Page tree
Skip to end of metadata
Go to start of metadata

1.1. Surface Winds

1.1.1. Mean Winds

Typically the winds in the boundary layer are backed and decreased from the geostrophic values through the retarding action of surface friction.  Frictional retardation varies according to the differing surface features – quite high over rugged terrain or forests, moderate over open grassland or farmland, small over featureless tundra or deserts.  Over open water the friction varies according to the size of waves which in turn varies in part according to the wind itself – generally minimal over smooth water but significant where swell or wind waves are high (e.g. near major storms).  

Sub-grid drag processes have a major impact upon the momentum flux at the surface and the evaluation of low-level winds. However, it is difficult to establish appropriate values for individual locations or areas. The effect on the momentum transfer at the surface is parameterised by a "roughness length" for the various vegetation types or land usages and may well vary with time as crops grow or are harvested, with autumnal leaf fall, or with widespread forest clearances etc.  The setting of roughness lengths has substantial uncertainty.

Fig9.3.1:Typical atmospheric model wind profiles over surfaces with differing roughness lengths (z).

Observations of the “surface” wind are conventionally measured and reported as the wind 10m above the ground.  The lowest model levels (HRES level 137, ENS level 91) both lie close to this height and it would seem simplest to use these data directly as the forecast 10m wind. However, to evaluate the momentum flux at the surface, with its consequent effect upon wind speeds at the lowest model levels, a representative "surface roughness" (the roughness length parameter) is assigned to each grid square within the model formulation. This naturally impacts upon the wind at the lowest model levels. This roughness length for the grid square will ordinarily not be typical for an individual location within that grid box (unless the grid box is very uniform), so given the likely variability of vegetation and land use (e.g. see Fig9.3.2) one should not expect HRES level 137, ENS level 91 data to agree well with observed winds.  

Stations sited according to WMO guidelines (which ECMWF uses for verification) are located in open terrain, over short grass and with no obstacles in the vicinity; hence a low roughness length parameter is appropriate. However the roughness length parameter assigned to the grid square will usually be greater. Consequently observed winds at WMO station sites will often be stronger than those forecast at HRES level 137 and ENS level 91 (because of the influence of areas of greater roughness within the grid square). ECMWF addresses this discrepancy as described immediately below.

Fig9.3.2: An example of the variability of land surface within an IFS grid box illustrating the difficulty in assigning a representative roughness length for the whole grid square area.  A verifying anemometer at A could reasonably be associated with a roughness length appropriate to short grass, but an unconventionally placed anemometer at B might be associated with a higher roughness length due to proximity of larger vegetation.  Both locations would be affected by valley winds, blocked flow and possibly gravity waves.  The red lines show the extent of a very approximate 9km x 9km schematic HRES grid square.  An HRES meteogram is derived from 3 grid points surrounding the station which would include at least one over high ground.  An ENS grid square is 18km x 18km and covers four times the area.  An ENS meteogram is derived from the 4 grid points surrounding the station which in this example might not include any low-level grid points at all.  See Section on Selection of gridpoints for Meteograms for details.    

1.1.2. Forecast 10m wind output

ECMWF's standard "10m wind components" - surface parameters - differ fundamentally from wind components on the lowest model levels, even if those are nominally at 10m.  These "10m wind components" are diagnostic quantities generally computed not by using the roughness length of the tile itself, but instead assuming a roughness length for short grass (=0.03m), the surface over which (by WMO convention) winds should be measured.  ECMWF creates these fields using a 'blending height' (40m since 2013), by extracting the model wind speed there, and by then effectively extrapolating that down to the surface on the assumption of a 0.03m roughness length. The new speed is converted to velocity (and thereby u and v components) by replicating the wind direction on the lowest model level.  It is these wind values which are shown as surface winds on ecCharts, Meteograms etc.  If the tile roughness length itself is ≤0.03m (e.g. over desert or ocean) then the above adjustments are not made.

Whilst such adjustments should lead to better agreement with WMO station measurements, there are still, arguably, issues with the 10m wind output in regions with very irregular (very mountainous) topography, as illustrated on Fig 9.3.3.

Fig9.3.3:  Mean sea level pressure (MSLP) isobars and 10m wind from data time 00UTC 2 March 2011 T+12hr forecast verifying 12UTC 2 March 2011.  The 10m winds are unrealistically weak over the rugged Norwegian mountains.  Values of 10m/s might be realistic in sheltered valleys, but not on exposed mountain ranges.

1.1.3. Sub-grid drag processes Parameterisation of Surface Roughness - Horizontal scales <5km.  

A roughness length parameterisation (z) is used to obtain a realistic area-averaged turbulent drag according to the underlying surface.  This is used during the calculation of winds at the lowest levels.  Surface roughness:

  • over land, is dependent upon the surface vegetation (See Fig9.3.4).
  • over sea, the drag on the low-level air flow is modelled using the Charnock parameter.  This includes an aerodynamical roughness length that is a function of the 10m wind speed, but limited to a maximum value of that associated with 10m winds of 40m/s.  Read more information about coupling between the atmosphere and ocean waves in waves near tropical storms.

Also considered is a representation of drag from small scale orography.  Flow around steep orographic features gives temporary, variable and quickly fading atmospheric waves that lead to drag - Turbulent Orographic Form Drag (TOFD).

Fig9.3.4: Sub-grid drag mechanisms. Scales smaller than 5km. Roughness length parameters (z) indicate the relative effects of different surfaces.  It is important to note that roughness length (z) is only a parameter to use within the calculations; it does not represent typical heights of the obstruction. Parameterisation of Surface Roughness - Horizontal scales between 5km and the model grid resolution

To represent orographic features that are too small to be resolved by the model grid (9km HRES, 18km ENS) orographic parameters are derived from the height of valleys, hills and mountains, taken from a land elevation data set with ~1km resolution which is first averaged to ~5km resolution.

The orographic parameters are:

    • Standard Deviation (variability of surface height in the sub-grid valleys, hills and mountains within a grid box}.
    • Anisotropy (shape of high ground - ridge, circular, etc).
    • Angle (dominant orientation of the valleys, ridges, etc).
    • Slope (the mean slope of the sub-grid orography). 

These parameters are used as input within the sub-grid orography scheme that represents:

    • low-level blocking (strong drag at lower levels where the flow is forced around a mountain) - red arrows on Fig9.3.5.
    • orographic gravity wave effects (gravity waves are excited by the “effective” sub-grid mountain height, i.e. height where the flow has enough momentum to go over the mountain) - black arrows on Fig9.3.5

Fig9.3.5: Sub-grid drag mechanisms - topographic. For scales larger than 5km.

Sub-grid drag processes:

  • have a large impact on the large-scale circulation, at all timescales.
  • are responsible for known systematic circulation biases.

The orographic drag parameterisations are fairly simplistic and don’t necessarily behave well with differing model resolutions.

1.1.4. Verification of forecast winds

As highlighted above users should be cautious when verifying forecast winds against observed values, and in particular should be aware that:

  • Local surface winds depend strongly on local exposure. 
  • Surface winds vary on small space and time scales.
  • Sometimes verifying sites are not well exposed, particularly at unofficial locations.  Winds may not be measured as well in some directions as in others.
  • Near-surface wind forecasts have weaknesses in many mountain areas, due to the difficulty of parameterising interaction between the airflow and the highly varying sub-grid orography (see Fig9.3.3).
  • When interpolating, (as for meteograms) the positions and altitudes of IFS grid points surrounding a selected location of interest have a significant impact upon the forecast values.

1.1.5. Wind Gusts

10m wind gusts (Ugust) are computed in the IFS by adding a turbulence component, and (at times) a convective component to the 10m mean wind:

Ugust  = U10  +  (turbulence contribution)  +  (0.6 D ( max( 0, UMLA-UMLB ) ) )


  • U10 is the 10m wind speed.

  • D is a boolean variable indicating whether the deep convection scheme is active or not at the said time (1 if yes, 0 if no).
  • (UMLA - UMLB). The convective contribution is taken to be proportional to the wind speed difference between model levels A and B, where A and B are the model levels that best correspond, in a standard atmosphere, to pressure levels of 850 and 950hPa (so the same model levels are used for this everywhere at all times).
  • "max" means take the maximum of the two values bracketed, so here only if the higher level wind speed is greater than the lower level wind speed is this non-zero

Regarding the turbulence component, users should be aware that:

  • This scales with the surface stress, which in turn relates to roughness lengths. So the component tends to be much greater over rough surfaces than over smooth, leading to a larger ratio of gust to mean speeds over rough surfaces.
  • Because of the above, gusts tend to exhibit less land-sea contrast than do 10 mean wind speeds.

Regarding the convective component, users should be aware that:

  • In spite of there being a convective contribution in the above equation, which does help, extreme gusts associated with vigorous convection (e.g. MCSs or squall lines) are generally under-estimated, sometimes by a factor of 2 (e.g. 60+kt gusts when 30-40kt gusts predicted).
  • Spuriously large gusts can occur associated with dynamically active, strongly sheared warm fronts, particularly in cases where convection (even quite weak convection) has been identified in the lowest layers (say below about 850hPa).
  • Current resolution cannot represent the internal circulations of convective systems and empirical rules relating instability to gusts do not work in all cases.
  • Gusts can be as much as 2 to 3 times the mean wind speed, but this depends on stability of the atmosphere and roughness of the surface.  Users should consider local effects (e.g. funneling, particularly in stable conditions, katabatic flow, CAPE, CAPE-shear and convective activity).
  • ECMWF's CAPE and CAPE-shear EFI fields have been designed to help the forecaster to predict extreme weather related to deep moist convection; this includes forewarning of the potential for extreme convective gusts.
  • There can be an unrealistic localised reduction in gustiness over a relatively short distance over water (~2 grid lengths). Strong gusts may be underestimated just offshore (lake or sea) when the wind is blowing off the land.

1.1.6. Winds related to Intense Extra-tropical Cyclones

More extreme extra-tropical cyclones can give rise to swathes of damaging winds (Fig 9.3.6). Verification using many case studies has shown that in most cases both the mean winds and the gusts are reasonably well predicted, on average.  That is to say if the cyclone shape, central pressure and pressure gradients are well-forecast, then so are the winds, in general.  There is also no clear evidence to suggest that extreme cyclones are under- or over-deepened on average, though some case studies show that the onset of the phase of most rapid deepening can sometimes come a little too late (meaning, for a standard eastward-moving low, a little too far east).  In turn this tends to mean that the onset of very strong winds may actually be further west than predicted, but conversely those very strong winds may actually not extend as far east as in model output.  A particularly difficult strong wind swathe to predict is the one associated with the rare phenomena known as the "sting jet".  Evidence from a few cases suggests that IFS gusts are not biased but are representative of the real gusts experienced during a sting jet event.  However because this phenomena is intrinsically difficult to predict, and because the damage swathe associated with it is narrow, and typically short-lived, one has to expect large errors in forecasts for specific locations in potential sting jet scenarios.  Because of these factors and the often finely balanced nature of extreme cyclogenesis events (that relate to sting jets) one must expect some jumpiness in wind forecasts in these situations, even at relatively short ranges.  This is particularly true for HRES but also, sometimes, for the ENS.  Nonetheless, the ENS can help a great deal with highlighting the intrinsic uncertainty in gust forecasts in very cyclogenetic situations.  Users are also encouraged to make use of the cyclone database products when dealing with strong winds related to cyclogenesis events.  These products were designed, in part, with that particular use in mind.

Fig9.3.6: Conceptual model of the life cycle of an extreme cyclonic windstorm.  The areas prone to significant gusts are:

  • yellow zone: the warm jet or conveyor within the warm sector and/or at the cold front.
  • orange zone: the cold jet or conveyor within the cold sector on the northern flank (N Hem) of the depression.
  • red zone: the sting jet in the flow around the western flank of some depressions.

Note that most intense cyclones will not exhibit all 3 jet phenomena shown, for most only 1 or 2 will be present; the sting jet is by far the rarest of the three. Also note that the tracks of individual cyclones will naturally differ greatly from the one example shown here, in location and form. Many tracks will curve to the left, as shown, but some are much straighter even in the rapid deepening phase.

See cyclone database products for a key to colours used (for spots) in the cyclone life-cycle phases. Very small cyclones

The guidelines in Section 1.1.6 hold for most cyclones, but very small cyclones pose extra difficulties. They may be beyond the capabilities of the IFS horizontal resolution, particularly for the ENS.  A windstorm that hit Brittany during March 2017 (that was incorrectly labelled "Zeus" in web links) was one such example.  For this case there were systematic resolution-related differences in intensity between ENS and HRES. HRES developed a more intense windstorm, and was more accurate. There have been other cases too. And even very small cyclones can have a sting jet associated. Accordingly forecasters need to treat ENS output with particular care when there is potential for very small cyclones to develop - say with lateral dimensions of order 200km or less.

Additional Sources of Information

(Note: In older material there may be references to issues that have subsequently been addressed)

Updated/Amended 06/11/20 - Additions regarding drag

Updated/Amended 03/12/20 - Additions regarding gusts near warm fronts

  • No labels