volcanoes” from around 7 minutes: 


Allard, P., Aiuppa, A., Loyer, H., Carrot, F., Gaudry, A., Pinte, G., et al. (2000). Acid gas and metal emission rates during long‐lived basalt degassing at Stromboli volcano. Geophysical Research Letters, 27(8), 1207–1210.

Cigolini, C., Poggi, P., Ripepe, M., Laiolo, M., Ciamberlini, C., Delle Donne, D., et al. (2009). Radon surveys and real‐time monitoring at Stromboli volcano: Influence of soil temperature, atmospheric pressure and tidal forces on 222Rn degassing. Journal of Volcanology and Geothermal Research, 184(3–4), 381–388.

James, M. R., Lane, S. J., & Gilbert, J. S. (2000). Volcanic plume electrification—Experimental investigation of fracture charging mechanism. Journal of Geophysical Research, 105(B7), 16,641–16,649.

Mather, T. A., & Harrison, R. G. (2006). Electrification of volcanic plumes. Surveys in Geophysics, 27(4), 387–432.‐006‐9007‐2


Posted in Climate, Convection, Measurements and instrumentation, Volcanoes | Leave a comment

Convective self-aggregation: growing storms in a virtual laboratory

By: Chris Holloway

Figure 1: An example of convective self-aggregation from an RCE simulation using the Met Office Unified Model at 4km grid length with 300 K SST.  Time mean precipitation in mm/day for (a) Day 2 (still scattered), and (b) Day 40 (aggregated).  Note that the lateral boundaries are bi-periodic, so the cluster in (b) is a single organised region.  Adapted from Holloway and Woolnough (2016).

Convective self-aggregation is the clumping together of isolated convective cells (rainstorms) into organised regions in idealised computer simulations.  This storm clustering may not seem all that unusual, but it is surprising because self-aggregation occurs in simulations of “radiative-convective equilibrium” (RCE) in which boundary conditions are homogeneous (sea surface temperature [SST] is constant in space and time), there is no imposed mean wind, and planetary rotation is set to zero (e.g., Figure 1).  In other words, there is no external cause of the clustering of convection in self-aggregation (hence the “self” prefix).  Instead, internal feedbacks such as cloud-radiation interactions and surface-flux feedbacks are key (Wing et al.2017 and references therein).

Figure 2: Satellite estimates of average fractional cover vs total Cold Cloud Area for a given domain-mean precipitation rate (R) range and for ranges of the “SCAI” aggregation index between 0.00 and 0.35 (red, aggregated), between 0.35 and 0.70 (black, intermediate), and between 0.70 and 1.50 (blue, disaggregated); t stands for optical thickness. Shaded regions indicate the 90% confidence interval. (a)–(c) thick anvil, (d)–(f) optically thin anvil.  Adapted from Stein et al. (2017).

While self-aggregation is intellectually interesting, many scientists are sceptical of the relevance of this phenomenon for real weather and climate.  After all, the real world has plenty of inhomogeneity in surface temperature as well as rotation and vertical wind shear.  However, organised tropical convection in real-world observations shows many similarities to self-aggregated convection in idealised simulations: for more aggregated conditions the mean state has lower relative humidity, outgoing longwave radiation (OLR) is larger, and anvil cloud amount is reduced (Holloway et al. 2017 and references therein).  For instance, work at the University of Reading using satellite observations has shown that optically thin anvil cloud cover decreases as convection becomes more aggregated, which could have implications for climate (Figure 2).    More realistic convective-scale simulations of organised tropical convection (with observed SSTs, rotational effects, and wind shear effects) also provide evidence that cloud-radiation feedbacks act to maintain organisation and reduce the mean relative humidity (Holloway 2017).   Other real-world forms of organised tropical convection, including the Madden-Julian Oscillation (MJO), tropical cyclones, and the Intertropical Convergence Zone (ITCZ) all show cloud-radiative feedbacks and moisture-convection feedbacks that resemble processes important for convective self-aggregation in idealised computer simulations.

The potential impact of convective aggregation on climate is an area of active research and debate.  Some idealised computer experiments show stronger self-aggregation with warmer SSTs sea surfaces, but others do not (Wing 2019).  If aggregation were to increase with increasing SST, this would likely be a slightly negative feedback for global warming, meaning it would allow for slightly less warming for a given increase in carbon dioxide concentrations, but this is also an active area of research and debate.  Aggregation tends to be weaker and more variable in simulations that include coupled ocean models (e.g. Hohenegger and Stevens 2016, Coppin and Bony 2017), so this is another area that needs more extensive research. 

Studying convective aggregation enables the scientific community to generate and test hypotheses and isolate mechanisms about fundamental processes that are potentially important for convective organisation, but which can be difficult to disentangle in more realistic settings.  Even if studies eventually demonstrate how self-aggregation is not an adequate framework for some climate problems, this will also be a form of important progress.  The Radiative-Convective Equilibrium Model Intercomparison Project (RCEMIP) is bringing scientific institutions together to compare self-aggregation at different model resolutions, domains and SSTs in order to facilitate further research into this exciting topic.  At Reading, Met Office Unified Model convection-permitting simulations have been performed and submitted to RCEMIP in association with the joint NERC-Met Office ParaCon project which seeks to greatly improve the representation of convection in weather and climate models.   RCEMIP and other research efforts will increasingly apply new concepts emerging from idealised simulations to the complex interactions between convection, moisture, clouds, radiation, surface fluxes, circulations and climate.


Coppin, D., and S. Bony, 2017: Internal variability in a coupled general circulation model in radiative‐convective equilibrium, Geophysical Research Letters, 44, 10, 5142-5149,

Hohenegger, C., and Stevens, B., 2016: Coupled radiative convective equilibrium simulations with explicit and parameterized convection. J. Adv. Model. Earth Syst., 8, 1468–1482, doi:10.1002/2016MS000666.

Holloway, C. E., 2017: Convective aggregation in realistic convective- scale simulations, J. Adv. Model. Earth Syst., 9, 1450–1472, doi:10.1002/ 2017MS000980.

Holloway, C. E., A. A. Wing, S. Bony, C. Muller, H. Masunaga, T. S. L’Ecuyer, D. D. Turner, and P. Zuidema, 2017: Observing convective aggregation.  Surveys in Geophysics, 38: 1199. doi:10.1007/s10712-017-9419-1. 

Holloway, C. E., and S. J. Woolnough, 2016: The sensitivity of convective aggregation to diabatic processes in idealized radiative-convective equilibrium simulations.  J. Adv. Model. Earth Syst., 8, 166–195, doi:10.1002/2015MS000511.

Stein, T. H. M., C. E. Holloway, I. Tobin, and S. Bony, 2017: Observed relationships between cloud vertical structure and convective aggregation over tropical ocean.  J. Climate, 30, 2187–2207. 

Wing, A. A., 2019: Self-Aggregation of Deep Convection and its Implications for Climate. Curr. Clim. Change Rep., 5: 1.

Wing, A. A., K. Emanuel, C. E. Holloway, and C. Muller, 2017: Convective self-aggregation in numerical simulations: A review.  Surveys in Geophysics, 38: 1173.



Posted in Climate, Climate modelling, Numerical modelling, Tropical convection | Leave a comment

Modelling Ice Sheets in the global Earth System

By: Robin Smith

As Till wrote recently, our national flagship climate model (UKESM1, the UK Earth System Model) has been officially released for the community to use, after more than six years in development by a team drawn from across the NERC research centres and the Met Office. The most unique capability of the UKESM effort isn’t included in that release however: UKESM1 can also be made to interactively simulate the evolution of the massive ice sheets of Greenland and Antarctica.

On millennial timescales, the growth and decay of ice sheets play one of the most fundamental roles in determining the climate of the Earth – think of the ice-age cycles of the last million years. But ice sheets aren’t just for the paleoclimate people. Loss of mass from ice sheets accounts for around a third of the currently observed global mean sea level rise and their contribution is expected to increase and dominate the sea level budget in the coming decades and centuries (Church et al. 2013). The climate change impact from ice sheets isn’t limited to sea level rise either, with ice melt input to the ocean linked to a range of wide climate change problems (Golledge et al. 2019).

Back near the start of this project I wrote in this blog about plans for the ice sheets in UKESM and some of the challenges that we were facing. Since then we’ve implemented a system of online climate downscaling over ice sheets in the Met Office Unified Model, taught the NEMO (Nucleus for European Modelling of the Ocean) ocean model to move its boundaries (a little) as it runs without becoming unstable and built a whole framework of Python code to transfer fields between the BISICLES (Berkley Ice Sheet Initiative for Climate Extremes) ice sheet model and the rest of the climate system. The whole system adds additional layers of complexity to what is already one of the most sophisticated climate models in the world. It’s all still a bit rough around the edges, which is why this isn’t included in the main UKESM1 release, but there is finally a functioning coupled climate-ice model that can do things that no other state of the art climate model can do.

 Figure 1: Surface Mass Balance (SMB) (the balance between accumulating and melting snow) estimated for Greenland in UKESM (blue and black lines) compared with the output from a specialised regional climate model (red line, Noel et al. 2018). UKESM captures the observed downturn in SMB at the end of the 1990s, often linked with decadal variability in the North Atlantic. The coupled ice sheet component additionally models the dynamic flow and calving rate of the ice sheet to give a complete estimate of how the ice mass will evolve.

So, has it all been worth it? What are we going to do with this model now we’ve made it? Whilst there are many open questions around the stability of the large ice sheets and how they interact with the climate around them, our first goal will be completion of a set of coupled climate—ice simulations for the Ice Sheet Model Intercomparison ISMIP6, an international model intercomparison that will provide projections of 21st century ice sheet mass loss. Early results from UKESM with our new online downscaling for the ice compare very well to regional climate model results (Figure 1) and suggest that additional surface melt of Greenland alone could be adding another millimetre to global-mean sea level every year by 2050.

Figure 2: Pine Island Glacier on Antarctica is observed to be thinning and retreating rapidly, most likely in response to ocean warming underneath its ice shelf. With the interactive ice in  UKESM we can model how the ocean melts the shelf away (top right), the flow of the glacier supplying new ice to the region (top left) and predict the overall rate of retreat. These are early results but show promise.

The most exciting new science we want to tackle with the ice in UKESM1 sits at the other end of the globe, however. There are many theories about how the floating ice shelves that fringe Antarctica respond to changes in ocean conditions, and how the flow of the grounded ice upstream will respond, but the uncertainties are enormous. Estimates of the resulting contribution to global mean sea level rise at 2100 range from centimetres to metres (see Edwards et al. 2019 for a recent perspective). This is an inherently coupled problem whose physics simply cannot be understood by modelling any one part of the system in isolation. Like all complex problems, it’s also going to be very hard – there are poorly-observed, crucial details in each component that can significantly alter the final outcomes – and we’re not pretending that one model is going to lead us straight to the answers. For one thing, there are long-standing biases in the climate simulation of the Met Office models in the high southern latitudes that will need to be improved before we are really simulating the processes against the right background.  But with UKESM1 we’re now getting our hands on some tools that can start to see the coupled atmosphere-ocean-ice physics evolving together for the first time (Figure 2), and that’s a very promising development.


Church, J.A., and Coauthors, 2013: Sea Level Change. Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, Stocker, T.F., D. Qin, G.-K. Plattner, M. Tignor, S.K. Allen, J. Boschung,  A. Nauels, Y. Xia, V. Bex and P.M. Midgley, Eds. Cambridge University Press, 1137-1216

Edwards, T.L., M.A. Brandon, G. Durand, N.R. Edwards, N.R. Golledge, P.B. Holden, I.J. Nias, A.J. Payne, C. Ritz & A. Wernecke, 2019: Revisiting Antarctic ice loss due to marine ice-cliff instability. Nature, 566, 58-64,

Golledge, N.R., E.D. Keller, N. Gomez, K.A. Naughten, J. Bernales, L.D. Trusel & T.L. Edwards, 2019: Global environmental consequences of twenty-first-century ice-sheet melt. Nature, 566, 65-72,

Noël, B. and Coauthors, 2018: Modelling the climate and surface mass balance of polar ice sheets using RACMO2 – Part 1: Greenland (1958–2016). The Cryosphere, 12, 811-831,

Posted in antarctica, Arctic, Climate, Climate modelling, Cryosphere, Numerical modelling | Leave a comment

The Boundary Layer and Submesoscale Motions

By: Alan Grant

Science is an exciting career, although what you may consider to be exciting will depend on your field. Sometimes things get most exciting when what initially appears to be a frustrating problem turns into an interesting problem.

Our story starts with the decision to develop a new parametrization of the ocean surface boundary layer (OSBL). The main reason for this was to represent what is known as Langmuir turbulence (McWilliams et al, 1997), and as part of this, the depth of the base of the OSBL in the scheme would be determined using a prognostic equation. Simple models of the entraining boundary layer using a prognostic equation for the thickness of the boundary layer have been around for many years. However, a significant problem with the new scheme was to work out how to use a prognostic equation in a parametrization destined for use in a finite difference model. Eventually the new scheme was ready, and it was time to test it, initially using a one-dimensional model.

This scheme was developed as part of NERC’s Ocean Surface Mixing, Ocean Submesoscale Instabilities (OSMOSIS) project. As part of OSMOSIS, gliders were operated for a year in the area of the Porcupine Abyssal Plain (PAP, named after HMS Porcupine) observatory, off the continental shelf, to the south west of Ireland The scheme was assessed against this year of data.

Figure 1 : Comparison between mixed layer depth determined from the gliders (symbols) and the boundary layer depth from the model (red curve) during the Autumn. The start date of the simulation is 23rd September.

The first test looked at the deepening of the boundary layer which occurs during the autumn. Figure 1 shows that there is good agreement between the boundary layer depth from the model and the mixed-layer depth obtained from the gliders. Over the seventy days of the simulation the thickness of the boundary layer increases from about 50m to 150m, due to entrainment.

Figure 2 : Same as Figure 1 but for the winter. The start date of the simulation is 25th December

Encouraged by these results, we moved on to look at the performance of the model during the winter. Figure 2 shows the results of this comparison, and the reason for the frustration should be clear – during the winter the base of the boundary layer in the model gets much deeper than the mixed layer depth from the gliders. This deepening of the model boundary layer is due to entrainment, and there is little scope to simply tune the scheme to improve agreement, particularly given how well the scheme performs during the autumn.

This is where submesoscale motions enter in to our story. Submesoscale motions have scales between 100m to 10km, are generally associated with baroclinic instabilities, and are thought to restratify the upper ocean. While parametrizations of submesoscale motions are still crude, one developed by Fox-Kemper et al. (2008) is of particular interest. In this parametrization, the effect of the submesoscale motions is formulated in terms of a buoyancy flux. This makes it possible to directly couple the submesoscale parametrization to the equation for the thickness of the boundary layer in the new boundary layer scheme.

And so, our frustrating problem becomes an interesting problem, since we now need to think about how to couple parametrizations of two distinct processes. Preliminary tests of one way of doing this coupling are encouraging.

I forgot to mention, the new boundary layer parametrization has been implemented in the ocean model Nucleus for European Modelling of the Ocean (NEMO) where it gives overly deep boundary layers during the winter. The hope is that submesoscale motions will provide the solution to this problem.

References :

Fox-Kemper B., R. Ferrari, and R.W.Hallberg, 2008 : Parameterization of mixed layer eddies. I. Theory and diagnosis. J. Phys. Ocean., 38, 1145–1165, doi:10.1175/2007JPO3792.1

McWilliams, J.C., 2016: Submesoscale currents in the ocean. Proc. R. Soc. A., 472,

McWilliams, J. C., P. P. Sullivan, and C.-H. Moeng, 1997: Langmuir turbulence in the ocean. J. Fluid Mech.334, 1-30,

Posted in Boundary layer, Climate, Numerical modelling, Oceans | Leave a comment

Multi-fluids Modelling of Convection

By: Hilary Weller

Atmospheric convection – the dynamics behind clouds and precipitation – is one of the biggest challenges of weather and climate modelling. Convection is the driver of atmospheric circulation, but most clouds are smaller than the grid size and so cannot be represented accurately or at all. Consequently, all but the highest resolution models use convection parameterisations – statistical representations which estimate the heat and moisture that would be produced and transported by clouds if they were resolved. These parameterisations have become sophisticated, estimating the mass that is transported upwards and how this will influence the momentum, temperature, moisture and precipitation. Without these parameterisations climate models dramatically fail. However, there are still big problems with these parameterisations; they tend to produce unrealistic hot columns of air and the main regions of convection in the tropics are usually misplaced (see Figure 1). These are large heat sources for the global atmosphere and so errors in these locations have knock-on effects across the globe.

Figure 1: Fig 11 from Neale et al, (2013) showing “Annually averaged (a) observed precipitation from GPCP (1979–2003) and model precipitation biases (mm day 21 ) in AMIP-type experiments for (b) CAM4 at 18, (c) CAM4 at 28, and (d) CAM3 at T85, and in fully coupled experiments for (e) CCSM4 at 18, (f) CCSM4 at 28, and (g) CCSM3 at T85.”

There are two assumptions made in convection parameterisations that could be to blame for their poor performance. Firstly, there is no memory of the properties of convection from one time-step to the next. The convection properties are calculated each time step from scratch as if there had been no convection in the previous time step. Secondly, although there is a class of convection scheme called “mass flux”, convection schemes do not actually transport air upwards. They transport the heat, moisture and momentum but the distribution of mass in the vertical is not changed by the convection scheme. Removing these assumptions is quite tricky. You realise that what you need to do is solve the same equations of motion in clouds and outside clouds, but these need to be modelled separately because the clouds are such a small fraction of the total volume. This is the multi-fluid approach. Separate equations for velocity, temperature, moisture, and volume fraction are solved for the air in clouds and the environmental air, outside clouds. As the clouds and the environment are interwoven, we assume that they share the same pressure.

John Thuburn and I are working on making this approach work as part of the NERC/Met Office Paracon project to develop big changes to the way that convection is parameterised in order to remove some of the less realistic assumptions (Thuburn et al, 2018, Weller and McIntyre, 2019). As ever, it is proving more difficult than we expected. We knew that we would need to solve simultaneous equations for the properties in and outside the clouds and that these would need to be simultaneous with the single pressure. I naively thought that this would be sufficient and that my expertise in this area would make it possible whereas convection modellers are usually less familiar with this simultaneous solution procedure. However, it turns out that the multi-fluid equations can be unstable. They are easy to stabilise but the stabilisation can have the effect of making the two fluids behave as one which defeats the purpose. We need to make the two fluids move through each other. John has made some good progress on this (Thuburn et al, 2019).

The multi-fluid equations need transfer terms to transfer air in and out of the clouds. These are not a mystery. Traditional parameterisations predict these transfers, and these have been thoroughly validated and tested. However, the multi-fluid equations are sensitive to these transfer terms and do not behave in the same way as the traditional parameterisations. We also want to base the transfers on the sub-grid scale variability both inside and outside the cloud so that a cloud is formed if some of the air in a grid box is ready to rise and condense out water. There is plenty to do.

While the multi-fluids team from the Paracon project have been working on simultaneous solutions of equations for in and outside clouds, ECMWF thought that they would try a more direct approach – simply adding a term to the continuity equation based on the mass flux predicted by their traditional parameterisation (Malardel and Bechtold, 2019). This was also the approach taken by Kuell and Bott (2008). The problem with this approach is that it will be unstable if a large fraction of a grid box is cloudy. However, ECMWF and Kuell and Bott (2008) have not reported any stability problems, although the ECMWF approach did ensure that only a small fraction of each grid box is transported by the convection scheme. The results so far seem promising. However, to be able to increase the resolution and run with sufficiently long time steps so that the model is competitive, we will need the multi-fluid approach.


Kuell, V., and A. Bott, 2008: A hybrid convection scheme for use in non-hydrostatic numerical weather prediction models. Meteorol. Z. 17 (6), 775-783,

Malardel, S. and Bechtold, P. (2019), The coupling of deep convection with the resolved flow via the divergence of mass flux in the IFS. Q J R Meteorol Soc. https://doi:10.1002/qj.3528

Neale R.B., J. Richter, S. Park, P.H. Lauritzen, S.J. Vavrus, P.J. Rasch, M. Zhang, 2013: The Mean Climate of the Community Atmosphere Model (CAM4) in Forced SST and Fully Coupled Experiments. J. Climate., 26, 5150-5168,

Thuburn, J., G.A. Efstathiou, R.J. Beare, 2019: A two‐fluid single‐column model of the dry, shear‐free, convective boundary layer. Quart. J. Roy. Meteor. Soc.,

Thuburn, J., H. Weller, G.K. Vallis, R.J. Beare, and M. Whitall, 2018: A framework for convection and boundary layer parameterization derived from conditional filtering.  J. Atmos. Sci., 75 (3), 965-981,

Weller, H., W. McIntyre, 2019: Numerical Solution of the Conditionally Averaged Equations for Representing Net Mass Flux due to Convection. Quart. J. Roy. Meteor. Soc.,

Posted in Climate, Convection, Numerical modelling | Leave a comment

SuPy: An urban land surface model for Pythonista

By: Ting Sun

Python is now extensively employed by the atmospheric sciences community for data analyses and numerical modelling thanks to its simplicity and the large scientific Python ecosystem (e.g., PyData community). Although I cherish Mathematica as my native programming language (like Mandarin as my mother tongue), I can see I have coded much more in Python than in Mathematica over the past year for my urban climate research.

One of the core tasks in urban climate research is to build climate resilience, where climate information at various spatiotemporal scales is an essential prerequisite. To obtain such climate information, accurate and agile modelling capacity of the urban climate is essential. Urban land surface models (ULSM) are widely used to simulate urban-atmospheric interactions by quantifying the energy, water and mass fluxes between the surface and urban atmosphere.

One widely used and tested ULSM, the Surface Urban Energy and Water balance Scheme (SUEWS) developed by Micromet@UoR, requires basic meteorological data and surface information to characterise essential urban features (i.e., urban surface heterogeneity and anthropogenic dynamics). SUEWS enables long-term urban climate simulations without specialised computing facilities (Järvi et al., 2011; 2014a; Ward et al., 2016). SUEWS is regularly enhanced and tested in cities under a range of climates worldwide.

Figure 1: SUEWS-Centred workflow for urban climate simulations

The typical workflow of conducting a SUEWS simulation may consist of a few steps (Figure 1), where several pre- and post-processing procedures can in fact be easily done by Python. However, one inevitable step that often bothers me is that the change of a single parameter may lead to another loop of the above workflow, which is somewhat annoying and tedious as you need to switch back and forth between numerous applications again and again.

Given the glue-like ability of Python, I started the project SuPy (SUEWS in Python) to use Python as a central tool to build a SUEWS-back-ended urban land surface model since the development of SUEWS v2017b. After several months of development and testing, I’m very pleased to release SuPy (Sun 2019; Sun and Grimmond 2019) via PyPI that allows the easy installation with the following one-liner for all desktop/server platforms (i.e., Linux, macOS and Windows) with Python 3.6+:

python3 -m pip install -U supy

Figure 2: SuPy-aided workflow for urban climate simulations

Figure 3: SuPy simulation results of surface energy balance.

And the whole workflow in Figure 1 can now be done in a much simpler way (Figure 2) with the following code in Python for one stop to quickly perform a simulation and produce a plot of its results (Figure 3):

import supy as sp
#load sample data
df_state_init, df_forcing = sp.load_SampleData()
grid = df_state_init.index[0]
#run supy/SUEWS simulation
df_output, df_state_end = sp.run_supy(df_forcing, df_state_init)
#plot results
res_plot = df_output.loc[grid,’SUEWS’].loc[‘2012 6 4′:’2012 6 6’, [‘QN’, ‘QF’, ‘QS’, ‘QE’, ‘QH’]].plot()

Along with the software, we also have a dedicated documentation site to provide more information of SuPy (e.g, installation, usage, API, etc). In particular, to familiarise users with SuPy urban climate modelling and to demonstrate the functionality of SuPy, we provide three tutorials in Jupyter notebooks. They can run in browsers (desktop, mobile) either by easy local configuration or on remote servers with pre-set environments (e.g., Google Colaboratory, Microsoft Azure Notebooks). Those impatient tasters of SuPy can even try out the package in an online Jupyter environment without any configuration by clicking here.

The SuPy package represents a significant enhancement that supports existing and new model applications, reproducibility, and functionality. We expect SuPy will help guide future development of SUEWS (and similar urban climate models) and enable new applications of the model. Moreover, the improvement in SUEWS model structure and deployment process introduced by the development of SuPy paved the way to a more robust workflow of SUEWS for its sustainable success. To foster the sustainable development of SuPy as an open source tool, we welcome all kinds of contributions – for example, incorporation of new feature (pull requests), submission of issues, and development of new tutorials.


Järvi, L., C. S. B. Grimmond, and A. Christen, 2011: The Surface Urban Energy and Water Balance Scheme (SUEWS): Evaluation in Los Angeles and Vancouver, J. Hydrol., 411(3-4), 219–237, doi:10.1016/j.jhydrol.2011.10.001

Järvi, L., C. S. B. Grimmond, M. Taka, A. Nordbo, H. Setälä and I. B. Strachan, 2014: Development of the Surface Urban Energy and Water Balance Scheme (SUEWS) for cold climate cities, Geosci Model Dev, 7(4), 1691–1711, doi:10.5194/gmd-7-1691-2014

Sun, T.: sunt05/SuPy: 2019.2 Release, doi:10.5281/zenodo.2574405, 2019.

Sun, T. and Grimmond, S.: A Python-enhanced urban land surface model SuPy (SUEWS in Python, v2019.2): development, deployment and demonstration, Geosci. Model Dev. Discuss.,, in review, 2019.

Ward, H. C., S. Kotthaus, L. Järvi, and C. S. B. Grimmond, 2016: Surface Urban Energy and Water Balance Scheme (SUEWS): Development and evaluation at two UK sites, Urban Climate, 18, 1–32, doi:10.1016/j.uclim.2016.05.001


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

The future of spaceborne cloud radars, and some very specific questions about raindrops and snowflakes

By: Shannon Mason

Cloud profiling radars (CPRs) provide snapshots of the journeys of many billions of hydrometeors through the column of the atmosphere: from ice particles and liquid droplets in clouds, to the snowflakes and raindrops—mostly raindrops—that reach us at the surface, whether from gloomy stratus or towering tropical storms. CloudSat’s CPR, the first instrument of the kind in orbit, has now completed its remarkable decade-long stint as part of the A-Train of satellites, during which we learned a lot about where and how frequently clouds form, their vertical structures, and how often they precipitate (Stephens et al. 2018). However—and without wanting to sound ungrateful—to better understand the processes by which hydrometeors form, interact, and fall to the surface, we still have a lot of very picky-sounding questions about all those ice crystals and snowflakes, cloud droplets and raindrops, like: how many were there, roughly, and; what sizes and shapes did they come in? These microphysical specifics are critical to pinning down important details of the global hydrological cycle and radiation budget, as well as processes at much smaller scales.

Figure 1: The Doppler CPR aboard the upcoming EarthCARE satellite will have the capability to measure the fallspeeds of hydrometeors, providing insights into the size of raindrops and the structure of snowflakes. 

The next generation of CPRs will start to improve our answers to these questions, beginning with EarthCARE, which is due to launch in 2021 and will have the additional capability to measure the fallspeed of hydrometeors from the Doppler shift of the reflected radar beam. Our two papers on EarthCARE’s retrieval algorithms (Figure 1) have focused on how this Doppler velocity information can be used to make better estimates of precipitation:

  • The fallspeed of raindrops tells us their size, so we can distinguish tiny drizzle drops that fall slowly from larger, faster-falling raindrops. This allows us to resolve the growth of drops due to collision and coalescence with cloud droplets, or their shrinking due to evaporation (Mason et al. 2017).
  • The fallspeed of snowflakes can distinguish fluffy snowflakes from faster-falling particles that have captured liquid cloud droplets (“riming”), increasing their density—and this reveals where shallow layers of supercooled liquid may be hiding within deeper ice clouds (Mason et al. 2018).

Despite the challenges of measuring Doppler velocities on the order of 1 m/s from a spacecraft 400 km above the surface and travelling at 7 km/s, our work suggests that EarthCARE will help answer some of our questions about the sizes of raindrops and the structures of snowflakes.

Figure 2: Beyond EarthCARE’s 94-GHz Doppler radar planned for launch in 2021, the configuration of subsequent spaceborne radars is still under discussion. One important consideration is how much additional information can be gained from observing ice and rain at two and three radar frequencies. 

However, we often find we can make improved estimates of rain and snow by observing them at two or more radar frequencies simultaneously. The planning for CPR missions beyond EarthCARE is happening now, and dual- and triple-frequency as well as Doppler radars are being considered (National Academies of Sciences, 2018). It remains an open question what radar configuration (Figure 2) would provide the most information about raindrops and snowflakes, and the processes by which they grow and interact.

Radar measurements at multiple frequencies are especially useful for exploring the properties of larger ice particles and snowflakes, which have different signatures depending on their sizes and structures. Using ground-based radars in Finland—where we can test our remotely-sensed estimates against direct measurements of the snow at the surface—we’re currently quantifying how much information about snowflakes we can gain using three radar frequencies. Further insights about ice particles and processes will emerge from the PICASSO field campaign last winter and this spring (Westbrook et al. 2018), in which the FAAM aircraft directly samples ice clouds over southern England while being closely tracked by Doppler radars at four frequencies from the Chilbolton observatory in Hampshire.

The details of the insides of snowflakes and the sizes of raindrops may seem insignificant, but the insights we gain from these field experiments help to sharpen the science questions and techniques that will be used with the next generation of satellites. Following the success of CloudSat, EarthCARE and its successors will help constrain global estimates of the role of clouds and precipitation in the atmospheric energy and water cycles.


Stephens, G., D. Winker, J. Pelon, C. Trepte, D. Vane, C. Yuhas, T. L’Ecuyer, and M. Lebsock, 2018: CloudSat and CALIPSO within the A-Train: Ten Years of Actively Observing the Earth System. Bull. Amer. Meteor. Soc., 99, 569–581,

Mason, S. L., J. C. Chiu, R. J. Hogan, and L. Tian, 2017: Improved rain rate and drop size retrievals from airborne Doppler radar. Atmos. Chem. Phys., 17 (18), 11567–11589.

Mason, S. L., J. C.  Chiu, R. J. Hogan, D. Moisseev, and S. Kneifel, 2018: Retrievals of riming and snow density from vertically-pointing Doppler radars. J. Geophys. Res.: Atmos., 123, 13807 – 13834,

National Academies of Sciences Engineering and Medicine, 2018: Thriving on Our Changing Planet: A Decadal Strategy for Earth Observation from Space. Washington, D.C.: National Academies Press.

Westbrook, C., P. Achtert, J. Crosier, C. Walden, S. O’Shea, J. Dorsey, and R. J. Cotton, 2018: Scattering Properties of Snowflakes, Constrained Using Colocated Triple-Wavelength Radar and Aircraft Measurements, AMS 15th Conference on Atmospheric Radiation,


Posted in Climate | Leave a comment

The North Atlantic Oscillation and the Signal to Noise Paradox

By: Daniel Hodson

 The North Atlantic Oscillation (NAO) is a key driver of European weather. It is an Atlantic pressure dipole (Figure 1a) and varies over time, with some interesting long-term trends (Figure 1b).

The NAO directly affects EU climate and weather – rainfall, temperature and winds follow swings in the NAO. These lead to significant impacts on society (Palin et al. 2016 Bell et al. 2017), and, with the surge in European wind and solar power generation, we are more exposed to NAO variability than ever (Ely et al. (2013), Clark et al. (2017)).
Is the NAO predictable, or just random? Analysis of the observed NAO are equivocal (Stephenson et al. 2000). Since the NAO has such a significant impact on society, predicting it would be of great value.

Figure 1: The North Atlantic Oscillation a) Spatial pattern (1st EOF of observed DJF MSLP) b) spatial variation over time, and smoothed with 10 year running mean (red).

Well, now we can (partly). In 2014, the Met Office’s seasonal forecasting system (GloSea) successfully predicted the winter NAO, months in advance (Scaife et al. 2014 ) Such models use ocean and atmosphere observations to produce an ensemble of forecasts for the coming winter.  Scaife et al  2014 showed that GloSea ensemble mean NAO forecast is correlated with the observed NAO (~0.6 ) This now presents some useful skill.
However, the amplitude of the ensemble mean NAO is smaller than observed – by a factor of 3.

This conundrum, predicting the variability, but not the amplitude is the Signal to Noise Paradox  (Scaife and Smith 2018).

We decided to examine this paradox using an optimal detection technique (Sutton and Hodson 2003, Venzke et al 1999). This allows us to extract the leading forced or predictable modes from an ensemble of forecasts. These modes are essentially Empirical Orthogonal Functions EOFS, but use extra steps to correct the statistical biases.

The output of this analysis is a set of spatial patterns that show how the model atmosphere responds to the common forcings. This allows to find the forcings for each mode; and compare the strength of these modes to observations.

The December-January leading mode (Figure 2) is an NAO-like dipole pattern, whilst the second mode is a canonical El Niño Southern Oscillation (ENSO) pattern – the atmospheric response to El Niño.
Figure 2 G&H shows how these modes correlate with the underlying Sea Surface Temperatures (SSTs). The ENSO mode (F) is correlated with tropical Pacific SSTs – a classic El Nino SST pattern. However, the NAO-like mode (E) shows no large coherent regions of strong correlations over the oceans (G).

Figure 2: First (E) and second (F) predictable modes in the GloSea early winter (DJ) hindcasts. G) SST correlations with first mode (E). H) as G, but for F.

This confirms that the ENSO pattern is driven by the SST variations (and hence initial ocean conditions), but the NAO-like pattern appears not to be. What is driving this predictable mode? The only other remaining factor are the atmospheric initial conditions. The troposphere is too noisy for initial conditions to persist until Dec-Jan, but perhaps the initial conditions of the stratosphere could. Studies have shown that signals can propagate slowly downwards from the stratosphere, into the troposphere (Baldwin and Dunkerton 2000). Could these be the source of predictability of the NAO in this model? Previous attempts largely ignored accurately initialising the stratosphere from observations, but the GloSea model does. A recent study (O’Reilly et al 2018) with a different model suggests that the stratosphere may indeed be key.

We have extracted the predictable variations from the forecasts, but we can also assess the magnitude of these variations compared to observations. Figure 3 shows this comparison for the NAO-like predictable mode in December-January. It is clear that the response in the model is much weaker; further analysis shows that this NAO-like predictable mode is indeed ~3 times weaker than in observations (Consistent with Eade et al 2014).Applying the same techniques, we can show that the ENSO mode is also weaker than observed, but by a noticeably smaller factor (~1.8).

Figure 3: Comparison of the magnitude of the NAO-like predictable mode in A) observations and B) GloSsea hindcast ensemble.

This suggests that the weaker response of the predictable modes in this forecast model is the not the same for all modes – some modes appear to be driven more weakly than others. This may be because, ultimately, different atmospheric processes are involved in driving these modes. Some of these processes may be weaker in the models than they are in the real world. If we can improve our understanding of these processes, we may be able to improve our seasonal NAO forecasts even more.

A few years ago, forecasting winter European climate months ahead seemed implausible. But now we know that useful NAO forecasts were there all along, buried in the noise. Further research may lead to routine, skilful forecasts of the NAO, months, or even seasons ahead.


Baldwin, M. P. and T.J. Dunkerton, 2001: Stratospheric harbingers of anomalous weather regimes. Sci., 294, 581–584,

Bell, V. A., H. N. Davies, A. L. Kay, A. Brookshaw, and A. A. Scaife, 2017: A national-scale seasonal hydrological forecast system: development and evaluation over Britain. Hydrol. and Earth Syst Sci., 21, 4681–4691,

Clark, R. T., P. E. Bett, H. E. Thornton, and A. A. Scaife: 2017, Skilful seasonal predictions for the European energy industry. Environ. Res. Lett., 12, 024002,

Eade, R., D. Smith, A. Scaife, E. Wallace, N. Dunstone, L. Hermanson, and N. Robinson: 2014, Do seasonal-to-decadal climate predictions underestimate the predictability of the real world? Geophys. Res. Lett., 41, 5620-5628,

Ely, C. R., D. J. Brayshaw, J. Methven, J. Cox, and O. Pearce, 2013: Implications of the North Atlantic Oscillation for a UK-Norway renewable power system. Energy Policy, 62, 1420–1427,

O’Reilly, C. H., A. Weisheimer, T. Woollings, L. Gray, and D. MacLeod, 2018: The importance of stratospheric initial conditions for winter North Atlantic Oscillation predictability and implications for the signal-to-noise paradox. Quart. J. Roy. Meteor. Soc., 145, 131-146,

Palin, E. J., A. A. Scaife, E. Wallace, E. C. D. Pope, A. Arribas, and A. Brookshaw, 2016: Skillful seasonal forecasts of winter disruption to the U.K. transport system. J. Appl. Meteor. Climatol., 55, 325–344.

Scaife, A. A., A. Arribas, E. Blockley, A. Brookshaw, R. T. Clark, N. Dunstone, R. Eade, D. Fereday, C. K. Folland, M. Gordon, L. Hermanson, J. R. Knight, D. J. Lea, C. MacLach- lan, A. Maidens, M. Martin, A. K. Peterson, D. Smith, M. Vellinga, E. Wallace, J. Waters, and A. Williams, 2014a: Skillful long-range prediction of European and North American winters. Geophys. Res. Lett., 41, 2514–2519.

Scaife, A. A. and D. Smith, 2018: A signal-to-noise paradox in climate science. npj Climate Atmos. Sci., 1,

Stephenson, D. B., V. Pavan, and R. Bojariu, 2000: Is the North Atlantic Oscillation a random walk? Int. J. Climatol., 20, 1-18.<1::AID-JOC456>3.0.CO;2-P

Sutton, R. T. and D. L. R. Hodson, 2003: Influence of the Ocean on North Atlantic Climate Variability 1871-1999. J. Climate, 16, 3296–3313.;2

Venzke, S., M. R. Allen, R. T. Sutton, and D. P. Rowell, 1999: The atmospheric response over the North Atlantic to decadal changes in sea surface temperature. J. Climate, 12, 2562–2584.;2

Posted in Climate, Predictability, Stratosphere | Leave a comment