The objective of this study is to assess the impact of the aerosol direct effect on the forecasted meteorological parameters during the dust storms that affected the Sahara and eastern Mediterranean Basin in April 2012. To achieve that, the MACC-II global system was run with no dust aerosols, with the aerosol direct effect estimated from a climatology, i.e. in its pre-operational configuration, and with the aerosol direct effect estimated from prognostic aerosols. All runs were carried out with a TL511 horizontal spectral resolution which corresponds to a grid-box size of about 40 km. 60 vertical hybrid sigma-pressure levels were used, the lowest level being 17 m above the surface. The time step was 900 s.

In this section, forecasts of the REF and TOTAL experiments are evaluated against ground observations of radiative parameters. The TOTAL experiment is also compared to NOAER to assess the impact of aerosols on the radiative fluxes for this situation.

First, NOAER and REF experiments are compared to check the impact of the Tegen aerosol climatology on radiative fluxes. This climatology provides rather small values of dust AOD for this period, close to 0.3, and the impact on radiative fluxes is also rather small: in the order of 1 W m-2 for long-wave fluxes and 20–30 W m-2 for short-wave fluxes (not shown). Maximal and minimal temperatures are marginally impacted by this radiative forcing, by around 0.1 to 0.2 K.

Looking at the time series of surface parameters measured at Tamanrasset and Cairo (Fig. ), we can see that Tamanrasset was mostly affected by the dust storm from 12 to 14 April; high clouds were also present at times from 10 to 13 April. Cairo was impacted by high dust load mainly on 15 and 18 April.

Aerosols and clouds both impacted radiative fluxes on 10–13 April at Tamanrasset. Observations of the diffuse and direct components (not shown) of downward solar surface flux (DSSF) show a decrease of up to 400–500 W m-2 for the direct component, matched by an increase of up 200 to 300 W m-2 for the diffuse component during this period. Total solar radiation was 200 to 300 W m-2 smaller on 12 and 13 April compared to 14–18 April. Observed downward long-wave fluxes at the surface were on average around 60 W m-2 higher than simulated on these 2 days. Since this is true for both REF and TOTAL, a bias in the model cloud cover is the likely cause.

Long-wave downward radiation forecasted by TOTAL was 10–20 W m-2 larger than with REF on 12–13 April at Tamanrasset, and 20–30 W m-2 larger on 15 and 18 April at Cairo, showing that the aerosol burden provided by the aerosol scheme was greater than the values given by the Tegen climatology. This reduced a negative bias of more than 20 W m-2 (for REF) in the forecasted long-wave fluxes at Tamanrasset. The DSSF was lower with TOTAL by 50 to 100 W m-2 on 12–13 April at Tamanrasset, and by more than 250 W m-2 on 18 April at Cairo, reflecting the total extinction effect by the aerosol layer. On 14, 16 and 17 April, at Tamanrasset, the predicted dust AOD was very low, lower than the values provided by the Tegen climatology: this was reflected in the slightly higher forecasted DSSF by TOTAL on these days.

At the top of the atmosphere (TOA), outgoing long-wave radiation (OLR) was smaller for TOTAL by 5–10 W m-2 on 12–13 April at Tamanrasset and by up to 20 W m-2 at Cairo. The order of magnitude in OLR difference between REF and TOTAL is in agreement with the results of and . The difference is most important during daytime because, in the absence of clouds, OLR is driven mainly by surface temperature, which was lower for TOTAL as compared to REF. The overall impact of prognostic aerosols was rather small: on average, the OLR at TOA was only 2–3 W m-2 smaller for TOTAL. Short-wave radiation was also smaller at TOA for TOTAL because of increased columnar absorption over the bright desert surface. The difference reached 30–50 W m-2 at Tamanrasset on 12–13 April and more than 150 W m-2 at Cairo on 18 April. On average, short-wave radiation at TOA was 14–16 W m-2 lower for TOTAL. The values for the aerosol forcing in the shortwave and in the longwave are consistent with .

The short-wave forcing at the surface reached -300 W m-2 at the heart of the dust storm (Fig. ), and between -25 and -75 W m-2 at TOA with a minimum of -150 W m-2. These values are in agreement with values found in , , and (the latter study uses the same dust emission scheme as this study) for the surface values. At TOA, this study found mostly negative values above the dust storm during the day, and positive during the night, as opposed to positive values in the cited studies (except for , whose results are consistent with the results presented here). This could be caused by differences in single-scattering albedo and in the surface reflectivity. This could also be due to the fact that after 36 h of forecast time, the dust storm was located above the slightly darker surfaces of central and eastern Egypt; the dust plume was then brighter than the surface underneath, as shown by Fig. . The long-wave forcing was larger for 24 h than for 36 h of forecast time, because the dust load was larger then, as shown by Fig. . It exceeded 50 W m-2 at the surface, where the dust load was highest. At TOA, the forcings were smaller and lay in the 10–30 W m-2 range above the dust storm. These values are close to values found by when using the refractive index, and to values found in , and .

The difference between radiative forcing at the surface and at TOA defines the net atmospheric forcing (); it reached 200 W m-2 in the shortwave during the day, and between -15 and -40 W m-2 in the longwave. This means that the radiative forcing in the shortwave provoked a heating of the atmospheric column above the surface, while the forcing in the longwave provoked a smaller cooling of the atmospheric column. These values are consistent with the case studies of and .

Radiative efficiency is defined as the aerosol radiative effect per unit aerosol optical depth (e.g. ). Figure  shows the radiative efficiency corresponding to the radiative forcings shown by Fig. . They ranged from -100 to -150 W m-2, with minima of -200 W m-2 in the shortwave at the surface, which is close to the results shown in and . At TOA, the short-wave radiative efficiency was -20 to -50 W m-2, with a very localized minimum of around -170 W m-2 N of the Nile Delta over the Mediterranean Sea. These results are close to but contradict those of who find mostly positive values. The very low values over the sea can be explained by the lower surface albedo: and showed a clear negative correlation between surface albedo and aerosol radiative forcing. The difference between our results and the results of for short-wave radiative efficiency at TOA could be explained by a difference in the surface albedo over the desert. Our results are also in broad agreement with the regional averages compiled in , once the long-wave component is also taken into account.

The radiative efficiency of dust in the longwave is much smaller as compared to the shortwave. At surface, it ranges from 10 to 30 W m-2 in the dust storm. Higher values behind the storm, reaching 50 W m-2 in Cyrenaica, are probably artifacts provoked by the low values of AOD there at this time. At TOA, the radiative efficiency lies in the range of 5 to 20 W m-2. The regions with highest AOD are collocated with the largest long-wave radiative forcing (see Fig. ); however the efficiency is lower there. This matches the results shown in Fig. 9 of : the link between AOD and long-wave radiative forcing at the surface is not a linear one but rather a logarithmic one (for surface). At TOA, the relationship between AOD and long-wave radiative forcing is weak.

In this section, the impact of the solar aerosol–radiation interaction on meteorological parameters and dust production is investigated. Figure  shows observed (when available) and forecasted meteorological parameters, dust production flux and 550 nm AOD at Cairo and at the Siwa Oasis, which lies at 29∘12′ N, 25∘29′ E, for the REF and the SW experiments. The latter location was chosen because it was affected by the dust storm from the morning of 17 April to the afternoon of 18 April, whereas Cairo was mainly affected around midday on 18 April.

The strong radiative forcing in the shortwave (see Fig. ) had a notable influence on maximum temperatures which are up to 3∘ lower for the SW experiment on 18 April at Cairo, and 2–3∘ lower on 17 April at Siwa. This increased a small negative bias, from -0.1 K for REF to -0.4 K for SW at Cairo, and from -0.8 to -1 K at Siwa. As the surface is less hot during the day with SW, the sensible heat flux also decreased by up to 150 W m-2 on 18 April 2012 at Cairo and up to 50 W m-2 at Siwa on 17 April. The impact was smaller at Siwa because of the timing of the dust storm which occurred during the night of 17–18 April.

Lower maximum temperatures and sensible heat flux increased the stability of boundary layer (BL), similarly to the process described in and . This provoked a decrease of wind speed at 10 m during daytime, by up to 1 to 1.5 m s-1 on 17 April at Siwa and on 18 April at Cairo, with little impact on scores.

Dust production was smaller with SW because of lower wind speed at the surface, by 25 % lower at midday on 18 April at Cairo, and 15–25 % on the second half of 17 April and also at midday on 14 April at Siwa. Dust production was around 30 times larger at Siwa than at Cairo during the storm, because of higher sustained winds so that the absolute impact on dust production was much larger for Siwa: the difference between REF and SW was around 20 times larger at Siwa than at Cairo. This also shows that the dust layer was mainly advected at Cairo, while it was both advected and produced at Siwa.

Lower dust emissions brought an overall decrease of AOD with SW. The impact was rather small at Cairo, which was farther from dust sources than at Siwa; dust AOD was on average 0.03 lower for SW, and up to 0.2–0.3 lower on the morning of 18 April. At Siwa, closer to the main dust emitting regions, dust AOD was lower by 0.05 on average, and by more than 0.5 on the afternoon and evening of 17 April. A bit further to the north-west, closer to the heart of the dust storm, the AOD difference reached nearly 1. Changes in wind speed can be caused by a combination of the following factors.

Synoptic causes: the pressure gradient changes the surface geostrophic wind.

The horizontal gradient of surface pressure affects surface geostrophic wind via the geostrophic wind equation (): Ug(0)=kfρ×∇zp(0). The horizontal gradient of temperature affects the vertical gradient of geostrophic wind via the thermal wind equation (): VT=Ug(1)-Ug(0)=Rfln⁡p0p1k×∇pT¯. VT is the so-called “thermal wind”, the difference between geostrophic wind at altitudes (0) and (1). R is the specific gas constant for air, f is the Coriolis parameter, k is the vertical unit vector, ρ is the air density, (0) denotes the surface and (1) a specific height above. The subscript p on the gradient operator denotes a gradient on a constant pressure surface, and the subscript z a gradient on a constant altitude surface.

The thermal stratification of the boundary layer affects winds at the surface via a modification of the turbulent momentum exchange coefficients, which are calculated in the MACC-II system using the Monin Obukhov similarity theory (more details can be found at https://software.ecmwf.int/wiki/display/IFS/CY40R1+Official+IFS+Documentation). A decrease of the vertical gradient of temperature is associated with a decrease in sensible heat flux at the surface (see Fig. ) and, as described in detail in , by a decrease in surface turbulent heat and momentum exchanges, and thus in lower wind speed at the surface.

To better understand the interaction between dust and meteorology and to help distinguish between the different causes for the changes in wind speeds brought by SW, Fig.  shows the difference between SW and REF for a set of meteorological parameters, for a 36 h forecast starting on 17 April 2012, 00:00 UTC, close to the local solar maximum. The region with lower 2 m temperature was nicely collocated with the region with high AODs (as shown on Fig. ). Temperature at 850 hPa was also generally lower for SW as compared to REF, but by a smaller margin as compared to 2 m temperature: 0.5–1.5 K against 1–3 K at 2 m for regions with AOD above 1. This differential impact of surface and 850 hPa temperature affected the thermal stratification of the planetary boundary layer (PBL) and was one cause for generally lower wind speed at 10 m. A band of higher surface wind speed and dust production lay at the west of Lake Nasser, showing that the modification of the thermal stratification of the atmosphere was not the only phenomenon that impacted winds and dust production.

Figure  also shows the difference between SW and REF for mean sea-level pressure and wind speeds at 925 hPa. Wind speeds at 925 hPa are less influenced by surface properties and should be more representative of the large-scale component of wind speed. They difference between REF and LW was smaller for 925 hPa winds than for surface winds. This suggests that the synoptic factors brought a smaller contribution than other factors to the decrease of surface winds with SW. Surface pressure was everywhere higher with SW, by 0.2 to 1 hPa in general. The distribution of the differences is quite uniform except just in front of the storm; it is well collocated with the area where 2 m temperature is significantly lower with SW. As a consequence, surface geostrophic wind, a good measure of the synoptic component of wind, generally did not differ much between SW and REF except locally at the edge of the storm.

Figures and show that the area where surface wind speed was higher for SW corresponds to an area of important horizontal thermal gradient associated with the cold front that was causing the dust storm. From east to west, 2 m temperature decreased by more than 10 K in no more than a few hundred kilometres along the 22∘ N parallel. The western part of this high gradient area was heavily impacted by the reduced incoming solar radiation: 2 m temperature there was up to 2 K lower with SW. The eastern part lay in front of the dust storm and was not yet affected: the dust load there was not very high and temperatures were reduced by only 0.5 to 1 K. The differential impact of the dust layer on 2 m temperature thus increased the horizontal gradient in this region by more than 1 K.

This is confirmed by Fig. , which presents a cross section of surface pressure, temperature and wind speed along the 22∘ N parallel. The horizontal pressure gradient between 29∘ and 30∘ E was slightly larger for SW than for REF: 2 hPa/100 km against 1.7 hPa/100 km; this brought an increase in surface geostrophic wind of about 1.5 m s-1, which is not enough to explain the increase in 10 m wind speed at the same place, of more than 2 m s-1 at the surface and 3 m s-1 at 925 hPa. Between 29 and 30∘ E, the horizontal thermal gradient was 5 K/100 km for REF, against more than 6 K/100 km for SW. At 925 hPa, it reached 5 K/100 km for REF, against 4 K/100 km for SW. This translates into an increase of geostrophic wind between the surface and 925 hPa of more than 2 m s-1 between REF and SW at this location. Winds at the surface, 925 and 850 hPa (not shown) in this region, are all between 2 and 4 m s-1 stronger with SW. This is in contrast with the general decrease of winds associated with the changes in thermal stratification, which concerns only the surface and not higher regions.

To sum up, the aerosol–radiation interaction in the shortwave is at the origin of two feedbacks between aerosol and meteorology: a negative one that is driven by the differential changes between temperature at the surface and at the top of the PBL, which in turn increases thermal stability of the PBL, and decreases surface winds and dust production. This feedback was documented by and . A local positive feedback occurs at the edge of the dust layer, where during daytime the horizontal temperature gradient was locally increased by the differential impact of the dust layer on surface temperatures. This increase in horizontal gradient in turn locally increased geostrophic wind to 925 hPa and higher, as well as surface wind, and thus dust production. This local feedback may also be the cause of the local increase in wind speed noted at the edge of a large dust plume in Fig. 6b of .

Surface winds are marginally impacted by changes in surface geostrophic winds brought by surface pressure changes. They are widely decreased by changes in the thermal stratification, as already noted by . They are locally increased by changes in geostrophic winds above the surface caused by differences in the horizontal temperature gradient. As Fig.  shows, the overall difference SW–REF for dust AOD was negative, which means that the negative feedback driven by vertical stratification factors is predominant compared to the local positive feedback driven by the “thermal wind”.

In this section, the impact of the thermal aerosol–radiation interaction on meteorological parameters and dust production is investigated. Figure  shows observed (when available) and forecasted meteorological parameters, dust production flux and 550 nm AOD at Cairo and at the Siwa Oasis for the REF and the LW experiments.

The level of 2 m temperature was higher during the nights with the LW experiment, because the dust aerosol layer emits downwards in the longwave and increases downward long-wave radiation (cf. Figs.  and ). With high dust load, the difference reached up to 1.5 K for Cairo, and up to 2 K for Siwa. This helped to reduce a cold bias for night-time temperatures at Siwa, from -0.8 to 0.4 K over the considered period. At Cairo, the daytime temperature on 18 April was also significantly higher with LW, by 0.5 to 1 K. This translated into slightly larger sensible heat fluxes, by about 30 W m-2. At Siwa, night-time cooling on the night of 17/18 April was significantly reduced and even partially reversed. At the beginning of that night, the heat fluxes indicate a cooling surface for REF and a warming one for LW.

The impact on surface wind speed was small at both Cairo and Siwa. Dust production, however, which follows a cubic function of surface wind, was significantly larger with LW, by up to 15 % at midday on 18 April at Cairo, and by up to 20 % in the evening of 17 April at Siwa. AOD was only marginally affected at Cairo, which is farther from the main dust emitting regions; the difference was significant at Siwa, where AOD maximum during the night of 17–18 April was increased by nearly 0.5. Figures  and show that on the night of 17–18 April, the area where AOD was larger by more than 0.3 with LW was extensive and well collocated with the area with high dust load and larger downward long-wave fluxes at the surface.

Figure  shows the difference on 18 April, 00:00 UTC, between LW and REF for a set of meteorological parameters as well as dust production. The level of 2 m temperature was larger for LW over most continental surfaces, by 0.3 to 0.5 K in regions where the dust load was not very important, and by 1 to 2.5 K in regions where dust AOD exceeded 2. This difference was caused by emission in the thermal spectrum by the dust layer, which lies mostly below 850 hPa. Temperatures at 850 hPa were affected in a different measure: over most areas they were slightly lower for LW. Over a band that corresponds to the cold front, 850 hPa temperature was higher by 0.2–0.5 K. This different impact of long-wave aerosol–radiation interaction between surface and 850 hPa affected the stability of the PBL strongly.

A less stable boundary layer with LW was associated with slightly stronger winds at the surface, by 0.3–1 m s-1 over most areas. The area of higher 850 hPa temperatures with LW was associated with weaker surface winds, by 0.5 to 1 m s-1. At 925 hPa, the pattern of wind change was more complex, with a marked dipole pattern, which suggests that the speed of cold front was increased by the LW experiment. This could also explain the dipole in 850 hPa temperatures. Generally higher temperatures at the surface were associated with lower surface pressure, by 0.5 to 1 hPa. As the decrease in pressure happened behind the cold front, in a region where pressure is building up after the front, this resulted in a slight reduction in the pressure gradient and thus surface geostrophic winds (not shown) behind the cold front. The synoptic impact of aerosols–long-wave radiation on wind speed was thus negative for this situation, but it was generally small in intensity, except at the edges of the dust storm, where the pressure gradient was notably increased. This is clear in the differences in wind speed at 925 hPa, which shows a band of positive values that is exactly collocated with the edge of the zone where surface pressure is different between REF and LW.

The widespread increase in surface winds translated into mostly larger dust emissions. The notable exception was the area with lower surface wind speeds that lies just before the cold front. As a consequence, dust AOD was generally larger for LW (see Fig. ); values were however significantly lower just before the cold front.

As done for the SW aerosol forcing, to better understand the phenomena taking place around the cold front, Fig.  presents a cross section of various meteorological parameters along the 22∘ N parallel. The pressure gradient along 22∘ N was modified by LW, the difference in pressure gradient reaching 0.3 hPa/100 km between 22 and 23∘ E. This difference translates into a surface geostrophic wind that is around 1.5 m s-1 lower (Eq. 1) at this location with LW. The level of 2 m temperature gradient was not very different between REF and LW; at 925 hPa, the radiative impact of aerosols in the longwave had a more differentiated effect on temperatures: the temperature gradient between 22 and 23∘ E was smaller by 1 K/100 km with LW. This reduced the “thermal wind”, i.e. the difference between geostrophic wind at the surface and at 925 hPa by around 1.5 m s-1 with LW (Eq. 2). The impact of the decrease of surface geostrophic wind between 22 and 23∘ E is clear on 10 m winds, which were significantly lower there, and on winds at 925 hPa, which were also lower in this region with LW and very similar elsewhere in the cross section.

To sum up, the aerosol–radiation interaction in the longwave is at the origin of two feedbacks between aerosol and meteorology: a positive one that is driven by the differential changes between temperature at the surface and at the top of the PBL, which in turn decreases the thermal stability of the PBL and increases surface winds and dust production. This mechanism is the symmetrical opposite of the one described for the SW experiment and by . A local negative feedback occurs at the edge of the dust layer, where during night-time the horizontal temperature gradient is locally decreased by the differential impact of the dust layer on temperatures at 925 hPa, thus decreasing geostrophic wind above the surface. Dust production and AOD are likewise affected, thus enhancing this negative feedback. In contrast with the SW experiments, the surface geostrophic wind is significantly affected by surface pressure changes with LW. This translates into local increases in winds at 925 hPa and above.

In this section, the reference experiment is compared against TOTAL, which uses prognostic aerosols to compute aerosol–radiation interaction in both SW and LW spectra.

The impact on wind speed was of the same nature as SW and LW for day and night respectively, but reduced in amplitude. As for SW and LW, a positive (respectively negative) feedback developed at the edge of the dust plume, just before the cold front. As these feedbacks are symmetrical between SW and LW, they are strongly reduced when the two are combined. This is why the areas where the change of thermal winds affected winds and dust emission are the same between SW during the day, LW during the night, and TOTAL, but are much reduced in intensity. These structures have less impact on dust emissions and dust AOD than in the SW and LW experiments. The areas where winds were decreased (by 0.2–0.8 m s-1) or increased (by 0.5–1 m s-1) by the changes in thermal stratification of the PBL are however clearly visible; dust emissions reflect the changes in wind speed. The impact of these two conflicting changes appear to be of similar amplitude: total AOD changes were much smaller for TOTAL as compared to SW or LW (see Fig. ).

Vertical profiles of dust, wind speed and temperature before and after the passage of the cold front are shown in Fig. . Dust is mainly confined to the boundary layer, the top of which lay at around 800 hPa at 15:00 UTC and just above 950 hPa at 03:00 UTC. The impact of the meteorology on dust emissions is again clear on this plot: dust mixing ratio was 20 to 50 % lower with TOTAL at 15:00 UTC, and slightly larger at 03:00 UTC. The impact of aerosols on temperatures was evident: light scattering occurred in the dust layer and reduced temperatures by 1–2 K below 925 hPa at 15:00 UTC for TOTAL and by a smaller amount, less than 0.5 K, between 800 and 925 hPa. At 03:00 UTC, thermal radiation from the dust layer provoked a small increase in temperature for TOTAL very close to the surface, below 970 hPa. Above that height, the atmosphere was cooler with TOTAL, because the dust layer absorbed part of the radiation from the surface. Winds for TOTAL were slightly weaker at 15:00 UTC at the surface, and slightly stronger at 03:00 UTC. Above the surface, winds were mostly stronger at 15:00 UTC, by up to nearly 1 m s-1 at 750 hPa. At 925 hPa with TOTAL, there was a small temperature inversion because the aerosol layer cooled the atmosphere below that height. Associated with this small temperature inversion was a significant increase of wind speed, by around 1 m s-1. At 03:00 UTC, winds were stronger just above the top of the PBL, by around 1 m s-1.

Clear-sky nights are generally characterized by very stable PBLs over the desert since the heat capacity of sand is small compared to other soil types. This very stable PBL is in turn at the origin of nocturnal low-level jets (NLLJs). NLLJs in north Africa can be formed by different mechanisms; here the driving mechanism is an inertial oscillation (), which compensates for the low value of surface winds caused by surface friction and by very high PBL stability by a low-level jet that lies below the top of the PBL, with wind speed above the geostrophic wind values. NLLJs are an important driver for dust emission in north Africa (). studied the feedback between dust and NLLJs and found that the intensity of NLLJs was locally enhanced by the radiative impact of a dust layer on the shortwave: a more stable boundary layer leads to a more intense NLLJ. This led to locally stronger winds during daytime, during the moment of the breakdown of NLLJs. On the other hand, higher nocturnal surface temperatures under the layer of dust associated with the long-wave effect of dust decreased significantly the PBL stability, which could have a weakening impact on NLLJs. The winds profiles of Fig.  at night show a strengthening of the NLLJ with TOTAL, which is consistent with the findings of . In this situation and for this phenomenon, the impact of dust on solar radiation was predominant over the impact on long-wave radiation.

To sum up, TOTAL is a composition of LW and SW: the mainly positive feedback between dust and meteorology associated with LW and the mainly negative feedback associated with SW co-exist and also impact each other. This completes the mechanism described in and , who concentrated mainly on the SW radiation–aerosol feedback. The local feedbacks before the cold front, driven by horizontal thermal gradients, neutralized each other and were thus much smaller in amplitude in TOTAL as compared to SW and LW. This shows that the timing of the storm, and whether it is primarily affected by the dust–short-wave or long-wave radiation interaction, is of great importance to understand how the dust layer impacts meteorology and vice-versa. In this case, the dust–radiation interaction had little impact on the synoptic situation, i.e. the motion of the highs and lows as well as the movement and intensity of the cold front that caused the storm. Cycling experiments, with a meteorological analysis that is provided by the NRT MACC-II system, are not the best tool, however, to assess the synoptic impact of dust–radiation interaction. Assimilation runs provide a better insight into this issue.

Figure  shows the differences between the experiments TOTAL_ASSIM and REF_ASSIM for 2 m temperature, 10 m wind speed, dust production and dust AOD at 550 nm for the runs starting at 00:00 UTC on 17 April 2012, 24 and 36 h forecasts. For 2 m temperature, the magnitude of the changes brought by interactive aerosol–radiation interaction was similar in the experiments with and without assimilation. This is true for both the SW and LW dust–meteorology feedbacks. However, the impact on surface winds and dust production was more important with assimilation runs. This different behaviour can be explained by the fact that the timescales involved were different for the surface temperature and for wind speed adjustments to the radiative forcings. The heat capacity of sand is low, which makes the thermal inertia of desert soils small as well; surface temperature adjusts quickly to a change in the radiative fluxes. As a consequence, the fact that the analysis takes dust–radiation interaction from an aerosol climatology or from interactive aerosols into account does not have such a large impact, since in any case surface temperature will adjust quickly to the radiative forcings during the forecast. However, for winds, the adjustment takes more time since the changes are driven by vertical and horizontal temperature gradients, and the changes concern the whole boundary layer (see Fig. ). As a consequence, it appears that taking the interactive dust–radiation into account in the analysis of TOTAL_ASSIM through the first guess (the previous forecast used in the 4D-Var to build the initial conditions) enhanced the feedback between radiation and surface winds as compared to TOTAL. The weight of the first guess in the analysis was amplified by the fact that the dust storm of 17–18 April 2012 occurred in a region where both meteorological and total AOD observations are sparse.

The impact on dust production is clear: the difference between TOTAL_ASSIM and REF_ASSIM was more marked, at 00:00 and at 12:00 UTC, than the difference between TOTAL and REF (not shown). However, the feedbacks associated with dust–short-wave and long-wave radiation interaction, even if they were more intense as compared to TOTAL, appear to neutralize each other: in the dust storm, the difference in AOD seldom exceeded 0.1 between TOTAL_ASSIM and REF_ASSIM. A few hundred kilometres east of the storm, the impact was not negligible, with dust AOD being reduced by 0.2 to 0.3 with TOTAL_ASSIM, at 00:00 and 12:00 UTC.

In this study we highlighted a series of interactions between aerosols and boundary layer meteorology, driven by the short-wave and long-wave radiative forcing of mineral dust. In the shortwave, lower maximum temperatures increased lower atmosphere stability, which brought in turn a decrease in wind speed and in dust production through saltation processes. Locally, at the edge of the dust plume, the short-wave forcing perturbed the horizontal temperature gradient and geostrophic winds associated with a cold front. Local increases in surface pressure gradient also brought local significant increases in surface geostrophic winds. These two processes led to sharp increases of surface wind and of local dust production.

The impact of the dust layer on long-wave radiation brought opposite feedbacks: warmer temperatures at night decreased the stability of the PBL, thus strengthening surface winds and dust emissions. Contrasted heating of the mid-boundary layer at night decreased the horizontal temperature gradient at the edge of the dust plume. Associated with lower pressure gradients, this brought lower geostrophic winds at the surface and higher, which in turn led to local decreases in wind speed and dust production.

The SW radiation–dust interaction was more pronounced than the LW radiation–dust for radiative fluxes and efficiency. For surface temperature and dust production, the two feedbacks were of a comparable amplitude. This highlights how important accurate forecasts of the timing of the storm are, since depending on the local time of the dust lifting episodes, the interactions between aerosol and boundary layer meteorology are of a very different nature.

The dust–boundary layer meteorology feedbacks were amplified in assimilation runs, because they were also taken into account in the initial conditions of both aerosols and temperature. Since the considered region does not have many observations of both temperature and total AOD from MODIS, the first guess had an unusually large relative contribution in the initial conditions. Although the synoptic situation was not much affected by the radiative forcing of the prognostic aerosols, we report a generally positive impact – up to a 48 h lead time on the 2 m temperature and forecasts of surface radiative fluxes.