A global data set of vertical profiles of dust concentrations was obtained from IASI cloud-free measurements using the Mineral Aerosol Profiling from Infrared Radiances (MAPIR) algorithm. MAPIR version 4.1, which was used in this work, is fully described in ; therefore, we only provide a short summary of the main scientific features of the algorithm here. MAPIR operates in three retrieval windows in the TIR spectral window: 905 to 927 cm-1, 1098 to 1123 cm-1 and 1202 to 1204 cm-1. All radiative transfer calculations are done with the Radiative Transfer for TOVS (RTTOV; version 12) and are performed in-line for each retrieval. Dust aerosols are parameterized as spheres with a log-normal size distribution with a median radius of 0.6 µm and a geometric standard deviation of 2, corresponding to an effective radius of 2 µm. The dust aerosol refractive index (RI) is taken from the Gestion et Etudes des Informations Spectroscopiques Atmosphériques GEISA; and the HIgh-resolution TRANsmission database HITRAN; dust-like RI based on measurements by and .

The MAPIR retrieval is based on the optimal estimation method OEM; and iteratively adjusts a state vector, composed of eight variables: the surface temperature (∘) and the dust aerosol concentration at the centre of seven 1 km thick layers from an altitude of 0.5 to 6.5 km. The a priori dust aerosol vertical profiles are derived from the LIdar climatology of Vertical Aerosol Structure (LIVAS) monthly 1∘×1∘ climatology, obtained from CALIOP data from 2007 to 2014.

After the retrieval of mineral dust, a general quality control on the retrieval results tests the goodness of the fit: the root mean square of the spectral residuals (between the measured spectrum and the modelled spectrum after the retrieval) must be lower than 1 K. In addition, the total AOD at 10 µm should be 5 at most, and the retrieved surface temperature must be between 200 and 350 K. The two latter tests are designed to reject cloudy spectra that were not detected by the EUMETSAT IASI cloud product.

Meteosat Second Generation (MSG) Spinning Enhanced Visible and InfraRed Imager (SEVIRI) measurements at infrared wavelengths centred around 8.7, 10.8 and 12.0 µm were used to infer multi-annual information on dust source activity . As described by , 15 min dust red–green–blue (RGB) images were analysed manually for identification and individual backtracking of dust plumes to their point of origin, which then was recorded as the initial dust source. By applying this visual backtracking method, an hourly (2006: 3-hourly) data set on dust source activation frequencies (DSAFs) was compiled covering the 4-year period from March 2006 to February 2010.

The 10 m wind speed fields used to test whether local atmospheric conditions fostering dust emission were fulfilled were obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis data, through the Climate Data Store (https://cds.climate.copernicus.eu/, last access: 3 December 2019). The data are available at an hourly time resolution.

The land cover and vegetation (normalized difference vegetation index – NDVI) data were obtained from the corresponding European Space Agency Climate Change Initiative (ESA CCI) land cover project (http://www.esa-landcover-cci.org, last access: 22 June 2016). The land cover data are available globally at a 300 m spatial resolution and as 5-year averages for three epochs centred on the years 2000, 2005 and 2010. We used the last one as it best covers the time period of the IASI data. The NDVI climatology describes the natural variability of vegetation cover, snow cover and fire scars (burned areas) at a weekly time resolution and, thus, provides further constraints on the soil erodibility potential.

The surface (upper few centimetres) soil moisture data were obtained from the ESA CCI soil moisture project (http://www.esa-soilmoisture-cci.org, last access: 22 October 2018), version 4.2 . This data set was generated using active and/or passive microwave space-borne instruments and covers the 36-year period from 1979 to 2016. For this work, we use the data set from the active instruments (on board ERS-1, ERS-2 and Metop-A) because (1) this ensures that the soil moisture was retrieved from measurements at about the same time as the IASI (Metop-A) overpass time, and (2) among the three data sets, the one from active measurements is the only one expressed in percentage saturation, which is easier to use in this framework. The other two are expressed in volumetric units (m3m-3). Here, information on soil moisture is used to exclude humid areas as potential dust sources.

For the use of the MAPIR data in dust sources studies, two additional selection criteria are applied in addition to the standard retrieval quality check: (1) only the data above land surfaces are considered, and (2) only data with sensitivity to the near-surface layer are considered. The term “near-surface layer” in this paper refers to the MAPIR 1 km retrieval layer closest to the surface. This additional filtering on near-surface sensitivity is the second step in the schematic shown in Fig. .

The sensitivity of the retrieval to any layer can be obtained from the averaging kernel (AK) calculated during the OEM retrieval. The AK is a square matrix of the size of the state vector, describing the sensitivity of the retrieval to each parameter of that state vector (diagonal elements) and the cross-sensitivity between the different state vector parameters (non-diagonal elements). In this study, the need for sensitivity of the retrieval to the near-surface layer is assessed by using the diagonal value of the AK corresponding to that layer (further referred to as AKsurface). The ideal situation would be AKsurface = 1, meaning that the retrieved concentration in that layer is fully independent of the a priori value. However, in reality, this is very rare, and the retrieved concentration depends partly on the a priori and the other layers.

The trace of the AK related to the concentrations, here the sum of seven values, represents the number of independent pieces of information one can obtain from the retrieval. It is about 2 at best for IASI TIR retrievals . Therefore, we have set a minimum threshold at 0.25 for AKsurface. Below this threshold, the sensitivity of the TIR retrieval to the presence of dust aerosols in the near-surface layer is too low, and the retrieval, although generally good, does not contain information about that specific layer. These data are excluded from the source analysis.

Figure shows the monthly average fraction of IASI retrievals per day passing both the basic quality control (step 1) and specific sensitivity check (step 2) for the entire 9-year period considered, from 2008 to 2016. In these averages, each day is given the same weight. White spaces in Fig.  indicate that no retrieval passes the two filters during the month, showing, for example, no retrieval sensitivity to the near-surface layer in the central Sahel and south of the Sahel during the summer. This is due to the very low dust AOD in the area in question during that period, with an especially low dust concentration in the near-surface layer, leading to lower sensitivity in that layer of the retrieval performed in the logarithmic space, as explained in .

As explained in Sect. , the MAPIR retrieved dust concentration in the near-surface layer depends on the real concentration in that layer but also partly on the a priori assumption. Therefore, the retrieved concentration in a single layer should not be used quantitatively but still gives evidence for the presence of dust in that layer. To assure robustness, a threshold for the MAPIR near-surface layer dust concentration can be defined, above which it is considered reliable. This is step 3 in the method schematic in Fig. . Here, this threshold was set to a concentration of 50 particles cm-3, or a 10 µm AOD of 0.2 (equivalent visible 550 nm AOD of 0.35) in the near-surface layer. For comparison, used a visible total column AOD threshold of 0.5 and used a visible total column dust AOD threshold of 0.2 to identify dust sources from MODIS AOD estimates. Our selected near-surface layer concentration threshold is a stronger constraint than the above-mentioned total column thresholds, but it ensures that these are real events.

Considering that the availability of data for this analysis (steps 1 and 2) is not homogeneously distributed in space and time (as shown in Fig. ), the number of days with near-surface dust detection is divided by the number of days with available data. The result for the years 2008 to 2016 is shown in Fig. . White areas refer to data unavailability due to low near-surface sensitivity, as in Fig. .

The MAPIR 3D dust data obtained after steps 2 and 3 allow for the inference of near-surface dust presence. However, there can be several reasons for the presence of dust near the surface : (1) it was just emitted there (or in the vicinity), (2) it was advected at low altitudes, or (3) it belongs to a descending dust layer due to either gravitational settling or turbulent down mixing. In order to filter for dust emission events, we apply further criteria: we use ancillary data to evaluate if local conditions foster plausible dust emission and, thus, support the link between near-surface layer dust and dust emission events. In particular, we consider surface wind speed, land cover, soil moisture and the vegetation index. All of these parameters help to determine if the observed near-surface layer dust events can plausibly be local emission events, although, obviously, none of the parameters render the analysis completely certain.

Figure shows the monthly all-day results of the combined analysis (corresponding to the final green box in the schematic in Fig. ), highlighting dust source areas for the years from 2008 to 2016. The figure clearly emphasizes the seasonal cycles of dust emissions with the most frequent emissions occurring during the summer (June, July and August) except for in the Bodélé Depression area (red star in Fig. 5), where the dust emission frequency peaks during the winter period (November to March).

The most striking and predominant hot spot for dust emission is the Bodélé Depression area in central Chad (red star in Fig. 5), south of the Tibesti Mountains (brown 7 in Fig. 5). It is active throughout the year with a maximum during the winter and a minimum during the summer, which agrees well with the seasonal cycle reported in the literature . Our analysis further highlights a high fraction of days with plausible dust emission during the winter located west of the Bodélé Depression (red star in Fig. 5) towards the Aïr Mountains in eastern Niger (brown 6 in Fig. 5).

A second major dust emission hot spot illustrated by our combined analysis is situated in the central Sahara, in the area of the Adrar des Ifhogas (brown 4 in Fig. 5) and the Aïr Mountains (brown 6 in Fig. 5). This area is very active during late spring and summer, especially from June to August, while it is quiet during the rest of the year, which is consistent with information from recent literature e.g., even though the precise emission locations vary from one study to another.

Regarding the third dust source, a high fraction of days with plausible dust emission is identified during the summer in central Algeria, illustrating a dust source region located north of the Hoggar Mountains (brown 5 in Fig. 5) and south-east towards the Libya–Niger–Chad border. It can be characterized as a series of individual small hot spots. reported central Algeria to be a low-occurrence natural dust source (based on MODIS data). The area close to the Libya–Niger–Chad border has also been reported to undergo moderate dust source activation by .

A fourth dust source area observed is the central region of eastern Sudan, in the Nile River basin, with a maximum in July and August, which is again in agreement with recent literature . Besides individual hot spot areas, a significant fraction of days with plausible dust emission is observed almost everywhere in the Sahara during late spring and summer. Smaller, less pronounced hot spots occur in north-west Algeria (south of the Atlas Mountains – brown 2 in Fig. 5), in Mauritania and north-west Mali. These areas are also described as dust sources in the literature .

All area highlighted here potentially also include a downwind dust transport component. This component is either linked to the transport between sunrise at about 06:00 LST and the measurement time at about 09:30 LST or to the fact that late-afternoon convection events are detected a bit later (at 21:30 LST). The implicitly included downwind transport component results into a shift between the actual source and the detected plume and, thus, in a downwind offset of the assigned source location.

The Sahel (green box in Fig. 5) is a known dust source region, mostly discussed with regard to human activities in a sensitive desert margin region . showed that dust is mostly present over the Sahel during the winter (December to April) but that it is mainly suspended dust transported from Saharan sources, which accumulates and remains in suspension due to wind convergence over the Sahel on the northern side of the monsoon trough. The whole northern Sahel is highlighted as a dust source in , although with a low DSAF (less than 5 %).

Our method sheds new light on the study of Sahel sources. Indeed, the method presented here intends to isolate local dust emissions from transported dust. In the Sahel, the additional criteria linked to surface conditions are extremely meaningful, allowing one to account for the seasonal cycles of vegetation and humidity in addition to the significant changes in winds. What appears in Fig.  as a hard cut-off between plausible and implausible source areas is mostly the result of the NDVI and soil moisture filters applied.

The three most dominant Sahel dust hot spots identified with our method are located in the northern part of the Sahel in the desert margin region: (1) the Mali–Niger border area, with a maximum in June–July; (2) southern Mauritania, with a maximum in June and July; and (3) Mali east of the Mauritanian border, also peaking in June and July.

Three ground-based stations were deployed in 2006 in the framework of the African Monsoon Multidisciplinary Analysis (AMMA). This set of stations is called the “Sahelian Dust Transect”. The stations are found in Banizoumbou (south-west Niger, blue 3 in Fig. 5), Cinzana (south Mali, blue 2 in Fig. 5) and M'Bour (west Senegal, blue 1 in Fig. 5). Using surface PM10 and wind measurements at these stations, and concluded that the enhanced levels of near-surface dust concentrations at these locations during the dry season is mostly due to dust transport from other sources, although some local emissions may contribute towards the end of the dry season (May to July). showed that emission events are rare at Cinzana and Banizoumbou (on less than 2 % of days) and very short in time (maximum 3 h), with a maximum during April to July. Two of our identified Sahelian dust hot spots are around two of the Sahelian Dust Transect stations: Cinzana in south Mali (blue 2 in Fig. 5) and Banizoumbou in south-west Niger (blue 3 in Fig. 5), where our analysis also highlights possible local emissions in spring and early summer. Quantitatively, our estimated fraction of days with plausible dust emission is much larger than the few percent of dust source activation days reported by the stations.

As mentioned in Sect. 1, dust emissions have a known diurnal cycle and, fortuitously, IASI measurements are obtained at interesting times, close to both emission maxima (morning and late afternoon to night). The difference between the morning and evening fraction of days with plausible dust emission may provide information about the main activation mechanism of each emission area and/or period. Figures and show the monthly fraction of days with plausible dust emission for morning (09:30 LST) and evening (21:30 LST) IASI measurements respectively, over the years from 2008 to 2016. We remind the reader that the selection based on the presence of sufficiently high wind speed is done on the 12 h preceding the IASI measurement in this combined analysis. The comparison of Figs.  and leads to three main conclusions. First, the Bodélé Depression (red star in Fig. 5) appears almost exclusively as a morning source, linking it to the LLJ breakup emission mechanism as reported in . Secondly, the area to the south-west of the Bodélé Depression is almost exclusively observed as an afternoon/evening plausible source and might be low-altitude transport from the Bodélé emissions. Thirdly, most of the other major source areas detected are observed both as morning and afternoon/evening hot spots, although the afternoon/evening occurrences are lower. also showed significantly less dust source activation during the afternoon and evening than during morning.

When analysing the monthly fraction of plausible emission days for each year individually (figures not shown), large inter-annual variability becomes obvious. A robust long-term trend analysis would therefore require time series longer than 9 years. This kind of study will be possible in the future, as the IASI mission is currently planned until the mid 2020s, with a scheduled continuation beyond with the successor IASI-New Generation (IASI-NG). A first glance at the temporal evolution of dust source activity is given in Fig. , which shows 3-year aggregations of our full 9-year analysis: 2008 to 2010, 2011 to 2013 and 2014 to 2016. The three epochs show similar patterns of dust emission hot spots, although there is a significant variability in the number of plausible emission days. A clear decrease in the frequency of occurrence of dust emission is observed for the central Sahara and Sudan during summer. The activity in the Bodélé Depression seems generally stable in the winter and increases slightly during autumn. Additional differences are observed between the different time aggregations, clearly showing that comparison between different dust sources studies should also consider underlying differences linked to the time period that was analysed. Therefore, robust long-term trends would require longer time series given the high inter-annual variability.

There are common source patterns in both analyses: the Bodélé Depression (red star in Fig. 5), the area between the Hoggar (brown 5 in Fig. 5) and Tibesti mountains (brown 7 in Fig. 5), west of the Aïr Mountains (brown 6 in Fig. 5), east of the Adrar des Ifhogas (brown 4 in Fig. 5), around the Tademaït Plateau (brown 3 in Fig. 5), Jabal Akhdar (brown 9 in Fig. 5) and Sudan. Differences are also observed: the northern Mali and Mauritania area and the Saharan Atlas Mountains (brown 2 in Fig. 5) do not stand out from the IASI automated combined analysis. On the other hand, central Libya and the West African coastal sources do not stand out from the SEVIRI manual tracking analysis. Overall, the IASI automated combined analysis identifies several large areas as “diffuse semi-hot spots”.

Both analyses agree regarding the most obvious features of the diurnal and seasonal cycles (see Figs. and ): large intense hot spots during the summer (comprising all highlighted areas in the previous section), the Bodélé Depression (red star in Fig. 5) being active throughout the year but only in the morning, and an intense summer afternoon/evening source area west of the Aïr (brown 6 in Fig. 5) and east of the Adrar des Ifhogas (brown 4 in Fig. 5). In this seasonal analysis, the sources in northern Mali and Mauritania stand out in the IASI automated combined analysis during the summer as well as in the SEVIRI manual tracking analysis (in contrast to the results from the all-year analysis in the previous section). The other differences mentioned in the previous section also appear here. Additionally, Sudan really stands out from the IASI automated combined analysis during the summer afternoons/evenings, whereas it is mostly inactive in the SEVIRI manual tracking analysis.

Most of the differences observed in the two previous subsections can be explained as a consequence of the differences between the source identification methods.

A non-negligible part of the differences, where larger source areas are identified from the IASI automated combined analysis (the most striking example is the Sahel area during the summer), originate from the fact that this method counts big events more than once, identifying each pixel matching plausible emission conditions as a potential source. This most probably leads to an overestimation of the emission occurrences, as the low-altitude transport after emission is also recorded as a distinct plausible emission event if the local conditions are met. On the other hand, the SEVIRI manual tracking identifies only one pixel and time stamp as the source for each event and probably underestimates the continued emissions along the low-altitude plume transport if the wind remains strong enough. The fact that the Sudan source area stands out in the evening IASI automated combined analysis is the result of late morning strong winds (after the IASI morning overpass); as a consequence, these events are recorded as evening events even thought they occur between 09:30 and 12:00 LST. This is supported by two pieces of evidence. First, if the time limit for the existence of high wind speeds is set to 6 h prior to the overpass time instead of 12 h in the IASI automated combined analysis, these events are no longer observed; therefore, they are linked to late morning strong winds. Second, local emissions are observed in Sudan between 09:00 and 12:00 LST in the SEVIRI manual tracking.

The more scattered appearance of the IASI automated combined analysis results is probably linked to the fact that more small plumes are identified in an automated analysis.

The absence of the West African coastal sources from the SEVIRI manual tracking analysis can be linked to the cooler and humid inflow from the ocean, rendering the SEVIRI RGB images much less sensitive to the presence of dust .

The Saharan Atlas is a place where IASI has very little sensitivity to the near-surface layer; therefore, most retrieval results are rejected from the source analysis.

Finally, we do not have an explanation for the central Libyan area only being highlighted in the IASI automated combined analysis; this specific spot is not a well-known source area. As the area is identified constantly over time (see Fig. ), it is either an artefact linked to the local surface properties (infrared emissivity) being misrepresented in the ancillary data used in the MAPIR retrieval, or it is a newly identified dust emission source.