Remote sensing of dust over land based on visible wavelengths is hampered by the fact that land surfaces are highly variable in nature and are characterised by a wide range of albedos. Large geographical variations in the reflective properties of land areas occur due to variations in geology, climate and land use. The reflective properties of land also vary seasonally with vegetation cover, and may vary on longer timescales due to environmental changes or changes in the patterns of human activity. For these reasons measurements of dust levels over land have to date relied upon relatively sparsely distributed ground-based measurements of visibility, aerosol optical depth or particle concentration. Any large-scale studies of interactions between mineral dust and climate require data which are relatively contiuous in space and time.
Firslty, the dust layer will reduce incoming solar radiation, cooling the land surface and thus reducing the outgoing longwave radiation (OLR) from the land. Secondly, if the dust particles are of a size comparable to the wavelengths of the measured IR radiation, some of the IR radiation emitted by the land will be absorbed by the dust layer, attenuating the OLR signal. Part of the IR radiation absorbed by the dust layer will be converted to kinetic energy as the layer is heated, and part will be re-emitted. Radiation will be re-emitted in all directions, with only a fraction continuing upwards towards the top of the atmosphere (TOA). IR radiation re-emitted downwards may cause a warming of the atmosphere under the dust layer, particularly at night. During the day, this warming will be offset by the reduced OLR from the ground surface as a result of reduced solar insolation. This effect will also be apparent as the ground cools after sunset. Any warming under a dust layer should therefore be most pronounced in the early morning before sunrise, when the ground has lost most of the previous day's heat. By contrast, warming of the atmosphere at the altitudes of dust transport will be present throughout the day due to absorption and re-radiation in the IR within the dust layer (Alpert et al., 1998). Heating of dust-laden air is likely to be important for generating and sustaining the dry convection which gives rise to certain types of sandstorm (Chen et al., 1994). When dust is transported aloft, the levels at which the dust exists will be generally cooler than the ground surface. The component of the measured IR signal originating from the dust layer will therefore represent a lower temperature than that characterising the land surface. The higher the altitude of the dust layer, the lower this temperature will be (neglecting the effect of temperature inversions).
The temperature depression (with respect to clear-sky IR radiance) resulting from the presence of dust in the atmosphere, and caused by the processes described above, may be used as an index of dustiness (Tanré and Legrand, 1991). Modelling studies suggest that a dust layer between the surface and the 500 hPa level, with an optical depth ranging from 0.5 to 3 (superimposed on a desert aerosol background of 0.2) results in a measured IR temperature depression of 10-30 K (Carlson and Benjamin, 1980). As yet unpublished modelling studies (Legrand 1999, personal communication) have also addressed the impact of dust on the temperature and IR radiance at the TOA. These indicate that, for a surface albedo of 0.25 and an aerosol optical depth in the visible range of d = 1, both coarse (> 1 m m) and fine (< 1 m m) dust cause a decrease in both temperature and IR radiance during the day, and an increase in these quantities at night. The daytime temperature decrease is greatest for fine dust, whereas the daytime decrease in radiance is greatest for coarse dust. At night, the greatest increase in both temperature and radiance occurs in the presence of coarse material. The simulations represented dust as a layer between the ground surface and 1.5 km. When the dust was represented as an elevated layer, the differences in the results were negligible for fine dust, and small for coarse dust.
The Infra-Red Difference Dust Index (IDDI)
The Infra-Red Difference Dust Index (IDDI) is a 1º x 1º gridded dataset providing an indirect measure of the presence of atmospheric aerosols over the land areas of Africa and parts of the Middle East for the years 1984-1993. It has been developed at the Laboratoire d'Optique Atmosphérique (LOA) de Lille, at L'Université des Sciences et Technologies de Lille in France (Legrand et al., 1994; Brooks and Legrand, 1999). The IDDI is a record of the reduction in brightness temperature of the Earth-atmosphere system due to the presence of aerosols, and is constructed from METEOSAT infra-red channel data. The processing involved in the construction of the IDDI is detailed below. In its present form the IDDI is a monthly-mean climatology of brightness temperature reductions due to atmospheric aerosols January 1984 to December 1993. These monthly means have been constructed from "instantaneous" daily measurements taken at midday, which have themselves undergone pre-processing. (The original METEOSAT data are measured along a number of traverses. It takes some twenty five minutes for all the traverses to be completed, so the image is created between approximately 11:35 and 12:00 UTC each day.) Over the arid and semi-arid regions of northern Africa, the dominant aerosol types are soil-dust particles originating in the Sahara Desert and the Sahel, so the IDDI may be interpreted as representing airborne dust of desert and semi-desert origin over these areas.
The data used to construct the IDDI dataset are radiometric count data measured by the infra-red (IR) channel of the METOESAT satellite daily between 11:35 and 12:00 UTC. Full-resolution radiometric count (RC) data are converted to B2 format prior to distribution to users of the METEOSAT data by the European Space Agency (ESA), from Darmstaadt in Germany.
The value representing a single B2-format grid square, or cell, is obtained by sampling full-resolution data at one line in every six and one pixel in every six. Thus an original "image" consisting of 2500 x 2500 "pixels" is converted to a 416 x 416 pixel image. The satellite resolution is 5km x 5km at the sub-satellite point; the resolution of the sampled B2 image is therefore some 30 km. Preprocessing is performed to ensure that a given pixel always corresponds to the same geographical area.
The METEOSAT radiometer measures radiative energy flux in Watts. These values are converted into radiances by dividing the area of the radiometer aperture by the solid angle subtended by the instrument's field of view. The radiometer on board METEOSAT compares each radiometric measurement with radiometric measurements of an internal "blackbody" on board the satellite. The emissivity of this blackbody is not exactly unity, and will vary slightly with time. The stability of the internal blackbody is monitored by comparing in situ measurements of SST with the satellite SST measurements (corrected for atmospheric effects using radiosounding data). Thus changes in emissivity of the blackbody can be accounted for.
A 256 (columns) x 275 (rows) pixel window, covering Africa and parts of the Middle East, is extracted from the B2 fields of RC data. Data are corrected for the daily variations in sensitivity of the IR channel using the satellite calibration reports (Legrand et al., 1994). Matching of the radiometric scales of the successive METEOSAT satellites allows RC data to be converted to brightness temperature (BT) data, using a unique relationship between a target black body temperature and the radiometric counts derived for the single, homogenised scale (Legrand et al., 1994). The brightness temperature associated with the measured radiance is the temperature at which a target black body would cause such a radiance to be measured. An increase of one radiometric count is equivalent to a rise in brightness temperature of 0.5 K over the useful range of 260 - 340 K. BT and RC data are then archived in ASCII format; the BT data are also stored as binary images in order to facilitate quick visual validation of the data using the Sphinx graphics software developed at the LOA. Both BT and RC data are processed to produce a daily version of the IDDI; the final monthly IDDI climatology is produced from the daily BT IDDI data.
The next stage is to create reference images representing clear-sky conditions for consecutive, non-overlapping periods whose duration is short enough to eliminate seasonal effects but long enough to ensure that clear-sky or near clear-sky conditions characterise at least one measurement for each pixel. In order that these criteria are fulfilled, a 15-day reference period is used. For a given pixel, the maximum in the daily values of BT/RC within a given reference period is assumed to represent the value characteristic of the target area in question in the absence of cloud and dust haze for that period. (Even if the atmosphere is not completely free of dust aerosols, the reference image may be interpreted as representing a minimum, background dust loading and therefore fulfils its intended purpose insofar as detection of episodic dust events is concerned.) Daily "difference images" are then constructed by subtracting the measured daily values from the values of the reference image for the period in question. In the case of the brightness temperature data, difference values indicate the temperature depression for each pixel due to the presence of atmospheric aerosol particles or cloud.
In order to use the IDDI as a means of measuring atmospheric dust, cases where the reduction in BT (or RC) is due to the presence of cloud must be identified and either masked or removed. Such cases are identified using the spatial coherence method (Coackley and Bretherton, 1982). Each processed B2 cell is tested for cloud contamination by examining the mean and standard deviation of the surrounding 3x3-pixel block. The distribution of the means and standard deviations of all the 3x3-pixel blocks is plotted and threshold values of the mean and standard deviation are assigned which define the cloud-free area in the resulting diagram. Cells which lie outside the cloud-free area are masked by assigning to them a code indicative of cloud presence. The initial estimates of the appropriate threshold values are modified by the cloud identification algorithm in order to minimise the errors in this process. The resulting cloud masking is reasonable, although cloud contamination (where particularly high IDDI values result from the misinterpretation of cloud as dust) is apparent in some cases, particularly at the edge of large cloud fields. However, such contamination is generally confined to the equatorial regions. Over the more arid Sahel and Sahara the degree of ambiguity in the IDDI is minimal. The application of the spatial coherence method to the IDDI data is described fully in N'Doumé (1993).
Finally, in order to create the daily IDDI climatology, the IDDI data are converted from the satellite grid to a geographical grid with a resolution of 1º x 1º, via a procedure which effectively averages the values within 1º x 1º blocks while transforming the projection.
The IDDI has been validated using visibility measurements from thirty-nine stations in northwest Africa covering the whole of 1984 (Legrand et al., 1994). The majority of these stations lie between 10º and 20º N and all lie between the West African coast and 20º E. A few of the stations lie well into the Sahara, and two are situated just south of 10º N. The validation was performed by calculating the IDDI for the cells of the 3x3 cell arrays centred on the stations. Seven categories of visibility (measured at 1200 UTC) were defined, after the elimination of cloudy cases. The IDDI data were classified into seven groups corresponding to the visibility categories. The mean and median of the IDDI for each visibility category were calculated, and the resulting values plotted against visibility (Figure 3.2).
Visibilities of less than 10 km are considered to be associated with the presence of atmospheric dust, and visibilities below 5 km are associated with severely dusty conditions. The plot of IDDI versus visibility (Figure 3.2) indicates that IDDI values above 9.8 counts correspond to dusty conditions as defined by the above criteria, with values above 14.6 counts resulting from severely dusty conditions. (The median, rather than the mean, is used in order to suppress the influence of outliers in the data.) Temporal and spatial changes in these threshold values due to seasonality and geography in the regions containing the meteorological stations are minimal, and threshold values of 10 and 15 counts have been chosen to represent conditions of dustiness and severe dustiness respectively (Legrand et al., 1994). Visibilities of 10 km and 5 km have been associated with values of AOD of 0.3 - 0.5 and 0.6 - 0.9 respectively. Legrand et al. (1994) take the mid-points of these ranges to suggest that an IDDI of 10 counts (or 5 K) corresponds roughly to an AOD of 0.4, and an IDDI of 15 counts (7.5 K) to an AOD of 0.75.
The IDDI images, which represent brightness-temperature reduction due to aerosols over land, have also been compared with images representing the distribution of vertical aerosol optical depth (VAOD) over the eastern tropical Atlantic. The VAOD images are produced from METEOSAT visible-channel data which allow the detection of Saharan/Sahelian dust plumes over the Atlantic ocean. This comparison is made in the vicinity of the coast of northwest Africa in order to examine the continuity of the dust plumes as identified over land with the IDDI data and over ocean with visible-channel data (Legrand et al., 1994). Although the units of measurement are different in the land and ocean cases, an appropriate choice of scale results in good continuity across the African coastline, indicating consistency between these two methods of identifying airborne dust.