Vertical column densities (VCDs) of NO2 and HCHO can be retrieved from nadir-viewing satellite instruments, by analysing solar backscatter radiances in the UV–VIS (ultraviolet–visible) spectral range. Differential Optical Absorption Spectroscopy DOAS;, which exploits characteristic narrow absorption structures, is typically used for the analysis.

Tropospheric VCDs are usually derived in a multi-step process e.g.. First, total slant column densities (SCDs) are retrieved, i.e. the integrated concentrations along the effective light path, by fitting the measured spectrum with a model taking into account all other absorbers in the atmosphere. Second, tropospheric SCDs are derived by subtracting the stratospheric column (NO2) or a latitude-dependent bias estimated over the Pacific (HCHO). Third, the tropospheric SCDs are then translated to tropospheric VCDs. The conversion of SCDs to VCDs is usually performed by dividing the SCDs by a so-called air mass factor (AMF) . The AMF is derived from radiative transfer simulations taking into account information of ground albedo, aerosols and clouds, the vertical profile of the trace gas and the satellite viewing geometry e.g..

Except for very low tropospheric trace gas amounts, the tropospheric AMF dominates the uncertainty of tropospheric trace gas observations from space, e.g. caused by insufficient knowledge of the trace gas profiles, aerosols, cloud properties and cloud cover . For observations of trace gases in the boundary layer, which are the focus of this study, the main effect of clouds is that they shield the atmosphere below. Thus, in the presence of clouds the retrieved trace gas absorptions (SCDs) are usually decreased compared to clear-sky conditions. During the conversion to VCDs, the AMF compensates for this effect, which, however, might lead to over- or underestimations of the trace gas column densities if the state of the atmosphere is not known precisely. As the transition between days with and without precipitation generally corresponds to a change in cloud cover, cloud effects need to be investigated. Therefore, satellite measurements retrieved under low and high cloud fractions are studied in detail using cloud information, derived from FRESCO+ for GOME-2 and SCIAMACHY and OMCLDO2 for OMI.

SCIAMACHY aboard the ENVISAT satellite was operated from 2002 to 2012. It had a ground pixel size of about 30 km × 60 km (VIS) to 30 km × 120 km (UV). GOME-2 aboard ESA's METOP-A satellite, launched in 2007, has a ground pixel size of about 40 km × 80 km. OMI on NASA's Aura platform, which was launched in 2004 has a ground pixel size of 13 km × 24 km at nadir and increasing pixel sizes to the far ends of the 2600 km wide swath. In our study, the two outermost pixels are screened out to remove the pixels with the largest viewing angles and lowest spatial resolution. The local overpass times for the three satellite instruments at the Equator are about 09:30 a.m. for GOME-2, 10:00 a.m. for SCIAMACHY and 01:30 p.m. LT for OMI.

For NO2, the products GOME-2 TM4-NO2A version 2.3, SCIAMACHY TM4-NO2A version 2.3 and OMI DOMINO version 2.0.1 are used . For HCHO the products GOME-2 version 12, SCIAMACHY version 12 and OMI version 14 are used . Data products are provided freely by the Tropospheric Emission Monitoring Internet Service (TEMIS) via http://www.temis.nl/.

Differences among the trace gas data products from the three satellite instruments are expected due to, among others, the calculation of the AMF, their different ground pixel size, local overpass time, cloud products used, the diurnal cycle of cloud conditions and the covered time period. Furthermore, the diurnal cycle of the instantaneous NOx lifetime and emissions might also cause systematic differences between SCIAMACHY and GOME-2 on the one hand, and OMI on the other.

Uncertainties of tropospheric NO2 VCDs result mainly from uncertainties of the stratospheric correction (about 2×1014 moleccm-2) and tropospheric AMFs (about 35–60 %) .

The estimation of precipitation on a daily global scale is facilitated through the combination of radar, passive microwave, VIS and infrared (IR) sensors aboard low-Earth orbiting as well as geostationary satellites. In this study, three different products are used that employ such a blended precipitation scheme. All three data sets agree in their spatial resolution (0.25∘ × 0.25∘) and provide data in 3-hourly time steps, i.e. 12:00 UTC covering the period 22:30 to 01:30 UTC and so on. They are briefly described below.

The Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) 3B42 Version 7 data set combines observations made by the TRMM satellite with other satellite systems, as well as land surface precipitation gauge analyses when possible. The passive microwave data, which has a strong physical relationship to the hydrometeors that result in surface precipitation, are collected from the Microwave Imager (TMI) on TRMM, Special Sensor Microwave Imager (SSM/I) on Defense Meteorological Satellite Program (DMSP) satellites, Advanced Microwave Scanning Radiometer – Earth Observing System (AMSR-E) on Aqua, and the Advanced Microwave Sounding Unit-B (AMSU-B) on the National Oceanic and Atmospheric Administration (NOAA) satellite series. The IR data, for the TMPA are collected by the international constellation of geosynchronous satellites. Additionally, data from TMI and the precipitation radar (PR) on TRMM are used as a source of calibration. The whole TMPA algorithm is constructed in four steps: (i) microwave precipitation estimates are calibrated and merged. (ii) IR precipitation data are produced using the calibrated microwave results. (iii) Then, the microwave and IR precipitation estimates are combined filling missing data. (iv) Lastly, rain gauge data are incorporated for the final product. For a detailed explanation of the TMPA algorithm see .

A similar approach is used for the CMORPH product CPC MORPHing technique;, which uses passive microwave information from SSM/I, AMSU-B, AMSR-E and TMI. The main difference to TMPA is that data gaps are treated differently by transporting rainfall features via spatial propagation information, which is obtained from geostationary satellite IR data .

PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks; ) assimilates IR precipitation estimates from geosynchronous satellites. These estimates are then calibrated using microwave precipitation from low Earth orbit satellites. It differs from the other above-described precipitation algorithms as its calibration technique involves an adaptive training algorithm that updates the retrieval parameters when microwave observations of precipitation become available .

Inter-comparison studies show good agreement with ground-based precipitation observations for these data products e.g. and references therein, which is, however, variable for different geographic regions, surface types and rain intensities.

In our study, we apply each precipitation product individually to differentiate between days with or without rain fall. From the comparison of the corresponding results we find that the uncertainties and differences among the precipitation data sets have only minor effects on the obtained results (see Appendices , ).

The processes of nitrification and denitrification, which govern sNOx fluxes, are closely related to the soil water content . In situ measurements of soil moisture are sparse and difficult to extrapolate to broad geographic regions due to their highly heterogeneous nature. Combined satellite measurements of soil moisture overcome this issue by providing global coverage on a daily basis. Although the absolute value of soil moisture from merged satellite products has large uncertainties, relative variations triggered by precipitation events, should be evident in the time series.

Here, we use data from the ESA Soil Moisture CCI (Climate Change Initiative) ECV project, which merges level 2 soil moisture data derived from multiple satellite sensor products in order to construct a consistent long-time data set. Among the list of sensors that are included, are the C-band scatterometers on board of the ERS and METOP satellites and the multi-frequency radiometers SMMR, SSM/I, TMI, AMSR-E, and Windsat. The data sources include active (scatterometer) and passive (radiometer) microwave observations acquired preferentially in the low-frequency microwave range.

The data sets described above provide the basic information used in our research study. To evaluate and understand other influences on the retrieved trace gas levels; however, further atmospheric and environmental parameters are considered.

The algorithm described above, is applied to the full spatial extent covered by the TRMM/TMPA precipitation data set (-180 to 180∘ longitude, 50 to -50∘ latitude).

Figure a displays the number of valid OMI observations on 1.25∘ × 1.25∘ grid pixels that fulfil the selection criteria, i.e. 60 days of drought and at least 2 mm of precipitation on Day 0. For most regions in the world enough data points are found for our analysis; exceptions are regions with no pronounced seasonality in rainfall (e.g. tropical rainforests, North America, Europe) and regions where rain occasionally falls during the dry season (Southeast Asia). Our algorithm is not optimized for those regions.

The days prior to the first rain are assumed to represent a background level of NO2. Figure b depicts a background NO2 map from OMI measurements obtained by averaging VCDs of the Day -10 to -2.

To examine variations in trace gas columns due to rain events, the enhancement of NO2 VCDs on Day 0 are considered with respect to the background. Figure c shows the spatial distribution of these absolute differences for OMI. In Fig. d, data points within 2 times the standard deviation, σ, of the background variation in the respective grid cell are screened out. Furthermore, as the uncertainty from spatial representativeness becomes the dominant uncertainty contribution if only few valid satellite pixels per grid cell are available , at least 50 % of possible data are requested per grid pixel in order to be considered in this analysis. The corresponding results for NO2 VCDs observed by GOME-2 and SCIAMACHY are similar to Fig. d, but are more affected by noise due to poorer statistics (see Appendix ).

The most eminent features are the high enhancements of OMI NO2 column densities on Day 0 in the distinct band of the Sahel region around 15∘ N. Single grid pixels in this distinct band exceed absolute enhancements of 1×1015 moleccm-2 over background.

Similar enhancements in the Sahel were also observed by . In the south-western part of Africa as well as over Australia spatially coherent enhancements are also present. Small-scale, local enhancements are found also, e.g. over India (also investigated by ), regions nearby the Caspian Sea, the Middle East or China. An important finding is that there are no clustered reductions in NO2 VCDs on Day 0. Since we do not apply any land–sea mask, oceans serve as control regions for our algorithm: no significant differences in NO2 are found over the vast majority of oceans area. However, over the Mediterranean sea and in proximity to coastal regions over oceans, small-scale enhancements in NO2 VCDs are detectable, which might be related to advection.

The applied algorithm considers all data regardless of the season. Analysing the data based on different periods of the year reveals local enhancements in NO2 VCDs in semi-arid areas matching dry-to-wet season transitions in these geographic regions (Fig. ). In April–May–June, Fig. 4a, the narrow band of the Sahel again is characterized by a mean enhancement of NO2 of ∼ 1×1015 moleccm-2. This time period corresponds to the start of the rainy season in the Sahel after a long dry spell of 3–4 months. In September–October–November, Fig. 4b, the strongest peaks in NO2 VCDs are observed in south-western Africa representing the start of the local wet season.

The Sahel region represents a transition zone between the savannah in the south and the Saharan desert in the north. It is characterized by a strong seasonality in rainfall governed by the north–south movement of the inter-tropical convergence zone (ITCZ). The northward movement of the ITCZ starts in March and the northernmost position is reached, at 15∘ N, in August. In the following four months the Sahel region receives about 90 % of its mean annual precipitation . The subsequent dry season begins in October and ends gradually with the start of the next wet season in April/May/June (AMJ period).

Previous studies e.g. argue that in this distinct geographic band pulsed soil emissions of NOx, which can be detected from space, occur at the beginning of the wet season in spring. Our findings support these previous studies and delineate this narrow band from 10 to 18∘ N and from the west coast of Africa essentially spanning the whole width of the continent, as shown in Fig. a. The pronounced sNOx features in the band of the Sahel during the AMJ period enable a more detailed investigation of pulsed soil emissions.

We restrict our detailed analysis to the central and eastern part of the Sahel region (0–30∘ E, 12–18∘ N), similar to previous studies i.e.. The western part of the Sahel shows a slightly weaker NO2 response to rain pulses, which might be related to different inter-annual variability patterns and seasonal cycles of precipitation regimes .

In order to depict the general behaviour of trace gas responses and to improve the statistics, 4 years (2007–2010) of the AMJ period are averaged for the eastern part of the Sahel. Figure  depicts the evolution of multiple environmental variables around Day 0 averaged over the study region. The precipitation amounts from the three different products, indicated in grey shades in each panel of Fig. , generally agree in their relative variation, showing little rain before Day 0, a heavy rain event on Day 0 and slightly higher precipitation after the first rainfall event compared to Day -10 to Day -1. The discrepancies among TMPA, CMORPH and PERSIANN products indicate that some rain events might be missed or assessed differently by the individual data products. Nevertheless, considering CMORPH or PERSIANN data as trigger leads to comparable responses in trace gases around the first day of rainfall (Appendix ).

The immediate wetting of the dry surface on Day 0 is captured well by satellite observations of the volumetric soil moisture content as seen in Fig. a. After the initial wetting of the soil, the moisture content drops quickly during the following 3 days due to infiltration and evaporation. Similar behaviour is observed for total column densities of water vapour from GOME-2 in Fig. b. Water vapour content in the atmosphere gives insight into the ambient humidity and may indicate impending rain events, as humidity in the atmosphere typically rises prior to precipitation. On Days -10 to -2 water vapour steadily builds up in the atmosphere and peaks 1 day before the initial rain event. On Day 0 and the following 3 days, the water vapour column densities drop, which is probably caused by, on the one hand, the removal of atmospheric water by precipitation and, on the other hand, by transport of dry air masses over the study area. The results for NO2 VCDs using the three satellite instruments (OMI, GOME-2 and SCIAMACHY) show a consistent enhancement around the first day of rainfall (Fig. e). This points to a strong source linked to the dry–wet transition, i.e. the precipitation trigger. However, the magnitude of the enhancement varies for the three instruments. The standard error of the mean (SME) value is indicated for each instrument and is generally smaller for OMI, which has the best statistics.

The average absolute enhancement in the Sahel for GOME-2 NO2 on Day 0 compared to the background levels is about 5×1014 moleccm-2, whereas OMI and SCIAMACHY only observe an absolute enhancement of ∼ 3 and 4×1014 moleccm-2, respectively.

The different magnitudes of the enhancements cannot be solely explained by differences in overpass time or pixel size as GOME-2 and SCIAMACHY are similar in both aspects. Furthermore, higher emissions are expected in the afternoon, i.e. at OMI overpass time, when the temperature is higher. The corresponding SCDs, however, (see Fig. e) indicate that the differences seen in the NO2 VCDs are mainly caused by differences in the AMF calculation for the three data products.

Note that the enhancement is about 5×1014 moleccm-2 on average, while it was shown in the previous section that for single 1.25∘ boxes the absolute enhancements can be as high as ∼ 1×1015 moleccm-2. Smaller grid pixels show enhancements of up to ∼ 4×1015 moleccm-2, and for single events even larger enhancements are found.

Another striking feature, similar to the results for soil moisture, water vapour and precipitation, is the generally higher NO2 VCDs during the 10 days following the first rainfall event compared to the background levels before Day 0. In Sect.  and , this important finding is studied more in detail by analyzing the NO2 levels after Day 0 depending on wind conditions and the precipitation on Day +1 and beyond.

As indicated in the introduction chapter, HCHO emissions from soils were found in several laboratory and field experiments e.g.. In Fig. f we also analysed HCHO VCDs from OMI, SCIAMACHY and GOME-2 for potential pulsed emissions triggered by precipitation. The time series of HCHO for the three instruments, however, show no significant enhancement around the day of the first rain event. Possible reasons are the low signal-to-noise ratio for HCHO observations or very low emission rates.

Soils from different regions and land types are differently affected by precipitation and vary strongly in their microbial composition, nitrogen availability and pH values, which presumably leads to strong differences in emission fluxes from soils. In the following study, the ESA GLOBCOVER land cover classification is used to characterize different land cover types. The data set is scaled down from the initial 300 m resolution to the 0.25∘ × 0.25∘ grid using a most-common-value method. The resulting land cover map is shown in Fig. a.

For different land cover types, both the NO2 response on Day 0 and the background level of NO2 vary systematically, see Fig. 6c. The observed NO2 background VCDs per land cover type in the Sahel are mainly governed by biogenic emissions from soils and biomass burning. Anthropogenic activity and related emissions such as domestic fires or fertilized fields are at a very low level, and originate mostly from the southern, more populated part of the Sahel .

Systematic variations among the different land cover types are captured well: barren land, for example, shows the lowest levels of NO2 compared to all other land cover types. Barren land relates to deserts with very low nitrogen input resulting in low sNOx, even after wetting. This land type is also associated with fewer rainfall events, see Fig. 6b. Mosaic land covers (a mixture between various vegetation types, grassland, cropland or forest), refer in this area to the loose term savannah delineating the transition zone between tropical forests and deserts. Savannahs can comprise various land cover sub-types and are characterized by distinct dry and wet periods with strong vegetation density and productivity during the wet season in summer. It is expected that savannah and cultivated land used for agriculture show strongest responses to initial rain events due to their higher potential for soil emissions. Figure c confirms these hypotheses: the largest NO2 enhancements are found for cropland and savannah; grassland shows a significant, but smaller response; and the driest land cover type (bare area) shows only slightly enhanced NO2 on Day 0.

In this section, we investigate the effects of possible additional sources of NOx such as fire or lightning and systematic errors in the satellite retrieval due to, e.g., changes in cloud fraction. To minimize the influence of these effects, our algorithm excludes measurements where lightning, fires or an effective cloud fraction > 20% are detected.

In the reference case analysis presented in the sections above, NO2 VCDs show an enhancement with respect to the background, i.e. the average NO2 VCDs before Day -1, not only on Day 0, but also on Day +1, Day +2 and Day +3. It is assumed that the background NO2 VCDs are not influenced by the precipitation-triggered sNOx pulsing event and, thus, can be used as reference to derive an absolute enhancement in NO2 VCDs predominantly induced by pulsed soil emissions. However, it is shown in Figs. e and c, e that the NO2 VCDs after the pulsing event (Day +3 to Day +10) are consistently higher compared to the background before Day 0. This could be related to inflow of soil NOx from adjacent pixels, which receive the first rain shortly after. However, initial tests showed that the number of such incidents is quite small (less than 5 %). For this reason, and because the effect is possibly offset by sNOx advection out of the pixel of interest, it is assumed that the effect of inflow is not dominant. As we focus on events with strong emissions (compared to background), a “smearing effect” by advection leads to a consistent underestimation of the peak and subsequent emissions. Thus, we conclude that inflow cannot explain the enhancement for 14 days after the first rain event. Furthermore, it remains unclear to what extent the enhanced NO2 VCDs are affected by a possible underlying change in the background.

This leads to four interesting questions. (i) Are the slightly enhanced NO2 VCDs after Day 0 related to the pulse on Day 0? (ii) In case the enhanced NO2 VCDs after Day 0 are (partly) caused by other sources, what effect does a background correction have on the retrieved absolute enhancements in NO2 VCDs around Day 0? (iii) Is continuous precipitation the cause for the enhancement after Day 0? (iv) In case the enhanced NO2 VCDs after Day 0 are only related to the pulsed rain event, can we give quantitative estimates on these “continuous” soil emissions?

Figure a depicts the reference case analysis for the Sahel region with respect to 120 days around the first day of rainfall after the drought period. The NO2 VCDs observed by the three satellite instruments show consistent patterns in the spike around Day 0 and still slightly enhanced NO2 VCDs 60 days after Day 0.

To investigate whether the increased NO2 VCDs after Day 0 are related to the pulsed rain on Day 0 or caused by a general change of the background NO2 VCDs (e.g. related to a seasonal variation), we try to estimate the temporal evolution of the background (not affected by a pulsed rain event) in the following.

In a first attempt NO2 VCDs are averaged over all grid pixels located at the same latitude as the identified rain events assuming latitudinal homogeneity in NO2 background VCDs. This assumption is justified in the Sahel due to the latitudinal distribution of its land cover types governing NO2 VCDs (the corresponding results are depicted in Fig. b). While compared to Fig. a, a much smoother temporal evolution is found (because more data are averaged), but apart from the much smaller spike on Day 0, very similar values can be seen in both panels. This confirms our assumption that the main part of the increase of the background value is not caused by the precipitation on Day 0. However, the spike in NO2 VCDs around Day 0 is still evident because this averaging method still considers the initial pixel with its adjacent neighbours, which are probably affected by either the overall precipitation pattern or spatial aliasing effects during the gridding of the NO2 data products. Thus, in Fig. c a 10 pixel buffer is additionally applied to the algorithm. Screening out such pixels leads to a time series of NO2 without the distinct spike around Day 0.

The time series of NO2 VCDs retrieved using the two latitudinal averaging methods is denoted as the prevailing background and is subsequently subtracted from the reference case analysis (Fig. ). In the first case, without the application of an additional buffer (Fig. d), absolute enhancements of 0.43×1015 moleccm-2 for GOME-2 and SCIAMACHY are found on Day 0. Also a steady increase in NO2 VCDs several days prior to Day 0 is observed. Although the pronounced spike in NO2 VCDs decreases rapidly in the days following Day 0, it lasts several weeks until the VCDs reach a minimum (but stays still slightly higher than on Day -60 to Day -20). A similar behaviour is observed for the case study with buffer screening (Fig. e). Here, absolute enhancements of 0.4×1015 moleccm-2 for OMI and 0.62×1015 moleccm-2 for GOME-2 and SCIAMACHY are found on Day 0.

From these results we conclude that the slightly enhanced NO2 VCDs after Day 0 are related to the precipitation on Day 0 at or close to the considered location. As our focus is on the quantification of the emission pulse triggered by the first rain of the wet season, it still needs to be clarified whether the enhanced NO2 VCDs after Day 0 are induced by the initial precipitation on Day 0 or by continuous precipitation during the following days. To answer this question we extract time series with the additional selection criterion of 3, 5, 10 and 20 days of no precipitation following Day 0. Figure a depicts the OMI NO2 VCDs for 0, 3, 5, 10 and 20 days without precipitation after Day 0. The intercomparison of these time series is not straightforward as the background values vary for each case indicating that the time series are captured at different dates throughout the April–May–June period. Longer drought periods after Day 0 are more likely at the very beginning of the wet season (April), whereas more constant rainfall dominates at a later stage, e.g. in June. In this time period background NO2 VCDs generally increase in the Sahel. For this analysis we assume that the impact of the different dates only affects the background and not the enhancement on Day 0 as the selection criteria (precipitation and drought length thresholds) presumably have the largest effect. In order to analyse the enhancement around Day 0 only, we apply the above-described latitudinal background correction to each time series individually, which reduces the influence of the background drastically. Figure b,c show the corresponding background NO2 VCDs derived without and with the aforementioned buffer screening applied. Naturally, the absolute change between the different background time series is smaller compared to Fig. a as they represent averages over all pixels on the same latitude. The changes on Fig. c are even smaller because the pixels in proximity to the triggered rainy pixel are excluded from the averaging. Finally, Fig. d and e depict the differences between panel a and panels b and c, respectively. As expected, the enhancements in NO2 VCDs around Day 0, are more pronounced for cases including a 10 pixel buffer during the background correction. Although the absolute enhancement on Day 0 is almost identical for the different cases, the time series still differ in the observed background NO2 VCDs. These systematic differences, seen in Fig. d and e, hint at more complex variations in the background, which are not entirely resolved by our correction. They could also be related to the fact that for the cases with longer dry periods after Day 0 the probability of precipitation in the vicinity of the considered location is lower. This implies, again, also differences in space and time of the observations, which influences the observed NO2 VCDs.

Generally, we find that the NO2 VCDs stay enhanced after Day 0 for a period of about 2 weeks, almost independent from the duration of the dry period after Day 0. This gives rise to the assumption that the enhanced emissions are mostly caused by the initial rain event on Day 0. Here, only results for OMI are presented because it has best statistics. Nevertheless, the analysis with data from the other instruments leads to the same conclusion.

As could be seen in the global analysis in Fig. , large-scale hot-spots in NO2 VCD enhancements are not only detectable in the Sahel, but also in south-western Africa and in Australia. Subsequently, we present also the average NO2 VCDs around the first day of rainfall for all three satellite instruments in these two regions.

Figure a depicts the results for south-western Africa (17–23∘ E, 22–28∘ S) for a drought period (precipitation < 2 mm) of at least 60 days in the months September–October–November 2007–2010 representing the transition phase between the dry summer and following wet season. Compared to the Sahel reference case, the evolution of NO2 VCDs before and after the first day of rainfall increases and decreases more gradually without having a distinct spike on Day 0. This might be due to different environmental conditions such as soil type or lower statistics because of the much smaller spatial extent. It is, thus, more difficult to estimate the absolute enhancement compared to a defined background level. The difference between highest and lowest NO2 VCDs in the full time series is ∼ 0.5×1015 moleccm-2 for all three instruments.

Figure b shows the analysis results for NO2 VCDs for the central part of Australia (120–145∘ E, 22–31∘ S) for the time series from 2007 to 2010. Since the seasonality in rainfall in this region is less pronounced, the full time series is considered in the analysis. The well pronounced spikes show an absolute enhancement of ∼ 0.3×1015 moleccm-2 for the three instruments, which is comparable to the findings from the Sahel and south-western Africa.

Soil emissions of trace gases are not only limited to the specific days we investigated in this study (based on the selection criteria for the temporal evolution of precipitation), but can play an important role in the atmosphere during specific seasons and throughout the whole year . Global chemistry models considered the mean seasonal behaviour of soil emissions in the past, but they were insensitive to rapid changes on a daily basis, e.g. during the onset of the wet season. Pulsed emissions of sNOx have been recognized to be a significant short-term local enhancement, which can be parametrized in GCMs as shown by for the GEOS-CHEM model. The latter study investigates pulsed soil emissions of NOx in the Sahel region and finds a comparable magnitude (49 % relative increase of OMI NO2 VCDs on Day 0) and length of the pulsing event (1–2 days) following the first rainfall. Here, we provide further evidence supporting that and other reported studies, i.e. the observed enhancements in NO2 VCDs are consistent for multiple instruments and are not introduced by retrieval errors or interfering sources.

For the Sahel region we find significant mean enhancements on the first day of rainfall in April–May–June averaged over four seasons (2007–2010) of ∼ 4×1014 moleccm-2, as observed by OMI, after the prolonged dry spell. However, we see much stronger enhancements for single pixels on the original fine resolution grid (0.25∘) of up to ∼ 4×1015 moleccm-2. Considering these values as upper and lower limits for the sNOx enhancement on Day 0, we can estimate emission fluxes. The top-down emission flux for NOx can be inferred from the NO2 VCD by mass balance: E=ΩNO2/τNO2 with ΩNO2, being the tropospheric NO2 VCD and τNO2 the lifetime. The lifetime for τNO2 is mostly determined by the oxidation of NOx to HNO3 in the boundary layer and typically ranges between 4 and 10 h in the tropics . Consequently, we find that the emission flux of nitrogen (N) for pulsing events, considering an assumed NO2 life time of 4 h, in the Sahel is between 6 ngNm-2s-1 for a conservative estimate and 65 ngNm-2s-1 on the upper limit on Day 0. This is in line with findings from , who gave an estimate of 20 ngNm-2s-1 for rain-induced sNOx pulses in June in the Sahel. Furthermore, field studies suggest emission fluxes of nitrogen of about 2–60 ngNm-2s-1 , which covers our estimated upper and lower limits. Previous studies also note the strong dependence of soil emissions on temperature. We conducted initial tests using ECMWF soil temperature data (see Appendix ), but found no clear temperature dependence of sNOx emissions. This is probably due to the systematic relation between temperature and the seasonal cycle, which affects the spatio-temporal selection of the data, e.g. in April more southern pixels are selected and in June more northern pixels. In consequence, we indicate the need for more detailed investigation on how these pulsed emissions are affected by different soil temperatures.

After the first direct emission pulse, emissions remain at an enhanced level for a period of about 2 weeks as described in the previous chapter with approximately 3.3 ngNm-2s-1 averaged over the Sahel. Analogous to the pulsed emission on Day 0, this value probably represents a lower limit because of the rather low spatial resolution of the extracted time series. These emissions are almost independent from the period of dry days after Day 0, which indicates that they are potentially caused by the initial rain on Day 0 and not by subsequent precipitation the following days. Possible advection effects into or out of the considered pixels may thereby raise or lower the retrieved sNOx fluxes. The analysis based on different dry phases after Day 0 also changes the probability for first rain events in pixels in close proximity. As we do not find strong differences in the emissions for different dry phases after Day 0, we conclude that the inflow of NOx from adjacent pixels is not the dominant source for the enhancement after Day 0. In contrast, a systematic underestimation of the emissions is likely due to advection out of the central pixel. In sum, we estimate that the integrated emissions after Day 0 are potentially of the same order of magnitude as those from the first emission pulse on Day 0.

Peak emissions of sNOx pulses typically occur on the scale of 1–3 days in accordance to our results for the Sahel, South Africa and Australia showing peak emissions shortly after the first re-wetting. Some studies, on the other hand, measure peak emissions several days after the first re-wetting of the soil, e.g. 7 days as observed in field by . Our algorithm does not specifically distinguish between such cases by taking average time series after the first precipitation event. Single pixels within the regions we investigated may exhibit peak emissions several days after the initial precipitation, which would, however, not be resolved by our analysis.

Our study focuses on the quantification of pulsed soil emissions and determines the NO2 enhancement on Day 0 and the following days with respect to a sophistically determined background. However, the seasonal pattern of the determined background, i.e. the NO2 enhancement at the onset of raining season compare, clearly indicates that it is mainly driven by microbial emissions from soils as well: from pulsed emissions discarded by our strict selection criteria or continuous emissions during wet season. Note that the seasonal pattern of NO2 over the Sahel, as shown in Fig. , can neither be explained by biomass burning nor lightning. Figure c of the manuscript shows that the background NO2 VCDs are about 0.9×1015 molec cm-2. This is about 0.17×1015 molec cm-2 higher than background in winter. Thus, in addition to the pulsed emissions quantified above, a mean background of 0.17×1015 molec cm-2 can be attributed to soil emissions as well. These estimates are based on TMPA precipitation data. For other precipitation products (CMORPH or PERSIANN), results change only slightly (see Appendix ).

In summary, we discriminate between soil emissions within: (a) 1–3 days (initial peak), (b) 14 days and (c) several months (background during the wet season). The separate quantification of soil emissions belonging to these three categories might also be adopted in model parametrizations of soil emissions. However, further research needs to be conducted on how these emission categories vary for different regions worldwide.

In this section we quantify the total soil emissions, both due to pulsed emissions and background, for the Sahel region. For the pulsed emissions on Day 0 (category a) and the following 2 weeks (b), the fluxes estimated above are multiplied by the area of the investigated region (0–30∘ E, 12–18∘ N). The statistics of our analysis in the Sahel suggest that on average one large pulsing event (after 60 days of drought) occurs within a single pixel in the April–May–June period. Scaling up the Day 0 emissions results in 1.2 and 12 GgN, considering the lower and upper flux estimates estimated above. Analogously, the emissions over the following 2 week period add up to 8.8 GgN. Together this sums up to 10.1–20.8 GgN emissions due to pulsing. As mentioned above, the observed increase of the background in the AMJ period of 0.17×1015 molec cm-2 is mainly driven by microbial emissions from soils as well. When integrated over the complete April–May–June period, these seasonal soil emissions correspond to 46.4 GgN (again based on a NOx lifetime of 4 h). Consequently, the emissions due to pulsing contribute about 21–44 % additionally to seasonal soil emissions for the Sahel and dominate the local NOx concentrations on the particular days.

determined top-down total soil emissions from GOME-2 measurements of about 400 GgN for North Equatorial Africa (0–18∘ N) in June alone. Our estimated total soil emissions of nitrogen (56.5–67.2 GgN for AMJ) are smaller, but are determined for a smaller region as well, which makes a direct comparison difficult.

For all analysed study regions and individual grid pixels showing significant NO2 spikes on Day 0, we also find enhancements in the NO2 VCDs 1 to 2 days before the first day of rainfall. The phenomenon is especially pronounced for the study region in south-western Africa. The finding of enhanced NO2 VCDs before Day 0 stands in contrast to the general expectation that soil emissions, e.g. of NOx, are only caused by the initial rain event and the subsequent wetting of the soil. However, absolute humidity shows a steady increase several days before the first rain event. Thus, reasons for the early increase of NO2 VCD may be an increase in atmospheric moisture content and dewfall, a misclassification of rainfall intensity by the precipitation algorithms, transport of NO2 from neighbouring regions or spatial aliasing effects during the gridding of the satellite observations, i.e. the overlap of the ground footprint onto multiple grid boxes of the precipitation products. The latter two can occur if the ground pixel observed by the satellite overlaps with two or more grid pixels of the precipitation products or vice versa. This error is difficult to estimate, but should be more pronounced for larger satellite ground pixels, i.e. from SCIAMACHY, and less for instruments with smaller footprints, i.e. OMI. The fact that all instruments observe similar enhancements already on Day -1 indicates that this possible error source is not the dominant cause for the early increase in NO2 VCDs. The use of three instruments for detecting NO2, each having different overpass times during the day, also makes it less likely that a temporal mismatch of the precipitation and trace gas products, as described in the “Methodology” section, leads to the enhancements on Day -1.

In Fig. b it is shown that water vapour (H2O) VCDs in the atmosphere retrieved by GOME-2 increases continuously for 10 days before the first precipitation event and peaks on Day -1. We speculate that the moist air over the extremely dry top soil layer induces initial sNOx emissions, despite the fact that the soil is not directly wetted by rain. Also an enhanced dew formation and water adsorption potential, which are both important sources of water in semi-arid areas, have a major effect on microbiological activity . The probability for nightly condensation over drying soils increases with higher absolute humidity. Although observations of these quantities are sparse, measurements in Israel, Jordan and in South Africa hint at contributions of 12 to 40 mm water per year in semi-arid regions and references therein. Transport of polluted air from the tropics northwards presumably also leads to enhanced NO2 VCDs before Day 0. However, this effect is expected to be quite low as the enhancements before Day 0 are also seen for cases dominated by northerly winds coming from the Sahara, which are generally associated with lower NO2 VCDs (see Fig. ).