Overseas Visit Report

AVIRIS Science & Applications Workshop
JPL/NASA
8 - 11 February 1999

Cindy Ong
CSIRO Exploration & Mining
Alex Held
CSIRO Land & Water

Background

My request to attend the AVIRIS workshop was for the purpose of developing my skills in the analysis and processing of remotely-sensed hyperspectral data for vegetation characterisation and BRDF correction, especially within the mine environment. I also took the opportunity to visit some of the key people in the hyperspectral field.

NASA JPL AVIRIS science and applications workshop (jointly written with Alex Held)

The AVIRIS Earth Science and Applications workshop was held on February 8 to 11, 1999 at the NASA Jet Propulsion Laboratory. The workshop started with a one day introductory course on the topic of Imaging Spectroscopy. Four researchers ranging from those involved in instrumentation, applications and software presented overviews of their work. The discussions included aspects of imaging spectroscopy and specifics of the AVIRIS sensor used by NASA and most US investigators. This sytstem is by far one of the best understood and refined imaging spectrometers, with exceptionally good calibration stability, signal:noise ratio and cross-scene uniformity. Robert Green, the principal AVIRIS scientist, has developed good routines for use of some of the water vapor absorption bands in the AVIRIS imagery itself to derive the necessary atmospheric correction parameters to produce reasonably corrected images. Other effects such as aerosols are corrected for, on a scene-basis by looking at the radiance changes across sharp bright-dark edges in the imagery (e.g. buildings with shadows), and then adjusting the parameterisation in the Modtran code until a smooth water vapor image is obtained across the edge. Recent discoveries of the use of a 1380 nm band for thin cirrus cloud detection, are also being implememented in the correction routines. We found that this session gave a good background and set the scene well for the rest of the workshop.

The period from the 9th to 11th was dedicated to papers presented by investigators who have acquired AVIRIS spectral images for their scientific research and applications.

Presentations that were of most interest to us were:

Atmospheric correction

Robert Green said that atmospheric correction can only be derived if the radiances are calibrated absolutely at the 1 to 5 % level. It appears that all the atmospheric corrections algorithms currently used (mainly Green's and Gao's) leave artefacts in the corrected data. The group at USGS overcomes this by intensive (1 to 2 person month) field calibration of an invariant site and adjusting to this. Most other groups also collect calibration data simultaneous with overflights whenever this is possible. (Notably, a number of investigators felt frustrated at the impossibility of closly scheduling of the AVIRIS overflights and calibration/validation field campaigns by their teams).

Joe Boardman and others use software called EFFORT to 'clean' up any remaining sharp edges or unexpected spectral features caused by sensor effects of limitations in the atmospheric correction procedures. The US Airforce is also developing an atmospheric correction routine called FLAASH based on MODTRAN (4), but this is not available to the general public yet.

An interesting output from Green's atmospheric correction is leaf water content (or as they refer to as liquid water as different from column water vapour), which could be a useful parameter for deriving parameters such as plant physiological status, stress and fire hazard assessments.

Vegetation

In contrast to previous years, where geological applications featured most prominently, ecological land/water applications of AVIRIS data dominated at least one of the three days in the workshop, suggesting that biological sciences are starting to see the value of hyperspectral data to their research as well.

Dar Roberts' group at UC Santa Barbara have developed the 'multiple endmember spectral mixture analysis (MESMA)', based on a linear combination of endmembers in which the number and composition of endmembers are allowed to vary for each pixel and the best model with the lowest RMS fit is chosen. It appears that huge computational power is needed to run the process, they are currently using a multi-processor supercomputer. Dar Roberts presented work where MESMA was applied to species mapping in the Santa Monica Mountains with very good results. From this the concept of regionally specific (and it's importance was emphasised) spectral endmembers was presented, as a much more logical alternative to the collection of large, wide-ranging vegetation spectral libraries. The need to also characterise the seasonal nature of vegetation spectral signatures was also highlighted during our discussion with Dar. The issue of field procedures for spectral endmember collection was considered to be an area of urgent attention. In some instances, collection of image-derived spectra from the new high resolution (3-4m pixel) AVIRIS data was found to be a good option for collection of spectral library data, for subsequent low resolution mapping. Separation of dominant species was found to be better at the higher AVIRIS resolution. MESMA was also applied to fire hazard assessment to obtain species composition, green vegetation fraction (GVF), non-photosynthesising vegetation (NPV). Fuel loads were obtained from the GVF and NPV, NDVIs and liquid water content derived from Green's atmospheric correction. This data and the detailed species maps derived by UCSB were very well received by the state fire service agencies.

Interesting aspects of active fire mapping were also presented by this group and Robert Green, where burning fires could be clearly detected through smoke at the 2.2-2.5 micron wavelengths.

John Gamon of California State University - LA, presented their work in the development of ecosystem flux models with water and pigment indices. Their ground-based work, at the leaf to canopy scale, indicated that several narrow band features can be used to detect plant physiological status. This study extends this work to the airborne scale. They found that the water absorption band feature, expressed as an index (905nm/972nm) is a good measure of canopy water content and it is also correlated to liquid water content derived from AVIRIS and seem to work across scales, ie. from ground base to airborne. The photo chemical reflectance index,

which gives an indication of xanthophyll cycle pigment activity, was found to be a very scale-dependent index. It works well on the leaf scale but may not at the airborne scale. It may be useful if individual shrubs can be measured.

Chris Van der Ven presented work conducted at Stanford University in the White-Inyo Mountains on the relationship between rock types, elevation and vegetation. Extensive field spectral and vegetation survey was carried out which indicated that there was a relationship between rock types and vegetation and elevation and vegetation. Only preliminary results from AVIRIS were presented.

Ray Kokaly presented results of work at the USGS mapping the biota of Yellowstone national park. Mapping was based on the Tetracorder algorithm (see definition below). Notably, in many of these studies, the use of ground reference reflectance panels is being used more and more for final refinement of the atmospheric correction process.

Some work was also presented in the use of AVIRIS data for water-based remote sensing. Laurie Richardson and Fred Kruse presented some work on the use of AVIRIS for mapping the type and mix of different suspended algae types in the Florida Bay. A variety of pigment absorbance signatures extracted from different algae were used as endmembers to drive a Spectral Angle Mapper classification. Results were quite good and showed interesting variability in the algae types in the different parts of the Bay. Bagheri and Dekker also presented their proposed approach for use of AVIRIS data for water quality mapping in the Hudson River, off New York.

Acid Mine Drainage (AMD)

Cindy was astounded to hear of the large number of abandoned land mines in the US. In Western USA alone there are approximately 200,000. Some years ago (1995), the USGS in collaboration with the USEPA found that AVIRIS data could be used as a tool for identifying areas of AMD potential. The USEPA has recently commissioned a huge study on this subject (Utah Abandoned Mined Lands). The main objective is to expand and evaluate the future use of hyperspectral data, to identify, measure, map and monitor the effects of mine waste releases on watersheds and other aquatic environments. 1.48 million acres of AVIRIS data was collected. A commercial-stakeholder venture was also established to demonstrate available commercial capabilities in hyperspectral imaging. Data were collected with SFSI, CASI and Probe1 (Hymap) with the goal of improving the understanding of the environmental application of hyperspectral data, developing common and reasonable standards for these instruments and verifying and expanding the reliability and use of hyperspectral imaging for environmental applications.

The two main approaches appear to be the USGS's approach of minerals mapping using the tetracorder algorithm and the commercially available packages such as ENVI. USGS's tetracorder is essentially a set of algorithms based on continuum removal and multiple spectral shape fitting. Mineral maps of sulfate bearing materials have been produced.

Other general

A mayor feature of this year's presentations, is that several groups presented data from the initial low altitude AVIRIS tests, carried out from a NOAA Twin Otter. Pixel sizes here ranged between 1.7 m and 4 m (see http://makalu.jpl.nasa.gov/html/lowalt.html for sample data).

We were pleasantly surprised with the output from the geometric correction routines written by Joe Boardman for the AVIRIS low altitude data. This was based on simple geometry and bundle adjustments using an INS-GPS. A D-GPS system and possibly incident-light sensor are being proposed for the '99 campaigns. As discussed above, many investigators are extremely excited about this new capability, due in part to the opportunity to measure and map more pure spectral features and smaller objects. From the point of view of processors who have spent months at a time picking enormous numbers of control points for accurate mapping, this would be a God-send and an INS-GPS was definitely be worth the investment. We believe that Integrated Spectronics are in the process of installing one for the HyMap which would also improve the geolocation capabilities.

An interesting feature of these annual workshops, is that opposing views of spectral un-mixing and analysis are presented, notably between the groups headed by Roger Clark, which advocate the collection of thousands of endmember spectra and use of the Tricorder software to identify these in the imagery, and Joe Boardman and friends, who advocate the extraction of endmembers from the imagery itself and use of pixel-purity index-type methodologies and limited pure endmember spectra for unmixing, For more details on all papers, contact details of the authors and copies of the proceedings, check the AVIRIS web site (above) in coming months.

Developments on the Warfighter1/Orbview4 sensor were presented. This is to be a small, inexpensive space-based hyperspectral sensor. It will have a 1m panchromatic band, 4m multi-camera, 8m hyperspectral sensor with 200 bands in VIS-SWIR and 80 in MWIR.

The Brazilians are developing a 20m resolution sensor, CBERSI with 80 bands in VIS and 160 in IR.

Switzerland and Belgium are developing a 400-2500nm sensor with 160 5nm bands in the VIS/NIR and 140 10nm bands in SWIR.

Alex also attended the AIRSAR workshop running concurrently at JPL. Some of the main aspects here included discussions around new improvements in the DC-8 navigation, SAR data processing throughput and some of the new experimental SAR modes being implemented, such as L-Band Quad-Pol 80 MHZ mode, Along-track interferometry and polarimertic interferometry. Interesting results using this later mode were presented by Ernesto Rodiguez, VanZyl and Robert Treuhaft, all from JPL. Their results and simulations suggest that this new technique has very good potential for improved plant canopy structure characterisation. Dr. Rodriguez is heavily involved in the deployment and future use of the planned (Sept. 99) Shuttle Radar Topography Mission, which will be using the previous SIRC/X system carried in previous Shuttle missions, but using a 60 m long extended antenna to enable the interferomertic capability during the 11-day shuttle mission.

The planned PACRIM 2000 mission was also mentioned and a number of papers from the '96 Pacrim were presented by some invited Asean investigators. Simon Hook (NASA Goddard?) also presented aspects of the MASTER (Mosis-ASTER) simulator which is being prepared for installation into the same DC-8 as AIRSAR and for use during PACRIM. The sensor will have 25 channels in the .4 - 2.5 um range, 15 channels in the 3 - 5 um range and 10 channels in the 7 -13 um range. With an anticipated pixel size of 5 -25 m, this data would be a very interesting complement to the AIRSAR data. !

On the second day of the workshop, more application papers were presented, including some for work in inter-tidal and coastal areas of various countries of the SE Asia region, covered during last PACRIM mission. We anticipate, that copies of the abstracts and papers presented during this workshop, would be available soon at the new NASA AIRSAR web site (http://airsar.jpl.nasa.gov/). Data from previous missions is also can be ordered through this site.

Visits

University of California, Davis Dr. Susan Ustin

Alex Held and Cindy Ong visited Dr. Ustin's lab and met some of her students. Unfortunately our visit occurred during project proposal submission deadline so time with her was at a premium. Nevertheless, aspects of the proposed ARIES work between Alex, Susan, Dar Roberts and John Gamon were discussed. We had a good opportunity to present our work to a handful of people in her faculty and students. Susan was impressed by the amount and diversity of the work that we conduct both at the MRRP and CLW. We shared some similar experiences with the video system as they have been working on a project involving a video system similar to the DMSV. At the time we were there they also received a request to look at plant architecture in a grapevines using a video system. One of Susan's staff had also developed a program that interactively looked at the different stage and outputs in Robert Green's atmospheric correction algorithm to help understand the sources of atmospheric correction errors.

California State University, Los Angeles. Prof. John Gamon. (Cindy)

I visited John Gamon and his staff at Cal State LA at a very busy period in their schedule but they were good enough to spend some time with me showing and discussing their work. John gave me a tour of their glass house and laboratories and explained some of the experiments they are performing at the leaf and canopy scale with optical instruments. Their work centres on the effects of environmental factors on plant physiology, structure, and productivity with a primary focus on factors controlling photosynthetic physiology. Their approach has been to apply optical techniques (involving remote sensing, spectrometry and image analysis) to evaluate plant productivity and physiological performance. One of the interesting off-shoots of the work was John's development of a VIS/NIR spectrometer, with accompanying accessories such as leaf clips, specifically designed for plant physiology studies.

This spectrometer is now produced by PP Systems and marketed as the UniSpec. John is very happy with this instrument and believes that most of his work could be satisfied with a VIS/NIR instrument. We also discussed his comparison of two (VIS-SWIR) commercial instruments. There are some sensitivities in the contents of this discussion so I will not go into details here. They are also working on projects that extend their work from the leaf/canopy scale to the larger airborne scale. While I was there they had recently received their AVIRIS data and had some preliminary results for a meeting the next week. I also took the opportunity to discuss some aspects of our work in the MRRP and he has given me some good ideas.

John Deeds and Ruth Hobbs, NOAA. (Cindy)

I managed to have a very quick chat with Chris Elvidge at the AVIRIS workshop and he arranged for me to meet his staff at his laboratories. I was interested in the work that he was doing in light detection and fire mapping. John and Ruth talked about their work in the applications of the DMSP data for global fire detection and detection of city lights as an indicator of population density. Ruth discussed the processes of implementing their work to an operational level. I was keen to see their work with turtles in Florida as this was in the same line as our work on the oil platforms. Unfortunately, they were not able to show me much as the work had not progressed as they are still awaiting funding.

Dr. Roger Clark & Mr. Sam Vance, USGS (Cindy)

Roger and Sam took me on a tour of the USGS laboratories and discussed aspects of the acid mine drainage and other vegetation mapping projects. They are very well equipped with an array of laboratory spectrometers (Beckman 5270 spectrophotometer, Nicolet model Magna 760 FTIR spectrophotometer and Mossbauer and two ASD FR field spectrometers which they used for their calibration of airborne data and other field collection. Roger believes the ASD to be the best field instrument in the market because of it's spectral resolution and stability. The crew reported that they have worked in temperatures as severe as the Pilbara and the ASD had performed well.

The USGS also have a very comprehensive library of samples including the complete collection of samples of Graham Hunt and Jack Salisbury, complete set of well-characterised samples from the Source Clay Minerals Repository, University of Missouri (26 samples), complete set of Wards Scientific rocks (1000 samples). Roger discussed the technical aspects of the airborne acid mine drainage project and Sam discussed the EPA perspective and it's interactions with the industry.

It was interesting to note the amount of time they are able to devote to field-work. For example, they spend 1-2 person months on a calibration site collecting hundreds of spectra.

Dr. David Ellerbroek, Montgomery Watson (Cindy)

Dave and I had discussions on commercial work at the minesites that we are jointly working on. We also discussed the Utah AMD projects. Dave is involved in the cleaning up of some of these minesites.

Prof. Ron Lyon and Mr. Chris Van Der Ven, Stanford University (Cindy)

My last meeting was with Ron and Chris at Stanford University. I was interested to see what they were doing as Ron is a mentor in this group and a lot of the group's philosophies have backgrounds that lead to him. Chris' work on establishing the relationship between elevation, rock types and vegetation is of interest to the group as we have been involved in some geobotanical studies. We have some hyperspectral data over the Pilbara site where we have worked. Chris discussed technical aspects of processing the data and showed me some results of his work.

At that stage they had more work to do and I will be keeping in touch with them to see their progress. Again I note with interest the amount of time they have devoted to field-work. They were able to conduct an extensive field campaign characterising rock types and vegetation species for the AVIRIS acquisition and are planning to do the same for this year's acquisition.

Copyright CSIRO 1999 ©