You are on page 1of 12

2380

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 47, NO. 7, JULY 2009

SSTL UK-DMC SLIM-6 Data Quality Assessment


Gyanesh Chander, Member, IEEE, Sebastien Saunier, Michael J. Choate, Member, IEEE, and Pasquale L. Scaramuzza

AbstractSatellite data from the Surrey Satellite Technology Limited (SSTL) United Kingdom (UK) Disaster Monitoring Constellation (DMC) were assessed for geometric and radiometric quality. The UK-DMC Surrey Linear Imager 6 (SLIM-6) sensor has a 32-m spatial resolution and a ground swath width of 640 km. The UK-DMC SLIM-6 design consists of a three-band imager with green, red, and near-infrared bands that are set to similar bandpass as Landsat bands 2, 3, and 4. The UK-DMC data consisted of imagery registered to Landsat orthorectied imagery produced from the GeoCover program. Relief displacements within the UK-DMC SLIM-6 imagery were accounted for by using global 1-km digital elevation models available through the Global Land One-km Base Elevation (GLOBE) Project. Positional accuracy and relative band-to-band accuracy were measured. Positional accuracy of the UK-DMC SLIM-6 imagery was assessed by measuring the imagery against digital orthophoto quadrangles (DOQs), which are designed to meet national map accuracy standards at 1 : 24 000 scales; this corresponds to a horizontal root-mean-square accuracy of about 6 m. The UK-DMC SLIM-6 images were typically registered to within 1.01.5 pixels to the DOQ mosaic images. Several radiometric artifacts like striping, coherent noise, and at detector were discovered and studied. Indications are that the SSTL UK-DMC SLIM-6 data have few artifacts and calibration challenges, and these can be adjusted or corrected via calibration and processing algorithms. The cross-calibration of the UK-DMC SLIM-6 and Landsat 7 Enhanced Thematic Mapper Plus was performed using image statistics derived from large common areas observed by the two sensors. Index TermsCalibration, characterization, Enhanced Thematic Mapper Plus (ETM+), geometry, Landsat, radiometry, relative spectral response (RSR), spectral bands, Surrey Linear Imager 6 (SLIM-6), Surrey Satellite Technology Limited (SSTL), United Kingdom (UK) Disaster Monitoring Constellation (DMC).

University of Surrey Satellites (UoSAT), Korea Institute of Technology Satellites (KITSAT-1 and -2), Portuguese satellite (PoSAT-1), Caractrisation de lEnvironnement Radioelectrique pour un instrument Spatial Embarque (CERISE), UoSAT-12, Thai-Microsatellite (TMSat), Surrey Nanosatellite Application Program (SNAP), and Tsinghua-1 [1], [2]. The DMC Consortium is made up of ve nations, each of which owns and operates a satellite and ground station. The members are Algeria, China, Nigeria, Turkey, and the U.K. Table I summarizes key specications of the constellation satellites. The objective of the DMC microsatellites is to provide a daily global imaging capability at medium resolution for rapid response disaster monitoring and mitigation. The project relies on an international consortium into which each nation or organization operates its own satellite and ground segments. Table II lists orbit characteristics of several Earth Observation missions and those of the UK-DMC. The DMC constellation satellite has an operation life of ve years [2], [3]. DMC International Imaging Ltd. (DMCii) handles the sale and supply of DMC data and international disaster response. Created after the launch of the satellite constellation, DMCii is in charge of data distribution and service creation. DMCii coordinates partners through a centralized mission planning system at SSTL and fullls customer requests. II. UK-DMC S ENSOR O VERVIEW Sensors onboard UK-DMC satellites differ and depend on satellite owner (Table I). The Extended Swath Imaging System sensor, also called Surrey Linear Imager 6 (SLIM-6), is used onboard UK-DMC satellites. The UK-DMC SLIM-6 sensor is a multispectral sensor operating in three spectral bands in the visible and near-infrared (NIR) regions of the electromagnetic spectrum. The UK-DMC SLIM-6 design consists of a threeband imager providing a spatial resolution of 32 m and a ground swath width of 640 km. The large swath of UK-DMC SLIM-6 is achieved by having two separate banks of cameras, each covering half the swath and overlapping at nadir, as shown in Fig. 1. Each sensor has a charge-coupled device (CCD) with a linear array of over 10 000 detectors. The satellite uses two separate arrays for each band, which overlap in ground projection to produce images with a swath width exceeding 640 km, using almost 20 000 detectors in total. The sensor optics is characterized by a very short focal length (150 mm) and an aperture diameter of about 90 mm; this conguration equips the sensor with a small detector size (6.5 m AC/AT) with respect to a 30-m ground sampling distance [2]. The UK-DMC has a eld of view (FOV) of 17.5 . The Landsat 7 (L7) Enhanced Thematic Mapper Plus

I. M ISSION P ROFILE HE DISASTER Monitoring Constellation (DMC) provides worldwide daily revisit capability for disaster response and delivers high temporal resolution imagery for many applications, including oil and gas supply continuity, cartography, environment, telecoms, agriculture, and forestry. The Surrey Satellite Technology Limited (SSTL) is a privately owned subsidiary of the University of Surrey, Guildford, England. Some key missions developed by SSTL are the ve
Manuscript received June 2, 2008; revised October 19, 2008 and January 2, 2009. First published March 27, 2009; current version published June 19, 2009. This work was performed under USGS Contract 08HQCN0005. G. Chander, M. J. Choate, and P. L. Scaramuzza are with Stinger Ghaffarian Technologies (SGT), Inc., contractor to the U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center, Sioux Falls, SD 57198 USA (e-mail: gchander@usgs.gov). S. Saunier is with the GAEL Consultant, 77420 Champs-sur-Marne, France. Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TGRS.2009.2013206

0196-2892/$25.00 2009 IEEE

CHANDER et al.: SSTL UK-DMC SLIM-6 DATA QUALITY ASSESSMENT

2381

TABLE I DMC SATELLITE SPECIFICATIONS

TABLE II UK-DMC AND OTHER SATELLITE MISSION CHARACTERISTICS

Fig. 1.

Layout of the UK-DMC SLIM-6 multispectral imager [3].

Fig. 2. UK-DMC SLIM-6 imager (640-km swath width) overlaid with scene footprints from other imaging sensors as summarized in Table II.

(ETM+) has a FOV of 7.2 . Because of the large FOV, the UK-DMC SLIM-6 images will have geometric distortions at the swath edge. Fig. 2 compares the UK-DMC SLIM-6 swath with the geographical coverage from other sensors.

Fig. 3 compares the swath width and spatial resolution of these sensors. Fig. 4 shows a comparison of the FOV of UK-DMC SLIM-6 and L7 ETM+ sensors. Table III summarizes the key specications of the UK-DMC SLIM-6 and L7 ETM+ sensors.

2382

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 47, NO. 7, JULY 2009

Fig. 3. Swath width along with spatial resolution. TABLE IV UK-DMC SLIM-6 PRODUCT ORGANIZATION

TABLE V UK-DMC SLIM-6 PRODUCT LEVEL

Fig. 4. Across-track spatial resolution along the FOV. TABLE III UK-DMC SLIM-6 AND L7 ETM+ KEY SPECIFICATIONS

UK-DMC onboard data storage capacity is 1.5 GB. Although the chip has potential for a 10-b dynamic range, only 8 b are stored and sent to the ground. Hence, the dynamic range is 256 gray-level values [2]. The data transfer is achieved using S-band with a data rate of about 5 Mb/s. The data volume that can be downlinked from UK-DMC is 1.5 GB per day at SSTL in Guildford, U.K. This data transfer capability is inferior to the one provided with communication in X-band. Satellites such as Landsat ETM+ and Systme Pour lObservation de la Terre (SPOT) HRVIR provide X-band transfer of about 2 75 Mb/s and 1 50 Mb/s. UK-DMC satellites are not able to transmit data to a ground segment through satellite relay. A 1.5-GB state recorder is mounted onboard the UK-DMC satellite, which is very small storage capability when compared to L7 ETM+ (378 GB) [3]. The UK-DMC provides a constant phasing between the satellites. Satellite orbital position is determined using a Global

Positioning System receiver, which provides the timing function. Orbit control is maintained using pressurized cold gas or a liqueed gas system (resistojet thrusters) for constellation phasing maneuvers and small orbit corrections. In general, the simplest attitude control is a passive stabilization, and it relies on either magnetic or gravity gradient methods. Simple and passively stable systems have a low pointing performance. Complex systems using reaction wheels are highly accurate pointing systems. Satellite stabilization techniques for controlling orientation or attitude of the spacecraft are classied into four categories: single spin stabilization, dual spin stabilization, three-axis stabilization, and gravity gradient stabilization [4]. UK-DMC platform attitude control is done through the use of gravity gradient method. The European Space Agency Environmental Satellite, SPOT, and Landsat are stabilized with a three-axis technique and offer a better pointing accuracy (about 0.015 ) compared to the microsatellites (DMC, Micro Lab) stabilized with a gravity gradient technique (about 2 ). A. UK-DMC SLIM-6 Data Products UK-DMC product from the SLIM-6 sensor is processed in GeoTiff Digital Image Map (DIMAP) format. The DIMAP format is the new format for SPOT products, introduced for the Spot 5 launch in May 2002 and developed with the Centre National dEtudes Spatiales. The DIMAP format is a public format for describing geographic data. Although it was specially designed for image data, it can also handle vector data. An example of the UK-DMC SLIM-6 product id DU000107T_L1T is summarized in Table IV. UK-DMC SLIM-6 processing levels listed in Table V are compliant with the commonly used

CHANDER et al.: SSTL UK-DMC SLIM-6 DATA QUALITY ASSESSMENT

2383

Fig. 5.

UK-DMC SLIM-6 (400 400 pixels) imagery over (a) Paris urban area, (b) industrial plant, and (c) crop elds.

Fig. 6.

Comparison of UK-DMC SLIM-6 spectral ranges with other sensors.

standard nomenclature. Fig. 5 shows sample full-resolution imagery over the following: 1) Paris urban area; 2) industrial plant; and 3) crop elds. The four-lane Normandie bridge that has a width of 23.5 m is shown near the industrial plant. When zooming in on the image data, the bridge is recorded in less than two pixels. The industry plant and the warehouse can be seen, but they are hard to recognize. B. Spectral Characteristics of the Sensor The UK-DMC SLIM-6 sensor provides three spectral bands covering the green (520600 nm), red (630690 nm), and NIR (770900 nm) electromagnetic spectrum regions. Fig. 6 shows a comparison of the UK-DMC SLIM-6 spectral ranges with other sensors. These spectral ranges are similar to the green, red, and NIR spectral bands of SPOT 5 High Resolution Geometric, Landsat 5 (L5) Thematic Mapper (TM), and L7 ETM+. Although the spectral range is similar, it does not mean that for a given wavelength, the spectral response of UK-DMC SLIM-6 band matches with the band of the L7 ETM+. The SLIM-6 spectral lters were manufactured by Barr Associates Inc., U.S., using the same materials and processes as the spectral lters on the ETM+ sensor. The spectral lter of each channel is protected by a fused silica radiation absorption window, which is positioned on the space facing side of the lter. The spectral lters are located in front of the camera lens. Fig. 7 shows a com-

parison of the relative spectral response (RSR) proles of L7 ETM+, Indian Remote Sensing Satellite (IRS-P6) Advanced Wide Field Sensor (AWiFS) and Linear Imaging Self-Scanner (LISS-III), China-Brazil Earth Resources Satellite (CBERS-2) High Resolution CCD Camera (CCD) the Infrared Multispectral Scanner (IRMSS), and UK-DMC SLIM-6 sensors. The two separate banks of UK-DMC SLIM-6 cameras have slightly different spectral responses. In red band, the spectral lters from the two banks were very similar and a single RSR prole was used to characterize the response. The RSR proles from the two banks are referred as primary (P) or secondary (S). They are sometimes referred to as Port and Starboard. The P is the master imager in each image pair and is usually in the left-hand side (hence, port side of the delivered image) [3]. Fig. 8 shows an evaluation that was performed to check the UK-DMC SLIM-6 ability to discern the typical surface materials. The green-band peak around 0.55 m (r tree curve) can be used to discriminate vegetation and identify man-made structures. The red band is ending just before the vegetation reectance drop (around 0.69 m) and can be used to discern bare soil from vegetation canopy. The NIR band is used for water body discrimination, and soil moisture and vegetation monitoring. Red and NIR bands are used for computation of spectral vegetation indices such as the widely used normalized difference vegetation index. Handling red and NIR bands is suitable for discrimination of crop canopy density.

2384

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 47, NO. 7, JULY 2009

Fig. 7. Comparison of the RSR proles of L7 ETM+, P6 AWiFS/LISS-III, CBERS-2 CCD/IRMSS, and UK-DMC SLIM-6 sensors.

III. G EOMETRIC A NALYSIS The geometric characterization involves understanding the processes by which pixel coordinates within an image can be mapped to known locations on the ground. Generally, this process involves the use of well-surveyed photo-identiable targets placed within a scene. Because this is primarily a function of what can be seen at a given resolution, different techniques are applied to different resolution ranges to ensure that methods exist to evaluate a wide range of sensors. In addition, these techniques seek to leverage existing technologies where possible. One place that this leverage is applied is with the use of some of the geometric assessment tools that were developed for the L7 ETM+ Image Assessment System (IAS) [5], [6]. The IAS is responsible for ofine assessment of image quality to ensure compliance with the radiometric and geometric requirements of the L7 satellite and the ETM+ sensor throughout the mission [7], [8].

Fig. 8. Typical spectral reectance curves and UK-DMC SLIM-6 spectral ranges.

The UK-DMC SLIM-6 spectral bands cannot be used for discrimination of clays (TM5, TM7), hydroxides, iron oxide (TM1 blue), or plant stress (TM5, TM7).

CHANDER et al.: SSTL UK-DMC SLIM-6 DATA QUALITY ASSESSMENT

2385

TABLE VI UK-DMC SLIM-6 DATA USED FOR GEOMETRY CHARACTERIZATION

A. Data Sets Used for the Study Eight UK-DMC SLIM-6 data sets acquired over the Railroad Valley Playa, Nevada (RVPN) geometric supersite were assessed in this paper. These images are listed in Table VI and were acquired with zero-degree viewing angle (nadir). The table also lists whether the digital orthophoto quadrangle (DOQ) data sets needed for the geodetic assessment of the data were available. Images were processed by UK-DMC International Imaging Ltd. and listed as being processed to a Level-1 terrain corrected (L1T) product. These products are processed with ground control and have the effects due to elevation removed from the imagery. GeoCover imagery was used as a reference source for registering the UK-DMC SLIM-6 data. The GeoCover program [9] created wall-to-wall orthorectied Landsat TM and ETM+ imagery. The imagery is listed as having a geodetic accuracy of better than 50-m root-meansquare error (rmse). Elevation data used within the processing were the global 1-km digital elevation models available through the Global Land One-km Base Elevation (GLOBE) Project [10]. All UK-DMC SLIM-6 images were resampled to a Universal Transverse Mercator projection using the World Geodetic System 1984 (WGS84) geographic coordinate system and had a pixel size of 32 m in both the X and Y map directions. B. I2I Assessment The image-to-image (I2I) characterization is usually performed to compare the relative accuracy between two images. One image is selected as a reference and another as the search image. Points/Image chips (small area of about 64 64 pixels) are selected from the reference image and are correlated with the search image. The coregistration results provide an insight to the relative accuracy of the search image with respect to the reference image. Plotting the points measured between the two images helps in assessing any systematic bias or higher order distortion between the image data. The I2I assessment provides numerical evaluation of the accuracy of common bands of temporally distinct images [7]. The I2I tool available within the IAS can be used to analyze other imagery as needed besides the ETM+ data. The IAS I2I tool assesses the geometric difference between two image les by performing normalized grayscale correlation on windowed image pairs between two data sets. Several criteria are used for determining if a single correlation measurement is successful. Some of these criteria include strength of the correlation peak and maximum allowable displacement. Offsets are measured as the peak location of the correlation surface. The offset is calculated to the subpixel level by tting a 3-D

Fig. 9. UK-DMC SLIM-6 and coregistered DOQ mosaic.

Fig. 10. Full-resolution window of resampled DOQ mosaic.

surface around the peak of the correlation surface. A student t-test is performed on the measured offsets to remove any outliers produced from the correlation process. Statistics calculated on the nal points kept from the I2I process provide an assessment of the geometric differences between the two images. The I2I comparison tool expects the image data sets being compared to have the same resolution. One worldwide reference system (WRS) wall-to-wall DOQ mosaic is often referred to as a geometric supersite. The Landsat program has a suite of geometric test sites (supersites) to evaluate geometric accuracy. The supersites are georeferenced images derived from a high-resolution source. These highresolution data sets consist of mosaics of DOQs over the extent of one standard WRS-2 Landsat image (187 km by 187 km). DOQs meet national map accuracy standards at 1 : 24 000 scale, which equates to a horizontal root-mean-square accuracy of about 6 m. This corresponds to an uncertainty of 1/5 of an ETM+ pixel at the 30-m scale. The DOQs are mosaicked to create a data set equal to one WRS-2 nominal swath. The L7 IAS I2I was used in performing a geodetic assessment of the UK-DMC SLIM-6 L1T data sets. The ground controls used for geodetic accuracy of the UK-DMC SLIM-6 data sets were the mosaicked DOQs. These mosaicked DOQs were subsampled to match the resolution of the UK-DMC SLIM-6 data. Fig. 9 shows the UK-DMC SLIM-6 images along with the coregistered DOQ mosaic images used for the geodetic accuracy assessment. The very apparent checkered pattern within the DOQ mosaic is due to the radiometric differences between individual DOQs. A full-resolution window of the DOQ mosaic can be seen in Fig. 10. There were two images for which no

2386

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 47, NO. 7, JULY 2009

TABLE VII GEODETIC ACCURACY OF UK-DMC SLIM-6 IMAGES

TABLE VIII B2B REGISTRATION RESULTS

Fig. 11. Vector residuals between UK-DMC SLIM-6 and DOQ mosaic image (DU0005eaT).

DOQs were available for a geodetic accuracy assessment. The images still had a band-to-band (B2B) registration assessment performed on the data sets. Comparison between the UK-DMC SLIM-6 images and the DOQ mosaics are listed in Table VII. The UK-DMC SLIM-6 images were typically registered to within 1.01.5 pixels to the DOQ mosaic images. The rmse measured between the UK-DMC SLIM-6 image and the reference DOQ data sets ranged from 7.03 to 23.53 m (0.220.74 pixels) in the X-direction and from 26.20 to 36.67 m (0.821.15 pixels) in the Y -direction. Fig. 11 shows a plot of vector residuals produced from the UK-DMC SLIM-6 to DOQ mosaic comparison. Vectors are scaled by a factor of 4000 so that they are more easily viewable within the imagery. In all six comparisons, there was a slight bias in the line direction or, more likely, along the ight path of the satellite.

C. B2B Assessment The B2B characterization is performed to ensure that the proper band alignment is provided for an image product. Typically, when using the IAS B2B assessment tool, a band alignment assessment registers each band against every other band (resample bands of higher resolution to coarse resolution). The B2B assessment provides a numerical evaluation of the accuracy of the band registration within an image. The B2B tool available within the IAS can be used to analyze other imagery as needed besides the ETM+ data.

The IAS B2B tool works similarly to the I2I tool. The B2B tool works by rst choosing an evenly distributed set of points between each band pair within the imagery. The process then performs a normalized grayscale correlation between the two bands at each location and follows the same procedures as I2I in determining subpixel location. After each band combination is measured, a student t-test is performed on each individual set of band pairs to remove any remaining outliers. Statistics can be calculated on the nal points kept to give an assessment of the band alignment within the image data. Plotting the residuals kept can help determine the kind of band alignment distortion that is present. For the B2B correlation process, data sets that contain bands of differing resolution must have one of the bands resampled so that image bands to be compared have matching resolutions. For multiresolution data sets, the differences in resolution must be of integer multiples. Bands of common resolution are created by reducing higher resolution bands to be of equal resolution to that of the lower bands. The two techniques used in the IAS for reducing the resolution of an image are the Gaussian Pyramid approach and an oversampled cubic convolution weighting function. Other approaches, such as using a weighting function that more closely resembles the modular transfer function (MTF) of the sensor, would be useful but require further investigation. The B2B assessment of the UK-DMC SLIM-6 data sets was also performed using the IAS B2B tool. Table VIII lists the B2B registration for all eight of the UK-DMC SLIM-6 images. Fig. 12 shows the residual vector plots for each band pair measured for all eight UK-DMC SLIM-6 images. Vectors are scaled by a factor of 3000 so that they are more visible within the imagery. The B2B assessment showed some bias within the band registration. Most often, the bias appeared to be associated with the registration of band 1 to band 3. In several cases, the registration of band 2 to band 3 also showed a noticeable bias.

CHANDER et al.: SSTL UK-DMC SLIM-6 DATA QUALITY ASSESSMENT

2387

Fig. 12. B2B registration residuals (DU0005eaT).

IV. R ADIOMETRIC A NALYSIS The UK-DMC SLIM-6 calibration team performs regular radiometric characterization and monitors the stability of the sensor. The calibration team has carried out the following projects: 1) acquisition of deep space image for dark current and relative calibration, just after satellite launch when gradient boom system is disabled; 2) acquisition of dark images over Pacic, snow scenes over DOM-C, and comparison to the deep space image; 3) monthly monitoring of relative calibration performed on snow scenes; and 4) a vicarious calibration exercise in collaboration with the University of Arizona on data sets acquired over RVPN test site to monitor the absolute calibration accuracy [3]. Radiometric analysis was performed on one SSTL UK-DMC SLIM-6 sensor image acquired over the Las Vegas, Nevada region, with an imaging time of 17:25:26 on 2004-07-19. The image used for the study was processed to create a Level-1 radiometrically corrected image (L1R) in TIFF format. A. Flat Detector There appears to be at least one at detector, which causes a stripe of bright data in the ight direction. This artifact stretches the entire height of the scene, although it is less visible in cloudy regions and more visible over dark terrain. In Fig. 13, it appears in all three bands but in slightly different locations; the red band is the major contributor to the stripe visible in the RGB image. This at detector has a reduced dynamic range, causing its response to be 34 DN higher than the adjacent detectors. The location and approximate digital number (DN) magnitude of this vertical stripe is NIR : Red : Green : pixel 15426 pixel 15416 pixel 15418 3 DN 4 DN 3 DN.
Fig. 13. Detector anomaly in each band and composite.

B. Coherent Noise All bands exhibit low magnitude banding, visible only in extreme stretches over homogenous terrain. Fig. 15 shows Lake Mead with an extreme linear stretch applied, displaying the banding. Considering that SLIM-6 is a pushbroom sensor, this banding appears to be a coherent noise source with a wavelength of about 270 scan lines. To examine the banding, an area of interest was cropped from the center of the lake, 380 detectors wide and 896 scan lines high, and then, the mean of each scan line was calculated. This created visible coherent noise patterns in all three bands, as shown in Fig. 16. Fig. 17 shows a plot of the Fourier transform of the water data, calculated in an attempt to isolate the coherent noise sources. Two major sources appear in all bands at wavelengths of about 270 scan lines and 25 scan lines. Other sources may be harmonics of these, or they may be separate sources of noise, and it is possible that some are caused by patterns in the underlying imagery of the lake itself. Access to nighttime or shutter data would be required for a full coherent noise analysis.

Quantitative assessments were performed on a uniform image part over the at detector. For a given line number, the pixel value of the three bands was extracted along the column number. In image coordinates, the location of the stripe differs from one band to the other, as shown in Fig. 14.

2388

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 47, NO. 7, JULY 2009

Fig. 14. Corrupted detector values along column number.

Fig. 15. Lake Mead (430 200 pixels) with linear stretch applied, displaying banding.

Fig. 17.

Power spectrum of 380-detector mean water data.

band in these data appears to be blurrier than the red and green bands. Fig. 18 shows an example of a highway whose details are much more distinct in the red and green bands than in the NIR band. Prelaunch or in-ight MTF is usually provided with the mission specication document because it is strongly associated with the ground resolution. However, no prelaunch MTF value is given for the UK-DMC SLIM-6 sensor and no scientic publications were found. D. Striping Individual detectors within an array typically do not possess similar gain and bias characteristics. This mismatch of detector response results in the appearance of striping in acquired images. Over clouds, the green band saturates at about 210 DN and exhibits minor detector striping, as shown in Fig. 19. This detector striping is about 4 DN in magnitude. The NIR and red bands saturate over clouds, but they saturate at 255, and they do not exhibit this detector variation. The fact that the green band saturates at 210 DN may point to a problem with the relative gain calibration in the green band and could indicate nonlinearity in the green band over high-radiance targets. If any specic band shows saturation, it is possible to change the integration times independently for each band to prevent striping. No low saturations were found in the provided data.

Fig. 16. 380-detector mean of RGB bands over water area.

C. MTF The ability of an imaging sensor to resolve the spatial objects can be assessed using MTF. The MTF is the normalized magnitude of the Fourier transform of the sensors point spread function. Specications can be dened by requesting certain minimum values of the MTF at critical spatial frequencies (e.g., Nyquist frequency). MTF is extremely difcult to quantify without a calibrated ground target, but qualitatively, the NIR

CHANDER et al.: SSTL UK-DMC SLIM-6 DATA QUALITY ASSESSMENT

2389

Fig. 18. NIR, Red, and Green bands (175 175 pixels), showing MTF differences.

E. Cross-Calibration With the L7 ETM+ The cross-calibration approach involved comparing image statistics derived from large common areas observed by the L7 ETM+ and UK-DMC SLIM-6 sensors [12][14]. The image data from both sensors were converted to absolute units of at-sensor radiance, which is the fundamental step in putting image data from multiple sensors and platforms onto a common radiometric scale [15][17]. The absolute radiance values are scaled to 8-b values representing calibrated digital numbers (Qcal ) before output to the distribution media. Conversion from Qcal in L1 products back to L requires knowledge of the original gain (Grescale ) and bias (Brescale ) rescaling factors. For L7 ETM+ data, the Qcal -to-L conversion is summarized in the L7 Science Data Users Handbook [18]. For UK-DMC SLIM-6 data, the Qcal -to-L conversion process is given by the relationship Qcal + Brescale Grescale Where : 255 Grescale = LM AX LM IN Brescale = LM IN . L = The scaling coefcients (rescale gain and rescale bias) are unique for every image, and these unique coefcients are stored in the metadata les that accompany every L1R and L1T product. The spectral radiance is provided in the units of W/(m2 sr m). A single sun elevation angle correction was applied, but no radiative transfer codes were applied for estimating gaseous transmittance, no bidirectional functions were estimated for taking into account surface roughness, and no spectral band adjustments were performed. Two different playa regions were taken for result conrmation. The results of the comparison for the spectral bands are shown in Fig. 22. The plots compare the at-sensor spectral radiance extracted from the UK-DMC SLIM-6 and L7 ETM+ data. Each data point on these plots represents an ensemble average of all pixels in a dened region for a given day and spectral band. The oneto-one line represents the idealized perfect agreement between the radiances obtained from both sensors for a particular band. The average percent differences in at-sensor radiance estimates obtained from the UK-DMC SLIM-6 and L7 ETM+ are within 18% for the green band, 32% for the red band, and 6% for the NIR spectral band. Additional work needs to be done to

Fig. 19. Detector saturation striping in green band (430 200 pixels).

Fig. 20. Odd/Even detector striping in the red band (100 40 pixels).

In addition to the saturation striping shown in Fig. 19, the red band exhibits a residual oddeven detector striping as shown in Fig. 20. This striping is about 2 DN in magnitude and is only visible in the imagery when an extreme stretch has been applied. It is visible over targets of any radiance, indicating a simple miscalibration of the relative detector biases, not a potential nonlinearity effect as in the saturation striping in Fig. 19. The dark detector bias appears to be above 10 DN for all bands. Fast Fourier transform (FFT) was performed on image chips extracted from the UK-DMC SLIM-6 image data set acquired, as shown in Fig. 21. An FFT 2-D representation highlights nonnominal frequency peaks occurring along northsouth axis. FFT results conrmed a persistent line-to-line miscalibration occurring on UK-DMC SLIM-6 green band. The histogram and statistics of the image chips are also summarized in Fig. 21. The shapes of the green- and red-band histograms and the standard deviation are very close. The histogram of the NIR band is wider than that of the other bands with higher standard deviation. The red band has some residual striping, but the magnitude of the striping artifact is less in comparison to the green band. The UK-DMC SLIM-6 NIR band seems to be free from the striping artifact.

2390

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 47, NO. 7, JULY 2009

Fig. 21. (Middle) Image chips, (bottom) histograms, and (top) respective 2-D FFT images.

Fig. 22. Comparison of at-sensor radiance (in the units of W/m2 sr m) measurements from large ground regions common to both UK-DMC SLIM-6 and L7 ETM+ sensors.

characterize the absolute differences between the two sensors over the entire mission. This study needs to quantify uncertainties due to spectral, spatial, temporal, and atmosphere differences and should be performed over multiple dates to yield more robust results. V. S UMMARY Geometric analyses were conducted on eight SSTL UKDMC SLIM-6 CCD sensor data sets acquired over the RVPN

area. The RVPN site is referenced to the WRS-2 as path 40 and row 33. The UK-DMC SLIM-6 images were processed with the latest geometric improvements to their processing system. The I2I characterization was performed to compare the relative accuracy between the UK-DMC SLIM-6 image and a reference DOQ. The UK-DMC SLIM-6 images were typically registered to within 1.01.5 pixels to the DOQ mosaic images. The rmse measured between the UK-DMC SLIM-6 image and the reference DOQ data sets ranged from 7.03 to 23.53 m (0.220.74 pixels) in the X-direction and from 26.20 to 36.67 m (0.821.15 pixels) in the Y -direction. The B2B characterization was performed to check the band alignment for the UK-DMC SLIM-6 images. The rmse band alignment offsets measured within the UK-DMC SLIM-6 data sets had values of up to 12.03 m (0.37 pixels) in the line direction and 20.03 m (0.62 pixels) in the sample direction. The preliminary results show that the data from the UK-DMC SLIM-6 sensor meet the specications and can be used to support remote sensing applications. Several radiometric artifacts like striping, coherent noise, and at detector were discovered and studied. Indications are that the SSTL UK-DMC SLIM-6 data have few artifacts and calibration challenges, and these can be adjusted or corrected via calibration and processing algorithms. The cross-calibration of the L7 ETM+ and UK-DMC SLIM-6 was performed using image statistics derived from large common areas observed by the two sensors. The average percent differences in at-sensor

CHANDER et al.: SSTL UK-DMC SLIM-6 DATA QUALITY ASSESSMENT

2391

radiance estimates obtained from the ETM+ and SLM-6 are within 18% for the green band, 32% for the red band, and 6% for the NIR spectral band. A full radiometric analysis of SSTL satellite data would require a dark or nighttime data acquisition and/or data from any dedicated calibration sources aboard the satellite. ACKNOWLEDGMENT Satellite data from the Surrey Satellite Technology LTD (SSTL) Disaster Monitoring Constellation (DMC) were provided under Technical Assistance Agreement UK-08.0000, ACIS Reference ID # 8512 between the USGS and SSTL dated September 29, 2005. Any use of trade, product, or rm names is for descriptive purposes only and does not imply endorsement by the U.S. Government. R EFERENCES
[1] M. Sweeting, 25 years of space at surrey-pioneering modern microsatellite, in Proc. 49th Int. Astronautical Congr., Melbourne, Australia, Sep. 1998. [2] Surrey Satellite Technology Limited (SSTL) Space Missions, May 2008. [Online]. Available: http://www.sstl.co.uk/ [3] The Disaster Monitoring Constellation (DMC) International Imaging Products, May 2008. [Online]. Available: http://www.dmcii.com/ [4] H. J. Kramer, Observation of the Earth and Its Environment: Survey of Missions and Sensors, 4th ed. New York: Springer-Verlag, 2002. [5] Landsat 7 Image Assessment System (IAS) Geometric Algorithm Theoretical Basis Document. Sioux Falls, SD: U.S. Geol. Survey EROS Data Center, 1998. 3.2 ed. [6] J. C. Storey, R. A. Mortt, and P. R. Thorson, Image processing on the Landsat 7 image assessment system, in Proc. Amer. Soc. Photogramm. Remote Sensing Annu. Conf., Portland, OR, May 1999, pp. 743758. [7] D. S. Lee, J. C. Storey, M. J. Choate, and R. W. Hayes, Four years of Landsat-7 on-orbit geometric calibration and performance, IEEE Trans. Geosci. Remote Sens., vol. 42, no. 12, pp. 28212831, Dec. 2004. [8] J. Storey and M. Choate, Landsat 7 on-orbit geometric calibration and performance, Proc. SPIE, vol. 4049, pp. 143154, Jun. 2000. [9] Earth Science Data Interface (ESDI) at the Global Land Cover Facility, Jan. 2008. [Online]. Available: http://glcfapp.umiacs.umd.edu:8080/esdi/ index.jsp [10] The Global Land One-km Base Elevation (GLOBE) Project, Jan. 2008. [Online]. Available: http://www.ngdc.noaa.gov/mgg/topo/globe.html [11] G. Chander and M. J. Choate, Geometric characterization of DMC, in Proc. JACIE, Civil Commercial Imagery Eval. Workshop, Fairfax, VA, Mar. 2022, 2007. [12] G. Chander, D. L. Helder, B. L. Markham, J. Dewald, E. Kaita, K. J. Thome, E. Micijevic, and T. A. Ruggles, Landsat 5 TM on-orbit absolute radiometric performance, IEEE Trans. Geosci. Remote Sens., vol. 42, no. 12, pp. 27472760, Dec. 2004. [13] G. Chander, D. J. Meyer, and D. L. Helder, Cross-calibration of the Landsat-7 ETM+ and EO-1 ALI sensors, IEEE Trans. Geosci. Remote Sens., vol. 42, no. 12, pp. 28212831, Dec. 2004. [14] G. Chander, M. J. Coan, and P. L. Scaramuzza, Evaluation and comparison of the IRS-P6 and the Landsat sensors, IEEE Trans. Geosci. Remote Sens., vol. 46, no. 1, pp. 209221, Jan. 2008. [15] D. L. Helder, B. L. Markham, K. J. Thome, J. A. Barsi, G. Chander, and R. Malla, Updated radiometric calibration for the Landsat 5 thematic mapper reective bands, IEEE Trans. Geosci. Remote Sens., vol. 46, no. 10, pp. 33093325, Oct. 2008. [16] G. Chander and B. L. Markham, Revised Landsat-5 TM radiometric calibration procedures, and postcalibration dynamic ranges, IEEE Trans. Geosci. Remote Sens., vol. 41, no. 11, pp. 26742677, Nov. 2003. [17] G. Chander, B. L. Markham, and J. A. Barsi, Revised Landsat-5 thematic mapper radiometric calibration, IEEE Trans. Geosci. Remote Sens., vol. 4, no. 3, pp. 490494, Jul. 2007. [18] Landsat-7 Science Data Users Handbook, NASA Goddard Space Flight Center, Greenbelt, MD, Feb. 28, 2006. [Online]. Available: http:// landsathandbook.gsfc.nasa.gov/handbook.html

Gyanesh Chander (M02) received the M.S. degree in electrical engineering from South Dakota State University, Brookings, in 2001. He is currently a Lead Systems Engineer with Stinger Ghaffarian Technologies, Inc., Greenbelt, MD, contractor to the U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center, Sioux Falls, SD. His primary responsibilities at EROS include satellite sensor characterization and calibration research to support ongoing radiometric projects. His current research focuses on cross-calibration between various sensors from different platforms for mission continuity, thereby providing consistent measurements of Earths surface features. For the past nine years, he has worked extensively with the National Aeronautics and Space Administration, European Space Agency, Indian Space Research Organization, and Instituto Nacional de Pesquisas Espaciais in the area of radiometric characterization and calibration of satellites and airborne sensors. He has played a pivotal role in the development of the Landsat Thematic Mapper Image Assessment System and the EO-1 Advanced Land Imager Image Assessment System. Mr. Chander is a member of the international Committee of Earth Observation Satellites and actively participates in the Working Group Calibration Validation and Infrared Visible and Optical Sensors subcommittee meetings. He is leading the Group on Earth Observations task DA-09-01a_2 to establish a catalog of prime candidate worldwide test sites for the postlaunch characterization and calibration of space-based imaging sensors.

Sebastien Saunier received the M.S. degree in applied mathematics from Jussieu Paris VII University, Paris, France, and the second M.S. degree in signal processing from cole Nationale Suprieure de lElectronique et de ses Applications School, Cergy Pontoise, France, in 1999 and 2001, respectively. For the past ve years, he has been an Earth Observation Application Engineer for optical sensor with GAEL Consultant, Champs-sur-Marne, France, in the frame of European Space Agency (ESA) European Space Research Institute (ESRIN) projects. His current activities are in the area of radiometric, geometric and spatial resolution, calibration, validation, and quality control.

Michael J. Choate (S92M93) received the B.S. degree in electrical engineering and the M.S. degree in engineering from South Dakota State University, Brookings, in 1990 and 1994, respectively. He is currently with Stinger Ghaffarian Technologies, Inc., Greenbelt, MD, contractor to the U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center, Sioux Falls, SD. His current work includes the geometric calibration and characterization of satellite and airborne instruments.

Pasquale L. Scaramuzza received the B.S. degree in physics from Drexel University, Philadelphia, PA, in 1990, and the M.S. degree in physics from Temple University, Philadelphia, in 1993. He is currently with Stinger Ghaffarian Technologies, Inc., Greenbelt, MD, contractor to the U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center, Sioux Falls, SD. He is a Radiometric Analyst working on the Landsat 7 image assessment system, where he performs the daily checks of the data and instrument calibrations, performs analyses on the radiometric performance of the instrument, and updates calibration parameters.

You might also like