Qualifying IMG tropical spectra for clear sky

Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

www.elsevier.com/locate/jqsrt

Qualifying IMG tropical spectra for clear sky G. Masielloa , C. Seriob; ∗ , H. Shimodac a

b

IMAAA=CNR, Tito Scalo, Pz, Italy INFM, Gruppo Collegato di Potenza, Sezione di Napoli, C.da Macchia Romania, Potenza, Italy c Earth Observation Research Center, NASDA Tokyo, Japan Received 15 March 2002; accepted 22 May 2002

Abstract The problem of cloud detection for the Interferometric Monitoring of Greenhouse Gases spectrometer has been addressed by considering a set of thresholding tests which takes full advantage of the high spectral resolution of the sensor. The methodology has been applied to a case study consisting of spectra recorded in the tropics on sea surface, although the scheme may be easily extended to other latitudes. The algorithm is very e4cient because it uses only the observed spectrum and no on-line radiative transfer calculation is needed. Based on this cloud detection scheme a set of clear-sky tropical spectra have been identi7ed to be used by the scienti7c community for further studies such as retrieval of atmospheric properties and high spectral resolution radiative transfer modeling. ? 2002 Elsevier Science Ltd. All rights reserved. Keywords: Infrared; Atmosphere; Clouds; Radiative transfer

1. Introduction The Interferometric Monitoring of Greenhouse Gases (IMG) [1] has =own on board on the Japanese Advanced Earth Observing Satellite (ADEOS) from October 1996 to June 1997. The instrument is a Fourier Transform Spectrometer which observes Earth’s emission spectrum at nadir view in three spectral bands from 3.3 to 16 m with a spectral sampling of 0:05 cm−1 and an apodized spectral resolution of 0:1 cm−1 [1]. The IMG data are of good spectral coverage of the Earth full disk and therefore provide new insights into remote sensing of atmospheric parameters (see, e.g., [2– 6]). However, one problem which is still hampering the full exploitation of IMG observations is the absence of information about cloud contamination in the sensor footprint. ∗

Corresponding author. DIFA, University of Basilicata, C.da Macchia Romania, 85100 Potenza, Italy. Tel.: +39-0971427-261; fax: +39-0971-427-271. E-mail address: [email protected] (C. Serio). 0022-4073/02/$ - see front matter ? 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 0 2 2 - 4 0 7 3 ( 0 2 ) 0 0 0 8 3 - 3

132

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

The Earth Observation Research Center (EORC) has developed a scene identi7cation procedure for IMG which uses the Ocean Color and Temperature Scanner (OCTS) [7], an imager which has been a companion instrument of IMG aboard ADEOS. The IMG footprint is co-located with the scene imaged through the OCTS visible channels. The OCTS co-location for the IMG observations is available at the EORC IMG home page: http://www.eorc.nasda.go.jp/AtmChem/IMG/. One limitation of the above procedure is that it is only available for day time and the presence of clouds has to be based on visual inspection of the OCTS imaged scene, which while powerful, since the human eye is quite able to capture cloud texture, is nevertheless a process subjective and therefore not immune from errors. Our primary objectives are to develop an IMG stand alone day–night cloud detection scheme, relying only on IMG observations, and to provide a well quali7ed set of clear-sky IMG spectra to be used for future studies concerning retrieval of atmospheric properties, high spectral resolution radiative transfer modeling, assessment of cloud detection schemes for high spectral resolution infrared sensors such as the American Advanced Infrared Radiometer Sounder (AIRS) [8] and the European Infrared Advanced Sounding Interferometer (IASI) [9]. A MATLAB code which implements the cloud detection scheme and the set of spectra used in this analysis are available from the authors on request. To attain the above goals we have devised a cloud detection scheme which incorporates the so-called hs-method developed in [10,11] and three new thresholding tests which are particularly sensitive to low and thin clouds. A validation data set of clear-sky IMG spectra has been build up by using colocated OCTS imagery. This validation set has been then used to evaluate the performance of the cloud detection scheme. The scheme relies mostly on spectral channels in the atmospheric window and is, therefore, primarily intended for IMG Band 3, which covers the wave number range from 600 to 2000 cm−1 . For the sake of brevity, we limit here to consider a case study consisting of sea surface tropical spectra. The extension of the methodology to other latitudes is straightforward and the cloud detection methodology can be used for any nadir view high spectral resolution infrared sensors. The paper is organized as follows. Section 2 describes the set of IMG data we have used and gives information about the IMG instrument in general. Section 3 describes the basic cloud detection methodology. Section 4 compares the results of the cloud detection scheme to that obtained through OCTS colocation. Conclusion are given in Section 5.

2. The IMG data set The IMG instrument is a Michelson interferometer which is operated with three infrared detectors covering the spectral ranges or bands: • Band 1: 3.3–4:3 m, • Band 2: 4.3–5:0 m, • Band 3: 5.0 –16:7 m. Each detector has its own 7eld of view of 0:6◦ , which yields a ground footprint of about 8 km2 . The detector for Band 2 is on the optical axis so that it looks directly at satellite nadir position, with the other two footprints side by side. The IMG Field of View and the relative position of the three footprints is shown in Fig. 1. The 7gure also shows a complete observation cycle of IMG which lasts 110 s during which the sensor acquires six diLerent observations ≈ 86 km apart. Each series

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

133

Fig. 1. Field of view geometry of IMG showing the footprint of the three bands (adapted by IMG Technical Report available at web page http://www.eorc.nasda.go.jp/AtmChem/IMG/).

of six spectra represent a fundamental unit of data produced by the IMG data processing team and will be referred to as an observation sequence. Band 1 has never operated during the operational life of IMG, so that data are available only for Bands 2 and 3. Because of the peculiar IMG Field Of View geometry, an independent cloud detection scheme for each band should be devised. However, thresholding tests for cloud detection rely mostly on atmospheric window channels which are mostly located in Band 3. For this reason, the scheme we present in this paper is intended only for Band 3. This should be kept in mind in the remaining of this paper, mostly when looking at the results presented in Section 4. The data used in this analysis have been calibrated with the calibration program version 005 and 006 (e.g., http://www.eorc.nasda.go.jp/AtmChem/IMG/). The spectra have been then apodized with a Gaussian function of half width at half height of 0:25 cm−1 and resampled at a rate M = 0:25 cm−1 . By this way the new data 7t the spectral characteristics of the IASI spectrometer [9] and can be used for analysis concerning such an instrument. The spectra have been recorded within the tropical belt and cover the period December ’96 to June ’97. Their geoghraphic location is shown in the map of Fig. 2. Each observation sequence is provided with the corresponding OCTS image showing earth location of the three IMG footprints. An example is shown in Fig. 3. 3. The cloud detection scheme The cloud detection scheme is based on a series of four tests which are here described. 3.1. The hs-method The hs-method has been developed in [10,11] and it was derived to take into account the high spectral quality which will become soon available with new generation infrared sensors such as AIRS and IASI [8,9], of which IMG has been a precursor.

134

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

90

Latitude (degrees)

60

30

0

−30

−60

−90 −180

−120

−60

0

60

120

180

Longitude (degrees)

Fig. 2. Map of the IMG tropical sequences used in this analysis.

Fig. 3. Example of OCTS frames co-located with the IMG footprint. The brightest box refers to the IMG Band 1, the box in the middle is the footprint of Band 2 and 7nally the third box refers to IMG Band 3. The sequence of observations is 1– 6 in the top to bottom order.

Basically, the scheme exploits the unique spectral signature of sea surface in the window region and needs a reference spectrum. In atmospheric window regions radiance spectra are quite sensitive to the emissivity of the underlying emitting surface. Clear-sky spectra are sensitive of course to the emissivity of the land or sea, whereas cloudy spectra are mostly dependent on the cloud spectral signature. Keeping this in mind, attention has been focused to some suitable atmospheric window (the atmospheric window 800–900 cm−1 has been considered to develop the hs scheme), and an index which is very sensitive to emissivity while being quite insensitive to other possible interfering parameters, such as temperature of the underlying emitting surface and water vapor concentration, has been de7ned. The index uses the observed spectrum and a suitable clear-sky (synthetic or measured) spectrum. We refer to [10,11] for any details on how to properly devise such an index and limit the discussion to the relevant de7nitions.

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

135

Let S1 () and S2 () be the observed and reference spectrum, respectively. Then, the 7rst operation involved to compute the hs index is to transform into brightness temperature (BT) the two spectra: Ti () = B−1 (Si ());

(1)

where i = 1; 2 and B is the blackbody Planck function, S() and T () are respectively the radiance (e.g. units of W=m2 cm−1 sr) and the BT spectrum, respectively. After BT conversion, T1 () and T2 () are standardized through the operation Hi () =

Ti () − Ti  si

(2)

with i = 1; 2 and Ti  and si the mean and standard deviation of Ti (), respectively. The mean and standard deviation are considered with respect to the wave number . Furthermore, the correlation and cross-correlation of the couple (T1 (); T2 ()) are now computed according to ci (j) (3) ri (j) = ci (0) and r12 (j) =


c12 (j) c1 (0)c2 (0)

(4)

respectively, where j is the lag. The covariance and cross-covariance functions, ci and c12 , respectively, are obtained by ci (j) =

N −j 1  Hi (k) · Hi (j + k) N

(5)

k=1

and N −j 1  c12 (j) = H1 (k) · H2 (j + k) N

(6)

k=1

with i = 1; 2; as usual, we have written H (kM) = H (k), with M being the sampling rate. Finally, the homomorphic degree of the two spectra is de7ned through the index: NL j=1 |r1 (j) − r2 (j)| hs = ; (7) Nl j=1 |r12 (j)| where NL is the number of lags for which the various correlation functions have been computed. Normalization of hs to the coherence function r12 will enhance diLerences in shape, resulting in a very e4cient test to detect inhomogeneities. The index hs will be zero for spectra, which are perfectly homomorphic, that possess homogenous features with clear-sky spectra, whereas it will tend to in7nity for incoherent functions. The characteristic of the index above is that it is based on covariance functions and therefore takes full advantage of the noise-averaging properties of correlation (see [10]). The use of correlation functions counts many applications to the identi7cation of atomic=molecular emission or absorption features (e.g. [12]).

136

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148 300

BT spectrum (K)

295 290 285 280 275

(a)

270 800

810

820

830

810

820

830

840

850

860

870

880

890

900

840

850

860

870

880

890

900

300

BT spectrum (K)

295 290 285 280 275 270 800

(b)

wave number (cm−1)

Fig. 4. IMG reference spectra used in the hs-test. Panel (a) shows the relatively dry reference spectrum, panel (b) shows the relatively wet reference spectrum.

To develop a cloud detection scheme on the basis of the hs index, we need to choose • a suitable cut-oL point for NL , • a suitable reference spectrum, • a suitable threshold, hso , so that the presence of cloud is identi7ed when the computed hs is greater than hso . For NL we have experienced that NL = 100 is enough to retain all the important characteristics present in the covariance functions. For the reference spectrum we have chosen two IMG tropical spectra, well quali7ed for clear sky on the basis of OCTS images. This choice, instead of a synthetic spectrum, minimizes biases due to forward modeling limitations. The two spectra are shown in Fig. 4 and corresponds to a wet and hot atmosphere and a relatively cold and dry atmosphere, respectively. The use of two reference spectra gives higher =exibility to the method. Although, hs works on the basis of standardized spectra, spectra with very diLerent water concentration content could be diLerent in shape because, e.g., of the spectral signature of water vapor continuum. In practice for a given observation we compute two values of hs, the 7rst one corresponding to the dry reference spectrum and the second to the wet reference spectrum. Let hsd and hsw the dry and wet index, respectively. The test is passed if min(hsw ; hsd ) ¡ hso :

(8)

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

137

BT spectrum

300

(a)

299 298 297 296 295 899.5

899.6

899.7

899.8

899.9

900

900.1

900.2

900.3

900.4

900.5

1167.6

1167.7

1167.8

1167.9

1168

1168.1

1168.2

1168.3

1168.4

1168.5

BT spectrum

300 299 298 297 296 295

(b) 1167.5

Emissivity

0.99 0.98 0.97

σ=900 cm−1

0.96 0.95 800

850

900

(c)

−1

σ=1168 cm 950

1000

1050

1100

1150

1200

−1

wave number (cm )

Fig. 5. (a) Example of IMG spectrum showing the super window 899.5 –900:5 cm−1 ; (b) the super window 1167.5 –1168:5 cm−1 is now shown; (c) sea surface spectral emissivity curve showing the position of the two super windows.

Finally, based on our previous experience with the hs method (e.g. [10,11]) a valuable threshold has been proved to be hso = 0:3. It is here important to stress that a suitable threshold may depend on the spectral resolution and type of apodization applied to the spectra. 3.2. Super window channel test The hs-test has proved to be e4cient for moderate and thick cloudiness. However, it may be fooled by thin clouds [11] such as high cirrus. An additional test, which may in part overcome this problem is based on super window channels, that is spectral channels which are only slightly aLected by gas line absorption. Two of these super windows are shown in Fig. 5. They cover the spectral ranges 899.5 –900:5 cm−1 and 1167.5 –1168:5 cm−1 , respectively and with central wave number w1 = 900 cm−1 and w2 = 1168 cm−1 , respectively. Let T (w1 ) and T (w2 ) be the BT values integrated over the two super windows, respectively:  ui 1 T (wi ) = T () d (9) ui − li li with i = 1; 2; T () the BT spectrum and u and l the upper and lower limit of the window, respectively. The idea, which goes back to Inoue [13], is to base a suitable test on the diLerence MTw = T (w1 ) − T (w2 ):

(10)

138

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

The location of the central wave number w1 and w2 over the seawater spectral emissivity curve [14] is shown in Fig. 5. We see that w1 has been chosen in such way to correspond to the maximum of the emissivity curve in the spectral range 800 –1200 cm−1 . Because of this choice, we expect the above BT diLerence to be positive in clear-sky conditions, over sea surface. However, because of the spectral dependence of the water vapor continuum, the sign is normally reversed and only for dry conditions we have that the diLerence is positive. The diLerence temperature de7ned in Eq. (10) measures the slope of the spectrum in the atmospheric window. This slope is very sensitive to the presence of ice particles such as those present in cirrus clouds (e.g. [15]). It is here to be stressed that conventional BT diLerence methods to detect or classify clouds are mainly based on the couple of window channels located at 830 and 900 cm−1 (see, e.g. [15]), whereas we use the couple 900 and 1168 cm−1 . This is because we want to convey in the scheme additional spectral information coming from the segment of the atmospheric window in between 1100 and 1200 cm−1 . The window 800–900 cm−1 has been already used in the hs-method, so that additional tests based on this window would add no fresh information to the cloud detection scheme. The range of possible clear-sky values for MTw has been investigated by computing synthetic spectra for tropical air masses. To this end we have used a representative set of tropical temperature and moisture pro7les, collected by NOAA=NESDIS from radiosonde ascent in the years 1988 and 1989. Based on these calculations we have that for clear sky 1 ¡ MTw ¡ 2

(11)

with 1 = −2 K and 2 = 1:6 K. This range is used to test IMG tropical observations for clear sky. If the computed MTw does not meet the above limits, the corresponding spectrum is =agged cloudy. 3.3. The 791 cm−1 CO2 split window test A further test for low cloudiness has been devised by exploiting the property of the weak CO2 Q-branch at 791 cm−1 . CO2 absorption yields a very well de7ned, sharp spectral feature centered at 791:75 cm−1 (see Fig. 6) in between of a window region with weak water vapor lines on the right-hand side. Because the CO2 mixing ratio is constant with altitude, an elementary calculation shows that the optical depth at the center of the CO2 line is proportional to the ground level pressure. However, the proportionality is altered because of water vapor continuum absorption whose strength depends on the H2 O concentration itself which is highly variable in time and space. This eLect can be highly reduced by diLerentiating the BT at the center of the line with a suitable BT outside the band, giving the split window diLerence (see Fig. 6) MT = T (o ) − T (i );

(12)

where i = 791:75 cm−1 is the in-band wave number and o = 790:5 cm−1 is the out-of-band wave number. One limitation of this method is that the constant of proportionality depends on the temperature pro7le itself. Nevertheless, for clear-sky MT assumes values which are typically greater than those characterizing cloudy conditions.

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

139

300

295

BT spectrum (K)

290

285

280

275

270

265

260 788

789

790

791

792

793

794

795

−1

wave number (cm )

Fig. 6. Example of IMG spectrum showing the CO2 Q-branch at 791 cm−1 . The depth of the CO2 line de7nes the quantity MT which is the basic ingredient of the test discussed in Section 3.3.

As in the previous section, based on synthetic calculation we have that MT for clear sky assumes values greater than 15 K, which gives us the following criterion to be satis7ed by IMG clear-sky observations: MT ¿ 3 = 15 K:

(13)

3.4. The spatial consistency and coherency test As it has been shown in Section 2, each IMG observation sequence consists of six spectra which are 86 km apart and spatially distributed, approximately, on a latitudinal line of 6 × 86 km = 516 km. On such a distance and over sea surface (at tropics), the clear-sky dynamic is expected to be very low. Sharp diLerences from one spectrum to the next are therefore unlikely and constitute a clue for cloudy conditions. To trap this possible situation we have devised a test which is applied at each IMG sequence at a time. After passing the three thresholding tests above on a given sequence, let us suppose that j spectra (j ¿ 1) have been =agged clear. At this point the maximum of MT (Eq. (12)) is computed. Let Mmax be the maximum, then the following additional clear sky criterion is set up: Mmax − MT (j) ¡ 4 :

(14)

140

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

The value of 4 has been found by trial and error by experiencing with a few clear-sky IMG sequences. We have found that the above diLerence is in the range 0:7–1 K. For the work here shown, the value 4 = 0:9 K has been used.

4. Cloud detection scheme evaluation As pointed out in the introduction section, the OCTS imager provides independent information for cloud discrimination in the IMG footprint. Using this information we can build up a set of IMG clear-sky spectra which can be used as a validation data set for our methodology. This approach has been taken in this study, although it turned out to be a very di4cult task. First, many OCTS images do not have enough contrast to clearly detect clear sky, which limits the number of cases to be included in the validation data set. Second, and mostly important, many IMG observations detected clear through OCTS, exhibits spurious spectral feature which are likely to be the results of phase mismatching in the calibration process. The spectra had, therefore, to be checked once at a time, even in presence of well contrasted OCTS images. An example of an IMG sequence showing spurious spectral feature is illustrated in Fig. 7. Although the OCTS image shows no presence of clouds for Observations 1 and 2, the

Spectrum (Watt/m2− cm−1 −sr )

Obs. 1 Obs. 2 Obs. 3 Obs. 4 Obs. 5 Obs. 6

0.1

0.05

0

800

1000

1200

1400

1600

1800

2000

wave number (cm−1)

Fig. 7. Example of a sequence of IMG spectra showing phase mismatching calibration. The right-hand part of the 7gure shows the six OCTS frames with the position of the IMG footprints (observation 1 is at the top, 6 at the bottom). The IMG sequence is # 329130.

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

141

0.14

0.12

2

1

Spectrum (Watt/m −cm − −sr)

0.13

0.11

0.1 Obs. 1 Obs. 2 Obs. 3 Obs. 4 Obs. 5 Obs. 6

0.09

0.08 840

845

850

855

860 865 wave number (cm−1)

870

875

880

Fig. 8. Further example of IMG spectra showing anomalies in the atmospheric region window. The anomaly here manifests as short pulses mostly evident in between the range 855–860 cm−1 . The right-hand part of the 7gure shows the six OCTS frames with the position of the IMG footprints (observation 1 is at the top, 6 at the bottom). The IMG sequence is # 334124.

sequence exhibits spurious spectral structure in the atmospheric window which are, for this case, surely due to phase mismatching. In other cases the spurious spectral features appear in a form of short pulses which distort the normal spectral structure expected for clear sky in the atmospheric window. An example is shown in Fig. 8. In the end, after a tedious and long scrutiny we sorted 166 OCTS clear-sky spectra and 196 OCTS cloudy spectra which form our validation data set. The clear and cloudy validation data set are shown in Tables 1 and 2, respectively. The cloudy set encompasses various kind of cloud contamination. Thick clouds, open and closed cloud cell formations, strati7ed clouds are present in the set. Cloudiness partly contaminating the IMG footprint for Band 3 was induced, as well. The four tests described in the previous section have been implemented in such a way to form an AND logic structure, that is they have to be simultaneously satis7ed for the given IMG sequence to be declared clear. Keeping this in mind, the validation data set has been passed through the cloud detection scheme. To begin with we discuss the results obtained for the OCTS clear data set. We found that 146 out of the total 166 passed the test. Twenty spectra were rejected as cloudy. This result was in part expected since the validation data set was prepared by visual inspection of OCTS images, a process

142

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

Table 1 List of IMG spectra classi7ed clear according to OCTS imagery

IMG ID#

OBS#

IMG ID#

OBS#

IMG ID#

OBS#

IMG ID#

OBS#

165829 165829 167228 167228 167229 187925 189226 189226 189825 189825 189825 190626 190626 190626 190626 190626 209026 209525 210924 210924 210924 229826 230628 231427 231427 231427 231628 231726 231726 231726 231726 231726 231826 232431 232524 232524 233725 233726 233924 235124 235124 235125

1 2 5 6 1 3 1 2 1 2 3 2 3 4 5 6 5 1 2 4 6 6 4 2 3 4 6 1 2 3 5 6 5 3 2 5 5 5 5 5 6 2

235125 289729 289729 291931 291931 292629 292629 292832 292832 293131 293131 293331 294032 294229 294831 294831 309626 309630 309630 309630 309630 309630 309631 309631 309631 310229 310229 310230 310430 310925 310929 325631 328730 329928 329928 330428 330428 330428 330430 330430 331031 331327

5 5 6 1 3 1 2 1 6 1 6 1 4 1 1 2 2 1 2 4 5 6 1 2 3 2 3 1 4 4 5 1 4 5 6 1 2 6 1 3 1 5

331327 331328 331328 331328 331329 331329 331928 331928 331928 334125 334127 334127 334127 334127 334128 334128 334330 334330 334330 335330 335330 335330 335330 335331 335331 335930 335930 337228 337330 352928 353629 353629 353829 354029 354429 354730 354731 354731 354828 354829 354829 356228

6 2 3 6 1 2 3 4 5 1 3 4 5 6 5 6 1 5 6 1 4 5 6 1 2 3 4 4 3 3 5 6 6 1 5 1 2 4 6 1 2 2

356228 356228 356228 356230 356230 356231 356231 356627 356731 356731 356831 357028 357028 357031 357627 357630 358028 372631 373024 373024 373231 373231 373629 373629 373629 373729 373829 373829 373829 373928 374030 374128 374431 374730 374730 375128 375130 375130 375227 375227

3 5 6 2 4 1 2 2 1 2 4 2 4 1 4 6 2 1 1 4 5 6 2 3 5 3 1 2 4 4 4 6 5 1 2 6 5 6 3 5

∗∗ ∗∗

∗∗ ∗∗

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

143

Table 2 List of IMG spectra classi7ed cloudy according to OCTS imagery IMG ID#

Obs.#

IMG ID#

Obs.#

IMG ID#

Obs.#

IMG ID#

Obs.#

167228 167229 167229 187925 189226 189226 189227 189227 189227 189825 189825 209026 209026 209524 209524 209525 209525 209627 209829 209829 210924 230425 230425 230628 230628 231826 231826 231932 231932 231932 232431 232524 232524 233724 233724 233924 233924 235124 235124 235124 290531 290531 291931 293131 293131 293131 293331 293331 293331

2 5 6 5 3 5 1 2 5 4 5 2 6 1 6 3 4 2 2 3 5 3 4 2 3 1 3 4 5 6 6 3 4 1 2 1 6 1 2 3 1 2 6 3 4 5 3 5 6

294032 294032 294032 294229 294731 294731 294731 294731 294731 294831 294831 309626 309629 310230 310230 310230 310230 310231 310231 310231 310429 310431 325631 325631 328730 328730 330428 330428 330430 330430 330430 331031 331329 334330 334330 335930 337229 337229 337330 337330 337330 352928 352929 352929 352929 352929 352929 352929 352930

1 2 3 5 2 3 4 5 6 3 4 6 4 2 3 5 6 2 3 4 3 4 5 6 1 2 3 4 2 4 6 4 5 3 4 6 1 2 1 2 6 4 1 2 3 4 5 6 1

352930 352930 353030 353230 353230 353230 353231 353231 353231 353231 353629 353829 354029 354029 354029 354029 354030 354030 354428 354429 354730 354730 354828 354829 354829 355527 355527 355527 355828 355828 355828 355828 355828 356230 356230 356231 356231 356830 356830 356830 357027 357028 357031 357324 357324 357531 357531 357625 357625

3 5 3 4 5 6 1 3 4 6 1 4 3 4 5 6 4 6 5 2 4 6 3 4 6 1 3 6 1 2 3 4 6 1 3 5 6 1 4 5 5 6 6 3 4 1 2 4 5

358028 358028 358028 358028 372631 372631 372631 372930 372930 373024 373024 373024 373231 373231 373629 373828 373928 373930 373930 373930 373930 373930 373931 373931 374030 374031 374031 374127 374127 374127 374128 374128 374128 374128 374128 374328 374328 374328 374328 374431 374730 374730 374730 375128 375128 375128 395231 395231 395231

1 3 5 6 4 5 6 1 6 2 5 6 1 2 6 3 1 1 2 3 4 5 2 4 6 1 5 3 5 6 1 2 3 4 5 3 4 5 6 3 3 4 5 1 3 4 3 4 6

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148 BT spectrum (K)

144

BT spectrum (K)

(a)

(b)

300 280 260 800

IMG # 167228 Obs. 6 850

900

950

1000

1050

1100

1150

1200

950

1000

1050

1100

1150

1200

300 280 260 800

Clear Sky Reference Spectrum 850

900

302

BT (K)

301

reference observation

300 299 298 900

(c)

950

1000

1050

1100

1150

1200

wave number (cm−1)

Fig. 9. IMG observation showing the signature of light cirrus contamination (panel a). The signature is evidenced by the slope of the straight line (shown in red) which connects the super window at 900 cm−1 to that at 1168 cm−1 . The slope is further evidenced by comparing the spectrum shown in panel (a) to the clear-sky reference spectrum shown in panel (b). The direct comparison between the two slopes is provided in panel (c).

which is not so e4cient for light clouds. Of the 20 rejected spectra, 13 exceeded the lower threshold of the window channel test discussed in Section 3.2. The failure could well be the result of thin cirrus contamination. This kind of contamination is hardly diagnosed on the basis of a visual check of visible imagery alone. To illustrate the possibility for thin cirrus contamination, Fig. 9 shows the spectrum slope for the observation 6 of the IMG sequence # 167228 for which we computed MTw = −2:89. The spectral signature of cirrus appears here to be quite evident when we observe the increasing linear trend visible in the atmospheric region. The comparison with a typical clear sky spectrum (also shown in Fig. 9) leads us to conclude that cirrus contamination is very likely to be the case for this observation. For the other seven rejected spectra, 7ve were found slightly cloud contaminated by a better check of the corresponding OCTS imagery. The remaining two were very close to the threshold values and no evident spectral signature of light cloud contamination could be evidenced by inspection of OCTS images and the spectra themselves. These two cases could be cases for which the scheme simply failed. For the validation cloudy data set, we found that the test diagnosed cloudy 195 out of the 196 total spectra. This better coincidence was expected since the classi7cation of cloudy spectra by visual inspection was easier because very well contrasted cloud scenes were privileged in the selection process. The only observation which was falsely detected clear is the # 6 of the IMG sequence

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

145

IMG #353829 305

300

295

BT Spectrum (K)

290

285

280 Cloud Test Results: 275

270

obs1: cloudy obs2: cloudy obs3: cloudy obs4: cloudy obs5: clear obs6: clear

265

260 800

820

840

860

880

900

wave number (cm−1)

Fig. 10. Results of the cloud detection scheme for the IMG sequence # 353829. The left-hand side of the 7gure shows the six spectra in the atmospheric window 800–900 cm−1 with the cloud detection results summarized in the legend. The right-hand part of the 7gure shows the six OCTS frames with the position of the IMG footprints (observation 1 is at the top, 6 at the bottom).

# 337330. For this case the test failed since the Band 3 footprint was partly contaminated by clouds. Apart from this case, the ability of the scheme to detect partly cloudy contaminated IMG footprints was found excellent. As an example, Fig. 10 shows the results for the IMG sequence # 353829. It is seen that the observations 1 and 2 which are partly contaminated by clouds are =agged cloudy upon passing the test. The observations 5 and 6 are correctly detected as clear. Again, it is here important to stress that the cloud test uses only information from the IMG Band 3, so that the results for the cloud test refers to that band. Unfortunately, Band 2 is not sensed through the same Field of View of Band 3, so that the results of the cloud detection cannot be extended to Band 2. A further example of the cloud detection scheme is shown in Fig. 11. The ability of the test to detect the low cloud contamination for the footprint of observation 3 is here really impressive. Finally, we want to show a case which highlights the good cloud penetration of IMG because of its small size footprint. Fig. 12 shows a sequence of observations recorded over a compact cirrus formation. Nevertheless, IMG is able to penetrate a relatively small hole in the clouds with the 7fth observation. This observation was =agged clear upon passing the test.

146

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148 IMG #335330 305

300

295

BT Spectrum (K)

290

285

280 Cloud Test Results 275

270

obs1: clear obs2: clear obs3: cloudy obs4: clear obs5: clear obs6: clear

265

260 800

820

840

860

880

900

wave number (cm−1)

Fig. 11. As Fig. 10 but for the IMG sequence # 335330.

5. Conclusions A cloud detection scheme for IMG has been described in this paper. The scheme is based on a set of thresholding tests which have been devised to take full advantage of the high spectral resolution provided by IMG. In the present version the scheme runs for sea surface only, although the extension of the method to land surface is straightforward. The algorithm has been applied to a set of IMG tropical spectra and the results compared to OCTS imagery which gives independent information about cloudy contamination in the IMG footprints. The algorithm has proved to be very successful in diagnosing the presence of clouds and can be reliably applied to other IMG sequences for which the OCTS co-location is not available. Even in presence of OCTS co-location, the scheme may add additional information, valuable to detect, e.g., thin cirrus cloud contamination which are not well contrasted in visible imagery. We have checked that the scheme may also detect the presence of spurious spectral features which are present in IMG spectra because of phase mismatching in the calibration process. For this reason, it is highly recommended to pass IMG sequences through the test presented in this paper even in presence of very well contrasted OCTS images. IMG spectra showing spurious spectral feature are, indeed, not unlikely, a point which should be kept in mind by people aiming at using IMG data, e.g., for retrieval or climatological studies.

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

147

IMG #355828 300

290

280

BT Spectrum (K)

270

260

250 Cloud Test Results

240

230

obs1: cloudy obs2: cloudy obs3: cloudy obs4: cloudy obs5: clear obs6: cloudy

220

210 800

820

840

860

880

900

−1 wave number (cm )

Fig. 12. As Fig. 10 but for the IMG sequence # 355828.

Acknowledgements Work supported by Italian Space Agency. References [1] Kobayashi H, Shimota A, Yoshigahara C, Yoshida I, Uehara Y, Kondo K. Satellite-borne high-resolution FTIR for lower atmosphere sounding and its evaluation. IEEE Trans Geosci Remote Sensing 1999;37(3):1496–507. [2] Amato U, Cuomo V, DeFeis I, Romano F, Serio C, Kobayashi H. Inverting for geophysical parameters from IMG radiances. IEEE Trans Geosci Remote Sensing 1999;37(3):1620–32. [3] Clerbaux C, Hadji-Lazaro J, Payan S, Camy-Peyret C, Megie G. Retrieval of CO columns from IMG=ADEOS spectra. IEEE Trans Geosci Remote Sensing 1999;37(3):1657–61. [4] Lubrano AM, Serio C, Clough SA, Kobayashi H. Simultaneous inversion for temperature and water vapor from IMG radiances. Geophys Res Lett 2000;27:2533–6. [5] Hadji-Lazaro J, Clerbaux C, Couvert P, Chazette P, Boonne C. Cloud 7lter for CO retrieval from IMG infrared spectra using ECMWF temperatures and POLDER cloud data. Geophys Res Lett 2001;28:2397–400. [6] Lubrano AM, Masiello G, Serio C, Matricardi M, Rizzi R. IMG evidence of chloro=uorocarbon absorption in the atmospheric window region 800–900 cm−1 . JQSRT 2002;72(5):623–35. [7] Shimoda M, Oaku H, Mitomi Y, Murakami H, Kawamura H. Satellite-borne high-resolution FTIR for lower atmosphere sounding and its evaluation. IEEE Trans Geosci Remote Sensing 1999;37(3):1484–95.

148

G. Masiello et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 77 (2003) 131 – 148

[8] Aumann HH, Strow L. AIRS, the 7rst hyper-spectral infrared sounder in support of operational weather forecasting. IEEE Aerospace Conference, Big Sky, 10 –17 March 2001, p. 345 –56. [9] Cayla F-R. IASI infrared interferometer for operations and research. In: ChTedin A, Chahine MT, Scott NA, editors. High spectral resolution infrared remote sensing for earth’s weather and climate studies. NATO ASI Series, I9. Berlin: Springer, 1995. p. 9–19. [10] Serio C, Lubrano AM, Romano F, Shimoda H. Cloud detection over sea-surface using auto-correlation functions of up welling infrared spectra in the 800 –900 cm−1 window region. Appl Opt 2000;29:3565–72. [11] Masiello G, Matricardi M, Rizzi R, Serio C. Homomorphism between clear-sky and cloudy spectra in the 800– 900 cm−1 window region. Appl Opt 2002;41(6):965–73. [12] Beer R, Norton RH. Analysis of spectra using correlation functions. Appl Opt 1988;7:1255–61. [13] Inoue T. A cloud type classi7cation with NOAA-7 split window measurements. J Geophys Res 1987;92:3991–4000. [14] Masuda K, Takashima T, Takayama Y. Emissivity of pure and sea waters for the model sea surface in the infrared window regions. Remote Sensing Environ 1988;24:313–29. [15] Stephens GL. Remote sensing of the lower atmosphere. Oxford, UK: Oxford University Press, 1994.