Recent global dust trend and connections to climate forcing



[1] Dust is an important indicator of climate change. In paleoclimate research, sediments bearing signals of dust deposition offer a rich archive for climate change history. However, the dust-climate link is very complex due to the various direct and indirect feedbacks in the Earth system. In this study, we examine two issues: (1) given the recent global warming, what are the dust variations, both globally and in key dust regions, and (2) what are the climate drivers behind the variations? Using synoptic data for the period 1974–2012, we analyzed the global trend of dust frequency and visibility-derived dust concentrations and their characteristics in key dust regions, including North Africa, the Middle East, Southwest Asia, Northeast Asia, South America, and Australia. We also examined the likely climate drivers for dust variations in the different regions by computing the correlations between the time series of dust and of major climate indices—the Multivariate El Niño/Southern Oscillation Index, North Atlantic Oscillation, and Atlantic Multidecadal Oscillation (AMO). It was found that over the period 1984–2012, the global mean (excluding North America and Europe) near-surface dust concentration decreased at 1.2% yr−1. This decrease is mainly due to reduced dust activities in North Africa, accompanied by reduced activities in Northeast Asia, South America, and South Africa. A significant negative correlation between Saharan dust and AMO was detected, and it seems reasonable to suggest that under present climate, the global dust trend is determined by the climate systems governing the Atlantic and North African regimes.

1 Introduction

[2] The link between dust and climate has attracted much attention of the climate research community. Direct dust-climate feedbacks have been identified [Rotstayn et al., 2011], and dust has been recognized to be a major player in, and sometimes even a driver for, various annual to decadal climate variations [Evan et al., 2006; Foltz and McPhaden, 2008; Evan et al., 2011]. On the planetary scale, the atmospheric general circulation determines the distribution of the arid regions and the global dust pattern [Prospero et al., 2002]. On synoptic scales, dust storms are generated by systems such as cyclones, fronts, squall lines, and monsoon troughs. In a wider role, dust in the climate system participates in a variety of physical, chemical, and biogeological processes that entail feedbacks with far-reaching consequences [Shao et al., 2010]. The importance of dust to the atmospheric radiation transfer, cloud physics [Tao et al., 2012], and atmosphere-ocean CO2 exchange [Martin and Fitzwater, 1988; Jickells et al., 2005; Maher et al., 2010] is widely recognized.

[3] Data show that dust event frequencies vary significantly on seasonal, annual, decadal, and even longer time scales, as reported in Shao and Wang [2003], Kurosaki and Mikami [2005, 2007], and Kurosaki et al. [2011] for Northeast Asia, Goudie and Middleton [2001], Knippertz et al. [2009], and Klose et al. [2010] for North Africa, and numerous studies for the other regions of the world. The interpretations of such findings are not without their difficulties, because dust variability is affected by a range of climatic and environmental factors. For instance, Zhou and Zhang [2003] analyzed the synoptic records of dust storms in northern China during 1954–2002 and found that dust frequency decreased from the early 1970s to the early 2000s. More recently, Kurosaki et al. [2011] compared the dust frequencies between 1990–1999 and 2000–2009 and found that dust frequency in Mongolia, Inner Mongolia, and Northeast China, rather than decreasing, increased by up to 5%, which is very significant because dust frequency in this region is typically 10 ~ 20%, i.e., dust is reported in 10 ~ 20% of the synoptic records. For some regions, it seems reasonable to relate the dust trend to the climate trend. For instance, the dust frequency in Northeast Asia appears to be negatively correlated with the Arctic oscillation (AO), with fewer dust events in the positive AO phase [Gong et al., 2006], while the dust frequency in eastern Australia is negatively correlated with the Southern Oscillation Index, with more dust events in the El Niño years [e.g., Marx et al., 2009].

[4] Past climate states, as evidenced by ice core studies, demonstrate a strong relationship between glacial stages, high dust concentrations, and low CO2 levels in every glacial cycle [e.g., Martínez-Garcia et al., 2009]. A possible explanation of such a relationship is the “Iron Hypothesis” proposed by Martin and Fitzwater [1988], i.e., the CO2 reduction in the atmosphere during glacial stages may be related to increased iron supply from enhanced dust deposition to the ocean. This hypothesis is yet to be fully tested with data, but it is neither convincing nor sufficient to infer global dust and climate change causality from isolated ice core measurements without a solid understanding of the climate drivers behind dust variability. Given the major role of dust in the global climate system, both in the present and the future, and particularly the recent increase in atmospheric CO2 concentration and global warming, it is thus important to understand the following: (1) recent global dust trends and dust variability in key dust source regions and (2) whether these can be linked to likely climate drivers or in turn be seen as a driver of climate events.

[5] The visibility data in the recent synoptic records may provide some answers to the questions, as demonstrated by Mahowald et al. [2007] for the period 1974–2003. Using such data, they derived a visibility index based on the fraction of observations with visibility less than 5 km (known as VIS5) and analyzed the trend of VIS5 against the trends of the dust control parameters (wind speed, precipitation, etc.). In this study, we continue the work of Mahowald et al. [2007] and analyze the recent dust history, using the synoptic data for the period January 1974 to December 2012, focusing on the two above-mentioned questions. We first describe the trends and spatial variability of the dust frequency and visibility-derived dust concentration and then analyze the characteristics of the dust patterns in key dust source regions, including North Africa, the Middle East and Southwest Asia, Northeast Asia, South America, and Australia. We also attempt to identify the mechanisms that underpin the dust-climate link and the possible dust-climate feedbacks in the different regions by examining the correlation between the time series of dust concentrations and major climate indices.

2 Data

[6] The global Met Office Integrated Data Archive System Land and Marine Surface Stations data [UK Meteorological Office, 2013], available from the British Atmospheric Data Centre (BADC), were used in this study for 39 years from 1974 to 2012. Other data sets are available from the BADC, including the UK data set that dates back to 1853 and a nonglobal data set that dates back to 1900. The data set used in the present study is the most homogenized and hence taken as being the most reliable.

[7] The data set consists of meteorological records reported at 6- or 3-hourly intervals in SYNOP (Surface Synoptic Observations) and METAR (Meteorological Aviation Routine Weather Report) codes. The records can be from either manned or automatic weather stations, but only those from the manned SYNOP stations are used. The dust event records in the present weather code (ww) are classified according to visibility into the following categories: (1) “dust in suspension” (ww = 6, widespread dust in suspension, not raised at or near the station at the time of observation; visibility is usually not greater than 10 km), (2) “blowing dust” (ww = 7, raised dust or sand at the time of observation, reducing visibility to 1 to 10 km), (3) “dust storm” (ww = 9 or 30–32, strong winds lift large quantities of dust particles, reducing visibility to between 200 and 1000 m), and (4) “severe dust storm” (ww = 33–35, very strong winds lift large quantities of dust particles, reducing visibility to less than 200 m). Two other types of dust activities are recorded in the ww code but are not related to a visibility classification, i.e., ww = 8 for “well-developed dust devils, but no sand or dust storm” and ww = 98 for “thunderstorm combined with a sand or dust storm.” In addition, dust events are recorded in the past weather code, i.e., pw1 or pw2 = 3 for “sand storm, dust storm, or blowing snow.” For simplicity, we group dust in suspension and blowing dust into the category of weak dust events, dust storm and severe dust storm into the category of strong dust events, and the rest as “other dust events.” (Note that pw1 or pw2 = 3 is counted as a dust event if the reported air temperature is higher than 3°C to exclude the possible blowing snow events). The frequencies of the weak, strong, and other dust events, fw, fs, and fo respectively, are defined as

display math(1)

where Nw is the number of events of a dust weather category (e.g., weak dust events) and Nobs is that of the total synoptic records. For example, if there are eight synoptic records for a given station on a given day, with three of which reporting weak dust, then Nw is increased by 3 and Nobs by 8. In this way of counting, dust event duration is not considered. The frequency of all dust events is then

display math(2)

[8] Stations that report less than once a day on average were excluded from the data analysis.

[9] Included in the data are also visibility records which, subject to the condition that a dust event is also reported, can be converted to dust concentration, c, using the following empirical function [Shao et al., 2003]:

display math(3)

where c is in µg m−3 and Dvis is visibility in kilometers. An assumption in this empirical relationship is that visibility during dust events only depends on dust concentration. The visibility-derived dust concentration is likely to have considerable uncertainty, because factors such as particle size distribution and air moisture are not accounted for. However, this quantity is useful as it combines the dust weather and visibility records to produce a globally consistent measure of dust trend and variability.

[10] Quality checks were applied to the data before they are used for dust frequency and concentration estimates. We encountered three problems with the data, which may have implications for our results. The first problem is that the number of records included in the BADC data set changed with time during the study period. If a station reports less than once a day averaged over a month, then the records of that station are excluded. Also, the duplicated records (which happen to be incorrectly stored) in the data are excluded. Figure 1 shows the number of monthly valid records from the manned stations (i.e., Nobs) for the globe and the different regions (see Table 1). Globally, two obvious step increases in Nobs were found, one in 1983 and the other in 1984. Nobs of January 1983 (~ 122 × 103) is 1.8 times that of December 1982 (~ 67 × 103), and Nobs of March 1984 (~ 569 × 103) is 4.4 times that of February 1984 (~ 129 × 103). In addition to the step changes, Nobs increased with time. For example, Nobs increased from 569 × 103 in March 1984 to 848 × 103 in December 2012. If individual regions are considered, more changes occurred for two main reasons. The first reason is that for the 1970s and early 1980s, only a subset of the operational stations worldwide was included in the BADC data set. In Appendix A, we have shown the included stations for the representative years of 1974, 1983, 1985, and 2012. For 1974, for example, only a small number of stations were included and no data from Northeast Asia were available. By 1984, however, a good coverage of stations was achieved. The second reason is that the number of manned stations in some countries decreased with their replacement by automatic stations. For example, the number of stations in Australia was 759 in 1974 and 634 in 1983 but 311 in 2012. This is also the reason for the step decrease in several regions in 1999.The changes in Nobs can lead to biases in the dust trend estimate. For instance, an artificial decrease in dust trend may result from a biased increase of weather records in dust-free areas. It is beyond our capacity to rectify the problems left in the BADC data set, but as will be shown later, we will provide error assessment whenever appropriate.

Figure 1.

Number of monthly valid records from manned weather stations included in the BADC data set for the entire globe and the different regions as defined in Table 1.

Table 1. Global Subregions for Dust Trend Analysis
RegionLower Left CornerUpper Right Corner
  1. a

    Excluded from the analysis due to possible inconsistent coding of present weather.

  2. b

    Excluded from the analysis due to the relatively small size of the region.

North Africa(30°W, 0°N)(34°E, 40°N)
South Africa(5°E, 35°S)(55°E, 0°N)
Middle East(34°E, 0°N)(65°E, 40°N)
Northwest Asiab(50°E, 35°N)(75°E, 50°N)
Southwest Asia(65°E, 5°N)(95°E, 35°N)
Northeast Asia(70°E, 30°N)(150°E, 60°N)
Australia(100°E, 50°S)(180°E, 0°S)
South America(85°W, 60°S)(30°W, 15°N)
North Americaa(160°W, 10°N)(50°W, 70°N)
Europea(20°W, 35°N)(50°E, 75°N)

[11] The second problem is that a ww code describes the most significant type of weather in the past hour, while visibility corresponds to the condition at the time of recording. As a consequence, two types of error might occur, namely, (Case 1) a dust event did occur in the past hour, but the visibility changed by the time of recording and (Case 2) at the time of recording, visibility was determined by dust, but dust was not the most significant weather in the past hour. For each and every category of the dust events (e.g., dust storm), we have computed the visibility probability distribution function using the BADC data, as shown in Appendix B. For dust in suspension and blowing dust, the range of the reported visibilities is consistent with the ww inbuilt visibility criteria for dust event classifications. For dust storms and severe dust storms, the visibility records were partially quantitatively, and generally qualitatively, consistent with the inbuilt criteria, i.e., strong dust events are more likely to be accompanied by low visibilities and weak events by high visibilities. For example, about 60% of the severe dust storms are related to visibilities smaller than 1 km. Our finding is consistent with the assessment shown in Table B1 for Australia. The remaining quantitative mismatch is expected to have a much smaller effect on the results of our study (i.e., dust concentration trend) as one might first suspect. Case 1 of the mismatch arises because some strong dust events move fast through a weather station or are short lived, but the higher (than criteria) visibility correctly reflects the fact that the (time-averaged) dust concentration at the weather station is lower (than that corresponding to criterion visibility). Case 2 of the mismatch does not often occur and results only in a negligible dust event underestimation. In conclusion, it remains a plausible approach to assume that the visibility-derived dust concentration is a good representation of the reality.

[12] The third problem is that inconsistencies in the present weather coding seem to exist. For example, there is a rapid decrease in the number of strong dust event reports from Europe in 1989. Furthermore, some northern European stations have an unexpected high number of dust reports in comparison to the adjacent stations, which is likely due to error in the data (Centre for Environmental Data Archival and UK Meteorological Office, 2012, personal communication). The same problem seems to exist for some North American stations (e.g., in Mexico), where few stations report much higher dust frequencies than the stations nearby. As Europe and North America have many synoptic stations, but only occasional dust activities, the inclusion of these two regions in our data analysis is unnecessary.

[13] To avoid the problems embedded in the data set, the following measures were adopted:

  1. [14] Unmanned weather stations were excluded.

  2. [15] North American and European records were excluded from the dust trend estimates.

  3. [16] Data for the period 1974–1983 should only be used with caution, due to the poor data coverage. If the data for this period are used, it is then specifically stated in this paper.

[17] The dust record is expected to show variability on various time scales from seasonal to multidecadal. The climate drivers for dust variability on the different scales are likely to be very different. Our main interest is to establish a large-scale overview of dust variability that is useful for studying its link to climate change. Therefore, our work is more focused on dust variability on the time scales from interannual to multidecadal, in which the dust-climate feedbacks and the large-scale atmosphere and ocean and land surface interactions are most significant. To this end, the correlation functions were computed between the time series of dust concentration and major climate indices, including MEI (Multivariate El Niño and Southern Oscillation Index), the NAO (North Atlantic Oscillation), and the AMO (Atlantic Multidecadal Oscillation), available from

[18] In comparison to Mahowald et al. [2007], the data used in this study cover the period 1974–2012, 9 years longer than the data they used. As large uncertainties in the trend analysis occur if the 1974–1983 data are included, our conclusions are mainly based on the 1984–2012 data. We also take the analysis further in using conditional sampling of visibility subject to dust report in the weather records. This sampling technique allows us to distinguish strong dust events from weak dust events and to avoid the contamination of visibility estimates due to other weather events (such as fog, rain, forest fire, and other anthropogenic activities). In addition, a statistical analysis is made to estimate the significance of the dust trend and the correlations.

3 Global Dust Pattern and Trend

[19] Figure 2a shows the global pattern of dust frequency based on the present weather reports ( fw + fs) and Figure 2b on all present and past weather reports (fw + fs + fo). The only significant difference is that dust frequency observed at some Australian stations is notably increased if the past weather records are included and the dust pattern there appears to be more consistent with expectations. Apart from such minor details, Figures 2a and 2b show the same features of the global dust patterns: North Africa, the Middle East, South West Asia, and North East Asia are the regions with high dust frequencies. The dust observed in the Caribbean is transported dust from the Sahara, as known from earlier studies [e.g., Prospero, 1999; Prospero et al., 2002; Liu et al., 2008], while the dust observed in Mexico may be partly related to the dust activities in the Chihuahua Desert. In South America, a region of relatively high dust frequency associated with the Patagonian sources exists in Argentina. In South Africa, a region of relatively high dust frequency associated with the Kalahari exists in the Botswana-South Africa region. Australia, perceived as a major dust source in the Southern Hemisphere, has in fact relatively low dust frequency over much of the continent, with the highest value being around 5–10% to the north to northwest of the Simpson Desert. The dust activities reported in Southern Europe and occasionally Central Europe are most likely due to the transported dust from the Sahara. The relatively high dust frequencies at two stations in Iceland are not expected but are plausible because wind erosion over the eastern parts of the island is active because of the frequent strong winds.

Figure 2.

(a) Global pattern of dust frequency estimated from the synoptic present weather records (fw + fs) for the period of January 1974 to December 2012. The analysis regions given in Table 1 are indicated as boxes. Europe and North America are included in this graph although later not used for the trend analysis. (b) The same as Figure 2a, but dust frequency is estimated from the synoptic present and past weather records (fw + fs + fo).

[20] Figure 3 shows the time series of the global monthly mean dust concentration, cg, which was estimated from visibility using equation (3) and then averaged over the valid stations. The 95% confidence intervals for the monthly means are also shown. For the period 1974–1983, the uncertainty in the cg estimates is large due to the rather small sample size (~ 20,000 per month), while for the period 1984–2012, the uncertainty is smaller due to a much larger sample size (~ 300,000 per month). The order of magnitude of cg is 10–102 µg m−3 (with the mean, standard deviation, and maximum being, respectively, 21.41, 15.30, and 125.45 µg m−3 for 1974–2012 and 17.91, 10.53, and 57.09 µg m−3 for 1984–2012). A decreasing trend of dust concentration was found. If the data from the entire period (1974–2012) are used, we find

display math(4)

where cg is in µg m−3 and x in year from 1974. If only the data of the period 1984–2012 are considered, then

display math(5)

where x is in year from 1984. The significance of the trend was tested by computing the trend-to-noise ratio T/σc with the standard deviation, σc, being estimated from the 12-monthly running mean and by comparing the resulting T/σc to z values [Schönwiese, 2006]. For the period 1974–2012, the decreasing trend is significant with 95% confidence, while for 1984–2012, the trend is significant with 80% confidence. Dust concentrations for weak and strong dust events were also computed, and it was found that weak events contributed 86% to the dust concentration for the period 1974–2012, while strong events contributed 14%. In terms of frequency, 97% of the dust events were weak events and 3%, strong events. Dust frequency also showed a negative trend for the periods 1974–2012 and 1984–2012 (not shown). It was found that the decrease in dust concentration was mainly caused by the decrease in dust event frequency, while the intensity of dust events remained steady.

Figure 3.

Time series of global monthly mean dust concentration and the corresponding 95% confidence interval (in error bars) for the period 1974–2012, together with the linear trends. If the data from the period 1974–2012 are used, then the trend is given by equation (4) (dashed line). If only the data for the period 1984–2012 are considered, then the trend is given by equation (5) (solid line).

4 Regional Dust Characteristics

[21] Our analysis suggests that the global (near surface) dust concentration, cg, has been decreasing at about 0.21–0.50 µg m−3 per year over the past four decades. To examine whether the dust trend shown in Figure 2 is uniformly global or only reflects the behavior of major dust sources, such as the Sahara, we divided the globe into 10 regions (Table 1 and Figure 2). The dust trend for each region was computed separately.

4.1 North Africa

[22] The time series of dust concentration for the period January 1984 to December 2012 for North Africa is shown in Figure 4a. Dust concentration in this region is several times that of the global average, with the monthly mean reaching over 300 µg m−3 in the peak dust seasons. The global dust trend is particularly consistent with the North African dust trend. During the relevant period, North Africa experienced a decreasing dust trend with

display math(6)
Figure 4.

(a) Time series of monthly mean dust concentration and the 95% confidence intervals for the period 1984–2012 for North Africa, together with the linear trend and 12-monthly running mean and (b) Atlantic Multidecadal Oscillation index together with the linear trend and running mean.

[23] We estimated the correlations between the dust variability and the key climate indices relevant for the climate variability of North Africa, including the MEI, NAO, and AMO. The AMO [Delworth and Mann, 2000; Knight et al., 2005] is the main cause of tropical Atlantic sea surface temperature (SST) fluctuations with a period of around 70 years [Goldenberg et al., 2001] and has a strong influence on the tropical Atlantic climate. A positive AMO phase corresponds with enhanced rainfall in the Sahel [Zhang and Delworth, 2006] and above-normal hurricane activity over the Atlantic [Goldenberg et al., 2001]. Wang et al. [2012] used a data set with records extending back to the 1950s and found a multidecadal covariability of North Atlantic SST, African dust, and Sahel rainfall (as well as Atlantic hurricanes). Low North Atlantic SST was accompanied by more African dust and less Sahel rain and high North Atlantic SST by less African dust and more Sahel rain. Our finding supports the claim that a mechanism for the positive feedback between SST, dust, and rainfall on multidecadal time scale exists.

[24] To compute the correlation, a normalized dust concentration was introduced, which is defined as

display math(7)

where c denotes the monthly mean dust concentration, cmax is the maximum of c, and inline image is the average of c over the study period. Figure 4 shows a negative correlation between the dust concentration and the AMO index. While dust concentration has a negative trend, the AMO showed a positive trend over the study period. Both trends are significant with 95% confidence with reference to the 12-monthly running mean. The correlation function is computed for various j values (j is the time lag in month) with

display math(8)

where ai is the AMO index for month i and σcn and σa, respectively, are the standard deviations of cn and a. If the global data for the period 1974–2012 are used, then the dust-AMO correlation is 95% significant (the null hypothesis that the two time series are uncorrelated has a probability of less than 0.05) with a correlation coefficient of −0.27 (±0.1 with 95% confidence). The most negative correlation occurred at j = 3 (i.e., 3 months) with a correlation coefficient of −0.33. No significant correlation between dust and the MEI is found (rj ranges between −0.04 and −0.07), as also reported in Mahowald et al. [2007]. There is no significant correlation between global dust and the NAO with the maximum of rj (in this case r2) being −0.1.

[25] If the North African data for the period 1984–2012 are used, then the dust-AMO correlation is significant with r0 = −0.28 and r1 = −0.30. Again, there is no significant correlation between dust and MEI (rj ranges between −0.01 and 0.01). There is also no significant correlation between dust and NAO with r0 being 0.05. We also computed the correlation between the observed dust concentration at Barbados and NAO index for the period 1964–2012 and found no significant correlation (not shown). We therefore cannot confirm the impact of NAO on the African dust as emphasized by Moulin et al. [1997]. The similarity in the global and North African dust to climate indices correlations is not surprising, because the North African dust explains more than 60% of the global dust variations.

[26] The above result is consistent with the finding of Foltz and McPhaden [2008] and Evan and Mukhopadhyay [2010]. They reported that the period of 1980–2006 was characterized by a significant increase in North Atlantic SST and a transition from a negative AMO phase to a positive AMO phase, accompanied by an increase in Sahel rainfall and a decrease in Saharan dust concentration. Evan and Mukhopadhyay [2010] used modern and historical data from meteorological satellites, in conjunction with a proxy record for atmospheric dust, to extend satellite-retrieved dust optical depth over the northern tropical Atlantic Ocean from 1955 to 2008. Their analysis revealed that during the 1950s and the 2000s, the annual mean northern tropical Atlantic dust optical depth (ADOD) was a minimum and during the early 1980s a maximum. It was suggested that a 50 year cycle to Atlantic dust cover may have existed. A comparison of the ADOD and AMO time series shows that they are negatively correlated. Further, Evan et al. [2012] studied the impact of dust on the tropical North Atlantic upper ocean temperature and circulation during 1955–2008, using an observation-based climatology of dust surface forcing and an ocean general circulation model. They reported that on decadal scales, dust-forced variability of ocean surface and subsurface temperatures are of a magnitude comparable to observed variability and on longer time scales, dust-forced SST anomalies vary in phase with the AMO. The study of Evan et al. [2012] thus also implies that tropical North Atlantic multidecadal variability and African dust variability are related. Using the dust concentration measurements at Barbados for 1965–1998, Prospero and Lamb [2003] found large year-to-year changes in Saharan dust transported across the Atlantic. They pointed out that the most dusty years in Barbados were associated with El Niño—Southern Oscillation (ENSO) events and the intense North African drought in the 1980s coincided well with the period when Barbados experienced high dust concentrations. Prospero [2011] reported that the trend and variability of the Barbados dust concentration are rather complex and are not clearly related to the AMO. Prospero's [2011] study does not contradict our results of a negative correlation between Saharan dust and the AMO. The dust concentration derived from visibility in our study was a proxy of dust emission from North Africa. In contrast, the variability of the Barbados dust is controlled both by dust emissions from North Africa and the flow patterns over the tropical Atlantic [Engelstaedter et al., 2008]. It is therefore not surprising that the Barbados dust variability is too complex to be simply related to a single climate indicator, such as the AMO.

[27] It is certain, based on the findings of the early studies and of our analysis, that the North African dust is an integral part of the Atlantic climate system, but the dust-climate feedback processes require further clarification. Evan et al. [2011] studied the impact of African dust on the ocean-atmosphere variability in the tropical Atlantic. They suggested that the Atlantic Meridional Mode is excited by variability in African dust outbreaks on interannual to decadal timescales and attributed the dust variability to land use change. Booth et al. [2012] even suggested that aerosol (including dust) may be a prime driver of the twentieth century North Atlantic climate variability. It appears plausible to suggest that AMO is a primary driver for the decadal-scale dust variability.

4.2 The Middle East and Southwest Asia

[28] The time series of dust concentrations for the Middle East and Southwest Asia during 1984–2012 are shown in Figure 5. The dust trend for the Middle East is slightly positive,

display math(9)

but the trend did not pass the trend-to-noise test with 95% significance and is less obvious when compared with North Africa. Superimposed on this weak trend is an evident decadal variation. In the 1980s, cme was relatively high but decreased from the 1980s to the early 1990s. Between 1993 and 2000, cme reached a minimum but increased thereafter. In addition to the AMO, the atmospheric and land surface factors affecting dust activities in the Middle East may be subjected to the influences of the Indian Ocean Dipole (IOD). In the 1990s, the Indian Ocean was characterized by two strong positive IOD events, one in 1994 with an IOD index of 2 and the other in 1997 with an IOD index of 3. According to Behera et al. [2005], the IOD has a paramount impact on rainfall in East Africa and possibly also in the Middle East. The successive positive IOD events in the 1990s may have been responsible for the low dust concentration in the East Africa-Middle East region.

Figure 5.

(a) Time series of monthly mean dust concentration and the 95% confidence intervals for the period 1974–2012 for the Middle East, together with the linear trend and 12-monthly running mean. The arrows mark the positive IOD phases in the 1990s, one in 1994 and the other in 1997; (b) The same as Figure 5a, but for Southwest Asia.

[29] The dust concentration time series for Southwest Asia is shown in Figure 5b. No significant dust trend can be detected here. Similar to the Middle East, Southwest Asian dust shows a noticeable decadal variation: cswa was relatively high in the 1980s, generally low in the 1990s, and high in the 2000s. The similarity in the interdecadal dust variability in both the Middle East and Southwest Asia decadal scale suggests the possibility of a common regional driver. Again, the successive positive IOD in the 1990s may be responsible for the low dust activities in Southwest Asia in the 1990s.

4.3 Northeast Asia

[30] The dust concentration time series for Northeast Asia is shown in Figure 6. The yearly averaged time series exhibits a significant negative trend with

display math(10)
Figure 6.

Time series of monthly mean dust concentration and the 95% confidence interval for the period 1974–2012 for Northeast Asia, together with the linear trend and 12-monthly running mean.

[31] On average, Northeast Asia had a relatively low dust concentration in comparison to the Sahara, the Middle East, and even Southwest Asia. This is because the dust season in Northeast Asia (in particular over the Gobi) is confined to the spring months of March to May. The dust storms in this region are mostly generated by fast-moving cyclones, and consequently, the mean dust concentration is relatively low. Southwest Asia had a dust frequency of 1.62 (averaged over 1984–2012), almost 3 times the dust frequency in Northeast Asia, which was 0.61. Dust activities in Northeast Asia both in terms of dust concentration and frequency have been decreasing since the 1980s and at the same time show strong annual changes. For example, 1984, 1988, 2001, and 2002 were years of high dust activities, while the 1990s was a relatively quiet decade. The decreasing dust trend detected in this study is consistent with reports that dust activities have been declining since the late 1970s [e.g., Zhou and Zhang, 2003].

[32] Northeast Asian dust activities are primarily driven by the cyclones affecting the Taklimakan and Gobi Deserts. Thus, in addition to the environmental control factors, such as vegetation cover and soil wetness, the variability of cyclone strength, frequency, and trajectory critically affect the dust activities in Northeast Asia. During the late twentieth century, there was a positive trend in the North Atlantic oscillation (NAO) accompanied by a north-eastward “shift” of the NAO action centers, which may be related to the recent increased greenhouse gas concentrations [Dong et al., 2011]. Gong et al. [2006] proposed a likely relationship between the Arctic oscillation (AO) and the frequency of the Northeast Asian dust events. There has also been a related claim of the weakening of the Siberian High between 1978 and 2001 [Panagiotopoulos et al., 2005], which may have caused a decrease in the dust frequency because the occurrence of cold surges in Northeast Asia is closely related to the strength of the Siberian High [Park et al., 2011]. Dust trends in Northeast Asia are also affected by the rainfall trend in Northwestern China. Records show that in this region, rainfall has increased over the past half century. The variability of Northeast Asian dust is thus controlled by a range of climatic, synoptic, and environmental factors that are yet to be thoroughly investigated.

4.4 Australia and South America

[33] The dust sources in the Southern Hemisphere are small compared with the major dust sources in the Northern Hemisphere. Dust sources from South Africa are negligible, but the dust activities there show a significant trend of csafr = − 0.04x + 1.49 (not shown). The time series of dust concentration for Australia and South America are shown in Figure 7. No significant trends were found. Australia has both low dust frequency and low dust concentration. Years of high dust activity are correlated with El Niño events (e.g., 1994, 2009), while years of low dust activity, with La Niña years (e.g., 1999–2002). The time series of dust concentration in South America is characterized by interannual variations. Also here, dust is related to ENSO events, with higher dust concentration in La Niña years (e.g., 1985–1986; 1995–1996). It appears, due to the influences of ENSO, that dust activities in Australia and South America are in opposite phases.

Figure 7.

(a) Time series of monthly mean dust concentration and the 95% confidence interval for the period 1974–2012 for Australia, together with the linear trend and 12-monthly running mean. Also plotted is the MEI; (b) The same as Figure 7a, but for South America without the MEI.

5 Discussion and Conclusions

[34] The overall results of our analyses are summarized in Table 2. The summary shows that over the 1984–2012 period, the global mean (excluding North America and Europe) near-surface dust concentration was 17.91 µg m−3, with an overall decreasing trend of 1.2% per year (with 80% confidence). The major dust regions are North Africa, the Middle East, and Southwest Asia, followed by Northeast Asia, South America, Australia, and South Africa. The negative global dust trend is significant due to the decreasing dust activities in North Africa and Northeast Asia.

Table 2. Summary of Global and Regional Long-Term Dust Weather Development for the Period 1984–2012


(µg m−3)


(µg m−3)


(µg m−3)

Std (µg m−3)

(Max, Min)


(µg m−3)

T/σc bTrend/Mean
  1. a

    y = ax + b; y in (µg m−3); x in year from the starting year.

  2. b

    significance level; S for significant, NS for not significant.

Global17.9115.772.1410.53 (57.09, 3.64)−0.21x + 20.931.51 (80% S)−1.2%
North Africa73.2466.896.3554.25 (329.83, 11.40)−1.08x + 88.781.76 (90% S)−1.5%
Middle East80.9074.346.5666.72 (401.60, 4.43)0.10x + 79.57−0.08 (NS)+0.1%
SW Asia32.4531.431.0241.85 (245.43, 0.09)0.02x + 32.25−0.06 (NS)−0.06%
NE Asia8.506.322.1710.61 (82.82, 0.05)−0.21x + 11.482.05(95% S)−2.5%
South America3.793.110.683.27 (25.19, 0.12)−0.04x + 4.300.69 (NS)−1.0%
Australia1.250.830.421.93 (23.22, 0.01)0.002x + 1.140.07(NS)+0.15%
South Africa0.910.700.221.04 (8.19, 0.01)−0.04x + 1.492.50(95% S)−4.3%

[35] A statistically significant negative correlation exists between the AMO and global mean dust concentrations. A similar relationship is apparent with Saharan dust emissions. Given the dominance of these regions in the total global dust load, it follows that the global dust trend is determined essentially by the climate systems that govern the climate variability in the Atlantic-North African region. For other regions, identifying major climate drivers is more problematic. For Southwest Asia, which is less affected by the AMO, no significant dust trend was found. The Middle East and Southwest Asia showed a coherent decadal variation: Dust activities were strong in the 1980s, weak in the 1990s, and again strong in the 2000s. The weak dust activities in the 1990s in these two regions appear to be possibly related to the successive positive IOD events. In detail, the decline in Northeast Asian dust events is closely related to the strength, frequency, and trajectory of cyclones and resultant rainfall trends over regions such as northwest China. The mechanisms that affect dust trend and variability in Northeast Asia are complex, but they may, nevertheless, find an overall forcing in AO trends. More apparent is the fact that ENSO is the major driver for Australian dust events, but as Australia is a minor dust source, this makes little difference to the global dust trend. Noteworthy is the fact that the Australian and South American dust trends appear to be opposite in phase, with South American events related to La Niña events, while El Niño states force Australian events.

[36] The results of our analysis show a general decrease in global dust loading over the period 1984–2012, and the global dust trend is largely driven by the dominance of the North African dust source. For this region, a direct dust-climate feedback is apparent. There are also other dust regions, for which direct climate drivers on dust processes are recognizable, but overall, the relationship between the climate drivers and dust responses must be further investigated by other means, e.g., statistical synoptics and remote sensing of land surface conditions. A further drawback of this study is that due to the relatively sparse distribution of the synoptic stations in the desert regions, dust activities in key dust source regions are probably underrepresented. To overcome this problem, an amalgamation of synoptic data and satellite remote-sensing data would be necessary.

Appendix A

Supplementary Information on the BADC Data Set

[37] The data set used in this study was compiled by the BADC over the years. Unfortunately, the data set consists of only a subset of the available synoptic records and is incomplete. As seen from Figure 1, there are several step increases in the amount of the data included. In Figure A1, the distributions of the weather stations included in the BADC data for four selected years are plotted. For 1974, only a small subset of the available data is included and the dust regions, such as North Africa and Northeast Asia, are clearly underrepresented. The station coverage is somewhat improved for 1983 and much improved for 1985. For 2012, the synoptic records quite well covered all parts of the continent, including the dust regions.

Figure A1.

Coverage of synoptic stations which report more than once per day on average over a month included in the BADC data set for 1974, 1983, 1985, and 2012.

Appendix B

Dust Code and Visibility

[38] In this study, dust concentration is estimated from the visibility records accompanying the present dust weather codes using an empirical relationship. However, the present weather codes describe “the most significant type of weather in the past hour,” while the visibility records the conditions at the time of the reports. As a consequence, a mismatch between the visibility records and the visibility criteria used to classify dust weather may occur for the following two cases: (1) dust is the most significant weather in the past hour, but visibility at the time of reporting has changed; (2) dust is not the most significant weather in the past hour, but visibility at the time of report is determined by dust. Do such errors indeed occur and do they significantly affect our conclusions? A summary of the dust weather codes in Australia for 1974–2010 and their correlations with independently measured visibilities is shown in Table B1.

Table B1. Summary of the ww Codes for Dust Weather in Australia for 1974–2010 and Their Correlation With the Independently Measured Visibilitiesa
  1. a

    In bold are the visibility classes the codes should relate to (source: T. O'Loingsigh, personal communication, 2013).

Code (Category)Independent VisibilityMatching Percentage Between Code and VisibilityComment
07, 08> 1 km99.1%Good match
(07: Dust or sand raised locally by wind; 08: dust devils but no dust storm)1 km ~ 200 m0.8%
< 200 m0.1%
09 (dust storm)> 1 km96.1%Poor match
1 km ~ 200 m3.3%
< 200 m0.6%
30–32 (moderate dust storm (DS))> 1 km18.4%At least 80% of these DS codes have DS-like visibility.
1 km ~ 200 m72.8%
< 200 m8.8%
33, 34, 35 (severe DS)> 1 km17.9%Half are matched by < 200 m visibility.
1 km ~ 200 m33.3%
< 200 m48.8%
98 (thunderstorm with dust)> 1 km74%A quarter has DS-like visibilities.
1 km ~ 200 m15.4%
< 200 m10.6%

[39] In our study, ww = 98 is not used for computing dust concentration. Table B1 shows that over all, the ww dust weather codes and the independent visibility measurements match quite well. For weak dust events, the consistency is as high as 99%. Also for dust storms and severe dust storms, the consistency is very good. For example, for severe dust storms, 48.8% are fully matched by the requirement of visibility < 200 m, another 33.3% have the visibility between 200 m and 1 km; only 17.9% have the visibility larger than 1 km. For ww = 09, the match is not as good, but the frequency of this code is low.

[40] We have computed the visibility probability distribution function for each and every dust weather category using the global data set, as shown in Figure B1. Our results are in general agreement with the Australian data shown in Table B1. The strong dust weather events are more likely to be accompanied by lower visibilities and weak events by higher visibilities. For example, about 60% of the severe dust storms events have a visibility smaller than 1 km. We can therefore conclude that while a mismatch between the visibility records and the visibility criteria used to classify dust weather indeed occurs, the visibility records and the visibility criteria are qualitatively consistent. The mismatch is expected to have very little effect on the results of this study. Case 1 of the mismatch arises because some strong dust events move rapidly through an observing weather station or are short lived. But the higher (than criterion) visibility correctly reflects the fact that the (averaged) dust concentration at the weather station is lower (than that corresponding to criterion visibility). Case 2 of the mismatch only results in a slight undersampling of the dust events.

Figure B1.

Cumulative probability of visibility for dust event categories dust in suspension, blowing dust, dust storm, and severe dust storm.


[41] We are grateful to Tadhg O'Loingsigh for providing us with Table B1 and other helpful remarks. We also wish to thank the anonymous referees for their valuable comments which helped to improve the manuscript.