GISdevelopment.net ---> AARS ---> ACRS 1989 ---> Digital Image Processing

Geometric and Radiometric Correction of MOS-1 Imagery in a Canadian Processing System

Brain Robert and Kelly Wiebe
Mac Donald Dettwiler & Associates Ltd.
13800 Commerce Parkway, Richmond, B.C.
Canada V6V 2J3



Abstract
The Marine Observation Satellite (MOS-1) has been operational and routinely acquiring data for over 30 months. As part of the MOS-1 verification project, the Canadian Centre for Remote Sensing (CCRS) has acquired over 160 orbits of MOS-1 data. Preliminary analysis of the raw data has shown that in order to effectively utilize the imagery, it must first be processed to correct for radiometric and geometric errors. To meet this need, Mac Donald Dettwile has developed a prototype MOS-1 correcting system capable of radio metrically and geometrically correcting raw MOS-1 data. This paper reviews the Canadian MOS-1 processing capabilities. it begins with a review of the MOS-1 sensors and mage characteristics. Following this, the radiometric and geometric correction algorithms of the Canadian proto type system are described. Finally example of the processed imagery are pre4sented.

1 Introduction
Since its lunch on February 19, 1987, the marine Observation satellite 9mOS-1) has been routinely acquiring data over north America. As part of the MOS-1 verificat6ion project, the Canadian Centre for Remote Sensing (CCRS) has tracked over 160 orbits of MOS-1 data from which test scenes have been chosen for sensor evaluation and application studies. preliminary valuation of the raw data imagery generated by the CCRS Quick look Transcription System (Butlin 89) indicated that in order to effectively analyze the data, it would first have to be processed to correct for radio metric and geometric errors. The goal of the current project is to provide corrected data for Canadian researches to assess the quality and utility of mjoS-1, and provide CCRS with limited moS-1 processing capabilities to meet further processing needs on an interim basis.

To meet this need, Mac Donald Dettwiler has developed a prototype radiometric and geometric correction system capable of processing data from all three MOS-1 sensors. The system, which is based on Mac Donald Dettwiler’s Geocoded Image Correction System 9gics0 engine, takes as input raw MOS-1 data on LGSOWG format computer compatible tapes (CCTs) generated by the CCRS Transcription System. Raw image data is absolutely radio metrically calibrated, geometrically corrected to a map projection, and output on LGSOWG format CCTs . The geometric processing resample the image to a map projection correcting for distortions introduced in the image acquisition process including sensor alignment, band mis registration, earth curvature and rotation effects, satellite orbit and attitude errors, and temperature dependent sensor characteristics. The radiometric processing converts the raw digital numbers 9dns) to physical units 9radiance or temperature0 correcting for temperature dependent sensor characteristics using both in flight sensor calibration data and preflight information.

This paper describes the Canadian processing of mos-1 data. The three mos-1 sensors are reviewed with emphasis on the quality of the raw image data. The radiometric and geometric correction algorithms of the Canadian prototype system are then discussed following which examples of processed imagery presented.

2 Image Characteristics
MOS-1 was designed to observed and monitor land, ocean and atmospheric processes. The satellite carries three sensors which observe the earth over a wide range of spatial resolutions in the visible, near-infrared, thermal infrared, and microwave regions of the electromagnetic spectrum. In this senses, MOS-1 1aqffords he remote sensing community with an ideal opportunity to study and compare data from a diverse set of sensors. As data is simultaneously acquired by all three sensors, the traditional problems associated with comparing multi sensor datasets taken at different times and under different environmental conditions are eliminated.

2.1 MESSR Imagery
The multi spectral Electronic Self-Scanning Radiometer 9MessR) was designed to observe the earth using four spectral bands in the visible and near infrared spectrum. The sensor contains two radiometers, systems 1 and 2, which point at fixed angles 9nominally 2.73o0 to either side of nadir. under normal operating conditions, imagery may be acquired from only a single system at a given time. Each MESSR is a push broom sensor incorporating a single 2048 lament CCD linear array in each spectral band. it has a nominal ground resolution of 50 meters acquiring data over a 100 kilometer swath to one side of the satellite track. Radio metrically, the MESSR’s four spectral bands are similar to those of the Lands at MSS sensor. This, together with the similar geometric resolutions, has sparked interest in comparing the quality of data acquired by the two sensors.

Evaluation of raw MES imagery has shown that the data contains a number of radiometric and geometric artifacts which hamper its analysis (Manore 89, Henry 89 a ). Significant band mis registration, caused by misalignment of the for detector arrays, is present in imagery acquired by both systems. The magnitude of this mis registration varies between system 1 and system 2, and may be as large as 6 pixels. In many applications such as land use classification, accurate band-to-band registration critical and any misalignment can introduce significant misclassification errors.

An analysis of six MESSR images was undertaken to characterize the magnitudes of the detector array displacements and their dependence on temperature. Using Band 1 as a reference, the spectral bands were correlated against one another at uniformly distributed grid points over image. The mean shifts in the along and across track directions were then calculated, and are shown in Table 1. These results, which are consistent with other5s reported in the literature (Manore 89, Henry 89 a), illustrate the magnitude of the band mis registration errors in the raw data.

Table 1. Mean band mis registration in raw MESSR data in the along and across track directions. The vales are reported in units of pixels.
MESSR System Misregistration From Band 1
Band 2 Band 3 Band 4
1
2
1.07/0.72
-0.12/0.58
2.25/-0.28
4.87/1.38
0.80/-0.86
1.05/-1.32

Analysis of the radiometric quality of raw MESSR data has identified artifacts introduced by the electronics of the sensor. Vertical striping, typical of linear array sensors, has been observed in all MESSR images studied thus far. This striping takes two forms: random striping caused by differences in the gains and dark current offsets of adjacent CCD elements, and periodic striping caused by the even-odd shift registers of the CCD. The random striping may be removed to a large extent by absolutely calibrating the data using NASDA supplied calibration coefficients. The periodic even-odd striping is not as easily removed.

2.2 VTIR Imagery
The MOS-1 Visible and Thermal Infrared Radiometer (VTIR) is designed to observe the temperature of earth surfaces (sea surfaces) and cloud tops, and ice and clod distribution using one band in the visible region and three bands in the infrared region of the electromagnetic spectrum. The VTIR is most similar to the NOAA AVHRR sensor , although it does have a number of distinct characteristics. It is mechanical scanning radiometer consisting of a single detector in each band with ground resolution of nominally 900 meters in the visible band, and 2.7 kilometers in the infrared bands. The VTIR provides 16 gain modes in each band. The imagery may be calibrated using data obtained while observing a built-in black body and deep space when the scanning mirror is not sweeping the earth.

The image quality of raw VTIR data is relatively good with only a few problems being reported in the literature. While the band mis registration is much better than for MESSR, there is still a 1 to 2 pixel misalignment between the different detectors. Radiometric artifacts may also be observed in the data in the form of horizontal striping particularly noticeable in VTIR Band 4,, as well as a small vertical striping noise almost periodic in nature.

2.3. MSR Imagery
The MOS-1 Microwave Scanning Radiometer (MSR) is designed to designed to observe snow conditions, water vapor in the atmosphere over oceans, and the amount of water in clod and sea ice by receiving microwave noise signals emitted from ground surfaces and oceans. The MSR is comprised of a conical scan system which employs an offset casse grain antenna common to both bands. Half the time required for one rotation of the antenna is used for earth observation while the remaining half is used for recording the calibration data of low and high reference temperature sources. The MSR produces data in two bands (23 GHz and 31 GHz) at two different integration times (47 msec and 10 msec) as a radiometric resolution of 10 bits. the ground resolution is nominally 32 kilometers in the two 23 GHz bands and 23 kilometers in the two 31 GHz bands.

Because of the conical scan system, the ground scanning pattern of the MSR is semicircular, leading to large geometric distortions in the raw imagery. Analysis of the radiometric quality of this data has shown the presence of horizontal typing in the imagery as well as random pixel dropouts, particularly prevalent near the top and bottom portions of the imagery. As the processing unit of MSR data is a full satellite pass, these dropouts are most likely caused by cross in communication with the satellite when it is near the horizon.

3. Radiometric & Geometric Correction
To allow for overlay and comparison of images from the different MOS-1 sensors, the raw data must first be corrected for geometric distortions introduced in the image acquisition process. Further, absolute radiometric calibration must be performed to convert the raw DNs to physically meaningful quantities This facilities meaningful comparisons of multi temporal and multi sensor data allowing the effective utilization and comparison of datasets from all three mOS-1 1sensors both amongst themselves and with other sensors such as the land sat MSS and TM sensors, the SPOT HRV sensor and the NOAA AVHRR sensor.

To meet these needs, Mac Donald Dettwiler has developed, on behalf of the Canadian MOS-1 Project Team, a prototype mos-1 processing system. The system, which is based on Mac Donald Dettwiler’s GICS engine with enhancements to handle the MOS-1 specific satellite and sensor systems, is capable of geometrically correcting data from algal three MOS-1 sensors to a standard map projection, and absolutely calibrating the raw DNs to physical units. The corrected imagery is output on standard LGSOWG format CCTs.

Geometric correction is performed in a two stage process. In the first stage, the correspondence between any given pixel in the input imagery and a point n the earth’s surface is established through modeling each stage of the image acquisition process. This correspondence, called the forward transformation, is based on sensor, satellite, orbit, earth and map projection models. in the second stage, the input imagery is resample to a regular grid usng a user selectable resembling kernel. The primary output of the system is an absolutely radio metrically corrected and credimatically geo referent products. Systematic refers to the fact that the images are geometrically corrected sing only a priori information such as the sensor geometry, ephemeris and attitude data: no ground truth is required. Geo referenced products are in a map projection oriented in the direction of the nominal satellite heading at the scene creator. Out put pixel spacing for the three sensors have been chosen such that hey differ by integral multiples from other sensors processed by the GICS system allowing for easy in ultisensor image overlay and comparison.

Radiometric correction is also performed in a two stage process. Input pixels are first converted to physical units (typically radiance) using telemetry data, as well as in flight and preflight calibration data. The corrected values are then converted back to digital numbers for output to CCT using either fixed or variable gains and offsets. These parameters are stored in the radiometric ancillary record of the corrected CCT, thus may be used to convert the corrected DNs back to physically meaningful units.

3.1. MESSR Products
The standard processing of MESSR data generates products with 50 meter pixel spacing which have been systematically geo referenced to either the Universal Transverse Mercator (UTM) or Lambert Conformal Conic (LCC) map projections. The process corrects for sensor alignment, band mis registration, earth curvature and rotation effects, satellite orbit and attitude errors and temperature dependent distortions of the CCD arrays.

As analysis of the geometric accuracy of the systematically corrected MESSR products was performed on a MESSR System 1 image of Hamilton, Canada acquired on September 28 1988 using seventeen control points distributed uniformly throughout the image. After removal of the along and across track bias errors (due to the systematic nature of the spacecraft modeling), the combined RMS error in locating accuracy was fond to be 44 miters, i.e, less than one pixel error. The scale accuracy of the product, measured using the same 17 GCPs, was found to have an RMS value of 0.001. This is similar to the scale accuracy reported for SPOT HRV products (Henry 89 a).

Significant effort was made to improve the band-to-band registration in corrected MESSR products. using the correlation data described in Section 2.1, the sensor model was refined so that the mis registration predicted using the sensor model agreed with the observed mis registration. To assess the quality of these model refinements, the band mis registration errors in for corrected MSSR products were assessed using the same inter-band correlation technique. The results, shown in Table 2, show the success of this technique as mis registration is less than a fifth of a pixe3l between all band pairs and for both systems in the corrected data.

Table 2. Mean band mis registration in corrected MESSR products in the along and across track directions. The values are reported in units of pixels.
MESSR System Mis registration From Band 1
Band 2 Band 3 Band 4
1 2 -0.06/-0.08-0.04/-0.10 -0.08/-0.07-0.05/-0.19 -0.09/0.010.02/-0.07

The radiometric correction of MESSR data includes absolute calibration of the raw DNs o radiance sing satellite telemetry data preflight measurements supplied by NASDA, relative calibration to remove residual striping and decompression of the 6-bit data. The calibrated radiance is converted to an 8-bit DN band/gain mode dependent radiance limits before output to CCT. As MS-1 has no in-flight absolute calibration capability, one can not directly monitor and track the radiometric degradaion of the sensor over time. however, work by Henry et. al. 9henry 89 b0 comparing MESSR imagery with near simultaneous SPOT MLA data has shown that it is possible to achieve absolute radiometric calibration with an uncertainty as low as 10 percent.

An example of a geometrically and radio metrically corrected MESSR System 2 image of over eastern Lake Ontario, Canada is shown in Figure 1. The image was acquired using low gain mode on may 31 1988, ands has been corrected to the UTM map projection. Qualitative analysis of this and other MESSR images suggests that the geometric resolution of MESSR data superior to that of Land sat MSS data, but not as good as Lands at TM data.


Figure 1. A black and white rendition of a geometrically and radio metrically corrected MESSR image over eastern Lake Ontario, Canada acquired on May 31 1988. The data was acquired using how gain mode by MESSR System 2.

3.2 VTIR Products
Standard processing of VTIR data generates systematically geo referenced products sin the LCC map projection which have been corrected for band mis registration, earth curvature and rotation effects, panoramic distortions, satellite orbit and attitude errors, and different spatial resolutions between bands. Two standard VTIR products are available is meet the needs of different users : a fixed size product based on 2000 lines of input data, and a variable sized product based on a full pass of input data. The corrected data is resample to a 550 meter grid for single scene products and a 1100 meter grid for full pass products which allows easy caparison with 11001 meter AVHRR data. The raw DNs are converted to radiance using both in flight calibration data obtained by viewing a reference black body and deep space, as well as preflight data. The corrected radiance is converted to an 8-bit DN using scene specific gains and offsets before being output to CCT.

An example of a geometrically and radio metrically corrected VTIR product of the east coast of North America required on September 8 1988 is shown in Figure 2. Geometric processing has corrected for the uneven line and pixel spacing and the large panoramic distortions present in the raw data.


Figure 2. A geometrically and radio metrically corrected VTIR thermal band (band 40 image of the east coast of North America acquired on September 8 1988. The image has been corrected to the Lambert conformal conic map projection with 550 meter pixel spacing. (copyright NASDA 1989)

3.3 MSR Products
Because of the unique characteristics of the MSR sensor, two types of products may be produced by the MOS-1 processing system. To meet the requirements of speciaqlized users, he system is caqpabl of generating an ASCH table from the raw MSR data which tabulates the location (in LCC and geodetic coordinates), the temperature, he time, and the solar conditions for each observation point. To meet the needs of more traditional image processing users, the system is also capable of generating raster MSR products. in both cases, the system corrects geometric cross due in earth curvature and rotation effects, satellite orbit and attitude errors, different spatial resolutions, and the sensor mixing characteristics. Radio metric correction f MSR data converts the raw DNs to brightness temperature using in fight calibration information. The correction has been further supplemented to allow the detection and correction of the random pixel dropouts. Figure 3 illustrate a set of nine geometrically and radio metrically corrected passes of MSR imagery acquired between the summer of 1988 and 1989. All datasets have been corrected to the LCC map projection with 3300 meter pixel soakings.


Figure 3. Nine geometrically and radio metrically corrected MSR band 1 images acquired between he summer of 19188 and 1989. (Copyright NASDA 1989)

4 Conclusions
Canadian MOS-1 data processing capabilities have been described. A prototype system has been developed by Mac Donald Dettwiler for CCRS capable of radio metrically and geometrically correcting data from all three MOS-1 sensors. The processing capabilities include ;
  • systematic geometric correction to a map projection modeling and removing distortions introduced by the sensor satellite, and earth:
  • absolute and relative radiometric corrections :
The system will provide CCRS with the capability to meet the processing requirements of Canadian MOS’-1 Project Team, and to meet additional MOS-1 processing needs on an interim basis.

References
  • (Butlin 89 ) T. Butlin, J. Lam, M. Manore, and R.A. O’Neil. The Canadian MOS-1 project. In Marine Observation satellite-1 International Symposium, pages 180-189, National Space Development Agency of Japan (NASDA), 1989.
  • (Henry 89 a) P. Henry and G. Begni. MESSR image quality. In The Third Symposium on MOS-1 Verification Program, page 435, Earth Observation Center, National Space Development Agency of Japan, 1989.
  • (Henry 89b) P. Henry and G. Begni. MOS-1 absolute calibration using SPOTI as an intermediate radiometer. In The Third Symposium on MOS-1 Verification Program, page 436, Earth Observation Center, national Space Development Agency of Japan, 1989.
  • (Manore 89) M. Manore and T. Butlin. Preliminary evaluation of typical Canadian scenes of MOS-1 imagery. In The Third Symposium on MOS-1 Verification Program, pages 191-200, Earth Observation Center, National Space Development Agency of Japan, 1989.