Summary of the International Forum on BRDF

Shunlin Liang (sliang@geog.umd.edu), University of Maryland, College Park, MD
Alan H. Strahler (alan@crsa.bu.edu), Boston University, Boston, MA

[Incorporating inputs from Fred Prata and submitted to EOC as a visit report
and information for the E_O_Crew]

The International Forum on BRDF (IFB) was held at the Argent Hotel in San Francisco on December 11-12, 1998. More than sixty people from nine countries attended this event. BRDF stands for bidirectional reflectance distribution function, and it generally refers to multiangle
remote sensing. The objectives of the IFB were to summarize recent research progress, to identify important future research topics and to determine their priorities. The IFB was structured in a panel discussion format. Eight panels were organized and each panel consisted of a series of themes. Theme chairs organized a group of researchers charged with writing a white paper on the theme and preparing a panel presentation(s). The set of white papers will be published in the special issue of the Remote Sensing Reviews.

The meeting identified five primary courses of action for the BRDF
community:

• Identify a set of key scientific questions to which multiangle data and modeling provide qualitative and quantitative advances over more traditional approaches. Organize case studies based on actual data or simulations to show the value added by multiangle data and to define the multiangle measurement requirements that are driven by critical science objectives.
  
• Explore different inversion techniques to retrieve geophysical, biophysical, or radiation parameters that are relevant to climate, environmental, and ecological sciences along with the requisite accuracies. Data fusion and assimilation from multiple sources with different spectral, spatial, temporal and angular characteristics should be emphasized.
  
• Continue development of simpler BRDF models, simplified from sophisticated physical models or semiempirical statistical models, that are suitable for kilometer-scale satellite observations over both homogeneous and heterogeneous landscape types.
  
• Develop a benchmark validation database that may be created from well designed field campaigns, laboratory measurements or model simulations.
  
• Strengthen graduate education programs emphasizing BRDF modeling and applications. Processing and analysis of EOS (Earth Observing System) multiangle remote sensing data will require many people with strong quantitative skills.

The IFB began with a welcome by the organizer, Shunlin Liang of the University of Maryland. After a brief overview of the IFB objectives, meeting formats and logistics, he presented his views on some of the BRDF top issues. The IFB's purpose is twofold. First, it provides an assessment
of the status and future problems of multiangle sensing science after a decade of directed research. Secondly, it prepares the way for the Second International Workshop on Multiangular Measurements and Models, in Ispra, Italy, September 15-17, 1999.

Alan Strahler of the Boston University reviewed BRDF historical developments, including previous meetings and prior research issues as they have evolved over the years. BRDF researchers began meeting informally as early as 1990, at the BRDF specialist workshop in Tempe, Arizona, which was sponsored by the Canada Centre for Remote Sensing.

Concerns at first largely centered on field and aircraft directional reflectance measurements and models. With subsequent meetings in Columbia, Maryland (1992); Beijing, China (1996); and College Park, Maryland (1997), attention became more focused on spaceborne applications of
multiangle sensing and on the information content of the multiangle signal. As we begin to enter the second decade of BRDF research, our most important mission is to translate the gains in understanding the physics of complex surface scattering into retrieval of physical and biophysical parameters that support global change assessment activities.

Science Drivers for BRDF Studies

Panel 1 on "Science Drivers for BRDF Studies" was chaired by Mike Barnsley of the University of Wales Swansea, UK. He first began to overview the scope and content of the panel. After discussing various science drivers and the `products' of BRDF models/data that may be appropriate, he organized this panel into two sessions. The first session was to identify the science drivers including two themes: earth radiation budget and ecological applications. The second session was to address the science drivers, including data normalization using angular information and estimation of biophysical properties directly from directional reflectance/emittance measurements.

There were several presentations in the theme of "Angular corrections to satellite data for estimating earth radiation budget." Alan Strahler of the Boston University first provided an overview of this topic and discussed its relevance to global change studies. Shunlin Liang of the
University of Maryland reviewed different angular models that convert radiance to flux and discussed the uncertainties in converting narrowband albedos to broadband albedos. Since surface broadband albedos also depend on atmospheric conditions, he raised the issue of producing broadband albedo products derived from one atmospheric condition to users who may
need albedo for many different atmospheric conditions. Fred Prata of CSIRO, Australia introduced their ground measurement network in Australia, and demonstrated the long-term trends and variations of the measured broadband albedos.

In the theme of "Ecological and biogeochemical benefits of multi-angle satellite observations: concepts and realization," Greg Asner of Stanford University first analyzed the major sources of uncertainty in large-scale ecological/biogeochemical research, and then discussed how multiangle
approaches can realize benefits for ecological/biogeochemical research.

In the theme of "Remote sensing data standardization using BRDF models: view and sun angle effects on vegetation indices (VI), leaf area index (LAI), and land cover classification," Wim van Leeuwen of the University of Arizona reviewed different sensors that will provide off-nadir observing capabilities and focused on remote sensing data standardization applications, including the use of BRDF models in global VI compositing, standardization of global VI's to constant sun and view angles, LAI derivation from standardized reflectances, and nadir equivalent reflectance values as input to land cover classification.

In the theme of "Biophysical parameter retrieval from multiangle remote sensing," Jing Chen of Canada Centre for Remote Sensing reviewed different approaches for deriving biophysical parameters from remotely sensed data.

Panel 1 discussions are summarized below:

(1) The BRDF community needs to more be aware of the broader environmental science (ES) drivers that underpin and justify BRDF research and development. We have, to varying degrees, lost sight of this point: much of the current debate in the BRDF literature is focused on the fine detail of model variants, inversion schemes etc. which, while important, is effectively unconstrained. We need to define our goals (and hence the methods required to reach these) in terms specific (ES) user requirements.

(2) The BRDF community needs to develop better links with its (current and potential future) `user' communities. We must re-cast our research in terms of specific ES questions and requirements, but it is critical that they must be questions posed by the ES community, not ones that we think that the ES community is interested in or necessarily just the ones that BRDF methods can easily address. Otherwise there is a danger that we might concentrate on issues that are considered `old hat' by the target communities (e.g., focusing on above-ground biodiversity, just as soil microbial biodiversity becomes the hot topic; or "missing carbon", just as the global nitrogen cycle is seen as more critical). If we are serious about this, we should invite more ecologists, climate modelers, etc., to the next and future BRDF meetings.

(3) We need to clarify the role/significance/value of data `normalization/standardization' methods. It could be that this approach is seen as inherently inferior to the derivation of biophysical properties directly by BRDF model inversion, or that the approach is widely accepted
and hence did not require further debate. The truth may lie somewhere in between these two extremes, but we did not manage to tease this out during the meeting. Further discussions reached the following consensus. The angle-normalized vegetation indices are the simplest ways of inversion to obtain biophysical parameters. These simple inversion methods are empirical, region and species specific. They therefore require field data to support the algorithm development. These methods are surely valid as long as they are calibrated with field data, therefore have
been widely used in climate and ecological studies and will continue to be so. More sophisticated BRDF inversion methods have potential to be general and reduces the requirements of field measurement data. However, the methods are far from mature, and more research is still needed.

(4) Do we need to show that BRDF approaches offer unique information about earth surface materials, or is it sufficient to demonstrate that the information provided by BRDF data is complementary to the multispectral signal and provides added value? That there is a need to show that BRDF approaches provide unique information about earth surface materials was accepted in the general debate without a critical discussion. The point is probably more important than it at first seems because our response to this question defines the way that we should direct our
research activities. Taken at face value, we must show not only that the BRDF hot-spot, bowl-shape, etc., relate directly to certain key biophysical properties that cannot be derived by other means, but also that these can routinely and unambiguously be retrieved from directional
reflectance measurements.

Three-dimensional (3D) Modeling and Simulations

Panel 2 on "Three-dimensional (3D) modeling and simulations" was chaired by Chris Borel of the Los Alamos National Laboratory. In the first theme of "Mathematical aspects of BRDF modeling: adjoint problem and Green's function," Yuri Knyazikhin of the Boston University and Alexander
Marshak of NASA/Goddard Space Flight Center (GSFC) reviewed the adjoint formulation of 3D radiative transfer and the Green's function concept in neutron transport and then discussed their applications in 3D radiation transport in vegetation canopies.

In the theme of "Monte-Carlo ray tracing simulations," Roger Davies of the University of Arizona reviewed the historical developments of ray tracing techniques particularly in cloud community. Philip Lewis of University College London, UK, discussed the progress and issues in the
land surface environmental applications.

In the theme of "Radiosity simulations," Chris Borel of the Los Alamos National Laboratory reviewed the historical developments of radiosity simulation techniques and applications in the vegetation canopy modeling, atmospheric correction and terrain modeling. Wenhan Qin of the University of Maryland presented applications of combining L-system and radiosity simulation techniques for vegetation canopies.

In the theme of "Recent advances in geometrical optical modeling and its applications," Jing Chen of the Canada Centre for Remote Sensing reviewed different geometric optical models and discussed his four-scale geometric optical model and applications. Xiaowen Li of the Chinese Academy of Sciences and Boston University discussed scaling issues in geometric optical models and pointed out that the reciprocity principle and Planck laws are not always valid at different spatial scales.

The discussions from panel 2 are summarized below:

(1) In the mathematical aspects of BRDF modeling, use of energy conserving techniques is important. Adjoint and Green's function methods are new powerful tools in the modeling of land surface scattering processes but need more development.

(2) Monte Carlo simulations are mature and are sufficiently powerful to simulate novel sensors such as cloud probing LIDARs. Due to computational complexity their use has been mainly as a benchmark for analytic radiative transfer.

(3) In radiosity simulations, the powerful combination of L-systems and radiosity allows studies of canopy structure effects in vegetative canopies in the near infrared where multiple scattering is important. The question of the need for spatial detail in modeling the scattering in vegetation is still unresolved.

(4) Geometric optical modeling has many practical applications for homogeneous surfaces but needs further development to model heterogeneous environments. The question of whether the Helmholtz reciprocity theorem is valid should be further investigated.

Simplified Radiative Transfer and Semiempirical Modeling

Panel 3 on "Simplified radiative transfer and semiempirical modeling" was chaired by Shunlin Liang of the University of Maryland. After a brief introduction to the panel composition and its rationale from the chair, Wenhan Qin of the University of Maryland presented the theme of
"simplified canopy radiative transfer modeling." He started with an introduction to radiative transfer formulation in both leaf and canopy levels, and then reviewed recent progress in the following five areas: finite size medium theory; hot spot effect consideration; role of non-leaf
organs (e.g. stems in crops and branches/trunks in forests); coupled atmosphere-canopy RT models; and RT model inversion and applications. Qin finally discussed a series of research and application-related issues.

In the theme of "Progress in surface particulate medium bidirectional reflectance modeling and applications," Shunlin Liang of the University of Maryland first demonstrated the need for estimating soil properties from remote sensing that are currently mapped from conventional techniques. He then reviewed the formulation and progress of radiative transfer and geometric optical modeling of particulate media and applications (e.g., soil and snow). He presented the research results on detectable depth of soils from visible and near-infrared spectrum. Different research issues were also discussed.

There were a series of presentations on the theme of "Inversion methods for physically-based models" led by Dan Kimes of NASA/GSFC. After Kimes gave the brief introduction to scientific rational for using physically-based models, general description of inversion problem, and current methods of inversion, Jeff Privette of NASA/GSFC focused on the traditional inversion methods. He discussed the mathematical formalism, different components of classical inversion problem,
different optimization algorithms and many research issues, such as speed vs. accuracy, model/parameter choice, data set quality effects etc. Yuri Knjazikhin of the Boston University focused on table look-up methods. He demonstrated that retrieval of vegetation parameters is an ill-posed problem. Regularization techniques need development in order to provide convergence of the retrieval algorithm; that is, the more measured and accurate information available will generate more accurate outputs. He also discussed the law of energy conservation as a basis for regularization and presented examples of global LAI and FPAR (Fractional Photosynthetically Active Radiation) fields derived from SeaWiFS (Sea-Viewing Wide Field Sensor), POLDER (Polarization and Directionality of Earth's Reflectances), Landsat (Land Remote-Sensing Satellite) and LASUR data. Abdelgadir Abuelgasim of NASA/GSFC focused on neural network methods. After describing the approach, he compared the advantages and disadvantages of the neural network methods with other inversion approaches and discussed research issues. Alex Lyapustin of NASA/GSFC presented his researches on the land surface albedo retrieval scheme. Finally, Dan Kimes of NASA/GSFC summarized this theme and pointed out the need for rigorous comparisons of the various inversion methods, future sensor studies, and development of new methods for handling ancillary information (e.g. topographic data, high spectral and spatial data, temporal and spatial BRDF data).

In the theme of "Bidirectional reflectance and albedo from semiempirical models: approaches, models and issues," Wolfgang Lucht of Boston University briefly reviewed different semiempirical models and their applications. He also discussed research issues in modeling, inversion, practical considerations and albedo derivations. Jean-Louis Roujean of CNRM, France, added more critical issues in semiempirical modeling and applications. Bernard Pinty of the EC Joint Research Centre, Italy, provided a historical review of the Rahman-Pinty-Verstraete model and its recent applications.

In the theme of "Directional variance of remote sensing images", Wenge Ni of Raytheon STX at Lanham, Maryland, reviewed a new concept, the bidirectional reflectance variance function (BRVF). She summarized applications of contextual information using nadir-view remote sensing imagery and then discussed the BRVF modeling using a geometric-optical approach and presented several examples of BRVF analysis using airborne and spaceborne remotely sensed data. Further research issues were also discussed.

In the last presentation of this panel, Gail Anderson of the US Air Force Geophysics Laboratory, discussed the requirements and recent implementations of the surface BRDF in the MODTRAN code that has been widely used in various remote sensing applications.

The issues from panel 3 were summarized below.

(1) Develop simplified radiative transfer models suitable for heterogeneous landscape. There have been interesting debates on the roles of semiempirical models and simplified radiative transfer models in various applications. By examining four major applications (boundary conditions, surface BRDF retrieval, angular normalization of remotely sensed data and biophysical parameter retrieval), we can see that currently semiempirical models are most suitable for the first three types
of applications, while the simplified radiative transfer models are better suitable for the last application. A consensus strategy is to retrieve surface BRDF from the remotely sensed data first using semiempirical models and then retrieve biophysical parameters later using simplified
canopy models. To effectively retrieve biophysical parameters directly from remotely sensed data, particularly with kilometer-scale resolutions, simple models are needed to account for surface heterogeneity. Studies are needed to focus on linkages of "effective" parameters in simple models
with spatial distributions of the surface corresponding parameters. Decent calibration/validation datasets are highly needed.

(2) Apply advanced inversion algorithms to retrieve important biophysical/geophysical parameters required by different ecological applications and surface modeling. Most current inversion algorithms are compromises between inversion accuracy and computational speed, such as searching look-up tables that are created from more sophisticated canopy models or inverting simplified radiative transfer models. Advanced inversion algorithms (e.g., neural networks, regression trees, etc.) may enable us to retrieve important biophysical parameters from more sophisticated models without further model simplifications, which may enable us to investigate the model invertability and sensitivity more realistically. Most current inversion algorithms rely on a specific satellite sensors New inversion algorithms should take advantages of information from multiple sources and produce more information by developing data fusion and data assimilation techniques.

(3) Explore high-order statistical moments of the surface radiation field and their linkages with surface structural information. Many inversion algorithms assume observations at different angles
are independent, while they are actually highly correlated. It is practically impossible to monitor the earth surface from every angle, and it is actually unnecessary to do so because of its high correlation. High-order statistical moments are also sensitive to surface structural information. Studies on statistical characterization of the surface radiation field will allow us to determine the best viewing angles at different bands and help retrieve surface structural information that
are highly needed in the ecological applications.

(4) Develop semiempirical models whose coefficients have explicit physical meanings suitable for surface modeling and various ecological applications. Semiempirical model parameters are not necessarily fully physically-based, but may contribute to represent a physical process or/and may appear as a grouping of physical parameters.

Hotspot Research and Applications

Panel 4 on "Hotspot research and applications" was chaired by Siegfried Gerstl of the Los Alamos National Laboratory. He first recognized and stressed that the hotspot region in BRDFs of vegetated land surfaces is the most information-rich subregion within a BRDF distribution so that
we may truly talk about a "hotspot signature" for all three-dimensional structured surfaces. Especially for vegetation canopies it has been shown by many model calculations, some recent measurements with airborne (ASAS (?? ), MAS, AirPOLDER, and AirMISR) and spaceborne sensors (POLDER), that the hotspot effect is indeed diagnostic for canopy structure and may allow the retrieval of such canopy structural parameters as leaf size and shape, tree crown size, and canopy height for low-LAI stands.

The present state-of-the-art in hotspot modeling and canopy architecture retrieval was reviewed by Wenhan Qin of the University of Maryland and Jing Chen of the Canada Centre For Remote Sensing. Data from airborne POLDER calibration measurements in the hotspot region were reviewed by Jean-Louis Roujean of CNRM, France, who also showed that the hotspot angular signature yields similar features when airborne and spaceborne data are compared, ranging from a few meters to several kilometers spatial resolution. Hotspot data from the MODIS (Moderate-Resolution Imaging Spectroradiometer) Airborne Simulator (MAS) were discussed by Sig Gerstl of the Los Alamos National Laboratory and show very remarkable detail in the BRDF distribution in the hotspot region, even without the application of atmospheric corrections. Gerstl discussed also the newly approved NASA satellite project Triana which will place a high-resolution multispectral imager at the earth/sun (gravity-neutral) Lagrange Point L-1 to provide continuous well-calibrated images of the full earth disk with spatial resolution of approx. 8km. Due to its distance from the earth (about 4 times the earth-moon distance) and its location along the line-of-sight between the Sun and earth, Triana will image the entire earth in the hotspot direction all the time, while the earth rotates under it. The hotspot angular distribution will be measured out to +-15 degrees by slow measurement around the L-1 point. Thus, Triana is truly a 'global hotspot imager' and is expected to deliver these data by early 2001. To quantitatively analyze and understand Triana data will require the acceleration and focusing of ongoing hotspot research with the hope that a global hotspot ecology may be developed with Triana data.

The following top research/action priorities were identified for hotspot R&D within the next few years:

(1) Simplified 3D hotspot models need to be developed that are capable of representing the three-dimensional aspects of realistic heterogeneous, km-size scenes. These models should allow the derivation of quantitative correlations of hotspot signature parameters to canopy structural parameters, and their scaling for different fields of view.

(2) Build an experimental data base for hotspot BRDF signatures from airborne measurements, like those from the MODIS Airborne Simulator, for solar zenith angles between zero and 90 degrees, covering the hotspot distribution function to at least +-30 degrees. This data base should
allow look-up tables to be built for terrestrial hotspot signatures by latitude, longitude, biomes, and season.

(3) Develop rigorous atmospheric correction algorithms for hotspot angular signatures that include absorption and multiple scattering effects by Rayleigh and Mie scattering from gaseous and aerosol atmospheric constituents.

BRDF Retrieval from Remotely Sensed Data

Panel 5 on "BRDF retrieval from remotely sensed data" was chaired by David Diner of NASA/JPL. He started with a set of questions about multiangle/BRDF remote sensing: Why should we do it? What does it mean? How should it be implemented technically? How do we insure its place in future missions?

In the theme of "BRDF retrieval from sequential multiangle observations", Mike Barnsley of the University of Wales Swansea, UK reviewed BRDF retrieval from different sensors that provide off-nadir observations one at a time (i.e., sequential multiangle observations), such as MODIS, AVHRR (Advanced Very High-Resolution Radiometer), SeaWiFS, SPOT-vegetation, CHRIS (Compact High Resolution Imaging Spectrometer), etc. Advantages and limitations were also discussed.

In the theme of BRDF retrieval from simultaneous multiangle observations", Marc Leroy of CESBIO, France, compared different inversion methods for those sensors that provide multiangle observations simultaneously (i.e., simultaneous multiangle observations), such as MISR (Multiangle Imaging SpectroRadiometer), POLDER, ATSR-2 and ASAS. Some common features and examples were also presented.

Diner then outlined a set of key scientific questions ("silver bullets") to which multiangle data and modeling provide critical and unique input, and organized the participants into groups to focus on key questions. He charged them to:

(1) identify the geophysical, biophysical, or radiation parameters that must be measured along with the requisite accuracies. Be able to state the consequences of not achieving these accuracies. (2) identify the angular sampling and coverage requirements;
(3) identify spatial, spectral, temporal, and supplementary model requirements;
(4) put together case studies based on actual data or simulations to show the value added by multiangle data. The detailed summary of these discussions will be described in an article to
be submitted to the Bulletin of American Meteorological Society. (see BRDF Future).

Modeling and Measurement of Thermal Angular Effects

Panel 6 on "Modeling and measurement of thermal angular effects" was chaired by Fred Prata of CSIRO, Australia. Chris Borel of the Los Alamos National Laboratory first outlined different modeling efforts, including vegetation models by J. Smith, Lee Balick, J. Norman, etc., DIRSIG modeling at RIT, Hyperspectral scene thermal modeling at LANL, and directional modeling of emissivity for sea surfaces by P. Villeneuve and F. Prata. Borel then presented recent some results
in temperature/emissivity retrievals. In the second talk in this panel, Fred Prata of CSIRO, Australia first clarified the meanings of parameters like land surface temperature (LST), emissivity, thermal BRDF and BEDF, then discussed field and laboratory measurements in thermal IR, temperature/emissivity separation algorithms. He pointed out that there remain many problems with surface temperature retrieval (particularly over the land) and some of these do relate to directional effects. He suggested to look at the state-of-the-science for LST retrieval and figure
out how much of the problem is due to directional effects, how much is poor experimental design (e.g. calibration/validation), how much is due to the atmosphere and how much is due to inadequate sensor performance. Prata also commented that while sensors like AVHRR, ATSR-2, Landsat, TIMS, Daedalus and others have been studied in detail, the new sensors (such as MODIS, ASTER, AATSR, GLI) will be the challenge for the future. Prata was followed then by Julienne Stroeve of the University of Colorado at Boulder who presented some interesting results that demonstrate the effects of snow/ice angular corrections on thermal energy balance studies.
Since this panel was quite small, they decided not to identify research issues and priorities on behalf of their community.

Measurements and Validation

Panel 7 on "Measurements and validation" was chaired by Charles Walthall of USDA/ARS at Beltsville. After brief panel charges by the chair, Stefan Sandmeier of NASA/GSFC reviewed different BRDF laboratory measurements for both vegetated and non-vegetated surfaces and major goniometer laboratories around the world. The advantages and disadvantages of laboratory BRDF measurements compared to field measurement were also discussed.

In the theme of "Measurements for biophysical and BRDF products: defining appropriate resolutions for validation," Charles Walthall gave a comprehensive discussion on a variety of issues related to field BRDF measurements, including objectives for BRDF research data collection, examination of field campaign models, brief history of major BRDF-friendly field campaigns, evolution trends of data collection on the surface, evolution of methods, outstanding issues and problems, and future activities.

Research Priorities from this panel are summarized below:

(1) Conduct fusion experiments (with multispectral and multiangle measurements) to address the uniqueness of multiple view angle (MVA) data relative to spectral nadir-only data. Parameterize the landscape and the atmosphere for use in models as an integral part of the effort. Few data sets exist that can be used to address MVA data value using both empirical and model-based approaches. There is also a lack of k nowledge about the spectral variability of directional reflectance of the earth's surface. A critical element of this includes specifying minimum parameters for models to address cause and effect, and parameter retrieval. New parameters describing ecological systems and structures may result from this effort. Protocols for measurements are needed to insure high quality data and to direct instrumentation development. Specification of surrogate measurements when optimum instrumentation is unavailable will increase the amount and quality of data. Recommended approaches include pairing investigators with modeling expertise with those having measurement expertise and/or ecological applications expertise.

(2) Conduct experiments with measurements from a range of temporal and spatial scales that address the changes of information accompanying different spatial and temporal resolutions. There is a concern that our understanding of the earth's surface is limited by our ability to
sample only small areas and short time scales with a reasonable level of confidence. It is therefore necessary to understand how information content changes as a function of the spatial dimension and as a function of time. This will increase our knowledge of earth systems and yield insights for selection of optimum spatial and temporal resolutions for earth observations.

(3) Identify levels of aggregation/clumping and heterogeneity of canopy and landscape elements and determine their significance. There appear to be levels of aggregation of earth surface features affecting radiant energy interactions. These structures are assumed to be ecologically significant. It is necessary to understand what canopy and landscape elements exhibit levels of aggregation, determine the dimensions of the aggregations and understand their ecological significance. This is a relatively recent finding that strongly questions assumptions about the degree of randomness in nature. Better knowledge of these issues will improve our understanding of earth systems, result in better models, optimize measurement strategies, and provide additional foundations for
specifying earth observation system resolutions.

BRDF Future

Panel 8 on "BRDF Future" was chaired by Alan Strahler of the Boston University. In a theme of education and outreach, Mike Barnsley of the University of Wales Swansea, UK, stressed the importance of BRDF education and outreach and demonstrated a web-based interface for
running the SAIL canopy model that can be essentially used for both undergraduate and graduate education. Charlie Walthall of USDA/ARC at Beltsville discussed a variety of issues, such as how we can increase awareness of BRDF research findings, how we can infuse BRDF into remote sensing applications, and how we can improve/expand remote sensing education so that BRDF research is an integral part.

Group discussion on BRDF future centered on short-term issues of promoting the value of multiangle remote sensing to the broader earth system science community, especially in the context of NASA's plans for future missions centered around science themes. No new mission specifically utilizing multiangle sensing except Triana is planned after the launch of MISR on the Terra-1 platform. The group expressed concern and decided to work harder to promote the contributions of multiangle sensing to retrieval of surface parameters essential to earth system science. It was agreed that an article be drafted and submitted to the Bulletin of the American Meteorological Society that would provide case studies and arguments for multiangle sensing, following the lead of Panel 5. Dave Diner agreed to lead the effort to draft the article with inputs from discussion leaders. It was also agreed that a summary of the meeting would be submitted to the Earth Observer. Continuing discussion confirmed five priority issues for future BRDF research, which were stated at the beginning of this meeting summary.

In January, 1999, a proposal to defray publication and distribution costs for the special issue of Remote Sensing Reviews was approved by Dr Diane Wickland, Program Manager of NASA's Terrestrial Ecology Program. The organizers gratefully acknowledge her contribution, as well as that of the Center for Remote Sensing of Boston University, which contributed substantially to help cover meeting costs. We would like to thank all panel chairs for their contributions of the panel summaries to this report.