Overseas Travel Report
1998 IEEE International Geoscience & Remote Sensing Symposium
(IGARSS'98)
Seattle, USA - 6 - 10 July 1998

Ian Grant
CSIRO Atmospheric Research



I attended the 1998 IEEE International Geoscience and Remote Sensing Symposium (IGARSS'98) in Seattle, USA, 6--10 July 1998. David Jupp from the CSIRO Earth Observation Centre (EOC) also attended. Well over half of the presentations were devoted to radar, but it is still a major international meeting in the optical remote sensing of the earth's surface and, to a lesser extent, atmosphere. Over the next three years IGARSS will be held in Hamburg, Hawaii and Sydney (2001). I have a hardcopy program and the proceedings on CD-ROM for this meeting, and also CD proceedings for the 1994, 1996, and 1997 meetings. The rest of this report summarizes the papers I saw and discussions I had that I think are of most interest to the EOC and particularly CAR.

AVHRR albedo and BRDF

My main priority was to learn what was being done in the areas of retrieval of surface albedo and BRDF (Bidirectional reflectance distribution factor - a quantitative description of the anisotropy of the surface reflectance) from satellite observations, particularly AVHRR, since getting results in those areas for continental Australia will be the focus of my work with the EOC over the next couple of years. Two presentations were devoted precisely to the generation of global albedo maps from AVHRR.

Nic Strugnell of Alan Strahler's group at Boston University described a scheme to generate albedo maps from AVHRR until high quality BRDFs are available from MODIS. He recognised that for BRDF work a new landcover classification based on vegetation structure is necessary, and so he established a seasonally-dependent mapping from an existing landcover classification into a BRDF classification. David Jupp (and previously Dean Graetz) have commented that vegetation structure is already well mapped for Australia but not for North America or other continents - so we have an advantage here. To each landcover class Strugnell will assign a BRDF shape from a single field measurement, then for each month scale the BRDF to fit a single AVHRR observation selected by a strict compositing and cloud-detection scheme. He thus assumes that as the state of a patch of vegetation changes the scale of the BRDF, and hence the albedo, may change but not the shape of the BRDF. He justified this by noting that darker surfaces generally have "flatter'' BRDFs - shallower bowl shape and less difference between forward scatter and backscatter -but I think the assumption needs validation and others I spoke to agreed.

I discussed with Garik Gutman and Ivan Csizar of NOAA/NESDIS their scheme to derive global albedo maps from AVHRR data (GAC composited or PATMOS). They use an empirical narrowband to broadband correction and the ERBE angular models to derive the albedo at the top of the atmosphere (TOA), and then use Li and Garand's parameterization (dependent only on solar zenith angle and atmospheric water vapour column amount) to convert to a surface albedo. They estimate the total relative error to be 15% and freely concede that many of the steps could be improved. They considered the alternative path of atmospherically correcting the radiances first, then doing the spectral and angular corrections at the surface but dismissed it because the spectral and angular reflectance properties of the surface are not sufficiently well characterised at present.

I came away from the meeting believing that the approach of fitting the BRDF to timeseries of non-composited AVHRR observations, commenced by Denis O'Brien and Ross Mitchell and to be continued by me, is potentially a significant improvement over the two approaches just described and very worthy of pursuit. The key is that the new approach will use what information is contained in the AVHRR observations regarding the BRDF of the surface at the location and time in question, rather than relying on a single field-measured BRDF shape, or on the fixed ERBE angular models (there are only two for clear land: one for desert, one for vegetated surfaces).

While the final MODIS BRDF/albedo algorithm (known as Ambrals) picks the best fitting of five sets of kernels, the at-launch processing will use only the RossThick-LiSparse kernel combination. Wolfgang Lucht told me that while the MODIS BRDF product was originally going to omit Australia due to processing limitations, there is a good chance now it will be included. He also showed me some recent work on simulations of expected errors in retrievals from sparse sampling, including the extrapolation of albedo to solar zenith angles away from those of the measurements. I showed him my recent work on the variation with solar zenith angle of the albedo of the Hay site.

Crystal Schaff of Boston University applied the Ambrals algorithm to combined AVHRR and GOES-8 data for New England and found that the retrieved albedo maps showed the spatial structure and ranking of brightness, and even BRDF shape, expected from a map of landcover class but was unable to make a more quantitative check. The prediction of the GOES radiance from the AVHRR observations alone was close but had noticeable deviations. When I asked Wolfgang Lucht how well the Ambrals BRDF fit was constrained by AVHRR observations alone he confirmed that they had no way of telling other than by looking at the reasonableness of results of trials. Alan Strahler presented work by Feng Gao of the Chinese Academy of Sciences on a new criterion for fitting BRDF models to angularly sparse data sets, based on Tarantola's inversion theory - it apparently works on as few as three view directions.

Andrew Hyman of Boston University presented field measurements to determine how well a single albedometer on a tower characterized a validation site in the face of spatial variability. Broadband albedo measurements and hemispherical photographs were acquired from a height of 4m at several points along transects through grassland with shrubs, and compared with albedos retrieved from AVHRR and POLDER. Hyman concluded that:

Other BRDF, POLDER

Stefan Sandmeier of GSFC presented high-spectral resolution BRDFs of grass and emphasised the variation with wavelength of the shape of the BRDF. The anisotropy of the BRDF was greatest at the wavelengths at which the reflectance was lowest (blue and red), as opposed to less anisotropy and higher reflectance in the green and near-infrared, which Stefan ascribed to the lesser occurrence of multiple scattering to smooth out the field of exiting radiance. In fact even though the reflectance went up and down across the spectrum there was a one-to-one correspondence between reflectance and BRDF anisotropy for the cases he showed. However, he said such a simple relation did not always hold, a fact he put down to wavelength-dependent transmission of the vegetation components.

I had a long discussion with Jean-Louis Roujean. A 10-day global albedo product from the 8~months of POLDER data is now available. Jean-Louis has commenced discussions with David Jupp regarding bringing the airborne POLDER to Australia. Thus he was interested in Australian sites uniform on a scale of about 10km or more; I told him of the Amburla, Hay and Tinga Tingana sites.

Jean-Louis suggested to me that since the ratio of direct to diffuse irradiance for clear sky changes with solar zenith angle, albedo measurements over a clear day may be separated into direct and diffuse albedos. This set me thinking about the cloudy-sky albedo at the CIGSN sites, to which I hadn't previously paid attention. It would be worthwhile looking for a stable albedo on uniformly overcast days and looking for a significant difference from the clear-sky albedo for the same surface. This would provide a test of the calculation of cloudy-sky albedo (roughly diffuse albedo) from clear-sky satellite observations as is done by, for instance, the MODIS algorithms.

Calibration on deserts

Yves Govaerts of EUMETSAT described his calibration of the Meteosat visible-band imager on African desert sites. He takes a fixed BRDF shape from a model for desert and the spectrum from airborne measurements. He claims a final accuracy of 5-10%. This is in preparation for the operational vicarious calibration of SEVIRI on MSG, the next generation of Europe's geostationary imager. I was surprised that SEVIRI will not have on-board calibration. I told Yves of recent Australian work on intercalibrating the GMS imager against AVHRR; he was concerned that the spectral mismatch of the two sensors would make intercalibration on bright desert doubtful, but was more confident about the use of (more spectrally flat) thick cloud as a target.

Francois Cabot of CNES, France, has used POLDER to characterized the BRDF of 20 uniform desert sites in North Africa and hence intercalibrate the VEGETATION, SeaWiFS and AVHRR sensors. The degradation of VEGETATION he finds is consistent with that from its on-board calibration. He plans to extend this work to SPOT-HRV, ATSR-2 and GOME.

EOS

A morning of talks was devoted to the EOS AM-1 platform, which won't be launched until at least a few months into 1999 because spacecraft control software is not ready.

The calibration of CERES on TRMM has been checked by comparing the ERBE-like processing of CERES with ERBE using the TOA longwave radiation budget over the tropical ocean as a reference. There is a systematic difference of 2% in the longwave calibration of which about half is expected to be due to El~Nino and most of the remainder to the inferior ERBE calibration. It was stated that the calibration of the shortwave part of the CERES TOTAL channel differs from ERBE by 1%.

Michael King described the MODIS algorithms to retrieve cloud and aerosol properties and atmospheric water vapour. The cloud mask is derived from 17 of the 36 bands and is threshold based so as to be fast enough to apply on a per-pixel basis. The land aerosol algorithms concentrated on bright desert and on dark dense vegetation identified in the
2mm and 3mm channels, in which the aerosol is most transparent. MODIS will provide CO2 slicing at the finest spatial resolution yet.

The MISR aerosol retrieval aims at fairly coarse characterisation of aerosols: particle size classed as small, medium or large; refractive indices as dirty or clean - these two parameters constitute the "underwear'' model shape - spherical or not; optical depth to an accuracy of 0.05 or 20%, whichever is bigger.

Aerosol

Brent Holben showed climatologies of aerosol measurements from several AERONET sunphotometer sites. The sites showed a marked variety in the distribution of points on scatterplots of aerosol optical depth versus angstrom exponent, which could be ascribed to the different dominant aerosol types. He showed seasonal variations in aerosol parameters.

Retrieval of aerosols over ocean from one week of data from the GOME spectrometer on ERS-2 are encouraging, but aerosol retrievals over land have been difficult and are unsatisfactory so far.

I was surprised when a talk from John Reagan's group described extensions to the Forgan plot method for analysing sunphotometer data. I had to go to Seattle to learn of Bruce's improvement on the Langley plot! By relaxing the assumption of constant aerosol amount and type throughout the morning or afternoon of observations to constancy in the microphysical properties only, all spectral channels can be calibrated with respect to an (already calibrated) reference channel.

Hyperspectral thermal

A session on hyperspectral thermal imaging focussed mainly on results from SEBASS, an airborne instrument constructed to explore the utility of hyperspectral thermal data. SEBASS covers 3-5.5
mm and 7.8-13.5mm with a resolution of l /d l @ 20, has a ground resolution of 0.5-3m, and has flown 12 campaigns since 1995. The spectral resolution will resolve the bands of surface materials and atmospheric gases other than CO2. Chris Borel of Los Alamos National Laboratory noted that there is hope that surface and atmospheric effects can be distinguished because in general surface emissivity spectra are much smoother than atmospheric absorption spectra.

J. Norman of the University of Wisconsin described simulations to retrieve the boundary layer temperature rather than the land surface temperature. V. Realmuto of JPL extracted from TIMS data over a Hawaiian volcanic plume images of the SO2 column abundance and LST which agreed with simultaneous data from the Airborne Emission Spectrometer (AES).

SST

W. McKeown of the Naval Research Laboratory described a differential absorption technique for profiling the skin sea surface temperature, using wavelengths of 2
mm and over, and covering depths to about 0.4mm. He described a system of eight buoys for monitoring radiometric SST with self-calibrating sensors atop 20-m masts. C. Donlon of the Southhampton Oceanography Centre (SOC) described the SoSSTR (Ship of Opportunity SST Radiometer) which is based on TASCO radiometers and calibration targets on a rotating arm, and is designed to be rugged for unattended operation. He then described the Infrared Scanning Autonomous Radiometer (ISAR) which will build on the experience of SOC and of the Rutherford-Appleton Laboratory's SiSTER radiometer. It will use Heimann KT15 radiometers because the TASCOs' calibration is demanding and they are susceptible to thermal shock. ISARs are expected to be deployed on buoys from (northern) spring 1999.

Lidar

K. Fischer of ERIM International described the "M10 Elastic Backscatter Lidar'', designed specifically to be a cheap, compact and robust atmospheric lidar which can run unattended in the field. It is housed in a waterproof rugged packing case of ~1m3 volume. It currently uses green light but an eye-safe version is being considered. In tests in 1997 it returned an aerosol profile every 15 minutes for most of 5 weeks.


Disclaimer