Instrument Performance of the Ocean Color Im ager

The Ocean Color lmager is an electro-optic remote sensing instrument with a 702 km swath width and 800 m resolution on board the ROCSAT-1 satellite that was launched in January 1999. It is the first push-broom type spectral-radiometer for measuring radiance from ocean surfaces and at­ mospheric scattering in six visible and near infrared bands for application research in ocean optics. Described in this paper are OCl's major charac­ teristics, parameters and pre-launch and post-launch performance test re­ sults including SNR, MTF, image sharpness, band to band spectral regis­ tration and anti-blooming performance.


INTRODUCTION
The first satellite of the Republic of China on Taiwan, ROCSAT-1, was launched in Janu ary 1999. The Satellite has three scientific payload instruments including the Ocean Color Imager (Fig. 1), or OCI, which is a spectral radiometer as well as a nadir looking push-broom imager that can acquire data of radiance from the ocean surface and atmospheric scattering in six visible and near infrared bands for application research in ocean optics ..
Since the late 1970' s, space-borne ocean observation instruments, such as CZCS (1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986), OCTS (1996), and SeaWiFS (1997, have acquired many high quality images for accessing the ocean's role in the global carbon cycle, elucidating the magnitude and vari ability of the primary productivity by phytoplankton, and determining the distribution and timing of spring booms. By using rotating mirrors as the scanning mechanism, their major characteristics and parameters ( Table 1 ) include, among others, a very wide field of view (FOY) for observing large areas and a very high signal to noise ratio (SNR) to provide reliable data for estimating water-leaving radiance from the ocean surface. · TAO, Vol. I 0, No. 4, December 1999  ··· · ··, ··;· , .. , ·· .... : . · caifiera ' · · '•·· · H:��•<f' The push-broom scanning mechanism has been utilized to make remote sensing instru ments on several land observing satellites, such as the SPOT series and IRS, of which major characteristics and parameters include very high ground resolution with relatively narrow POV and low SNR. (Table 2) To simplify interface development and reduce risks in instrument engineering, no moving parts were used to make the three payload instruments on the ROCSAT-1. Consequently, instead of a rotating mirror, the OCI employed a push-bloom scanning mecha nism, and achieving very high SNR, very wide FOY, and accurate radiometric calibration became tasks with unprecedented challenge in developing this kind of remote sensing instru ment.

MISSIONS AND BAND SELECTION FOR SCIENTIFIC OBSERVATIONS
The ROCSAT-1 program, including the OCI, is managed by the National Space Program Office (NSPO) of the Republic of China. The ROCSA T-l orbit is at an altitude of 600 km with a 35° inclination angle, and is not sun synchronous. This orbit was selected with the consideration of maximizing contact time with the ground station in Taiwan and fulfilling the necessity of another payload, Ionospheric Plasma Electrodynamics Instrument (IPEl), to gain more operation time near the equator. The OCI was originally planned to take images between 9:00 AM and 3:00 PM local time and the acquired data would be available to all interested scientists for doing research subjects in: -mapping the pigment distribution in the low-latitude oceans and generating surface spectral data;   Table 3) that are similar to, but fewer than the SeaWiFS'. Since there is little water leaving radiance from the ocean surf ace in the spectral regions near 670 nm and 869 nm, B5 and B6 are mainly for measuring atmospheric scattering with the measured data to be used for atmospheric correction. The data acquired from the other bands are from both atmospheric scattering and reflectance from the ocean surface. After correction, by subtracting the esti mated radiance of atmospheric scattering, water leaving radiance values can be obtained and used to compute the pigment distribution near the ocean surface by using the validation and calibration algorithms developed by the science team.

SYSTEM ARCHITECTURE
The OCI' s hardware was built by the NEC Corporation and consists of three modules: the Optical Unit (OU), the Analog Electronic Unit (AEU), and the Main Electronic Unit (MEU). Its functional block diagram is shown in Fig. 2 .
The OU consists of four camera heads with seven focal planes each with a Thomson TH7811 linear array CCD device for detecting the input radiance. Each camera head has an eight-lens telecentric telescope subsystem (Fig. 3) with a 19.5 mm focal length and a 60. 7° field of view ( FOY) to give a 702 km swath width at an altitude of 600 km. Two focal planes (B 1/B3, B2/ B4, or B5 / B6) share a camera head and B7 uses a stand-alone camera head with a lens subsystem similar to that of the others.
The TH7811 linear CCD device has 1728 cells each with a 13µ m x 13 µ m photosite that can transform optical radiance into electric current. The cells in each device are arranged to give 832 double cell pixels plus 64 single cell pixels in the center region of the array. At an altitude of 600 km and with a 19.5 mm focal length, 6.9 km/s ground track speed and 115.8 ms   integration time, the nadir-looking push-broom mechanism gives -800m x 800 m ground resolution for the double cell pixel and -400 m x 800 m ground resolution for the single cell pixel.

RESULTS OF PERFORMANCE TESTS
The results of selected pre-launch and post-launch performance tests are described in the following sections.

Signal To Noise Ratio
To assure good image quality, various techniques were developed to minimize noise in the signal chain and achieve very high SNR values (Table 4). These techniques included: •widening the telescope aperture to enhance the input signal; • keeping the CCD operation temperature at -10°C to lower the thermal noise; • using a clamp and sample circuit to perform correlated double sampling; • good shielding to block electro-magnetic interference as much as possible; and � digitizing the signal into 12 bits to make the quantization noise insignificant.

841
The pre-launch SNR was measured using a very stable integrating sphere as the light source and the results, given in Fig. 4, indicated that the SNR values would be at least 650 with mean input radiance and higher than 300 even with relatively weak input radiance, -1/5 of the corresponding mean radiance. Such high performance suggested that the sensor's observation time could be expanded from 9:00-15:00 to 7:30-16:30 local time at least.
We also performed post-launch SNR measurements by analyzing radiance data from a selected region in an acquired image with (apparently) uniform total radiance. Its radiance value was measured to be -0.45 times that of the mean value and was > 570 times the noise level. This SNR value was similar to, but slightly lower than, the pre-launch SNR value, -600 at the same radiance. The difference could be attributed to non-uniformity of the radiance from the selected region. An example of comparison is shown in Fig. 5.

OCI-
The pre-launch data showed that the output signal was -600 times greater than the standard deviation, or noise, and the post-launch data showed that SNR was -578.
where Ir, Ia and Iw represent Rayleigh scattering, aerosol scattering and water-leaving radi ance, respectively, and t is diffuse transmittance. The value of Iw is used to estimate the amount of pigment in the ocean and it is desirable to calculate this value by removing the atmospheric components, i.e., Rayleigh scattering and aerosol scattering, from the total input radiance. Before launch the OCI was calibrated with a standard light source and the signal chain of the calibration process could be expressed as follows: [Radiance from the Standard Light Source] """ (The OCI) """ [Digitized Output Signal] where the light source was a well calibrated integrating sphere with 97% absolute accuracy and 99.5% uniformity. The calibration results provided useful parameters for computing total input radiance from the OCI's digitized data, as follows: (2) where x was the OCI' s digitized signal and P, Q, R, and S were calibration parameters that varied from band to band and from pixel to pixel. Unlike other space-borne ocean observation instruments, such as CZCS, OCTS, and SeaWiFS, that utilized rotating mirror architecture with only one photodetector to measure the input radiance for each spectral band, the OCI had 896 CCD pixels (detectors) to measure the input radiance for each spectral band. It took a great deal of effort to achieve so many values for P, Q, R, and S for the pixels by performing radiometric calibration. Since Iw generally constitutes less than 20% of It, a small uncertainty in the calibration parameter due to non-uniformity of the standard light source could have a big impact on com puting lw, and, moreover, degrade the image quality of the pigment distribution significantly because the relation between Iw and pigment concentration is largely logarithmical. Specifi cally, we have found that a 0.5% uncertainty due to non-uniformity of radiance from the stan dard light source would lead to a -2.5% uncertainty in the computed Iw, and eventually a -15% uncertainty in the computed concentration of pigments, such as chlorophyll-a. As a result the image quality would be degraded by non-periodic vertical stripes (Fig. 6) arising from uncertainty of the derived pigment concentration originated from inaccuracy of relative calibration among the CCD pixels due to non-uniformity of radiance from the standard light source.
This phenomenon, the presence of non-periodic vertical stripes originating from non uniformity of radiance from the standard light source, has not been found in ocean color imag ery from any other space-borne ocean observation instruments because, instead of several hundred CCD pixels, they have used only one photodetector for each spectral band. The phenomenon has not been seen in images provided by any space-borne land observation in struments, including those with a push-broom scanning mechanism such as SPOT and IRS, because the data processing for land imagery is much less sensitive to It uncertainty than that for ocean imagery. It is therefore a task with unprecedented challenge to improve the OCI's image quality by modifying the calibration parameters to eliminate the non-periodic vertical stripes. We have very carefully analyzed many post-launch data taken by the OCI over regions with uniform radiance ( SNR2500) and used the results to modify the original calibration parameters. The modified parameters, with less than 0.2% inaccuracy in relative calibration among the CCD pixels, would lead to a -1 % uncertainty in Iw computation and a -5% uncertainty in pigment

845
concentration computation. By using the modified parameters the processed image is of better quality and has far fewer visible stripes (Fig. 7).

Modulation Transfer Function
The square wave modulation transfer function (MTF) requirements for each band at the Nyquist frequency are specified in Table 5. Each telescope subsystem was carefully designed to achieve simulated performance with static MTF better than 0.72. With unavoidable align ment eITors, assembly eITors, crosstalks, and simulated satellite motion effects, the instrument performance neve1theless had good MTF measurement values, between 0.47 and 0.71 ( Table  Table 5  Before launch, we measured image sharpness by using a bright object with sharp edges. The measured results, given in Fig. 8, showed good performance that was compatible with the measured MTF. We also performed post-launch measurements by selecting vertical coast lines and measuring the spatial variation of the radiance across the lines (Figs. 9 & 10). The measured results, given in Fig. 11, showed sharpness similar to that of the pre-launch results and suggested that the MTF values should be similar, too.

Band to Band Spectral Registration
With a 60.7°FOV and 7 focal planes for 7 spectral bands ranging from blue to near infra red, making an opto-mechanical hardware with high accuracy in spectral registration among the bands is a great challenge. To achieve accuracy better than 0.5 x IFOV, tremendous efforts were made in optical design, component fabrication, alignment and assembly to mini mize error sources such as mis-alignment and aberration.
The pre-launch test results, given in Fig. 12, showed that the registration distribution was within ±0.36 x IFOV in the along track direction and within ±0.48 x IFOV in the cross track direction. By analyzing in-flight images taken by the OCI we obtained post-launch spectral registration results (Fig. 13) which showed that the distribution was within ±0.40 x IFOV in the along track direction and within ±0.25 x IFOV in the cross track direction.

Blooming and Sun-Glint Effects
When the OCI is used for observation, some objects such as land and cloud are much brighter than the ocean. Anti-blooming is a CCD feature that can effectively reduce the charge  spill over to the neighborhood when some CCD cells are illuminated by extremely bright incident light. By setting the saturation voltage at 6.5 Von TH7811 CCD chips, a significant anti-blooming feature as well as a very high response linearity over the desired dynamic range were both achieved. The anti-blooming performance for each spectral band was measured by using a bright stripe as the observation object with 20 x IFOV width, and brightness that pro vided 6 times the saturation radiance. The results showed that blooming effects were limited to only �5 x IFOVs, or �10 CCD cells (Fig. 14).
We analyzed the OCI' s post-launch imaging data and found that radiance of B6 from land and cloud was 20 and 30 times stronger than the specified mean radiance if image acquisition time was at noon. As a result, even with the anti-blooming feature, the blooming effect could not be easily suppressed for B6 in noontime imagery, especially in the right neighborhood of a large piece of cloud or land. (Fig. 15) Because the OCI is always viewing in the nadir direction, it is not easy to avoid sun-glint, solar radiance directly reflected from the ocean surface, and its image quality is hence often degraded by this effect if the image acquisition time is around noon, i.e., between 10:30 AM and 1:30 PM local time. ( Fig. 16) It is nevertheless possible to significantly reduce this effect provided the imaging schedule is carefully planned. (Fig. 17) Spectral Registration @ Pixel #895, Post-Launch

7 Dark Signal vs. Temperature
It was found that the OCI' s digitized output signals were always greater than zero even when its total input radiance was zero. The positive values, termed dark signals, were attrib uted to offsets and thermal agitation in the CCD cells as well as noise f r om the CCD cells, analog electronics and AID conversion. Dark signals were temperature dependent, and could vary f r om band to band as well as from pixel to pixel. We have analyzed many dark signal data and found that the signal value could be estimated by: where Dark Signal (in digital count) = Circuit Offset+ Rn • 2 T10 ( 3 ) "Circuit Offset" = 34.3± 2 .6, had different values for different CCD chips; "Rn" = 7 .1±1.6, varied from pixel to pixel; "Q" = 8.9±1.2°C, varied from chip to chip; and "T" was the CCD temperature in °C. Post-launch dark signal data from early orbit through August 1999. The averaged dark signal value in August 1999 was higher than the early orbit as well as the pre-launch value by -4 digital counts.

853
We have also measured post-launch dark signal data and found that the value has in creased gradually since April 1999. (Fig. 18) In August 1999, the averaged dark signal value was higher than the pre-launch value by -4 digital counts. Since the environment of CCD electronics was thermally well controlled to stay within 10±2°C, it was unlikely that the CCD temperature could be as high as I 3°C to cause the measured difference of the dark signal. The actual cause is currently regarded as an electronics aging effect, but needs to be investigated in the future. Because the OCI's digitized signal is -1800 with mean input radiance, 4 digital counts is equivalent to 0.22% of the mean signal. It is desirable to take this value into account ·when we generate the calibration parameters to assure the reliability of data process.

CONCLUSION
A spectral radiometer without any moving parts has been built and tested. It is the first space-borne remote sensing instrument with a push-broom scanning mechanism for ocean observation and is a nadir-looking imager with a 60.7° FOV and 800 meter ground resolution at an altitude of 600 km. The results of selected pre-launch and post-launch performance tests have been described. Because of the high performance with reasonable SNR value at rela tively weak input radiance, the OCl's observation time could be expanded from 9:00-15:00to 7:30-16:30 local time at least. To avoid degradation of image quality, noontime imaging should be carefully planned because otherwise the OCI' s response could often be saturated by the blooming effect when viewing a large area of land, cloud, or sun-glint region. Since computation of pigment distribution is very sensitive to the uncertainty of radiometric param eters that arises from non-uniformity of radiance from the pre-launch calibration light source, the OCI's image was degraded by non-periodic vertical stripes. We have modified thecalibra-tion parameters and used them to process data and achieve images with improved quality.