Section 9.4 Visibility

Visibility within IFS

As a weather hazard, fog is extremely important, but a difficult variable to predict.  The visibility diagnostic includes information on the reduced visibility in fog, but correctly predicting very low visibility is dependent on predicting the correct dynamic and thermodynamic conditions in the boundary layer.  These can be highly variable in space and time and are often tied to orographic features that are not resolved by the atmospheric model.  For these reasons a probabilistic approach using the ensemble members is more beneficial.

The visibility diagnostic is an experimental product introduced in 2016 and the quality of this product is undergoing continued evaluation with the aim of improving the diagnostic and its usefulness.  It should be used with caution; expectations regarding the quality of this product should remain low.  Visibility products are currently made available so that they be evaluated and also to allow users to gain experience.

Visibility in the atmospheric model is defined as the near surface horizontal visibility within the lowest 20m layer above the surface and is calculated using an exponential scattering law.  The visual range is taken to be the distance at which, despite extinction of light by the atmospheric model fog, it would be possible to distinguish between a theoretical object and the model's foggy background.  In the model, the contrast between object and foggy background is taken as 2% difference in luminosity (a liminal contrast of 0.02).  The extinction coefficients are calculated for the contributions from precipitation and cloud water droplets in the lowest layer of the model and also the presence of climatologically and seasonally varying aerosol species. 

The current technique for visibility forecasting has several limitations; it uses a fixed particle size for cloud and precipitation particles, and the effects caused by local deviations of the aerosol fields from climatological values and the interactions of fog and aerosol particles are not modelled yet.  Horizontal resolution is relatively low, and the lowest level in the vertical is 10m so capturing the detailed extent and composition is difficult, especially in rugged areas.  Where fog is forecast due to cloud water drops in the lowest model layer, visibility can fall below 500m.  Visibility is still a relatively new product and undergoing continuing assessment.  Initial perceptions are that:

• radiation fog tends to be rather too dense, to form too slowly and clear too quickly (by about 1-3hr),
• hill fog is rather under-represented,
• visibility in precipitation falls rather too low.

The extinction coefficient of clean air is taken to be equivalent to a visibility of 100km so values can be no greater than this, and in general visibility in clean air seems rather too great.

Visibility calculation using a ‘tuned’ CAMS aerosol climatology was introduced in mid 2017 in Cycle 43R3 and may alter or reduce the deficiencies in visibility outlined above.

It should be re-iterated that this is a preliminary implementation of a visibility diagnostic.  Users are advised to keep themselves updated about visibility products through the ECMWF Newsletter and web site. 

Fig9.4-1: An example of HRES forecast fog distribution on right hand chart compared with the observed visibility on left hand chart (visibility as in coloured scale).  Forecast run data time 00UTC 23 January 2017 T+30 verifying at 06UTC 24 January 2017.  Thick fog, less than 100m visibility in places, caused problems over southern England.  Many flights were cancelled and others delayed.  In the early morning hours the fog was more widespread over England and Benelux countries.  HRES forecast visibility was too poor in some areas but provided good guidance to forecasters regarding areal extent of the fog risk. 

Fig9.4-2: Fog probability in the early medium-range from an earlier forecast run, data time 00UTC 21 January 2017 T+78 verifying at 06UTC 24 January 2017.  The ENS also highlighted England and Benelux countries as areas where fog was likely.

An example of fog prediction by HRES

However, HRES/Ensemble Control can give good guidance on the development of fog and signal to users the possibility of otherwise unexpected hazardous conditions.  Sea or coastal fog is relatively rare in June in the Mediterranean but HRES was able to predict well ahead of time a significant event in June 2022.  

Fog banks formed as a weak westerly flow brought relatively moist surface layer air across anomalously cold sea surface temperatures near the south coast of Sicily.  HRES captured the development and eastward advance of the fog banks particularly well.   Previous HRES forecasts showed a strong signal for fog banks in the general southern Sicily area, even at four days lead-time.  These gave a useful warning to coastal craft and recreational sailing, especially given the unexpected nature of the event.  Little if any fog was predicted for any other parts of the Mediterranean.

Fig9.4-3Hour-by-hour visibility HRES predictions around Sicily 16 June 2022 00 to 24 UTC (HRES DT 16 June 2022 T+0 to T+24).  The colour scale shows the visibility in m.  Note the eastward movement of the fog bank with some detail shown of the changes in area.  Some areas with visibility <100 m (purple) are evident.

Fig9.4-4: Mediterranean sea surface temperatures around Sicily (HRES) at 00 UTC 16 June 2022.  The colour scale shows the temperatures in ºC

Fig9.4-5: Example of visibility meteogram showing forecast visibility in cumulative frequency form.  Colours represent forecast visibility ranges (see scale). 

Additional Considerations: Fog and Freezing fog

Prediction of fog depends strongly on the simultaneous and accurate prediction of several parameters in the lowest layers of the atmosphere on a spatial scale smaller than the IFS (scale of m rather than km).  Forecast values of temperature and moisture are of prime importance of course, but these in turn depend upon correct assessment and prediction of surface parameters (e.g. soil moisture and the efficiency of nocturnal radiative cooling) and the precise low-level wind and associated turbulent mixing.  In particular the detail of orography is very important in governing local winds (e.g. katabatic winds) and stagnating air in valleys. 

Pinpointing regions of very low visibility is very challenging for IFS, possibly even more so in cases of freezing fog where humidity mixing ratios are smaller and moisture (water and ice) interactions are more complex.  Fog (or freezing fog) may be very patchy in areas where IFS forecasts fog to occur, even those where a high probability of fog is predicted.  It can be more extensive than IFS indicates, but equally patchy fog may not be indicated at all.  

For these reasons jumpiness in forecasts of fog and freezing fog should be expected.  And results from a series of forecasts may well not converge towards a correct distribution or "intensity" of fog, particularly at the lowest visibilities.  At present, there is no clear evidence of any systematic differences in the IFS handling of fog versus freezing fog, although it remains possible that there are some (e.g. biases).

Fog forecasts are quite difficult to verify in the absence of satellite imagery since conventional observations are reports of fog at an individual location, not an area. Also, patchy fog may not be identified.

Sub-grid variability in visibility (due to patchy fog) can potentially be "greater" than it is in for any other meteorological parameter.  Users should always remember that raw IFS forecasts are delivered on the model grid resolution (e.g. currently ~9km boxes for ENS).  Compare this with the dimensions of a fog patch, which may be ~10m.  These dimensions differ by 3-4 orders of magnitude.

Fig9.4-6: Forecast visibility at 30hr lead time (left) and 6hr lead time (right) with verifying observations, in a freezing fog situation.  Although there is a good deal of consistency between the forecast fog distribution, there is slightly more fog forecast on the T+6 forecast chart than on the T+30 chart.  However, consistency should not necessarily be taken as a good guide to the true extent of fog.  The large area of fog over north France is captured quite well,  but over Britain observations of fog do not in general match either of the forecast locations at T+30 or T+6 (e.g. East Anglia).   However, fog is often quite patchy and may not be captured by the observations (observations are reports of fog at an individual location, not an area).    Additionally it should be noted that forecast visibility charts show poor visibility where cloud covers hills.

Fig9.4-7: Forecast probability <200m visibility at 30hr lead time (left) and 6hr lead time (right) with verifying observations for the same freezing fog case as in Fig9.4-6.  In this example the areas of low/moderate probability capture the observed 200m visibility fairly well over England and Wales, and the probability distribution sharpened up as lead time reduced (e.g. over East Anglia), though not clearly in a manner that verified better.  A chart of probability of fog with visibility <200m (or other low thresholds) should only be taken to be very roughly indicative of those areas most at risk.

Users should recognise that:

• Fog can be very patchy and 10km-scale models (such as IFS) cannot be expected to handle this well. 
• Areal extent of fog is broadly captured by HRES and ENS forecasts but at best these are only indicative of areas at risk.  Such predictions do not preclude occurrence outside of the apparent "risk areas".
• Local skill in forecasting the occurrence of fog (e.g. for a specific site) can be very low.  Local skill in predicting very low visibilities (e.g. meeting motoring warnings threshold criteria) can also be very low.  Beware!
• Jumpiness in fog forecasts (ENS and HRES) must be expected.  Results may well not show convergence towards the correct solution.  Forecasts with shorter lead times will not necessarily be more skilful that those from longer lead times.

Errors in forecast near-surface data associated with cases of thick fog.

A bug introduced in cycle 47r3 in October 2021 has resulted in a tendency for 2m dewpoint and relative humidity (RH) values to sometimes drop to unrealistically low values (e.g. 30% RH) once fog has been predicted or diagnosed in model output. This process is most apparent when dense fog is forecast (e.g. visibility < 100m).  This is clearly not physically realistic and indeed wrong - RH should be close to 100% in such a situation.  The cause has been found to be positive feedback between two interacting and imperfectly represented mixing processes in the new moist physics scheme.  The problem has been added to Known IFS forecasting issues and a fix has been prepared with implementation in the next IFS upgrade expected late in 2022.

Fig9.4-8: An example of incorrect temperature and dew point forecasts in a case of predicted fog.  Imperfectly modelled mixing processes near the surface induce errors in 2m temperature and humidity.

Visibility reduction due to sand or dust

Major components in the forecast and evaluation of the IFS visibility product are the temperature and moisture content of the airmass in question.  In arid regions 2m temperatures can be high (at least during the day) and the humidity extremely low.  In consequence the forecast visibility will generally remain high (15km or greater).

However, significant reductions in visibility can occur in arid regions (e.g. deserts or other areas after prolonged drought conditions).  Raised sand or dust occurs when winds lift amounts of sand and dust from bare, dry soils into the atmosphere.  Strong or gusty winds can raise large amounts of dust and sand resulting in observations of drifting sand or dust, or blowing sand or dust, or sandstorm.  Strong and gusty surface winds associated with outflows from large convective storms can also induce raised dust and sand.   Remnant dust, previously lifted nearby or elsewhere, can continue to impair visibility even in areas where surface winds are, or have become, relatively light.   Visibility can be reduced to 5000m or less in dust haze, but visibility associated with raised dust or sand can fall to below 1000m, and in severe cases can be reduced to 100m or less.   It is important to recognise the IFS visibility product will not identify poor visibility associated with these phenomena.

ecCharts offer forecast charts of aerosol optical depth and dust aerosol optical depth.  These, in effect, supply information regarding the reduction of transmittance of light by particulate matter in the lower layers of the atmosphere.  The charts can be used as an indication of areas where visibility is likely to be reduced because of dust or sand in the lower atmosphere, particularly where the winds are forecast to be strong and/or gusty.  There is no direct link between aerosol optical depths and visibility reduction due to rising sand and dust but they are strongly indicative of areas at risk, particularly where winds are forecast to be strong.  This is particularly helpful in data sparse areas (e.g. the Sahara).

Fig9.4-9: ecChart of NW Africa, T+57 VT 09UTC 10 March 2019, DT 00UTC 8 March 2019 showing forecast 10m winds, visibility, aerosol optical depth.  "Dust aerosol optical depth", though not shown on the map, is shown on the time series and probe diagrams.  Also shown are edited METAR reports for the same time (International station code (e.g. DAUA = Adrar), wind (e.g. 010/29 = N'ly 29kt), visibility (e.g. 0300 = 300m), present weather (e.g. BLSA = blowing sand) - visibility observations below 800m are shaded in red on the diagram.

The dark red area indicates where relatively high aerosol optical depth is forecast.  Within this area visibilities are, with one exception, below 3000m.  Three stations where wind strength is above 20kt report visibility as 600m or below in blowing or drifting sand (the previous day Ghardaia (DAUG) reported sandstorms continuously between 1030UTC and 1600UTC with winds generally around 35kt and gusts to 46kt).  Stations in the surrounding orange area report lighter winds and visibility above 3500m in haze.

The pin (near 34N 05E) locates an upslope elevated area downwind of marshy lakes (the Chott Ech Chergui).  Some high-based showers were forecast in the area suggesting a local reduction in visibility from precipitation or local increase in humidity near the surface.   Forecasts of fairly high aerosol and dust aerosol optical in the area are not responsible for the IFS forecasts of locally reduced visibility.

Fig9.4-10: Meteosat IR Channel 10 (approx 11μm to 13μm).  09UTC 10 March 2019.  Some indication of the extent of the dust/sand is given by the light grey area on the satellite imagery.

Fig9.4-11 ecChart as Fig9.4-9 with the Meteosat IR Channel 10 shown in Fig9.4-10 superimposed.  This compares well with the area where higher aerosol optical depth are forecast.    

Fig9.4-12: ecCharts time series and probe data for Ouargla (DAUU) and Ardra (DAUA) taken from the ecChart presentation of forecast data based on DT 00UTC 8 March 2019 (as for Fig9.4-H). The METARs at 09UTC 10 March 2019 show the very reduced visibility at these stations.  The time series shows the IFS forecast visibility varies with time but nowhere approaches the observed visibility.  The aerosol optical depth and dust aerosol optical depth forecast data gives greater indication of reduction in visibility due to dust and sand.  However there is no direct link between the aerosol optical depths and visibility reduction due to sand and dust but they are strongly indicative of areas at risk particularly where winds are forecast to be strong.