Chapter 14 AVIRIS and Related 21st Century Imaging Spectrometers for Earth and Space Science Robert O. Green, Jet Propulsion Laboratory, California Institute of Technology Contents 14.1 Introduction 336 14.2 AVIRIS and the Imaging Spectroscopy Measurement 338 14.2.1 The AVIRIS Imaging Spectrometer Characteristics 339 14.2.2 The AVIRIS Measured Signal 342 14.2.3 Range of Investigations Pursued with AVIRIS Measurements 345 14.2.4 The AVIRIS Data Archive and Selected Imaging Spectroscopy Analysis Algorithms 346 14.3 Objectives and Characteristics of a Spaceborne Imaging Spectrometer for the Moon 348 14.3.1 Objectives of the Moon Mineralogy Mapper 348 14.3.2 Characteristics of the M 3 Imaging Spectrometer 348 14.3.3 Prospects for the M 3 Imaging Spectrometer Data Set 351 14.4 Objectives and Characteristics of a Future Spaceborne Imaging Spectrometer for the Earth 352 14.4.1 Objectives of an Earth Imaging Spectrometer for Measuring the State of Terrestrial and Aquatic Ecosystems 352 14.4.2 Characteristics of an Ecosystem Focused Earth Imaging Spectrometer 353 14.4.3 Roles for High-Performance Computing 354 14.5 Acknowledgments 356 References 356 Imaging spectroscopy (also known as hyperspectral imaging) is a field of scientific in- vestigation based upon the measurement and analysis of spectra measured as images. The human eye qualitatively measures three colors (blue, green, and red)in the visible portion of the electromagnetic spectrum when viewing the environment. The human eye-brain combination is a powerful observing system, however, it generally pro- vides a non-quantitative perspective of the local environment. Imaging spectrometer 335 © 2008 by Taylor & Francis Group, LLC 336 High-Performance Computing in Remote Sensing instruments typically measure hundreds of colors (spectral channels) across a much wider spectral range. These hundreds of spectral channels are recorded quantitatively as spectra for every spatial element in an image. The measured spectra provide the basis for a new approach to understanding the environment from a remote perspective based in the physics, chemistry, and biology revealed by imaging spectroscopy. The measurement of hundreds of spectral channels for each spatial element of an image consisting of millions of spatial elements creates an important requirement for the use of high-performance computing. First, high-performance computing is required to acquire, store, and manipulate the large data sets collected. Second, to extract the physical, chemical, and biological information recorded in the remotely measured spectra requires the development and use of high-performance computing algorithms and analysis approaches. This chapter uses the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) to review the critical characteristics of an imaging spectrometer instrument and the corresponding characteristics of the measured spectra. The wide range of scientific research as well as application objectives pursued with AVIRIS is briefly presented. Roles for the application of high-performance computing methodsto AVIRIS datasets are discussed. Next in the chapter a review is given of the characteristics and mea- surement objectives of the Moon Mineralogy Mapper (M3) imaging spectrometer planned for launch in 2008. This is the first imaging spectrometer designed to acquire high precision and high uniformity spectral measurements of an entire planetary-sized rocky body in our solar system. The size of the expected data set and roles for high performance computing are discussed. Finally, a review is given of one design for an Earth imaging spectrometer focused on investigation of terrestrial and aquatic ecosys- tem status and composition. This imaging spectrometer has the potential to deliver calibrated spectra for the entire land and coastal regions of the Earth every 19 days. The size of the data sets generated and the sophistication of the algorithms needed for full analysis provide a clear demand for high-performance computing. Imaging spectroscopy and the data sets collected provide an important basis for the use of high- performance computing from data collection to data storage through to data analysis. 14.1 Introduction Imaging spectroscopy is based in the field of spectroscopy. Sir Isaac Newton first separated the color of white light into the rainbow in the late 1600s. In the 1800s, Joseph von Fraunhofer and others discovered absorption lines in the solar spectrum and light emitted by flames. Through investigation of these absorption lines, the linkage between composition and signatures in a spectrum of light was established. The field of spectroscopy has been pursued by astronomers for more than 100 years to understand the properties of stars as well as planets in our solar system. On Earth, spectroscopy has been used by physicists, chemists, and biologist to investigate the properties of materials relevant to their respective disciplines. In the later half of the © 2008 by Taylor & Francis Group, LLC AVIRIS and Related 21st Century Imaging Spectrometers 337 25002200190016001300 Wavelength (nm) 1000700400 0.0 0.1 0.2 0.3 0.4 0.5 Reflectance 0.6 0.7 0.8 0.9 1.0 Kaolinite Conifer Grass Broad Leaf Sage_Brush NPV Gypsum Jarosite Dolomite Hematite Figure 14.1 A limited set of rock forming minerals and vegetation reflectance spec- tra measured from 400 to 2500 nm in the solar reflected light spectrum. NPV cor- responds to non-photosynthetic vegetation. A wide diversity of composition related absorption and scattering signatures in nature are illustrated by these materials. 20 th century Earth scientists developed spaceborne instruments that view the earth in a fewspectralbandscapturingaportion of the spectral informationinreflectedlight.The AVHRR, LandSat, and SPOTare importantexamples of this multispectral approachto remote sensing of the Earth.However, the few spectral bands of multispectralsatellites fail to capture the complete diversity of the compositional information present in the reflected energy spectrum of the Earth. Figure 14.1 shows a set of measured reflectance spectra from a limited set of rock forming minerals and vegetation spectra. A wide diversity of composition-related absorption and scattering signatures exist for such materials. Figure 14.2 shows these selected reflectance spectra convolved to the band passes of the LandSat Thematic Mapper. When mixtures and illumination factors are included, the 6 multispectral measurements of the multispectral Thematic Mapper are insufficient to unambiguously identify the 10 materials present. In the 1970s, realization of the limitations of the multispectral approach when faced with the diversity and complexity of spectralsignatures found onthe surface of the Earth lead to the concept of an imaging spectrometer. The use of an imaging spectrometer was also understood to be valid for scientific missions to other planets and objects in our solar system. Only in the late 1970s did the detectorarray, electronics,computer, and optical technology reach significant maturity to allow design of an imaging spectrometer. With the arrival of these technologies and scientific impetus, the Airborne Imaging Spectrometer (AIS) was proposed and built at the Jet Propulsion Laboratory [1]. The © 2008 by Taylor & Francis Group, LLC 338 High-Performance Computing in Remote Sensing 25002200190016001300 Wavelength (nm) 1000700400 0.0 0.1 0.2 0.3 0.4 0.5 Reflectance 0.6 0.7 0.8 0.9 1.0 Kaolinite Conifer Grass Broad Leaf Sage_Brush NPV Gypsum Jarosite Dolomite Hematite Figure 14.2 The spectral signatures of a limited set of mineral and vegetation spec- tra convolved to the six solar reflected range band passes of the multispectral LandSat Thematic Mapper. When mixtures and illumination factors are included, the six mul- tispectral measurements are insufficient to unambiguously identify the wide range of possible materials present on the surface of the Earth. AIS first flew in 1982 aswell as in severalsubsequentyears as atechnologyand science demonstration experiment. Concurrently with the development of the AIS a role for high-performance computing was identified and pursued [2]. The AIS instrument had limited spectral coverage as well as limited spatial coverage. Even as a demonstration experiment, the success of the AIS led to the formulation of the proposal for the Airborne Visible/Infrared Imaging Spectrometer. This next generation instrument was specified to measure the complete solar reflected spectrum from 400 to 2500 nm and to capture a significant spatial image domain. The broader spectral and spatial domain of this full range instrument continued to grow the role for high-performance computing in the field of imaging spectroscopy. 14.2 AVIRIS and the Imaging Spectroscopy Measurement The Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) [3, 4] measures the total upwelling spectral radiance in the spectral range from 380 to 2510 nm at ap- proximately 10 nm sampling intervals and spectral response function. Figure 14.3 shows a plot of the AVIRIS spectral range in conjunction with an atmospheric trans- mittance spectrum. Also shown for comparison are the spectral response functions © 2008 by Taylor & Francis Group, LLC AVIRIS and Related 21st Century Imaging Spectrometers 339 250022001900160013001000700400 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Wavelength (nm) Atmosphere Transmittance Transmittance AVIRIS 224 Contiguous Spectral Channels Landsat TM 6 Multispectral Bands Figure 14.3 AVIRIS spectral range and sampling with a transmittance spectrum of the atmosphere and the six LandSat TM multi-spectral bands in the solar reflected spectrum. of the multispectral LandSat Thematic Mapper. With AVIRIS a complete spectrum is measured with contiguous spectral channels. Across this spectral range the atmo- sphere transmits energy reflected from the surface, except in the spectral regions of strong water vapor absorption centered near 1400 and 1900 nm. These strong water vapor absorption regions are used for cirrus cloud detection and compensation. Mea- surement of this complete spectral range allows AVIRIS to be used for investigations beyond those possible with a multispectral measurement. In addition, measurement of the full spectrum allows use of new, more accurate, computationally intensive algorithms that require high-performance computing. In the spatial domain, AVIRIS measures spectra as images with a 20 m spatial resolution and an 11 km swath with up to 1000 km image length from NASA’s ER-2 aircraft flying at 20 km altitude. On the Twin Otter aircraft flying at 4 km altitude, the spatial resolution is 4 m with a 2 km swath and up to 200 km image length. Figure 14.4 shows an AVIRIS data set collected over the southern San Francisco Bay, California, from the ER-2 platform in image cube representation. The spectrum measured for each spatial element in the data set may be used to pursue specific scientific research questions via the recorded interaction of light with matter. 14.2.1 The AVIRIS Imaging Spectrometer Characteristics The full set of AVIRIS spectral, radiometric, spatial, temporal, and uniformity charac- teristics are given in Table 14.1. These characteristics have been refined and improved since the initial development of AVIRIS based upon the requirements from scientists © 2008 by Taylor & Francis Group, LLC 340 High-Performance Computing in Remote Sensing Figure 14.4 AVIRIS image cube representation of a data set measured of the south- ern San Francisco Bay, California. The top panel shows the spatial content for a 20 m spatial resolution data set. The vertical panels depict the spectral measurement from 380 to 2510 nm that is recorded for every spatial element. TABLE 14.1 Spectral, Radiometric, Spatial, Temporal, and Uniformity Specifications of the AVIRIS Instrument Spectral properties: Range 380 to 2510 nm in the solar reflected spectrum Sampling 10 nm across spectral range Response FWHM < 1.1 of sampling Accuracy Calibrated to 2% of sampling Precision Stable within 1% of sampling Radiometric properties: Range 0 to maximum Lambertian radiance Sampling 16 bits measured Response > 99% linear Accuracy > 96% absolute radiometric calibration Precision (SNR) As specified at reference radiance Spatial properties: Range 34 degree field-of-view (FOV) Sampling 0.87 milliradian cross and along track Response FWHM of IFOV < 1.2 of sampling Temporal properties: Airborne As requested 1987 to present Uniformity: Spectral cross-track > 99% uniformity of position across the FOV Spectral-IFOV > 98% IFOVs uniformity over the spectral range © 2008 by Taylor & Francis Group, LLC AVIRIS and Related 21st Century Imaging Spectrometers 341 250022001900160013001000700400 0 200 400 600 Signal-to-Noise Ratio Radiance (uW/cm^2/nm/sr) Signal-to-Noise Ratio Reference (0.5 at 23.5z) 800 1000 1200 1400 30 25 20 15 10 5 0 Wavelength (nm) Figure 14.5 The 2006 AVIRIS signal-to-noise ratio and corresponding benchmark reference radiance. using AVIRIS data. Of particular importance has been the improvement of the signal- to-noise ratio. An increased signal-to-noise ratio has been a critical factor enabling more advanced algorithms and sophisticated analysis approaches. Figure 14.5 gives the 2006 AVIRIS signal-to-noise ratio at the specified AVIRIS reference radiance. The AVIRIS reference radiance was specified in the original AVIRIS proposal as the radiance from a 0.5 reflectance surface illuminated by the sun at a 23.5 degree solar zenith angle through the standard mid-latitude atmospheric model. The current AVIRIS signal-to-noise ratio is 10 to 20 times greater than when the instrument first flew in 1986. Of special importance for valid physically based imaging spectroscopy science is the uniformity of the imaging spectrometer measurement. Two aspects of unifor- mity are critical. The first is cross-track spectral uniformity. The spectral cross-track uniformity requirement is that each spectrum in the image have the same spectral calibration to some percentage near 100%. For AVIRIS, the spectral cross-track uni- formity exceeds 99% because each spectrum in the image is measured by the same spectrometer. This is inherent in the AVIRIS whiskbroom imaging spectrometer de- sign. For the (approximate) 10 nm spectral sampling of AVIRIS, this 99% uniformity assures that the spectral calibration is the same for all spectra measured in an image to the level of 0.1 nm. Excellent spectral cross-track uniformity is required for all analysis algorithms that are applied directly to all spatial elements in an image. Some of the most powerful algorithms such as spectral dimensional analysis and spectral unmixing require near-perfect spectral cross-track uniformity. The second critical form of uniformity for an imaging spectrometer is spectral instantaneous-field-of-view (IFOV) uniformity. The IFOV is the sampling area on the surface for a single spatial element. Spectral-IFOV uniformity requires that the IFOV © 2008 by Taylor & Francis Group, LLC 342 High-Performance Computing in Remote Sensing Cross Track Sample Wavelength Figure 14.6 Depiction of the spectral cross-track and spectral-IFOV uniformity for a uniform imaging spectrometer. The grids represent the detectors, the gray scale represents the wavelengths, and the dots represent the centers of the IFOVs. This is a uniform imaging spectrometer where each cross-track spectrum has the same calibration and all the wavelengths measured for a given spectrum are from the same IFOV. for a given spectrum be the same for all wavelengths to some high percentage near 100%. This assures that the same area on the ground is sampled for all wavelengths measured in a spectrum. Again, because AVIRIS is a whiskbroom spectrometer, the spectral IFOV uniformity is high at better than 98%. Figure 14.6 depicts the spectral cross-track and spectral IFOV uniformities for a 100% uniform instrument. Several imaging spectrometers have been constructed with low spectral cross-track and low spectral-IFOV uniformities undermining their potential use. 14.2.2 The AVIRIS Measured Signal Understanding the detailed nature of the AVIRIS or any imaging spectrometer mea- surements is essential for appropriate analysis of the data. Figure 14.7 shows the reflectance spectrum of a vegetation canopy. From this reflectance spectrum a wide range of plant composition and status information may be extracted. This informa- tion is contained in the molecular absorption and constituent scattering signatures recorded in the vegetation canopy spectrum. An Earth-looking imaging spectrometer such as AVIRIS does not measure re- flectance. AVIRIS measures the total upwelling radiance incident at the instrument © 2008 by Taylor & Francis Group, LLC AVIRIS and Related 21st Century Imaging Spectrometers 343 25002200190016001300 Wavelength (nm) 1000700400 0.00 0.20 0.40 Reflectance 0.60 0.80 Figure 14.7 Vegetation reflectance spectrum showing the molecular absorption and constituent scattering signatures present across the solar reflected spectral range. aperture. When flying on the NASA ER-2 aircraft, the aperture is looking down from 20 km. Figure 14.8 show the modeled [5, 6] radiance incident at the AVIRIS aperture for the vegetation canopy reflectance spectrum. This spectrum includes the combined effects of the solar irradiance, two-way transmittance, and scattering of the atmosphere, as well as the reflectance of the vegetated canopy. This is the radiance in 25002200190016001300 Wavelength (nm) 1000700400 0 5 10 Radiance (uW/cm^2/nm/sr) 15 20 Figure 14.8 Modeled upwelling radiance incident at the AVIRIS aperture from a wel-illuminated vegetation canopy. This spectrum includes the combined effects of the solar irradiance, two-way transmittance, and scattering of the atmosphere, as well as the vegetation canopy reflectance. © 2008 by Taylor & Francis Group, LLC 344 High-Performance Computing in Remote Sensing 224208192176160144128112 Channel (#) 9680644832160 0 500 1000 1500 2000 Signal (DN) 2500 3000 3500 4000 Figure 14.9 AVIRIS measured signal for the upwelling radiance from a vegetation covered surface. The instrument optical and electronic characteristics dominate for recorded signal. terms of power per area per wavelength per solid angle available to measure for the pursuit of imaging spectroscopy. As with any radiance measuring instrument, AVIRIS has optical components that collect the incident light and focus it on a detector. At the detector the incident light is converted to measurable signals that are amplified, digitized, and recorded. Figure 14.9 shows the AVIRIS recorded signal for the vegetation canopy upwelling radiance spectrum. The AVIRIS signal has no inherent radiometric or spectral calibration and is recorded as digitized number (DN) versus channel. The process of spectral and radiometric calibration in the AVIRIS data processing subsystem converts the measured signal to units of spectral radiance. Considerable effort has been expended over the years of AVIRIS’ operation to develop spectral, radiometric, and spatial calibration methods in the laboratory [7, 8]. A companion effort has been applied to validate the calibration of AVIRIS in the flight environment [9, 10]. Figure 14.10 shows the AVIRIS calibrated radiance spectrum of the vege- tation canopy target. Accurate calibration is an essential requirement for the spectra measured by AVIRIS or any imaging spectrometer to be analyzed quantitatively for science research or application objectives. If the objective of the investigation is the understanding of surface properties, the calibrated radiance spectra must be corrected for the effects of the atmosphere. Atmo- spheric correction generally includes compensation for the solar irradiance as well as atmospheric absorbing and scattering effects. Figure 14.11 show the atmospherically corrected spectrum for the vegetation canopy target. The portions of the spectrum lo- cated in the strong atmospheric water vapor absorption bands near 1400 and 1900 nm are lost due to the lack of a measurable signal. © 2008 by Taylor & Francis Group, LLC [...]... precision or signal-to-noise ratio For this imaging spectrometer, the precision is specified at a set of four reference radiances given in Figure 14. 15 The corresponding signal-to-noise ratios required are shown in Figure 14. 16 The signal-to-noise ratios specified here are high in order to pursue © 2008 by Taylor & Francis Group, LLC 354 High- Performance Computing in Remote Sensing TABLE 14. 5 Nominal Characteristics... pursuing research questions spanning a wide range of scientific disciplines References [1] G Vane, A F H Goetz, J B Wellman, Airborne imaging spectrometer: A new tool for remote- sensing IEEE Transactions on Geoscience and Remote Sensing, vol 22, pp 546–549, 1984 [2] A S Mazer, M Martin, M Lee, et al., Image-processing software for imaging spectrometry data-analysis Remote Sensing of Environrment, vol 24,... Analysis: Elemental and Mineralogical Composition, C M Pieters and P Englert, Eds., Cambridge University Press, New York, pp 145 –166, 1993 © 2008 by Taylor & Francis Group, LLC 358 High- Performance Computing in Remote Sensing [19] D A Roberts, M Gardner, R Church, et al., Mapping chaparral in the Santa Monica Mountains using multiple endmember spectral mixture models, Remote Sensing of Environment, vol... Reflectance (z23.5) SNR 0.50 Reflectance (z23.5) Signal-to-Noise Ratio 1000 800 600 400 200 0 350 650 950 1250 1550 1850 2150 2450 Wavelength (nm) Figure 14. 16 The signal-to-noise ratio requirements for each of the bench-mark reference radiances © 2008 by Taylor & Francis Group, LLC 356 High- Performance Computing in Remote Sensing TABLE 14. 6 Earth Ecosystem Imaging Spectrometer Data Volumes Period Volume Period... beyond the scope of this chapter Finally, given the diverse spectral signature content of high precision and high uniformity spectra, there is clear potential for the development of new algorithms and approaches for the extraction of valuable information from existing AVIRIS measurements © 2008 by Taylor & Francis Group, LLC 348 High- Performance Computing in Remote Sensing 14. 3 Objectives and Characteristics... Visible Infrared Imaging Spectrometer (AVIRIS) Remote Sensing of Environment, vol 44, pp 127 143 , 1993 [4] R O Green, M L Eastwood, C M Sarture, et al., Imaging spectroscopy and the Airborne Visible Infrared Imaging Spectrometer (AVIRIS) Remote Sensing of Environment, vol 65, pp 227–248, 1998 [5] A Berk, L S Bernstein, D.C Robertson, MODTRAN: A moderate resolution model for LOWTRAN 7 Final report, GL-TR-0122,... spatial sampling in the along-track and crosstrack directions For each 70 m advance of the image swath in the orbit direction around the moon, a full set of 600 cross-track spectra will be read out from the © 2008 by Taylor & Francis Group, LLC 350 High- Performance Computing in Remote Sensing TABLE 14. 3 Spectral, Radiometric, Spatial, Temporal and Uniformity Specifications of The M3 Imaging Spectrometer... mission, final © 2008 by Taylor & Francis Group, LLC 352 High- Performance Computing in Remote Sensing calibration and validation the complete data set will become available Throughout the M3 mission and following, the imaging spectrometer measurements will be used to characterize and map the lunar surface composition and to assess and map mineral resources at high spatial resolution to support future missions... Reflectance 0.7 N-Olivine Olivine . Group, LLC 342 High- Performance Computing in Remote Sensing Cross Track Sample Wavelength Figure 14. 6 Depiction of the spectral cross-track and spectral-IFOV uniformity for a uniform imaging spectrometer improved since the initial development of AVIRIS based upon the requirements from scientists © 2008 by Taylor & Francis Group, LLC 340 High- Performance Computing in Remote Sensing Figure 14. 4. spectral and spatial domain of this full range instrument continued to grow the role for high- performance computing in the field of imaging spectroscopy. 14. 2 AVIRIS and the Imaging Spectroscopy Measurement The