Atmospheric Science Letters (2001) Volume 1doi:10.1006/asle.2000.0021

Simulations of time-coincident, co-located measurements from ENVISAT-1instruments for the characterization of tropospheric aerosols:

A sensitivity study including cloud contamination effects

Francesca Torricella1*, Elsa Cattani1, Marco Cervino1, Vincenzo Levizzani1 and Maria JoaÄo Costa1,2

1Institute of Atmospheric and Oceanic Sciences, (ISAO-CNR), Via Gobetti, 101, 40129 Bologna, Italy2University of EÂvora, Department of Physics, R. RamaÄo Ramalho 59, 7000 EÂvora, Portugal

1530-261

* Corr

Abstract: A sensitivity test of the aerosol properties retrieval is conducted using nadir viewing instruments on board the ENVISAT-

1 platform trying to reproduce the natural variability of the aerosol ®eld. The analysis highlights problems arising in the

retrieval when aerosol loading, relative humidity, and cloud coverage are simultaneously varied.*c 2001 Royal Meteorological Society

Keywords: Aerosol, space-based retrieval methods, ENVISAT-1 mission.

1. INTRODUCTION

Satellite-based instruments aimed at studying the scattering constituents of the atmosphere, speci®cally aerosol and clouds, are designed to

include several spectral channels with better spectral and spatial resolution than classical instruments conceived for ocean colour, sea-

surface temperature, atmospheric chemistry or land cover. Indeed, the large spectral range from ultraviolet to infrared, which is required to

detect and characterize such variable atmospheric components successfully, is very imposing for a single instrument. In principle each

spectral range would require in fact different sensors, observation techniques, and spectral resolution. The traditional use of sensors

mounted on board polar satellites for the characterization of tropospheric aerosol is now receiving new impetus from their synergetic use

with sensor mounted on geostationary satellites. The spectral resolution and coverage of the former are combined with the high repetition

time of the latter for an improved characterization of the aerosol ®eld at a time scale appropriate for appreciating the rapid variations in the

aerosol ®eld (Costa et al., 2000).

It is widely recognized that proper cloud clearing is a mandatory a priori requirement to perform an accurate aerosol retrieval, so that the

use of simultaneous IR measurements is crucial. The ENVISAT-1 mission, planned to be launched in June 2001, carries, among other

X *c 2001 Royal Meteorological Society

esponding author: F. Torricella. E-mail: f.torricella@isao.bo.cnr.it

Figure 1. Example of superimposition of AATSR, MERIS and SCIAMACHY swaths for one ENVISAT-1 descending orbit overpassing the Cape Verde

islands. Filled box: MERIS. Open box: SCIAMACHY. Hatched box: AATSR.

sensors, the nadir-viewing SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY), the MEdium

Resolution Imaging Spectrometer (MERIS) and the Advanced Along Track Scanning Radiometer (AATSR). High quality, calibrated data

from UV to IR will be simultaneously available from the same region of the Earth (see Figure 1, which shows the nadir swaths of the three

instruments). These instruments, although independently operated and aimed at different atmospheric studies, can be regarded in principle

as an hyper-instrument with enlarged spectral coverage, due to the spatial overlap of their swaths and the time coincidence of the

measurements. Because of the different spatial resolutions of the three spectrometers (see Section 2), a single measurement of the hyper-

instrument encompasses the largest ground pixel among AATSR, MERIS, and SCIAMACHY, i.e. it has the SCIAMACHY's spatial

resolution of 30 � 60 km2. The measurements taken at the wavelengths of the various channels of the remaining two instruments at higher

spatial resolution enter the hyper-instrument measuring process by spatial averaging. A sensitivity analysis is presented that assesses the

performances of a generic ®tting method (Torricella et al., 1999) in retrieving the aerosol characteristics when adding to the MERIS

channels the UV measurements of SCIAMACHY and the near infrared (NIR) of AATSR. The effect of a thick water cloud partially

obstructing the ®eld of view of the instrument is explored, both analysing its potential in disturbing the aerosol retrieval, and its effects on

AATSR IR channels (centred at 11 and 12 mm) not included in the hyper-instrument channels. The analysis of the thermal IR channels is

by no means a cloud cover retrieval: it rather shows their response to the same scenarios sensed by the hyper-instrument, including changesof relative humidity and aerosol loading.

2. INSTRUMENTS AND MEASUREMENTS

MERIS, SCIAMACHY and AATSR are spectrometers/spectroradiometers that measure the radiation re¯ected/emitted by the atmosphere±surface system in the UV±VIS±NIR±IR part of the electromagnetic spectrum. Each one has its own capabilities for the detection andcharacterization of cloud and aerosol ®elds. AATSR (Murray et al., 1999) is a dual-view, self-calibrating, VIS and IR radiometer. It measuresre¯ectance/brightness temperatures at the top of the atmosphere (TOA) for land-surface, atmosphere, clouds, and ocean studies. AATSRobserves the same scene along the direction of the orbit track ®rst at an incident angle of 558 (forward scan) and 150 s later at an angleclose to the nadir. The swaths are curved and 500 km long; the one at nadir is composed of 555 pixels, 1 � 1 km2 at the centre. MERIS(Rast et al., 1999) is an along-track, push-broom imaging spectrometer conceived primarily for oceanographic and at the same time foratmospheric and land observations. MERIS works with up to 16 spectral bands, whose width and position are programmable in-¯ight inthe 0.390 to 1.040 mm spectral range, and a spectral resolution of 2.5 � 10ÿ3 mm. MERIS data are characterized by two spatial resolutions:full resolution (FR) with a pixel size of 300 � 300 m2 at nadir, and a reduced resolution (RR), achieved by a combination of 4 � 4 adjacentpixels across-track and along-track, resulting in a resolution of 1200 � 1200 m2 at nadir. SCIAMACHY (Bovesmann et al., 1999) is amoderate resolution imaging spectrometer, conceived for studies of the chemistry and physics of the Earth's atmosphere. SCIAMACHY hasthree measurement modes, nadir, limb and solar/lunar occultation, but in this work only the nadir mode is used. In this case, each full scancovers an area of approximately 30 km along-track times 960 km across-track, with a typical spatial resolution of 30 � 240 km2. The spatialresolution can be improved up to 30 � 60 km2 in selected spectral windows (e.g. UV) to investigate key atmospheric constituents, such asclouds and aerosols, at their appropriate spatial scales. As to the hyper-instrument concept, we regard it as having the lowest spatialresolution (SCIAMACHY's 30 � 60 km2, compared with 1 � 1 km2 and 0.3 � 0.3 km2 at nadir for AATSR and MERIS, respectively) andthe narrowest across-track swath (AATSR, about 550 km, see Figure 1); the simultaneous use of all the three instrument is thus in principleguaranteed. Since many AATSR and MERIS measurements cover one single SCIAMACHY ground pixel acquisition, AATSR and MERISdata spatial averages will be required. However, in the following the instruments are considered to look at nadir with the same observinggeometry and at the very same time. Slight differences in the observing conditions are accounted for when considering the overall errorassociated with the method.

3. THE SIMULATED SPECTRA

The radiative transfer model MODTRAN (Berk et al., 1998) was selected for the simulations of the ENVISAT spectra. The model computesthe intensity of radiation re¯ected/emitted by the atmosphere accounting for all the main physical processes, and including the standardvertical pro®les of the atmospheric components of paramount importance. The inputs used in the simulations are summarised in Table 1.Aerosol spectral optical properties over the whole range 0.3±12.5 mm were derived from the aerosol database of Levoni et al. (1999) andsupplied to MODTRAN as parameters from the user. The URBAN aerosol class at 7 humidity levels (relative humidity, RH from 0 to 99%)was selected. The choice yields to a transition from a moderately absorbing type (dry) to a non-absorbing type (humid), the latter beingassociated to almost spectrally independent extinction properties. Moreover, an increase of RH re¯ects in an increase of the size of the

Table 1. Input values for creating the set of simulated spectra

Variable Symbol Number of Values Selected Value(s)

Observer Zenith Angle y 1 08Sun Zenith Angle y0 2 158, 458Spectral Ground Albedo Ag(l) 2 Ocean, Sand

Relative Humidity (%) RH 7 0, 70, 80, 90, 95, 98, 99

Aerosol Optical Thickness at 550 nm tA 9 0, 0.01, 0.05, 0.1, 0.2, 0.5, 1, 2, 3

Cloud Fraction (%) f 4 0, 5, 10, 20, 30

aerosol particles, thereby altering the size distribution. The larger the aerosol particles because of humidity, the stronger their effect on

longer wavelengths: the scattering ef®ciency of atmospheric particulate peaks at a wavelength proportional to the effective size of the

particle. This approach allows to cover a number of different aerosol regimes, depending on RH and on the aerosol optical thickness at

0.550 mm (tA), without the need of ®xing for instance the aerosol complex refractive index. In fact the increasing RH level reverberates on

both refractive index and size distribution according to HaÈnel's theory (Levoni et al., 1997). The aerosol was con®ned to the 0±2 km

atmospheric layer and tA was varied as in Table 1 so as to represent aerosol loadings from almost clear to very turbid atmospheres. The

observation geometries are those of the typical nadir-viewing sensors. Two kinds of surface spectral lambertian albedoes were adopted, the

dark sea surface (avoiding sun glint conditions) and the heavily re¯ecting sand, to emphasise the different behaviour of absorbing and non-

absorbing particulate above non-re¯ecting or intensely re¯ecting surfaces. These two kinds of re¯ectance are typical of surfaces (large

oceanic and desert areas) that can be considered fairly homogeneous also at the coarse spatial resolution of 30 � 60 km2. The boundary

temperature was set to 308C for sand and 158C for sea. The aerosol-loaded spectra were linearly combined with cloudy spectra to simulate

scenarios with different cloud covers (f, see Table 1). The cloudy spectra were obtained by selecting stratus/stratocumulus clouds with an

optical thickness of 50 in the visible range and top height of 8 km. This schematic representation of clouds is acceptable and instrumental

within the present analysis, in which clouds are regarded as a contaminating effect that can distort or even to destroy the aerosol

information one wants to retrieve. The spectral radiances obtained from MODTRAN using the described inputs and settings are regarded

as high resolution spectra, i.e. they simulate the spectral radiance incident upon the instruments' entrance slit. A software was developed to

compute the radiance output of the three instruments by convolving the high resolution spectra with the suitable slit functions of the

various channels. Table 2 reports the list of the channels of the three instruments that compose the spectral coverage of the hyper-

Table 2. Central wavelengths (mm) of the selected channels for the three instruments. The bandwidths are also

reported in brackets (�10ÿ3 mm)

Selected Channels

SCIAMACHY 0.364 (1.0) 0.373 (1.)) 0.394 (1.0) 0.424 (1.0)MERIS 0.753 (7.5) 0.775 (12.5) 0.890 (10.0)

AATSR 1.6 (300.0) 3.7 (300.0)

instrument. SCIAMACHY channels affected by gas absorption features were not considered. The resolution was degraded to 1.0 nm so asto smooth solar spectral features. The nine channels at variable resolution cover a spectral range from 0.364 to 3.7 mm. The two thermal IRAATSR channels at 11 and 12 mm have been separately processed (see section 4).

4. ANALYSIS TOOLS

After computing the radiances measured by the hyper-instrument for the scenarios described in the previous section, the correspondingspectra were analysed by computing the w2 distance between one of them, acting as pseudo-measured (RM(li , SURF; y0 , t0A, RH0, f 0 )), andall the remaining RC(li , SURF; y0 , tj

A, RHk, f l) according to equation (1):

w2j;k;l �

XNwl

i�1

�RM�li; SURF; y0; t0A;RH0; f 0� ÿ RC�li; SURF; y0; tjA;RHk; f l��2

RM�li; SURF; y0; t0A;RH0; f 0�2 �1�

where Nwl � 9, j � 1, . . . 9; k � 1, . . . 7; and l � 1, . . . 14, see Table 1 for the symbols. The w2 is analogously and separately computed forthe two AATSR IR channels (centred at 11 and 12 mm). These two channels are not included in the spectral array of the hyper-instrument,because they carry little information on aerosol loading. Nevertheless they are separately analysed to assess their variability incorrespondence to the selected scenarios. Not surprisingly, they show a marked sensitivity to the cloud presence and a reduced, but notzero, response to the other two parameters. The w2 term is not divided by the error value s2 because s2 is assumed to be the same for allchannels. In any practical application of the method such a simplifying hypothesis must be reconsidered, and the true standard deviationassociated with the real measurements. In the present context the assumption of constant s2 would not alter much the conclusions drawnfrom the sensitivity analysis. Analysis of the shape of the w2 function, depending on f, tA , RH, and on the surface type (sea or sand),provides an idea of the robustness of a generic pseudo-inversion method based on the ®tting of measurements to the computed spectra.The ideal condition is that the w2 function has a single minimum and that the shape of the function near the minimum is steep enough forthe retrieval of the aerosol parameters with acceptable uncertainty. The w2 hyper-surface is represented as contour plots on planes thatintersect it at a ®xed value of one of the parameters. The minimum of the w2 function determines the retrieved parameters relative to thepseudo-measured scenario. In any real condition the absolute minimum value of w2 will not be identically zero due to various error sources(model and measurement errors). The most appropriate way of examining this kind of plot is by selection of a contour level asrepresentative of the overall error inherent to the method and/or measurements (we adopt the 10% threshold, corresponding to w2 � 0.01).All the triplets (f, tA , RH) corresponding to the spectra whose w2 value is lower than 0.01, cannot be discriminated. Contour levelsencompassing small areas well centred around the true value imply that the retrieval has good chances of retrieving precise and accurateparameters.

5. DISCUSSION AND CONCLUSIONS

The results for some relevant cases are presented as contour plots of w2 depending on the parameters (f, tA , RH) (Figures 2±5). Thefollowing conclusions could be drawn from the analysis of all the simulated cases. Small cloud fractions (f 04 10% over the sea andf 04 5% over the sand) are retrieved from IR data only if a dry aerosol is present over the scenario, both over sea and sand. For humid

Figure 2. Example of contour plots of the w2 function depending on tA , RH and f for a pseudo-measured scenario characterized by y0 � 158, sand,

t0A � 3, RH0 � 99%, f 0 � 5%. w2 values are derived from IR measurements only (see text).

aerosols there is a complete trade off between humidity and cloud fraction, especially for high values of tA (Figure 2). Spectra characterizedby low cloud fractions and high values of RH are not distinguishable from the pseudo-measured spectrum as well as from spectracharacterized by high cloud fractions and low RH values. For higher values of f 0 the retrieval is more accurate, because even the highestvalue of RH cannot match high values of f 0. For instance f 0 � 30% (Figure 3) leads to retrieved f 4 25% (no upper limit is given because

Figure 3. The same as Figure 2 but for y0 � 158, sea, t0A � 3, RH0 � 0%, f 0 � 30%.

our analysis is restricted to f 4 30%) with the actual value depending on t0A. The only relevant exception are the scenarios with dry aerosolover the sand, for which the cloud fraction retrieval remains problematic also for f 0 � 30%.

For low values of the optical thickness the cloud fraction is in general well retrieved, except for high cloud covers over sand. Our resultssuggest that in the IR even low aerosol loadings (t0A � 0�2) can be mismatched with partial cloud covers (f 5 30%), especially if the aerosolis highly humidi®ed. Cloud fractions as low as 5% are found to destroy the aerosol information content of such large spectral interval if theaerosol loading is small (t0A 5 0�5) at the selected accuracy level. Assuming an overall accuracy of 3% (corresponding to the 10ÿ3 isopleth)

Figure 4. The same as Figure 2 for y0 � 158, sand, t0A � 2, RH0 � 0%, f 0 � 0%. w2 values are computed using the hyper-instrument channels.

will, however, allow the retrieval of the aerosol characteristics also at f 0 � 30%. For clear scenarios we found that the hyper-instrument is

able of retrieving small t0A over the sand only if RH0 � 0% while high aerosol loadings (t0A 4 1) are retrieved over both sand and sea, with

smaller errors over sand if the aerosol is dry and over sea if the aerosol is humid (see Figures 4 and 5). For instance, t0A � 3 is retrieved over

sand as 42.7 if dry and 40.9 if humid and, over sea as 42 if dry and 42.8 if humid. The relative humidity is undetermined for small t0Abut is very well retrieved (DRH 5 10%) for t0A 4 1.

Figure 5. The same as Figure 2 for y0 � 158, sea, t0A � 3, RH0 � 99%, f 0 � 0%. w2 values are computed using the hyper-instrument channels.

As mentioned before, the different spatial resolutions of the three instruments force the hyper-instrument to have the worst spatialresolution, thus loosing one of the most relevant characteristics of AATSR and MERIS, the high spatial resolution. Likewise, the AATSRcapability of sensing the same air-mass under two different viewing angles (useful to distinguish, for instance between humidity andaerosol) is not exploited in the present approach, although no information is lost. The information gathered at the ®ner spatial scales(AATSR cloud mask, relative humidity from MERIS, etc.) and exploiting all the speci®c characteristics of the two instruments can be easily

included in the method as constraints on the free parameters (for instance on the cloud fraction f ). Several improvements of the methodcan be devised at the present stage, as for instance the extension of the analysis to other aerosol types and surface albedoes. Moreover, thetypical scenario of the given geographic area can be preselected from scene climatology.

The conclusions drawn from the present analysis are applicable to any spectrometer covering such a large spectral interval. In the speci®ccase of ENVISAT-1 instruments the best approach for the analysis of non-uniform scenarios will probably consist in separately processingSCIAMACHY data (i.e. the data at the coarser spatial scale) in order to derive some information on aerosol type prior to process MERISand AATSR data. For instance the UV SCIAMACHY measurements allow to derive a TOMS-like Aerosol Index (Torres et al., 1998), usefulto discriminate the presence of absorbing aerosol over the scenario.

Acknowledgements

We would like to thank Pierre Viau from ESA-ESTEC for supplying the CFI-ESOV software needed to display ENVISAT-1 instruments'swaths. MJC was ®nancially supported by the Subprograma CieÃncia e Tecnologia do 28 Quadro ComunitaÂrio de Apoio. We would like tothank three anonymous Referees for useful comments on an early version of this paper.

REFERENCES

Berk, A., Bernstein, L. S., Anderson, G. P., Acharya, P. K., Robertson, D. C., Chetwynd, J. H. and Adler-Golden, S. M., 1998. MODTRAN cloud andmultiple scattering upgrades with application to AVIRIS. Remote Sens. Environ., 65, 367±375.

Bovesmann, H., Burrows, J. P., Buchwitz, M., Frerick, J., Noel, S., Rozanov, V. V., Chance, K. V. and Goede, P. H., 1999. SCIAMACHY: Missionobjectives and measurement modes. J. Atmos. Sci., 56, 127±150.

Costa, M. J., Cervino, M., Cattani, E., Torricella, F., Levizzani, V. and Silva, A. M., 2000. An update of a GOME-METEOSAT method for aerosol opticalthickness determination and classi®cation Proc. The 2000 EUMETSAT Meteorological Satellite Data Users' Conference, EUMETSAT, Bologna,29 May±2 June, 420±427.

Levoni, C., Cervino, M., Guzzi, R. and Torricella, F., 1997. Atmospheric aerosol optical properties: a database of radiative characteristics for differentcomponents and classes. Appl. Opt., 36, 8031±8041.

Murray, J., Bailey, P., Birks, A. and Smith, D., 1999. ATSR-1/2 User Guide, edited by C. Mutlow, 29 pp, http://www.atsr.rl.ac.uk.Rast, M., BeÂzy, J. L. and Bruzzi, S., 1999. The ESA medium resolution imaging spectrometer MERISÐa review of the instrument and its mission. Int. J.

Remote Sensing, 20, 1681±1702.Torres, O., Barthia, P. K., Herman, J. R., Ahmad, Z. and Gleason, J., 1998. Derivation of aerosol properties from satellite measurements of backscattered

ultraviolet radiation: Theoretical basis. J. Geophys. Res., 103, 17099±17110.Torricella, F., Cattani, E., Cervino, M., Guzzi, R. and Levoni, C., 1999. Retrieval of aerosol properties over the ocean using Global Ozone Monitoring

Experiment measurements: method and application to test cases. J. Geophys. Res., 104, 12085±12098.