22nd ASTER Science Meeting

Summary

The latest ASTER Science Team Meeting in Tokyo encompassed several issues and developments from both American and Japanese scientist working with ASTER data. Radiometric calibration of Level 1 products and QC issues related to processed Level 2 surface products figured high in the discussions of the meeting. It is apparent that fine tuning of the processing of Level 1A to Level 1B (radiance at the sensor) and to Level 2 is still being undertaken at ERSDAC (Japanese Space Agency) and NASA/EDC (EROS Data Center - USGS). However case histories presented at the public workshop on the May 23rd indicated that useful geological and environmental information is being extracted from ASTER datasets as presently supplied by ERSDAC and EDC.

ERSDAC stated that out of 413,874 ASTER scenes collected from the 15/5/2002, 319,555 scenes have been processed into Level 1A, of which 150,393 meet the “cloud-free threshold” of 0-20% and 74,646 have been further process into Level 1B by 10/5/2002. Australia was second after the US for downloading ASTER data volumes during April 2002 followed by several European countries. The emphasis continues to be on the Global Mapping operation which is now 60 % complete according to ERSDAC. By comparison only 4 % of STAR (Science Team Acquisition Requests) orders have been entirely completed.

ERSDAC reported that their validation of Level 2 products (independent of the USGS-EDC/NASA Level 2 products) is behind schedule although their D-stretch and DEM products have been released in May. In particular ERSDAC’s surface radiance and temperature-emissivity TIR products are still in the process of being validated. New versions of Level 1B registration were being produced from Level 1A data to refine for cross track mis-registration between bands however the differences are subtle and sub-pixel. The last product to be validated from the USGS EDC site is the Surface Emissivity and Kinetic Temperature products, which should be finalised by June 2002. There is no plan to process DEMs from Level 1B data.

A status report on the ASTER instrument mentioned that the SWIR detector temperature were now being maintained within normal ranges and the coolers were expected to last for the project’s duration. The ASTER instrument is now expected now to last for a total of six years or more and will continue to supply data, funds permitting. The TIR and VNIR RCCs (Radiometric Calibration Coefficients) are to be updated soon with the possibility of the thermal bands changing once per 3 months, emphasising the importance of the checking calibration coefficients from the HDF format for each scene/granule. Cross talk (“leakage” between detectors) of SWIR bands is regarded as a significant problem by several US scientists, particularly noticeable for bands 5 and 9. In particular cross talk can handicap the reliable identification and comparison of spectral signatures with laboratory/field measurements or corrected airborne hyperspectral survey signatures. The correction of the ASTER data for water vapour is not possible with the MODIS estimates at present and default models are used for deriving surface products. Comparisons with Landsat reflectances or other airborne sensor data is also complicated by the use of the different WRC (World Radiation Centre) solar irradiance model for deriving surface reflectance products from both ASTER and MODIS radiance data.

The suite of programs used to convert Level 1B data to Level 2 surface products (Winicar), including surface reflectances and emissivities, will be unlikely to be released by the end of the year. This software is presently restricted to US scientists however a request has been submitted to the US Commerce and State departments to allow access to non US citizens although this will delay access until next year sometime.

EDC stated that they would start charging for ASTER data no later than 1st July 2002 and their prices would be comparable to ERSDAC (Y9,800; ~ $A 140 @ Y71 / $A) but probably cheaper for on-line downloading.

Calibration and QC issues

Cross Talk

The issue of cross talk between ASTER SWIR bands was discussed at length during the ASTER Science Meeting and the following description of the problem and examples have been obtained both during the discussions in Tokyo and subsequent email contacts, particularly with Kurt Thome (Assoc. Prof. Optical Sciences, University of Arizona) and Larry Rowan (USGS, Reston, Va.).

The primary band causing most of the cross talk problem is band 4. What they believe is happening is that some of the incident photons onto the band 4 detector plane are reflected (not a surprise since even a perfect, single detector can reflect as much as 30% of the incident light). The problem is that there are no baffles or other
structures blocking light from band 4 bouncing around to the detectors for the other bands. Basically, the detectors are arranged in a rectangular geometry and the top of the rectangle contains all of the filters for all of the bands. The bottom contains the detectors for all bands (and all of the detectors have the same spectral response). Once an incident photon enters, it is basically trapped until some detector collects it. If the band 5 detector gets a photon through the band 4 filter, cross talk occurs.

The basic problem is that the solar output in band 4 is considerably higher than the other SWIR bands. Hence, even a small number of band 4 photons leaking out can have a big effect in the other bands. The effect is largest in bands 5 and 9 because those detectors are physically the closest to the band 4 detectors. The correction at this point is assumed to be an offset based on the pixels location in the scene. Essentially, a Gaussian distribution is drawn around the pixel of interest in band 9 (for example). The band 4 scene is then examined to determine the radiance of each pixel within the Gaussian, and then the contribution due to cross talk from each of these pixels is determined by the radiance of the pixel and the Gaussian value acting as weighting function.

Larry Rowan of the USGS has done comparison studies between ASTER Level 2 surface reflectance data (EDC) and NASA’s AVIRIS airborne hyperspectral sensor (resampled to ASTER bandpasses) of the Cuprite hydrothermal deposit for several different units as shown within the attached file “cup-spectra_Larry_Rowan.pdf”. In particular these figures show the resampled AVIRIS spectral signatures (upper left), the standard surface reflectance EDC product (AST_07) (lower right), the ASTER Level 1B data corrected by ACORN software using the MODIS derived water vapour estimates (upper right), while the lower left figure shows ASTER data processed with ACORN and MODIS (as for the upper right) but also cross talk corrected as developed by ERSDAC. These figures show that the discrepancies between the AVIRIS and ASTER data is a combination of several effects, including cross talk, the assumed solar irradiance curve and atmospheric water vapour uncertainty. According to Kurt Thome, discrepancies between the default MODTRAN solar irradiance curve used commonly used by Landsat (or airborne sensors) and the ASTER assumed WRC solar curve can result in differences of approximately 8%. Larry Rowan concludes that ASTER’s Band 5 reflectance can be 2-10% too low while its Band 9 can be 20-30% too high. Kurt Thome predicts that mis-calibration of Band 5 will be corrected (assuming similar solar curves) using the final version of the Cross Talk software being developed by ERSDAC. This software will hopefully be available by or soon after August 2002 and possibly incorporated as part of the EDC ASTER processing. Accurate calibration of ASTER band 9 however is considered problematic because of the uncertainty of water vapour estimates for atmospheric correction.

TIR striping and Temperature-Emissivity Separation

Two types of striping noise is apparent in the Level 2 emissivity product and occurs either as horizontal striping or diagonal stripping. The diagonal striping appears to be an artefact of the cubic convolution resampling filter applied in the process of Level 1A to Level 1B. A proposal was suggested to changeover from cubic convolution to bilinear nearest neighbour during Level 1B processing to avoid this artefact, although further investigations and acceptance by ERSDAC will be required.

It was also decided to change the MMD (Minimum Mean Difference) threshold assumed as part of the emissivity-temperature separation algorithm for water body discrimination from 0.0006 to 0.032, affecting the derived absolute emissivity values although not the spectral shape. It is anticipated that the surface TIR products (emissivities and kinetic temperatures) will be validated by June 2002. Validation field exercises presented by Tom Schummge (US Dept. Agric., Maryland), of emissivities within an arid area of New Mexico, indicated a 2% quantitative accuracy between field/lab emissivity determinations and ASTER derived values. Work done by Matsunago of the Japanese National Institute for Environmental Studies also indicated good comparison between NASA’s airborne TIMS (Thermal Infrared Multispectral Scanner) sensor and ASTER data at Cuprite. In particular the average difference between ASTER and TIMS derived emissivities was less than 0.014 emissivity units for ASTER data collected in 2000 and 2001, excluding band 10 which showed a discrepancy of up to 0.034 with TIMS estimates. Simon Hook (JPL) also presented results which showed favourable comparisons between vicarious temperature water measurements and ASTER derived surface temperatures within ± 0.2 o C and a NEDT (Noise equivalent temperature difference) of 0.15 o C

Radiometric calibration

On-going calibration studies into the performance of the VNIR, SWIR and TIR detectors were reported by both American and Japanese scientists. Onboard calibration studies of the VNIR using the two lamps indicated a degradation of Bands 1, 2 and 3 of 8 %, 6 % and 3% respectively. The deterioration in VNIR bands is considered to be due to contaminants from the rocket thruster but Japanese Science Team members did not consider it necessary to change the radiometric calibration coefficients as yet.

Similarly the SWIR detectors have been monitored for any change in radiometric performance by onboard calibration lamps but have shown a less than 2 % degradation with time and the radiometric calibration coefficients do not require revision at this stage. The main calibration SWIR issue are the effects of cross talk and fine tuning of the cross talk correction software is currently undertaken by ERSDAC.

The TIR detectors however have shown a trend in performance behavour when compared to onboard blackbody sources, particularly for bands 12 and 14. It is anticipated that the radiometric calibration coefficients will change in the near future and possibly at regular intervals in the future.

Concern was expressed at the delay of the updates of calibration coefficients for newly acquired ASTER data and also resolution of the SWIR cross talk issue. It was agreed that these would be resolved at the next ASTER Calibration Team meeting in July 2002.

Application Studies and projects

Larry Rowan presented two case studies using ASTER at Cuprite, Nevada, and also at Saindak, Pakistan. The Cuprite ASTER 1B data were calibrated using ASD field spectral measurements (convolved to ASTER bands) and matched filter partial unmixing was applied using scene derived endmember signatures. ASTER mapped host rocks and areas of hydrothermal alteration including units rich in alunite, kaolinite, opaline silica, muscovite, muscovite-chlorite, calcite/carbonate, salt playa and unaltered tuff.

Rowan’s study of the Copper porphyry belt in western Pakistan used Level 2 surface reflectance ASTER data to map alteration with a series of band ratios. No correction for cross talk was applied. Sericite was mapped using (B5+B7)/B6, Alunite and/or Kaolinite rich areas (argillic alteration) were mapped with (B4+B6)/B5 while Ferric oxides were mapped with B2/B1. These relative band depths of Bands 5 and 6 were used to map areas of porphyry alteration. Partial spectral unmixing was also attempted with this data after adjusting Band 9 with a gain factor of 0.90. This unmixing mapped areas rich in ferric iron, sericite, epidote-chlorite, carbonate and ferrous iron. These results have been submitted to Economic Geology for publication. Rowan also found that the emissivity data produced quite good results for well exposed areas with good spectral contrast however they gave poor results for areas with reduced spectral contrast, as in areas of extensive vegetation cover.

Shibata et al. of the Mitsubishi Materials Natural Resources Development Corp. mapped alteration zones associated with massive sulphide deposits in the sparsely vegetated northern part of Oman using decorellation stretching of various band combinations of Level 1B ASTER data. In particular Shibata et al. found that D-stretches of ASTER bands 1-4-8 and 8-6-4 highlighted the areas of alteration that included silicification, chloritization and epidotization. A D-stretch of bands 1-3-4 highlighted the gabbro and basalt units as well as the intensity of alteration (attached figure) while B4/B1 mapped Fe oxides. Shibata et al. also found it useful to drape the VNIR over a 3-D image of the ASTER DEM. Although the TIR bands mapped the gabbro units it was found that the SWIR (Level 1B) bands produced better results.

Ninomiya of the Japanese Geological Survey presented his results of processing TIR ASTER Level 1B bands using series of band ratios to map siliceous, carbonate and mafic-ultramafic rocks. His description was similar to the presentation made at the IGARSS 2001 Sydney Conference where siliceous rocks are mapped with the ratio (B11 * B11) / (B10*B12). A carbonate index can be mapped with B13/B14 however a correction for surface temperature variations is achieved by normalising the Level 1B data with the Brightness at the Sensor temperature ASTER Band 13 data (AST_04). A SiO2 content index can be also achieved using B12/B13. Ninomiya’s technique is more fully described in Proc. SPIE Vol. 4710, pp. 191-202, Thermosense XXIV, 2002, also listed in the following web site:
(http://spie.org/scripts/abstract.pl?bibcode=2002SPIE%2e4710%2e%2e191N&db_key=INST&qs=spie&s_type=paper)

Miyatake of the Metal Mining Agency (MMAJ) of Japan successively used mosaicked ASTER VNIR, SWIR and TIR Level 1B scenes for mapping limonite, alteration alunite / clay and silica content images for exploration within the Central Volcanic Arc of Myanmar (Burma). Flat field calibration of the ASTER data was performed using a spectrally flat quartz rich detritus area and SWIR and TIR data was registered to the VNIR with actual GCPs to correct for subtle inter-telescope mis-registrations. The limonite content images were interpreted from B2/B1 and then further density sliced. The clay minerals (kaolinite, muscovite) and alunite images were generated by spectral angle mapping techniques using field spectra measured adjacent to the Monywa Mine as SWIR reference pixels. The silica content images were produced with alpha residual derived emissivity estimates from the ASTER Level 1B TIR data, although no atmospheric correction had been applied previously. This silica content information was derived from MMAJ’s “K – value” defined by :