VAS solar channels

REMOTE SENSING OF ENVIRONMENT 22:73-101 (1987)

73

Calibration of NOAA-7 AVHRR, GOES-5, and GOES-6 VISSR/VAS Solar Channels

ROBERT FROUIN and CATHERINE GAUTIER California Space Institute, Scripps Institution of Oceanography, La ]olla, California 92093

The NOAA-7, GOES-5, and GOES-6 VISSR/VAS solar channels have been calibrated for the periods from October 1983 through January 1985 (NOAA-7, GOES-6) and from October 1983 through July 1984 (GOES-5). Space and the White Sands National Monument area in Mexico, whose reflectance properties are well known, are used as calibration targets. The shortwave reflected terrestrial radiance that is measured at satellite altitude is computed using a faidy accurate radiative transfer model which accounts for multiple scattering and bidirectional effects (Tanr6 et al., 1979). The relevant atmospheric characteristics are estimated from climatological data (ozone amount, aerosol size-frequency distribution, and refractive index) and observations at the nearest meteorological sites (water vapor amount, visibility). The approach produces accuracies of 8-13% depending on the channel considered. For both types of instruments, no drift in the solar channels in detected during the 15-month period. The gain changes, about 15% of the mean values, are largely attributed to inhomogeneities of the ground target (shading effects due to the presence of dunes). No systematic effect of the normalization procedure applied by NOAA to the raw VISSR/VAS data is detected. There is some evidence that the GOES-5 solar channels gradually deteriorated from March 1984 until the satellite failure in July 1984. Comparisons between gains determined in orbit and those before launch show that the NOAA-7 solar channels read higher by about 15%. The disparities, however, cannot be explained by model errors and must have occurred before the time period analyzed here.

1. Introduction

The Visible Infrared Spin Scan Radiometer/Vertical Atmospheric Sounder (VISSR/VAS) and the Advanced Very High Resolution Radiometer (AVHRR) aboard the meteorological satellites operated by the National Oceanic and Atmospheric Administration (NOAA) were primarily intended to provide frequent, high resolution, visible and infrared images for monitoring weather events and temperature gradients over large areas. SucesshtUy applied to cloud detection, the data in the visible and near-infrared bands have also proven suitable for other environmental applications, in particular surface albedo and vegetation index mappings (e.g., Rockwood and Cox, 1978; Tarpley et al., 1984; Tucker et al., 1984), and radiation budgets studies (e.g., Stephens et al., 1981; Gautier, 1982; ©Elsevier Science Publishing Co., Inc., 1987 52 Vanderbilt Ave., New York, NY 10017

1986). For these and other applications, accurate calibration of the satellite sensors is essential. A 5-5% (or better) accuracy is often desirable (Slater, 1984; 1985). If the calibration requirements are not met, not only will significant changes in the retrieved geophysical parameters not be detectable, but the results may also turn out to be unreliable and misleading (e.g., comparisons of different data sets become nonsensical). Careful preflight laboratory calibrations of the VISSR/VAS and AVHRR short-wave length channels were performed, respectively, by the Santa Barbara Research Center (Staff Members, 1980) and by ITT Aerospace (Lauritson et al., 1979; Koczor, personal communication). Transfer functions were established to calculate satellite level radiances from telemetered digital sensor signals. These functions, however, may not be 00344257/87/$3.50

74

valid for the data acquired in operations because the orbital satellite environment differs from the steady state laboratory environment, and in-flight changes occur, as observed with other radiometers. Common problems include zero offsets in the signal conditioning circuitry and degradation of the optics with time. It is therefore important to maintain an in-flight check-of-calibration for the short-wavelength channels. The AVHRR, unfortunately, contains no calibration capabilities other than space viewing for its visible and nearinfrared channels. Onboard facilities exist for the VISSR/VAS visible channel (i.e., an optical subassembly with reduced size aperture for directly viewing the sun), but they are not being utilized because of uncertainties in the characteristics of the optical elements. One has therefore to rely on indirect methods to calibrate the orbiting instruments. Various indirect calibration methods are available. One can utilize independent and coincident measurements from a suitably equipped high-flying aircraft (e.g., Smith and VonderHaar, 1980; Kriebel, 1981; Hovis et al., 1985) or by a calibrated satellite-borne instrument with similar spectral characteristics (e.g., Smith and Loranger, 1977; Brooks et al., 1984). Another possibility, perhaps the simplest way, as pointed out by Coulson and Jacobowitz (1972), is to use areas on the earth's surface as calibration targets (e.g., Koepke, 1982; Fraser and Kaufman, 1986). This last method is applied here to NOAA-7 AVHRR, GOES-5, and GOES-6 VISSR/VAS short-wavelength channels. We report on the results obtained for the period from October 1983 to January 1985.

R. FROUIN AND C. GAUTIER

2. Method of Calibration

First, a suitable calibration target on the earth's surface is selected. Then, the intensity of the radiation that is directed to space from the target to the satellite is calculated from the optical parameters characterizing the target and the intervening atmosphere, and is related to the sensor digital output. The procedure is repeated periodically to monitor temporal changes in the behavior of the instruments. The GOES-5 and GOES-6 VISSR/VAS visible channels comprise eight independent detectors, similar in spectral response [Figs. l(a) and l(b)] and sharing the same telescope and scan assembly. The analog signal from the detectors is digitized into 6-bit counts before its transmission to the ground, but 2 bits are added to each count at the NOAA Command and Data Administration (CDA) station in Wallops Island (i.e., users have access to 8-bit coded data with, however, only 6-bit accuracy). While the detectors respond linearly to incident radiance, the digitizing scale is not linear, but purposely adjusted to keep the signal-to-noise ratio linear with respect to digital count i.e., the digital count is proportional to the square root of the voltage response (Fig. 2). Thus, the radiance sensed by each VISSR/VAS visible detector varies as

L = GX 2 + I,

(1)

where X is digital count and G and I are calibration constants. It is important to emphasize that this relationship, which investigators normally use, is only approximate. At count value 48 on the 6-bit

1.0

GOES-5 VISSR/VAS

laJ (/3 Z

o

0.8

(/1 IaJ IZ

~> 0 . 6 t.-,< ._J Ld

cP"

a Ld N

0.4

-'i ~< 0 . 2 IZ: O Z

•

\

0.0 I

t

0.6

I

0.8

I0

1.

SPECTRAL WAVELENGTH 1.0-

B

(/zm')

OES- 6 VISSR/VAS

bJ U3 Z

°n

0.8

IZ

~ 0.6F

FIGURE 1.

o.4 r

i

1"4 '"

•,~ 0 . 2

o

0 .'6

'

0:8

'

1:0

SPECTRAL WAVELENGTH 1.0

NOAA-7 iI

LiJ (rJ Z

n°

~ \\

'

(#m~

- ~--

AVHRR

CHANNEL CHANNEL

1 2

\

0.8

\ \

n,"

> ._1 hi n~

L

0.6 L

0.4

S .J

~ 0.2 0 Z

0.0

I

0:6

\

'

0:8

I

SPECTRAL WAVELENGTH

[

1.0 (/a,m~

Spectral response of the solar channels.

76

R. FROUIN AND C. GAUTIER 5

++

+

MANUFACTURER

-- V=O.OO1226X2+O.02

++/~

/

4

/

v I-nD 3 }--

÷+++

o ..J

,,~

Z

+÷ ++

1

WSSR/VAS 0 110

210

310

410

510

610

DIGITAL OUTPUT (DECIMAL) FIGURE 2. Analog-digital VISSR/VAS transfer function: (+) manufacturer; (--) V = 0.001226X ~ +0.02.

scale, for instance, a 7% error in L is introduced when using the best fit of Fig. 2 for the analog-digital transfer function. Unlike the GOES-5 and GOES-6 VISSR/VAS visible channels, for which the spectral response of the detectors defines the optical bandwidth, the wavelength range of NOAA-7 AVHRR channels 1 and 2 [Fig. l(c)] is controlled by a filter. The voltage output of the detectors (one per channel), converted into 10-bit counts aboard the spacecraft, is linear as a function of sensed radiance (Lauritson et al., 1979), i.e., L = CX + I,

(2)

describes the relationship between counts and radiances. To determine the "constants" G and I, two calibration points (two X, L pairs) are needed. The White Sands National Monument area in New Mexico (hereafter referred to as White Sands), located at

32°52'N, 106°17'W and 1.22 km altitude, which is highly reflective, serves as one suitable high calibration point. First, the area is large, approximately 30 km in diameter, providing multiple sampling points and minimizing edge and environmental effects. Second, its reflectance properties are well-known and stable in time (Coulson and Jaeobowitz, 1972). Finally, favorable atmospheric conditions prevail in that region (minimum cloudiness, relatively low aerosol content). The area, however, is not horizontally uniform. A high water table keeps the very low parts permanently wet, modifying its reflectance. Part of the area is covered by dunes, and their shading is difficult to compensate for, thereby restricting the observations to conditions of small solar viewing zenith angles. The selection of small zenith angles is also important, as we shall see, to reduce uncertainties resuiting from unknown atmospheric characteristics. The second calibration point,

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA

77

at practically zero intensity value, is pro- parison with the solar component, so that vided by space. it can be ignored. Solar radiation in the 0.5-1.1 /~m As evidenced in Figs. l(a) and l(b), the spectral response of GOES-5 and GOES-6 wavelength range is scattered by air VISSR/VAS visible detectors may differ molecules and aerosols, and is absorbed in the long-wavelength half power point primarily by ozone, water vapor, oxygen, by as much as 0.1 /~m, which suggests and aerosols. Scattering and absorption that the calibration be effected separately processes interact complicatedly, but, forfor each detector. However, an average tunately, gaseous absorption can be calibration of the eight detectors is treated separately (Deschamps et al., accomplished for each satellite, using 1983; Tanr6 et al., 1985). The radiance average count data and average spectral measured by the space-borne sensor viewresponses [also shown in Figs. l(a) and ing the surface target can therefore be l(b)]. This procedure is justified because written with good approximation as the data which are made available to users by NOAA CDA have been L( M, Oo,O,eo) processed to eliminate displeasing stripes that arise from differences among the =fo~[FxEoxcOsOotgx(Oo,O) detectors. The output of the eight detectors is normalized to one of the detectors, xo' (U, Oo,O, )/,qax, (a) and the normalization is readjusted as necessary (approximately every couple of weeks). In other words, the data where )t is wavelength, 00 and 0 are calibrated one week may be out of respectively the solar and viewing zenith calibration the following week. Conse- angles, , is the relative azimuth angle, quently, our vicarious calibration of the F x is the spectral response of the detecGOES-5 and GOES-6 VISSR/VAS visible tors or filters, E0x is the extraterrestrial channels will be performed every 2 weeks solar irradiance (in Wm-2/~m-1), tgx is and, for convenience, we shall proceed in the transmittance due to absorbing gases, the same way for NOAA-7 AVHRR chan- and p~ is an equivalent reflectance of the target M. nels 1 and 2. The radiation geometry is depicted in Fig. 3. The equivalent reflectance p~ is 3. Radiative Transfer Model defined as 3.1. Model

In the spectral region of interest, 0.5-1.1 /~m, the sensed energy originates almost entirely from solar radiation which propagates through the atmosphere, is reflected by the surface target, and is finally transmitted to the space-borne sensor. The thermal radiation emitted by the atmosphere itself is negligible in com-

Oo,O,,) E0xcos 00 (4) where L~ is the radiance that would be sensed if the air molecules were nonabsorbing. The study deals with monochromatic quantities, but subscript ~ will be omitted hereafter to simplify the notation.

78

R. FROUIN AND C. GAUTIER SUN

SATELLITE

,s

/

TARGET FIGURE 3.

Radiation geometry.

The equivalent reflectance p* is modeled according to Tanr~ et al. (1979):

p*(M, Oo,O,,)

---po(M,

1-(p)(M,Oo,O,@)r ×[Ap(M,8o,O,~)+Bp(M,Oo,8,*) -C(p>(M, (5)

where Pa is the intrinsic atmospheric reflectance (i.e., the contribution of the signal backscattered into space without sudace reflection), r is the spherical albedo of the atmosphere, p is the bidirectional reflectance of the target, and ~ and ( p ) are average angular and environmental reflectances, respectively. In Eq. (5), A, B, and C are development coefficients characterizing the relative importance (or

weight) of reflectances p, ~, and (p). Detailed expressions for ~, (p), A, B, and C can be found in Tanr~ et al. (1979). 3.2. Solar irradiance

The solar irradiance incident at the top of the atmosphere is taken from Neckel and Labs (1984) and corrected for earth-sun distance during the course of the year. The total irradiance in the different channels, as obtained by convolving Neckel and Labs' (1984) data with the respective response functions, is given in Table 1. More energy flux is available in the spectral domain of the VISSR/VAS channels, because their peak of transmission is closer to the wavelength of maximum solar emission, and their bandwidth is relatively large. For AVHRR Channel 2, the larger filter bandwidth compensates for the smaller energy flux

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA TABLE 1 Extraterrestrial Solar Irradiance in the Different Channels and Integrated Filter Functions. SATEIaXm/Sra~SOR GOES-5 VISSR/VAS GOES-6 VISSR/VAS NOAA-7 AVHRR/1 NOAA-7 AVHRR/2

/~Fx dh

f~FxEoxdh

(/z m)

(W m - 2)

0.199 0.187 0.107 0.243

323.3 307.9 176.0 255.8

emitted by the sun in the near-infrared. In order to facilitate the conversion of total irradiances into monochromatic irradiances, we have also listed in Table 1 the values of the different filter functions integrated over wavelength. 3.3. Atmospheric functions

To evaluate tg, Tanr6 et al. (1985) parametrized gaseous absorption as a function of air mass and vertically integrated amount of absorber. They also established fairly accurate analytical formulas for po, r, A, B, and C which involve viewing geometry and atmospheric optical thickness (or turbidity). As an alternative, surface visibility, a parameter routinely measured at all weather stations, may be used instead of turbidity. This approach is selected for the present study. One should note, however, that the correspondance between visibility and turbidity is only a rough inverse proportionality (thick layers of aerosols may exist aloft with a very clear atmosphere at the surface). Became of unknown aerosol scattering properties at White Sands (i.e., size-frequency distribution, refractive index), we opted, among diverse aerosol models available, for the continental model of the International Radiation Commission (WCP, 1983). Although this model may not represent the actual conditions at White Sands, it certainly remains useful in setting limits to what might be reasonably expected.

79

3.4. Surface reflectance

The spectral reflectance of White Sands p was measured in the laboratory by Walraven and Coulson (1972) for different radiation geometries and for dry and wet conditions. In situ measurements were also carried out by Hovis (1966), but they are not applicable to the present study since they do not contain angular information. Figure 4 summarizes Walraven and Coulson's observations (1972). For dry sand, the spectral reflectance curve [Fig. 4(a)] is characterized by a rapid increase to 0.6 gin, followed by a slight dip around 0.7/~m, continued by a barely perceptible increase beyond 0.8 ~tm. For wet sand, the entire reflectance curve is displaced toward lower values by about 0.1, or 15%. The slope is steeper below 0.6 #m, and the depression at 0.7 /~m is more pronounced. Above 0.8 #m, as opposed to dry sand, the curve exhibits a negative declivity caused by water vapor absorption in the 0.94 /~m band. The shape of the curves below 0.6 #m is consistent with the general finding that, for sandy textured softs, the reflectance sharply drops in magnitude at wavelengths smaller than the average particle size. In interpreting Fig. 4(a), however, one should bear in mind that the portion of the curves extending beyond 0.9/~m results from a linear extrapolation of Walraven and Coulson's (1972) data. This questionable procedure is at least consistent with Hovis' measurements (1966), which exhibit a fairly linear wavelength dependence in the 0.8-1.0/~m domain. Important angular effects at 0.7 ftm are clearly seen in Figs. 4(b) and 4(c). Variations with the solar zenith angle (or equivalently with the viewing zenith angle because of the reciprocity principle)

0.9

00=15" 19=30*

LI.I (.)

z

I--

cb=9o"

0.8

bJ ..J h

~.

DRY

0.7

(/1 tm Z

~ 0.6 hi F--

0.5

o'.s

0'.6

o17

0'.8

WAVELENGTH

0'.9

1'.o

(/zm)

0.9

B

X =0. 7tzm 8=30*

hi C~

D R ~

dp=90*

~ 0.8 C) a.i _J I.i..

0.7 O3 1:3 Z

WET_ ~ - ~ -

0.6 L.IJ I--. -r

0.5

C)

'

2'0

r

SOLAR

0.9

,

ZENITH

6LO

ANGLE

L

8'0

(o)

h=O. ?/~rr~

80=15" 19=30"

LIJ

0

I-0 I.d --I l,

4,0

0.8

DRY 0.7

01 Z

0.6

/

k-r

~ 0.5 J 0

i

4I 0

RELATIVE

i

810

i

AZIMUTH

I 120 ANGLE

i

I 1 60 (°)

FIGURE 4. Reflectance of White Sands as a function of wavelength, solar and viewing zenith angles, and relative azimuth angle (alter Walraven and Coulson, 1972).

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA

are particularly significant for dry sand [Fig. 4(b)], i.e., 15% when passing from 0 ° to 6 0 ° for the specified conditions. The reflectance curve for wet sand is characterized by a nearly constant slope, and the overall effect is more moderate (only a 5% change in the 0-60 ° range). This behavior is similar at other wavelengths (not shown here): Generally, an isotropic distribution is observed at small zenith angles, all the more so as wavelength decreases. Almost no dependence on the relative azimuth angle is discerned for dry sand [Fig. 4(c)]. This is in agreement with the previous statement, because the zenith angles considered in Fig. 4(c) are small. At larger zenith angles a specular component superimposes the diffuse contribution, its importance increasing as the principal plane is approached (i.e., ~ = 180°). Even for solar and viewing zenith angles of 15 ° and 30 ° , however, the wet sand data [Fig.

81

4(c)] shows definite evidence of a specular component. The relatively strong influence of surface moisture upon White Sands reflectance must be taken into account in the radiative transfer calculations. Unfortunately, soil moisture is not currently measured at White Sands, and, anyhow, the usefldness of punctual or isolated soil moisture data, even recorded when the satellite views the target, is debatable. A correction is therefore estimated with the help of gravimetric measurements performed by Williamson (1975) during a 1 year period. These measurements are reproduced in Fig. 5 for two sites, one located in the flat portion of the area and the other in the gypsum sand dunes. Distinguishable in Fig. 5 is a seasonal dependence, with minimum values during summer (about 10%) and maximum values during winter (typically 30% in the flats and 20% in the dunes). Short-term

40 ~ D U N E S

35

- - - FLATS

30 Iii

W

25!

f

A

\\

!

\I

f~

5"5 2O 0

vv

15 10

if')

5 0

I

~)

1 0

J

I

1

200

L

I

300

JULIAN DAY (1974")

FIGURE5. Soilmoisturevariationsat WhiteSandsduringthe course of the year(alter Williamson,1975):(-) dunes;(---) fiats.

82

R. FROUIN AND C. GAUTIER

variations are also visible, and can be of importance (e.g., 20% over a few day period around Julian day 200 in the flats). The yearly averages are 21.4+5.3(lo)% in the flats and 17.8___3.2(lo)% in the dunes. Since the standard deviations for the 1 year period translate into surface reflectance changes of less than 0.6%, we conclude that the yearly averages adequately represent the actual moisture conditions at White Sands. We are aware, however, that this procedure is somewhat arbitrary because of the high spatial and temporal variability of surface moisture. Also, Williamson's measurements (1975), even though they correspond to clear or partly cloudy skies without rain, were taken only near noon (local), and we know nothing about the atmospheric events preceding these measurements. A better way to proceed would certainly consist of directly measuring the White Sands' reflectance at the very moment of the satellite pass, but such a practice is beyond the scope of this study. Although the distinction between fiats and dunes is not always apparent in the satellite data, we decided, for consistency, to restrict our analysis to the dunes area. In the simulations, consequently, the surface moisture is fixed at the yearly average value of 17.8%. Since Walraven and Coulson's data (1972) indicate that, in some degree, the reflectance of white gypsum sand varies linearly when the sand dries from a saturated state, the moisture effect at the intermediate value of 17.8% is simply computed by linear interpolation. 3.5. Average reflectances

The average reflectances ~ and (p) are laborious to evaluate because their calculation requires angular integrations and knowing precisely the spatial varia-

tions of the target reflectance. Fortunately, the difficulty can be obviated. First, A largely surpasses B and C in magnitude for clear atmospheres (i.e., for visibilities greater than 23 km) and nottoo-horizontal sightings (Tanr6 et al., 1979). In other words, the contribution of the signal directly reflected by the ground, which naturally involves the true reflectance of the target p, is dominant for sufficiently clear atmospheres. Second, r does not exceed the value of 0.2 at the wavelengths of interest ( h > 0 . 5 #m), even for considerable aerosol loadings. Third, ~ and (p) depart from p by less than 5% in most situations. These considerations, and the fact that ( ~ - p) and ( ( p ) - p ) are of opposite signs, strongly incline one to replace ~ and (p) in Eq. (5) by p. In so doing, the modeling of the apparent reflectance p* is drastically simplified and reduces to the case of a homogeneous Lambertian surface. The validity of the procedure is substantiated by Fig. 6, which shows the resulting error on p*. The error mainly depends on atmospheric visibility and zenith angles, and does not exceed ___1% in most situations. The apparent reflectance p* is either underestimated or overestimated, depending on visibility and radiation geometry. This interestingly indicates that, by neglecting the influence of the environment, the selection of clear atmospheres might not be optimum, particularly for those channels whose peak of transmission is located near 0.5 #m. Figure 6 reveals that AO*/O* is generally reduced at small zenith angles. Such angles are also favored to minimize errors arising from uncertainties in atmospheric characteristics (smaller optical path). In many cases, however, and this is especially true for a sun-synchronous satellite, it will not be possible to meet the conditions (visibility,

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA

83

~.=0.9,u,m

1.0

A=O.5#m

1.0

V=Skm

0.5

0.5 ~

0.0

0.0

*~ - o . 5

~. - o . 5

~" -1.0

~" -1.0

-I .5

-1.5

MOIST 0=30= (~=90

-2.0

MOIST 8=50 + •=go °

-2.0

°

I'0

2'0

3'0

4=0

5'0

()

6'0

1'0 2'0 3'0 4=0 5'0 SOLAR ZENITH ANGLE (°)

SOLAR ZENITH ANGLE (°)

A=O.5/~m

1.0

0.0

V = 2 3 k m ~

*~ -o.s -1

* ,,~ - 0 . 5 5-1.0

.o

-1.5

V=5km V=23km

V=5km

0.0

~"

).,=O.9g,m

1.0 0.5

0.5

6'0

I

-1.5

MOIST

MOIST

-2.0

-2.0

1'o

2'o

3'o

4=0

s'o

6'0

£)

1~0 2'0 3'0 4=0 5'0 VIEWING ZENITH ANGLE (°)

VIEWING ZENITH ANGLE (°)

1.0

1.0

0.5

0.5

6'0

X=O.9,u.m V=Skm V=23km

V=5km

0.0

.-. 0.0 V=23km

k -o.5

-o.s

~'-1.0 -1.5

-1.5

MOIST =15 ° 0%30°

8

-2.0 0'

'

MOIST 0o=15° e=3°°

-2.0 4'0

~

8'0

'

L I 120 1~ 0 RELATIVE AZIMUTH ANGLE (°)

I

0I

'

4'0

'

I I I 8L0 J 120 160 RELATIVE AZIMUTH ANGLE (°)

FIGURE 6. Error introduced in the modeling of p* by replacing ~ and ( p ) with p. Positive values indicate underestimation.

84

R. FROUIN AND C. GAUTIER

3.0

3.0

X=O.55~m

2.5

2.5

,---, 2.0

.--..,2.0

~=0.85~m

v

%.

%. 1.5

1.5~

~" 1.0 V=23km /

0.5

+

MOIST

0.5

MOIST

0.0

8=30° ~=go°

0=30 °

0.0

~=go° '0

2=0 3JO 4'0 5'0 SOLAR ZENITH ANGLE (°)

3.0 2.5

1'o

6'0

~

.

k

30

6'0

3.0

k=O.55~m ~

2'0

SOLAR ZENITH ANGLE (°)

m

k=O.85.u,m

2.5

,..---2.0

2.0

v

%` 1.5

k 1.5

1.o

"~" 1.o

0.5

0.5 MOIST

0.0

MOIST

0.0

[

1'0

20 3'0 4'0 5'0 VIEWING ZENITH ANGLE (°)

3.0

6'0

1'0

2.0

--- 2.0

k 1.5

%` 1.5

~'1.0

~'1.0 V=23km I

X=O.85/zm

4

I

I

V=23km I

'-I-

e=30°

0.0

8'0 . 120. . 160 . RELATIVE AZIMUTH ANGLE (°)

'

I

MOIST 8o= 15 °

e=30° 0'

V=5km

0.5

MOIST e . = 15 °

0.0

6'0

2.5

V=Skm

I

20 3'0 4'0 5'0 VIEWING ZENITH ANGLE (°)

3.0

X=O.55p, rn

2.5

0.5

V=23km I

+

4" 0

'

.

.

.

0

4' 0

'

8'0

'

120'

'

160'

RELATIVE AZIMUTH ANGLE (°)

FIGURE 7. Error in the modeling of p* for a Lambertian and homogeneous ground. The comparisons are made with exact calculations involving the successive orders of scattering method. Positive values indicate underestimation.

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA

85

coded on an 8-bit scale. The images were acquired at approximately 15-day intervals and cover the periods from October 3.6. Accuracy 1983 through January 1985 for GOES-6 and from October 1983 through July 1984 In order to assess the accuracy of our modeling of L, errors in the parametriza- for GOES-5 (until the failure of the sateltion of gaseous absorption and in Tanr6 lite). For each selected day, the data set et al.'s model (1979) should be consid- consists of three consecutive (i.e., half ered in addition to the errors that we hourly separated) clear sky images of just discussed above. According to Tanr6 White Sands and one image of space, all et al. (1985), the error on tg is smaller taken near the local noon. The NOAA-7 AVHRR data were chothan 1%, except at grazing incidence and observation angles (0o, 0 > 80°). Tanr6 et sen among orbit passes received in real al.'s model (1979) has not been validated time and archived at the Satellite Oagainst i n s i t u measurements, even in the ceanography Facility of the Scripps Insimple case of homogeneous and Lamber- stitution of Oceanography, San Diego. tian ground. Comparisons with exact They also cover the period from October calculations involving the successive order 1983 through January 1985. White Sands of scattering method indicate, however, generally falls within the acquisition range that for turbid atmospheres with visibili- of the ground station, but not always, and ties of about 23 km the model under- is often located near the east edge of the estimates p* by less than 1% in the wave- 2900-km-wide swath. The passage over length range of interest (Fig. 7). At X = the antenna site occurs at about 16:00 0.5/tm, this error happens to partly com- local time, which is relatively late in the pensate for the error introduced by our afternoon and, consequently, not optiapproximation of ~ and (O), but not at mum for calibration, especially during = 0.9 pm. In fact, by restricting our winter when the sun is low above the observations to reasonably clear condi- horizon. The direct readout High Restions (visibility of about 23 km), the olution Picture Transmission data absolute value of the overall error telemetered to the station are 10-bit in expected in our modeling of L should not precision (i.e., 1/1024) and the ground resolution at nadir is 1.1. km. exceed 2%. Figure 8 gives examples of White Sands images in the AVHRR and VISSR/VAS 4. Data short-wavelength channels. White Sands is easily distinguishable from the sur4.1. Satellite data sets rounding environment by its high brightThe VISSR/VAS data used for the ness (white tones). The shape of the area calibration were provided by the Space appears somewhat distorted from one Science and Engineering Center of the image to the other, a direct consequence University of Wisconsin, Madison, in the of the different viewing geometries. The form of full resolution i.e., - 1 . 5 km at demarcation between the flats and the the latitude of White Sands), 6-bit preci- more reflective dunes, clearly apparent sion (i.e., 1/64) 256×256 pixel images on the AVI-IRR images [Fig. 8(a)], is not radiation geometry) that correspond to the most desirable compromise.

86

R. FROUIN AND C. GAUTIER

FIGURE 8.

Examples of White Sands images in the AVHRR and VISSR/VAS solar channels.

discernible on the VISSR/VAS images [Fig. 8(b)]. The contrast, in fact, depends on the observation geometry. The VISSR/VAS images exhibit stripes, which illustrate differences among the detectors (see Sec. 2). This indicates that the destriping procedure applied by NOAA CDA is not completely efficient. Subimages of the White Sands dunes were extracted from the original satellite

data sets. Since no geometric corrections were applied to the data, the number of pixels contained in the subimages is not constant, but depends on the viewing geometry (we seek to include as many pixels as possible to improve the statistics of the calibration results). Since our objective for the VISSR/VAS is to perform an average calibration of the eight short-wavelength detectors, eight con-

CALIBRATION

OF

NOAA-7

AND

GOES-5

AND

87

-6 DATA

secutive lines form the subimages. The space counts at the beginning of the AVHRR scan lines that correspond to the White Sands viewing and the entire 256 × 256 pixel VISSR/VAS images of space constitute the zero intensity data set used in the computations. Average White Sands and space counts and their respective standard deviations are presented in Fig. 9. Over the study period, the average space count is practicaUy constant for AVHRR Channels 1 400

A

NOAA - 7

and 2 [Figs. 9(a) and 9(b)] and does not significantly change for the VISSR/VAS short-wavelength channels [Figs. 9(c) and 9(d)]. No appreciable shift or trend can be detected. The average White Sands count, on the contrary, varies strongly with time. The low frequency fluctuations reflect those of the solar irradiance at the top of the atmosphere since, for each satellite, the observations were taken at approximately the same time during the entire study period. In some in400

B

N OAA - 7 A VHRR/2

300

300

b

Z ::D 2 0 0 0 (,3

Z 200 0 C.)

100

100

SPACE

SPACE

0

i

SEP83

i

L

i

DEC83

i

i

,

MAR84

i

i

i

JUN84

,

~

i

SEP84

i

i~

0

DEC84

i

SEP83

i

i

i J - ~ l

DEC83

DATE

MAR84

I JUN84

i

i .J_~l

SEP84

i

DEC84

DATE

250

C

GOES-5 IrISSR/VAS

250

D]

tOES-6 VISSR/VAS

! i

200

200

~

I--- 1 5 0 Z 0 (")

~ W H T E

SANDS I-- 1 5 0 Z :E) 0 100

100

50

50 I~

SEP83

SPACE :: :~--I

. . . . .

DEC83

MAR84

JUN84

DATE FIGURE

9.

SPACE ~k-II I I I I l ~ - * ~ - ~

SEP84

DEC84

SEP83

DEC83

MAR84

11 ~ : : : z l

JUN84

DATE

White Sands and space brightness counts during the study period.

SEP84

DEC84

88

R. FROUIN A N D C. GAUTIER

~o ,

16o ............NOAA--?

140

._1 (D

z

100

i

---GOES--5 ..... GOES~6

i

N

1 20

<~

7-

8o

100 80

NOA~ - ? GOES--5 GOES-6

..,.........j'x,

/:

z "~ --r

I--

,,~

............ ----.....

o~

//"

./

60

"""""....,

//

I---

~,

6O

40

N

>

40

"~ _J LO

2O

n-

0

/

.~

_..I

.............

"~'iI11:>-"i;(

X/

2o

o

'>'' "--';~ 'dl

/

/

t

/'J

0 SEP83

DEC83

MAR84

JUN84

SEP84

DEC84

SEP83

DEC83

MAR84 JUN84

SEP84

DEC84

DATE DATE

~" 3.5 I c.:, 3 . 0 (D

............

NOAA-?

- -

GOES-5 GOES-6

.....

v

../'i

60

............

.VOA.4 -- ?

- -

GOES-5 GOES-6

.....

W ._1

50

~2.5

Z

Z

o

-I- 40

2.0 i/

rt~

1.5

.

", i ,.

,/[

I--

'', ,L.

0

1.0

~: 0 . 5 I.-

SEP83

\

z:/ v

DEC83

".i~ '/

,,,,,,--:.,/

~/'-:

MAR84

"~v..../'

JUN84

SEP84

~i.i N

30

~ z

2O

v

V ii

SEP83

DEC84

DEC83

MAR84

JUN84

:

',i SEP84

DEC84

DATE

,.-...32

............NOd.A-- 7 ----

GOES,-.5

..... ~o,~-~ /

I--

:~ /"\

/\

.31

\

/,

o .30 7 0

i"

V

IM

DATE

~-

ii i~

.29

.< L,J .28 Z O

"X \ , / /

N

o

.27 SEP83

i

i

i

DEC83

i

i

i

i

MAR84

J

i

JUN84

i

J

J

SEP84

i

i

i

i

DEC84

DATE F I G U R E 10. Viewing geometry and atmospheric chaxaeteristics at the moment of satellite overpass: A: ( - - . ) NOAA-7; B: ( - ) COES-5; (---) COES-6.

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA

89

stances, abrupt changes are observed at changed to about 38 °. The NOAA-7 about 2-week intervals (e.g., between 2 viewing zenith angle is subjected to more February and 8 February 1984 for important, somewhat erratic changes, GOES-5) or from one image to the next which in fact depend on the distance of during the same day (e.g., on 30 Decem- the satellite subtrack to White Sands (the ber 1984 for GOES-6). The spatial (hori- closer the satellite passes the area, the zontal) standard deviation is generally smaller the viewing angle). small (2-6 counts) for AVHRR Channels Successful capture by the tracking an1 and 2, but often exceeds 10 counts for tenna generally implies that NOAA-7 is the VISSR/VAS short-wavelength chan- passing westward of the target. Since, on nels, and may result from surface mois- the other hand, the data are acquired tttre variations, inhomogeneities in the when the sun is in the western quadrant, surface composition, local changes in the sun and the satellite are viewed by atmospheric properties, intrinsic instru- the target from about the same azimuthal mental noise, differences among the de- direction, thereby leading to small relatectors (for GOES-5 and GOES-6), and tive azimuth angles. On one occasion (30 instrmnent changes. November 1984), however, the satellite passed close to, but east of White Sands and a relative azimuth angle of nearly 4.2. Viewing geometry 158 ° was observed. For GOES-5 and The angles characterizing the inci- GOES-6, the relative azimuth angle dence and observation directions are dis- rapidly varies from one hour to the next. played in Fig. 10 for each satellite. Dur- This is simply due to the fact that the ing the study period, the solar zenith observations are taken near the local noon, angle varied from 10 ° to 60 ° for GOES-5 i.e., near the maximum solar elevation and GOES-6, and from 45 ° to 80 ° for during the day. The amplitude of the NOAA-7. The low frequency fluctuations fluctuations is larger during the summer exhibit a sinusoidal form, which is in- because the sun is higher at noon during timately related to the annual cycle of the the season. Note also that before the sun. The NOAA-7 solar zenith angles are failure of GOES-5, the GOES-5 and generally large, which makes their selec- GOES-6 relative azimuth angles vary in tion for an accurate calibration question- opposite direction during the day. The able. But we are limited to the afternoon reason for this is that the two satellites NOAA-7 passes, and they occur when the were viewing White Sands from direcsun is already low. tions located respectively east and west of The viewing zenith angles from the the target. Indeed, after the failure of geostationary altitude remain fairly con- GOES-5, the same type of variations as s t a n t (i.e., - 50 ° for GOES-5 and - 49 ° those observed for GOES-5 are encounfor GOES-6) until GOES-5 failed in July tered for GOES-6, except that the magni1984. GOES-6, stationed over the Equa- tude of the relative azimuth angle is torial Pacific at 135°W at the moment of smaller, a consequence of the larger the failure, was then maneuvered over GOES-6 azimuth angle (the GOES-6 subthe United States to the longitude of point on the earth is closer to the White 85°W, and the viewing zenith angle Sands meridian).

90

R. FROUIN AND C. GAUTIER

4.3. Meteorological parameters The vertically integrated ozone amounts were estimated from the meteorological atlas of London et al., (1976). At the latitude of White Sands, the values range from 0.27 cmatm in autumn to 0.32 cmatm in spring. The vertically integrated water vapor amounts were obtained from radiosonde observations at E1 Paso (31°48'N, 106°24'W), a meteorological station located at nearly the same altitude and longitude as White Sands. No corrections for the latitudinal distance between E1 Paso and White Sands (about 120 km) were effected. This procedure is roughly correct because the atmospheric disturbances usually travel eastward in that region. The atmospheric visibility is that visually observed at meteorological station HMN, a station very close (a few km) to the calibration site. Unfortunately, a standard value of 25 km was reported continuously during the entire study period. The actual visibility might in fact be much larger in many cases: Values of 100 km or more have been frequently observed (Slater, personal communication). Our deficient knowledge of the visibility at White Sands is indeed a potential source of error for the calibration, which can be estimated (see below) but not accounted for. At White Sands, the atmospheric pressure is considerably reduced because of the surface altitude. This was carefully accounted for in the computations of the Rayleigh optical thickness. 5. Results The satellite radiances computed for White Sands, with the data and the model discussed above, are presented in Fig. 11. They correspond to the brightness counts

of Fig. 9. Large variations on the seasonal time scale are observed of the order of respectively 80 and 60 W m -2 for AVHRR Channels 1 and 2 [Figs. l l(a) and ll(b)] and 80 W m -2 for the VISSR/VAS visible channels [Figs. ll(c) and ll(d)]. For consecutive times during a given day, the VISSR/VAS radiances vary by only a few W m -2 because the sun zenith angle does not change much during the 1.5-h period of the observations. The modeled radiances fairly well reproduce the seasonal or even monthly variations of the counts, but not those existing at smaller times scales, especially from one hour to the next in the case of GOES-5 and GOES-6. This is partly due to our method of specifying the atmospheric parameters, which does not account for hourly changes. Also, the short-term variations in the counts may result from instrument modifications, which indeed are not considered in the method. To evaluate the accuracy of the simulated radiances, the following uncertainties were assumed: 30% on the anisotropy factor of the aerosol phase function, 30% on the atmospheric visibility, 10% on the ozone amount, 20% on the water vapor amount, 50% on the soft humidity, and 1% on the surface reflectance (due to processes other than soil humidity). Uncertainties on other aerosol parameters, such as their vertical distribution, are not significant to the first order. The 10% uncertainty on the ozone amount corresponds to daffy fluctuations around the geographical monthly mean (London et al., 1976). The 20% uncertainly on the water vapor amount is typical of radiosonde observations. The 50% uncertainty on the soil humidity is representative of the annual variations reported by

C A L I B R A T I O N O F NOAA-7 AND GOES-5 AND -6 DATA

A

NOAA- 7 A VHRR/1

tw bO

91

NOA.A - 7

aVHnn/a

30

7

;

'

o4

T

!t

T

20 b..I

N .<

/

T ![

o z<

Z

N <

10

,

SEP83

,

l

~

DEC83

,

J

,

MAR84

i

i

,

JUN84

,

i

~----~,

SEP84

.

i

'~{

i

DEC84

\

10

SEP85

,

i

i

i

DEC83

DATE

60

i

J

IdAR84

,

i

,

i

JUN84

i

SEP84

,

,

DEC84

DATE C

GOES-5 wssa/vAs

D

toes-6 60

VISSR/'V,I S

l

I

U3 i

50

7

50

'"

40

~ z <

4O

z <

tw

50

3o L

SEP83

i

DEC83

i

i

MAR84

i

i

JUN84

i

i

SEP84

i

i

i

DEC84

DATE FIGURE 11.

i

SEP85

DEC85

i

MAR84

i

,

i

JUN84

i

,

SFP84

,

DEC84

L

DATE Modeled radiance from White Sands during the study period.

Williamson (1975). The 1% uncertainty on the surface reflectance aceounts for errors in Walraven and Coulson's measurements (1972). The 30% uncertainty on the asymmetry parameter is characteristic of differences between the extreme cases of maritime and urban aerosols. Finally, the 30% uncertainty on the atmospheric visibility is somewhat arbitrary because the visibility at the calibration site is poorly known. The question arises: Is the standard 25 km value consistently reported by meteorological station HMN

representative of the average conditions at the calibration site? If not, and recent observations tend to indicate that the visibility at White Sands is often larger ( > 100 km), the modeled radiances are consequently biased (i.e., underestimated) be several percent, more precisely by 2.5% if the average visibility is 100 km. Thus, except for the visibility, the uncertainties estimated for the atmospheric and surface parameters are reasonable, and would most likely maximize the overall error on the modeled radi-

92

R. FROUIN AND C. GAUTIER TABLE 2 Uncertainties on Model Parameters and Simulated Radiances

AL/L (%)

aP/P P,~,X~mTEa

Corresponding Errors on

(%)

VISSR/VAS

AVHRR/1

Anisotropy factor Visibility

30 30

0.5 0.9

0.7 1.1

0.6 1.0

Water vapor amount Ozone amount

20 10

0.2 0.6

0.2 0.7

1.2 < 0.1

Surface reflectance Soil humidity

1 50

1.5 2.2

1.4 1.9

1.3 2.1

antes. On top of these uncertainties, one should not forget the uncertainty on the radiative transfer model itself (about 2% for most situations encountered in the study; see See. 3.5). When introduced together in the model, the uncertainties discussed above yield the errors shown in Fig. 11, computed for all individual cases. These errors range from 8 to 13%, and might explain the inability of the method to retrieve the short-term variability in the brightness counts. To apprehend which are the most critical parameters for the calibration, the different uncertainties were considered separately with average conditions for the atmosphere. The results are presented in Table 2. Clearly, the most contributing errors are those on soil humidity, surface reflectance, and atmospheric visibility. The influence of the anisotropy factor is relatively limited, because the effect of this parameter on p~ on the one hand and on A, B, and C on the other hand [see Eq. (5)] partly compensate. For the ozone amount our treatment based on climatological data appears to be sufficient. For the water vapor amount, ff radiosonde observations are satisfactory for the VISSR/VAS visible channels and for AVHRR Channel 1, they yield an error of

AVHRR/2

about 1.2% for AVHRR Channel 2. This error, as most of the other errors, can be reduced, for instance, by using optical instruments on site (e.g., pyrheliometer for the turbidity, Voltz photometer for the water vapor amount). This would require, however, increased means and effort, and will be discussed in the final section as part of a global strategy to control the quality of the data acquired in orbit. From the White Sands and space average counts (Fig. 9), the modeled radiances (Fig. 11), and assuming that the space radiance is null, average calibration coefficients are calculated. This is done by comparing the modeled radiances with the corresponding AVHRR 10-bit counts [Figs. 12(a) and 12(b)] or VISSR/VAS 8-bits counts squared [Figs. 12(c) and 12(d)]. The correlation coefficient is better than 0.995 at the 99% confidence level for both types of instrument. Assuming that the calibration points should fit a straight line of the form (1) or (2), the best slope and intercepts estimates are computed using the principle of maximum likelihood, and are given in Table 3. In the computations, however, the linear regressions were forced through the space points, because the radiance of space

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA 25

A

CORR.=0.999

GA,rN=o.0$81 W.M-2.SR-I/COUNT INTERCEPT=--2.42 W.M-2.SR-~ RES. ERR.=0.53 W.M-I.SR -I

2O

93 35

30

×

25

0"}

N Z

lO

'~

5

x×

x/X

~'" 15

15

/

10

NOAA-7 AWRR/1 0'

'

1; 0

200 '

'

'

36o

'

i

400 '

.

NOAA-7 AVHRR/2 .

.

.

200

C

CORR.=0.997 GAJ'N=1.632X10 ~ W.M-2.SR-1/COUNT2 INTERCEPT=-I.01 W.M-2.$R-1

G~N= 1.617X10-5 W.M-:~.SR-I/COUN'r2 INTERCEFF=-0.81 W.M-=.SR-1 ~,~ RJ~S. ERR.=2.64 W.M-2.$R-I ~ / xXXx~ x xx

i

(f)

? 401

, 4 0

COUNT

cORR. =0.997

T'~'50

~ 3 0

.

100

o

COUNT

60

x

20

<

v

B

E ' .2~' f~-,-E'w.j'~~:;.:_",-I CORR.=0.998 G,MA'=0.1007 W.M-Z.SR-1/COUNT

, < /

RES. ERR.=2.78

I

-

/

W.M-Z.SR-I ~III~ XxX

I

x>~ x

D! /'

× Xx

30

7O ~, 20

Z

-~ 20 loi

10

GOES-6

VISSR/VAS 1

0

I

I

I

10000

I

20000

I

I

30000

I

VISSR/VAS

I

1

40000

0

COUNT2

I

I

10000

I

20000

I

I

30000

I

I

40000

COUNT 2

FIGURE 12. Count (or count squared) vs. radiance calibration curves. Corr, gain ( W m - 2 s r - l / c o u n t or W m - a s r - 1 / c o u n t 2 ) , intercept ( W m - 2 s r - l ) , res. err. ( W m - a s r - ] ) : A) 0.999, 0.0681, - 2.42, 0.53; B) 0.998, 0.1007, - 3 . 7 3 , 1.00; C) 0.997, 1.617 × 10 -3, - 0 . 8 1 , 2.64; D) 0.997, 1.632 × 10 -3, - 1.01, 2.78.

TABLE 3

Average Post-Launch Calibration Coefficients GAIN

SATEta~rr~/Sv_~SOB

(W m - z sr - ] / c o u m : a o n W m - Z s r - 1 / c o o N ' r 2b )

IrcrEncEPT (Wm-~sr -] )

GOES-5 VISSR/VAS GOES.6 VISSR/VAS NOAA-7 AVHRR/1 NOAA-7 AVHRR/2

1.617 X 10-3 +0.022× 10 -3 1.632 x 10-3 + 0.020 × 10 - 3 0.0681 _+0.0006 0.1007 + 0.0013

-0.8+0.6 -1.0+0.5 - 2.4 _+0.1 - 3.7 +_0.2

"AVHRR. bVISSR/VAS.

94

R. FROUIN AND C. GAUTIER

(equal to zero) is not subjected to uncertainty and, furthermore, the time variability of the space counts is small. The residual errors are small, i.e., about 5% of the mean radiance value for the VISSR/VAS visible channels, and only 3 and 4%, respectively, for AVHRR channels 1 and 2. This provides a good idea of the level of stability of the calibration coefficients, and tends to indicate that our assessment of the errors in the modeled radiances is pessimistic, but the residual errors do not include eventual biases. Note that Fraser and Kaufman (1986) found for the period from July 1981 through August 1983 a GOES-5 calibration which was only about 7.5% lower than the results reported in Table 3. 6. Discussion

One of the goals of the calibration is to assess whether the instruments are experiencing changes (e.g., drift) as a result of performance degradation in flight. This can be verified by examining the time evolution of the calibration coefficients, and by comparing the results with those obtained in the laboratory before launch. In the following, the cases of AVHRR Channels 1 and 2 and of the VISSR/VAS visible channels are discussed separately. 6.1. AVHRR channels 1 and 2

Figure 13 depicts the time evolution of the AVHRR Channels 1 and 2 calibration coefficients. In the figures, the error bars associated with individual points represent the differences due to spatial variations in the White Sands counts. There is no evidence of a systematic drift in any of the computed coefficients and no dramatic change is observed during the 15-month period of the observations. The

gains fluctuate somewhat erratically throughout the study period, by 11 and 15% ( l o ) of the mean values for Channels 1 and 2, respectively. The larger excursion of Channel 2 gain likely results from the higher sensitivity of this channel to water vapor (the effect of uncertainties on the water vapor amount is increased). These fluctuations, however, do not appear to be significant at the level of accuracy of our calibration; the errors in the modeled radiances may easily account for the observed variability. It is interesting to note that the gains are highly correlated to the viewing zenith angle. The correlation coefficients are - 0.77 (Channel 1) and - 0.83 (Channel 2), significant at the 99% confidence level. This intriguing feature cannot be attributed to errors in the parametrization of gaseous absorption, or even to the fact that absorption-scattering interactions are neglected, because absorption processes act quite differently in the two channels, which would not produce similar correlation coefficients. Also, when examining the model errors and their dependence on viewing geometry (Figs. 6 and 7), there is no evidence that an increase in viewing zenith angle would be reflected in a decrease in the gains. On the contrary, one would conclude that O* is slightly more overestimated at higher viewing zenith angles, yielding a positive correlation. One possible explanation is linked to surface reflectance inhomogeneities. In fact, shading effects are enhanced when the sun is low (recall that 0 > 50 ° for most conditions), which reduces the apparent target reflectance at small viewing zenith angles because incidence and observation directions have similar azimuth angles. In other words, O* is more overestimated as 0 decreases because more shaded areas

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA

95

.09 NOAA - 7

,.~ 7

o

z ~) r~

,08

°

f./3

~

T

!i

T

.06

o

T

L T

........

z .<

j

T

Ii

l

SEP83

.10~

.09

,

;

.... _"

t

;

o

; _

I i

_t . . . . . . . . . . . .

t

~

i

DEC83

i

i

i

i

i

~

MARB4 JUN84

i

~

i

i

SEP84

i

DEC84

SEP83

DEC83

MAR84 .JUN84 SEP84

DATE

-1.6[

DEC84

DATE

-2.0

NOAA-7

NOAA - 7 A VHRR/2

l AVHRR/I -2.C

i T

.08

I .05

.12

5rY .11 ~4

z <

NOA.A -- 7

AVHRR/2

.13

a VHnR/,

....................................

~"

-3.0

i

I' 13

•

'"

c~°

: I

{'{ {T

'

4'

{'

i

-2.8

---

i

{

wry

-4.0

{

{

{

{

{l

t

;

l

},

L

- 5 . 2 [, SEP83

. . . . . . . . . . . . . . . . DEC83

MAR84

JUN84

SEP84

i DEC84

DATE

--5.0

±

SEP83

t

J

1

DEC83

•

i

i

MAR84

i

,

~

JUN84

~

J

J

SEP84

L

~

L

DEC84

DATE

FIGURE 13. AVHRR calibration coefficients during the study period. Prelaunch values are represented by dashed lines.

occupy the instrument field of view; in the model these areas are implicitly assumed to be illuminated by the sun. The effect is almost impossible to quantify, however, because this would require precise knowledge of the shape and the geometry of the dunes and other spatial inhomogeneities; but it might realistically explain the irregular variations of the gains observed in Fig. 13. When compared to prelaunch gains, the average gains obtained in the present

study read higher by about 14% for Channel 1 and 15% for Channel 2 (Fig. 13). The differences are significant and cannot be explained by model errors since these errors do not contain biases of such magnitude. In fact, uncertainties in the atmospheric visibility likely introduce a negative bias (see Sec. 5) which would aggravate the disparities. Hovis (1982) pointed out several problems in the technique applied by ITT Aerospace to calibrate the AVHRR solar channels, from

96

R. FROUIN AND C. GAUTIER

the standpoint of the calibration in the laboratory with a sphere and from the fact that there still is no agreement on the value of the solar irradiance in the bands of interest. We used Neckel and Labs' solar data (1984) instead of Thekaekara's tables (1974), but this cannot quantitatively explain the discrepancies. A likely possibility is therefore that the solar channels degraded after launch. Since Fig. 13 indicates little or no drift, the solar channels might have deteriorated suddenly. It would certainly be worthwhile to check whether a degradation took place as a result of launch constraints. So far we have only analyzed changes in the Channels 1 and 2 gains. Comparing intercepts determined in orbit with those before launch (Fig. 13) shows that the post-launch values are lower by 2 W m -2 (Channel 1) and 3 Wm -2 (Channel 2). The time variations in the intercepts reflect those in the gains, but naturally are inversely correlated since two-point calibrations were performed. We recall here that the zero offsets (in counts) were found practically constant for the two channels, which is evidence of the system stability. 6.2. VISSR/VAS visible channels In order to assess the stability of the VISSR/VAS solar channels with time, average calibration coefficients were computed for each selected day of the study period. For a given day two-point calibrations were performed using space and each of the three White Sands observations, and the results were averaged to provide a daily value. Figure 14 shows the time evolution of the gains and intercepts obtained in that way. The error bars in the figure represent the variability of the coefficients during the 15 months

of the observations. The gain changes, about 12% (GOES-5) and 15% (GOES-6) of the mean values, are not as erratic as those of AVHRR Channels 1 and 2. Still, they remain within the uncertainty domain of our calibration (e.g., model errors, transfer function uncertainties, and the limitations due to the 6-bit digitization). We note, however, that the VISSR/VAS viewing zenith angles, unlike the AVHRR ones, did not vary much throughout the study period, except for GOES-6 immediately after the failure of GOES-5 (when GOES-6 was maneuvered to its new position). The solar zenith angle, on the contrary, varied noticeably and a correlation coefficient of 0.66, significant at the 99% confidence level, is computed between the GOES-6 gain and the solar zenith angle. In fact, at fixed or minimally varying viewing zenith angles, O* and, consequently, the gains are more overestimated as 00 increases because, here again, the instruments view more shaded areas. This might be the cause of at least part of the variability in the GOES-6 gain. Such a high correlation, however, is not observed for GOES-5. On the contrary, from March 1984 until the failure of the satellite, during which 00 decreased, an increase in gain is noted (Fig. 14). This reversed trend might therefore be associated with changes in the system that eventually lead to its collapse. It is interesting to note here that Fraser and Kaufman (1986) observed a 15% loss in the responsivity of SMS-2 solar channels in the last year of service of the satellite. No systematic effect of the normalization procedure can be detected. However, an abrupt increase in the GOES-5 gain (i.e., a loss of sensitivity) of about 17% of the mean value is observed be-

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA

ff-

GOES--5

ft.

VISSR:"VAS

I

~TSSR, VAS

~< 2.2

2.2

7--

Z

7

(Z)

(Z)

T

1.a

¢Y (/3

1.8 . . . .

?

7 ~

GOES-6

I

0

0

5

97

1.4-

~ " 1.4 Z

Z .,¢

r.5

i

SEPB3

DECB3

MAR84

JUN84

SEP84

DEC84

SEPB3

DEC83 M A R 8 4 JUN84

DATE 0.0

0.0

GOES--5

GOES- G VISSR qrAS

[

I

n,"

rY

co - 0 . 5

u') - 0 . 5

'7

'7

~E

v

- I .0

z

®

- I .0 (3_ Ld 0 13C I,I

L) n," i,i l.--

S E P 8 4 DEC84

DATE

VISSR /VAS

I.-13-

L

-1.5

- I .5

L

SEP85

i

i

i

DEC83

,

,

i

i

i

i

MARB4 JUN84

i

i

~

i

,~

S E P 8 4 DECB4

SEPB5

DECB3 MARB4 J U N B 4 SEPB4 DECB4

DATE FIGURE 14.

DATE Same as Fig. 23, but for VISSR/VAS.

tween 2 February and 18 February 1984, system in orbit at that time. Important which cannot be explained by shading variations are also observed during cereffects or inhomogeneities in the surface tain days, for instance on 4 October 1983 reflectance. The fact that on 3 March the for GOES-5 and 30 December 1984 for gain recovers its value of 18 February GOES-6, and correspond to those cases might be the manifestation of the normal- for which the White Sands brightness ization procedure or other intervention at counts vary considerably over a 0.5-h inthe ground station. Note that the phe- terval. Because the changes do not last, nomenon occurred just before the solar they cannot originate from the normalizachannels began to lose sensitivity, which tion procedure. A possible explanation is might therefore indicate that something a n error at the ground station that is drastic and irreversible happened to the immediately corrected for.

98

From the prelaunch calibration curves provided by the Santa Barbara Research Center and the actual scanner temperatures communicated by the NOAA GOES support branch, prelaunch gains and intercepts were computed and are displayed in Fig. 14 against postlaunch values. The scanner temperatures vary from about 31°C in winter to 10°C in summer, which translates into a significant increase in the sensitivity of the GOES-5 detectors (8%) and a slight decrease ( < 1%) in that of the GOES-6 detectors. Between certain detectors, however, deviations of up to 4% from these average figures may exist. The seasonal changes in the heating effect are not retrieved at all in the postlaunch values of Fig. 14; they are masked by the more important variations in the apparent reflectance discussed above. The thermal effect on the intercepts is obviously not visible. These, in other respects, depart from the prelaunch values by - 1 Wm -2 (GOES-5) and - 3 W m -2 (GOES-6), which generally represents a negligible contribution to the signal measured by the instruments. Our analysis thus indicates a fairly good agreement between pre- and postlaunch calibration coefficients, the discrepancies being largely attributed to spatial inhomogeneities in the surface reflectance. It is plausible, however, that the GOES-5 solar channels deteriorated during the period from March 1984 until the satellite failure.

7. Summary and Recommendations The potential of measurements from space of the solar radiation reflected from the earth-atmosphere system for environmental applications has been extensively

R. FROUIN AND C. GAUTIER

illustrated in the recent literature. Vegetation index, ground and cloud albedoes, surface insolation, and radiation budget at the top of the atmosphere represent an incomplete list of parameters that are discussed in terms of satellite observations in the visible portion of the solar spectrum. Most of the satellite instruments to be used in determining these parameters, unfortunately, are insufficiently characterized and calibrated, except those on the Earth Budget Radiation Experiment. Such is the case of the VISSR/VAS and A V H R R radiometers onboard the meteorological satellites operated by NOAA, which do not have active inflight calibration capabilities for the solar channels. A simple, probably the least expensive, way to routinely calibrate the solar channels of these orbiting instruments is to use areas on the earth's surface as calibration targets. This technique was applied in the present study to the AVHRR onboard NOAA-7 and to the VISSR/VAS onboard GOES-5 and GOES-6. The White Sands National Monument area in New Mexico, of well-known bidirectional reflectance, was used as reference standard. Observations taken with the sensors pointing at deep space served to fix the offset points on the count (or count squared) vs. radiance curves. The satellite level radiances were computed from the ground target characteristics and the optical parameters of the overlying atmosphere using a fairly accurate radiative transfer model, which accounts for multiple scattering and bidirectional effects. Atmospheric properties were determined either from climatological data (ozone amount, aerosol phase hmction) or from observations at the nearest meteorological site (water vapor amount, visibility). The

CALIBRATION OF NOAA-7 AND GOES-5 AND -6 DATA

calibration accuracy is believed to be in the range 8-13%, depending on viewing geometry, surface and atmospheric conditions, and channel considered. The results obtained for the 15-month period analyzed indicate no significant drift in the solar channels of both types of instrument. The gain changes (about 15% of the mean values) are largely attributed to spatial inhomogeneities in the calibration site (shading effects due to the presence of dunes) that are not accounted for in the modeling. There is some evidence, however, that the GOES-5 solar channels gradually deteriorated from March 1984 until the failure of the satellite in July 1984. Comparing gains determined in orbit with those before launch shows that the AVHRR solar channels read higher after launch by about 15% on the average. This apparent loss of sensitivity cannot be explained by model errors, whose bias component tends to aggravate the discrepancies. The accuracy of the calibration could certainly be improved by measuring in situ all the necessary parameters at the moment of the satellite overpass. This, however, would require increased means and effort. In other respects, the nonnegligible uncertainties introduced by the inhomogeneity of the target might be avoided by performing the calibration in the fiat portion of the White Sands area. A detailed validation of the radiative transfer model is also necessary. In order to accomplish this a complete data set on atmospheric properties and surface reflectance (including radiation measurements) has been recently acquired for the Yuma desert in Arizona. The Yuma desert, in fact, appears to be a near-homogeneous, suitable calibration target. Its use with White Sands would place more con-

99

fidence in the results. In any case, the calibration of the GOES VISSR/VAS and NOAA AVHRR solar channels should be addressed in the context of a global and concerted strategy (e.g., see IAMAP, 1981; WCP, 1986). In view of the requirements for many applications (in particular climate studies), the only way to ensure that the observed variability in the satellite-derived parameters is the result of changes on earth and not in the instruments is to calibrate the data acquired in operations on a continuous basis. As demonstrated in the present study, this can be achieved with our simple and relatively inexpensive method. Nevertheless, episodic (i.e., once or twice during the life of individual satellites) but intensive process-oriented campaigns that involve the utilization of high-flying aircraft should be conducted, even if they are onerous, for they constitute the optimum way of assessing the data quality.

The authors are indebted to D. Tanr~ of the University of Lille, France, who made available his radiative transfer model. The programming support of B. Dilulio and R. Wylie is gratefully acknowledged, and the authors thank R. Markworth for typesetting the manuscript. The work has benefited from discussions with R. Hoelter of the Santa Barbara Research Center, P. Slater of the University of Arizona, D. Forness of the NOAA GOES support branch and C. Whitlock of the NASA Langley Research Center. The detailed comments of an anonymous reviewer have improved the paper considerably. The research effort was supported by NASA under Contract NAGW-697 and by the California Space Institute.

100

References Brooks, D. R., England, C. F., Hunt, G. E., and Minnis, P. (1984), An intercalibration of METEOSAT 1 and GOES-2 visible and infrared measurements, I. Atmos. Oceanogr. Tech. 1:283-287. Coulson, K. L., and Jacobowitz, H. (1972), Proposed calibration target for the visible channel of a satellite radiometer, NOAA/ NESS Tech. Rep. ~ 2 . Deschamps, P. Y., Herman, M, and Tanr6, D. (1983), Mod61isation du rayonnement r6fl6chi par l'atmosph6re et al terre, entre 0.35 et 4 #m, Final Report, ESA Contract 4393/80/F/DD(SC). Fraser, R. S., and Kaufrnan, Y. J. (1986), Calibration of satellite sensors after launch, Appl. Opt. 25:1177-1185. Gautier, C. (1982), Mesoscale insolation variability derived from satellite data, 1. Appl. Meteorol. 21:51-58. Gautier, C. (1986), Evolution of the net surface shortwave radiation over the Indian Ocean during summer MONEX (1979): a satellite description, Mon. Wea. Rev. 114:525-533. Hovis, W. A., Jr. (1966), Infrared spectral reflectance of some common minerals, AppI. Opt. 5:245-248. Hovis, W. A. (1982), Calibration of the NOAA-7 AVHRR/2 shortwavelength channel, NOAA/NESS Tech. Memo. E/RA2:WAH. Hovis, W. A., Jr., Knoll, J. S., and Smith, G. R. (1985), Aircraft measurements for calibration of an orbiting spacecraft sensor, Appl. Opt. 24:407-410. IAMAP 18 (1981), Report of proceedings, III Scientific Assembly, Hamburg, August. Koepke, P. (1982) Vicarious satellite calibration in the solar spectral range by means of calculated radiances and its application to METEOSAT, Appl. Opt. 21:2845-2855. Kriebel, K. I. (1981), Calibration of the

R. FROUINAND C. GAUTIER METEOSAT-VIS channel by airborne measurements, Appl. Opt. 20:11-12. Lauritson, L., Nelson, G. L., and Porto, F. W. (1979), Data extraction and calibration of TIROS/NOAA radiometers, NOAA/NESS Tech. Mem. #107. London, J., Bojkov, R. D., Oltsuans, S., and Kelly, J. I. (1976), Atlas of the global distribution of total ozone, NCAR/IN/133+ STR. Neckel, H., and Labs, D. (1984), The solar radiation between 3300 and 12500 A, Sol. Phys. 90:205-258. Rockwood, A. A., and Cox, S. K. (1978), Satellite inferred surface albedo over Northwestern Africa, I. Atmos. Sci. 35:513-522. Slater, P. N. (1984), The importance and attainment of absolute radiometric calibration, Proc. SPIE Critical Rev. Remote Sens. 475:34-40. Slater, P. N. (1985), Radiometric considerations in remote sensing, Proc. IEEE 73:997-1011. Smith, E. A., and Loranger, D. (1977), Radiometric calibration of polar and geosynchronous satellite shortwave detectors for albedo measurements, Dept. of Atmos. Sciences, Colorado State Univ., Fort Collins (unpublished manuscript). Smith, E. A., and Vonder Haar, T. H. (1980), A first look at the summer MONEX GOES satellite data, Proc. 15th AIAA Thermophysics Conf., Snowmass, CO, 14-16 July. Staff Members (1980), Visible infrared spin scan radiometer atmospheric sounder system description, Santa Barbara Research Center, Goleta, CA. Stephens, G. L., Campbell, G. G., and Vonder Haar, T. H., Earth radiation budgets, 1. Geophys. Res. 86:9739-9760. Tanr6, D., Herman, M., Deschamps, P. Y., and de Leffe, A. (1979), Atmospheric modeling for space measurements of ground reflectances, including bidirectional properties, Appl. Opt. 21:3587-3594.

CALIBRATIONOF NOAA-7AND GOES-5AND -6 DATA Tanr6, D., Deroo, C., Duhant, P., Herman, M., Morcrette, J. J., Perbos, J., and Deschamps, P. Y. (1985), Effects atmosph6riques en T61&l&ection--Logiciel de simulation du signal satellitaire dans le spectre solaire, Proc. 3rd. Int. Coll. on Spectral Signahares of Objects in Remote Sensing, l_~s Arcs, France, 16-20 December. Tarpley, J. D., Schneider, S. R., and Money, R. L., (1984), global vegetation indices from the NOAA-7 meteorological satellite, /. Climate Appl. Meteorol. 23:491-494. Thekaekara, M. P. (1974), Extraterrestrial solar spectrum, 3000-61000 A at 1 A intervals, Appl. Opt. 13:518-522. Tucker, C. J., Gatlin, J. A., and Schneider, S. R. (1984), Monitoring vegetation in the Nile Delta with NOAA-6 and NOAA-7

101

-AVHRR imagery, Photo. Eng. Remote Sens. 50:53-61. Walraven, R. L., and Coulson, K. L. (1972), Measurements of the light properties of gypsum sand, contrib. Atmos. Sci., No. 7, Univ. of California, Davis. WCP 55 (1983), Report of the WMO Radiation Commission of IAMAP meeting of experts on aerosols and their climatic effects, Williamsburg, VA, 28-30 March. WCP 115 (1986), Report of the Workshop on Surface Radiation Budget for Climate Applications, Columbia, MD, 18-21 June. Williamson, L. E., (1975), Satellite calibration data, Annual Data Report, 1974, Tech. Rep. ECOM-DR-75-1, Atmos. Sciences Lab., White Sands Missile Range, NM. Received 10 September 1986; revised21 lanuary 1987.