Review of NASA's Distributed Active Archive Centers (1999)

The Earth Resources Observation Systems (EROS) Data Center (EDC) DAAC is hosted by the U.S. Geological Survey. Its mission is to manage and distribute data products generated by low-Earth-orbit missions of the EOS program that deal with land surfaces and land processes. At the time of the formal review, the DAAC had no experience with the EOSDIS Core System (ECS), which is intended to provide the distribution and subsetting capabilities for all of the DAAC's new data streams. The DAAC has since worked with the ECS contractor on its "mini-DAAC," and launch dates have been delayed by at least six months. The extra time should enable the DAAC to address, at least partially, the panel's main recommendations: (1) the DAAC should devote considerable effort to preparing for the Landsat 7 and AM-1 data streams, including develop-

ing contingency plans for inevitable initial difficulties with the ECS; and (2) the DAAC should develop a clearer vision for serving the needs of its scientific and, potentially much more numerous, applications users. This vision should include a plan for meeting the challenges of the EOS era, including major increases in data volume and product diversity, a broader user community, and concomitant increased complexity of user services.

The EDC DAAC was created in 1992 to manage data from Landsat 7 and other land remote sensing instruments. It is located within the EROS Data Center, which has been managing land processes data for two decades, but its holdings and operations are separate from those of the USGS. The creation of the EDC DAAC signaled a shift in the focus of the Landsat program. Previous Landsat satellites served primarily the applications community, a very large user group that is likely to increase substantially if data become available at lower cost and in near real time. Landsat 7, on the other hand, has been incorporated into the EOS program as a science instrument, and scientists will be the DAAC's highest-priority users (Box 5.1). However, the demands on the system will likely still be driven by the applications users.

When it was formed, the EDC DAAC did not acquire any heritage data sets. Its data sets, which are few in number but relatively large in size, include the Global 1-Kilometer Advanced Very High-Resolution Radiometer (AVHRR), Landsat Pathfinder, and the digital elevation model Global 30 Arc-Second Elevation Data Set (GTOPO30).

In the EOS AM-1 era, the DAAC will manage data from Landsat 7 and the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER). It will also receive Moderate-Resolution Imaging Spectroradiometer (MODIS) data from the GSFC DAAC and will distribute and possibly process MODIS land products. Data from these instruments, which are scheduled to be launched in 1999, will increase the volume of holdings by a factor of 20, from 9 TB to nearly 200 TB per year (Box 5.1).

To prepare for these data streams, the DAAC has requested an additional 50 ECS contractors (there are 38 ECS contractors at the DAAC in FY 1998), and it has begun readiness exercises. The DAAC has tested the algorithms for processing ASTER data and appears to be ready for the ASTER data stream. This is to be credited in large part to the ASTER instrument team, which was extremely diligent in porting its processing codes to the DAAC and testing them extensively. Readiness for MODIS will depend, in part, on the GSFC DAAC and the MODIS science team; at the time of the review, the schedule and products from the MODIS science team were still uncertain, and only limited tests of the algorithms had been completed.

Delays in the delivery of the ECS and uncertainties about its capabilities

have made preparations for the EOS data streams more difficult. The EDC DAAC relies exclusively on the ECS for managing the data, and a relatively high level of functionality will be required to ensure timely processing, subsetting, and distribution of data.

The Panel to Review the EDC DAAC held its site visit on November 24-25, 1997. The following report is based on the results of the site visit and e-mail discussions with DAAC personnel held in June and July 1998.

The data currently distributed by the EDC DAAC (Box 5.2) are widely used for a variety of scientific applications. The Landsat Pathfinder data, for example, serve as a rich resource for many aspects of global change studies related to the

land surface. The 1-km AVHRR data have been used by many researchers to characterize the land surface as well as to model the role of the land surface in earth system processes. GTOPO30 topographic data (Figure 5.1) are needed for both radiometric and geometric correction of sensor data. Use of these data and imagery has greatly facilitated research on topics ranging from land cover change to geomorphology to hydrology.

Three agencies are involved in Landsat 7—NASA, NOAA, and USGS. NASA is responsible for building and launching the spacecraft and instrument, and for building, installing, and testing flight operations, ground data reception, and processing, archive, and distribution systems. It is also funding the DAAC to provide Level 0 archive and distribution for NOAA and Level 1 product generation and distribution for NASA (see Table 1.1 for a description of processing levels). NOAA is responsible for coordinating the international ground station network and for performing flight operations, data acquisition, and preprocessing. Finally, the USGS is responsible for long-term archive and distribution. Such sharing of responsibility is a concern to the panel because no single entity has end-to-end responsibility for the program.

Nearly all EOS data sets will be formatted in Hierarchical Data Format (HDF). The land processes community has little experience with HDF-EOS, but

the DAAC could undertake efforts immediately, such as converting existing data sets to HDF-EOS for data distribution and exchange.

Currently all DAAC data are unrestricted, but the EDC DAAC is planning to buy and disseminate commercial Landsat data from foreign ground stations. Use of these commercial data will be subject to restrictions established by the commercial vendor, the Earth Observation Satellite Company.

At the time of the site visit, the EDC DAAC was scheduled to receive and process data from three remote-sensing instruments—Landsat 7, ASTER, and MODIS. Delays in the ECS, however, have led NASA to consider transferring data processing from the DAACs to the science teams on an instrument-by-instrument basis. If the DAAC and instrument teams adhere to the original processing plan, the DAAC will receive, archive, and disseminate Level 0 Landsat data for NOAA and will later produce and distribute Level 1 Landsat products for NASA. With regard to ASTER, the DAAC will receive Level 1 data from Japan and will archive and distribute it for NASA. The DAAC will also produce and distribute Level 2 products on demand on behalf of the U.S. ASTER Science Team. Finally, the DAAC will receive Level 2 MODIS data from the GSFC DAAC and will process, archive, and distribute Level 2+ MODIS land products on behalf of the MODIS science team.

To prepare for these data streams, the DAAC is testing Product Generation Executables (PGEs) for ASTER and MODIS, and integrating them into the ECS. The DAAC has received all the PGEs for ASTER and has integrated them as far as is possible in the current release of the ECS. Since the full, required capabilities of the ECS have not been delivered to the DAAC, the PGE-integration process is not yet complete. With regard to MODIS, only two of 33 PGEs have been delivered to the DAAC, and neither has been integrated with the ECS. In this case, the readiness of the EDC DAAC to produce MODIS land products is limited mainly by the readiness of the MODIS science teams and GSFC DAAC.

On the other hand, if some or all of the Landsat, ASTER, and MODIS instrument teams decide to process the data themselves, the DAAC's role will be limited to traditional data management tasks (e.g., dissemination, archive, user services; see Chapter 2, ''DAAC versus Data Center").

Reprocessing Level 1 and higher data to accommodate improvements in calibration and algorithms is necessary to keep data sets viable and internally

consistent. Anticipating that the instrument teams will want data to be reprocessed often, the DAAC plans to work with its Science Advisory Panel to set priorities on reprocessing.

As noted above, data sets held by the EDC DAAC are large and will have to be subsetted to make them manageable to the user community. Rather than developing its own subsetting tools, the EDC DAAC plans to use the subsetting capability of the ECS. ECS subsetting, however, will not be available until mid-1999 at the earliest, long after launch of the AM-1 platform. The impact of the DAAC's decision on users is described below (see "Data Set Preparation").

NASA and USGS have signed a Memorandum of Understanding stating that DAAC holdings will move to the USGS three years after the conclusion of the mission(s). The USGS has a line item in its budget to pay for the transition. The mission of the EROS Data Center includes long-term data storage, and the institution has successfully carried out this mission since its inception in the 1970s. Given the importance of maintaining the archive for global change science, the panel suggests that the DAAC begin to plan, at an early stage, for the transfer of data from the DAAC to the EROS Data Center for long-term archiving.

Much of the current and expected user community is now assumed to be drawn from the Earth science research community, and it is apparently assumed that the instrument teams for the EOS instruments are representative of this community. However, given the greater breadth and variety of the EOS instrument suite, the user community after the launch of the AM-1 platform and Landsat 7 is likely to be much larger and more diverse than has been the case for previous, less complex instrument sets. It is of critical importance that significant effort be made to predict the size, character, and makeup of this larger, more diverse user community and to assess its needs. With such a user profile, it should be possible to construct a data archive and access system that optimally fits the client community.

For example, the Landsat series of satellites has a unique history, one unlike other instruments in the EOS program. Early in the life of the Landsat program, NASA placed the emphasis of this program on applications rather than science per se. Much of the user community for Landsat is therefore composed of what

might be referred to as remote sensing practitioners (e.g., value-added commercial organizations; nonscientific federal, state, and local government officials; educators; the general public).

Throughout the 25 years of the Landsat program, substantial emphasis has been placed on ''data continuity" because of these users. It was not until the incorporation of Landsat 7 into EOS that Landsat came to be regarded primarily as a science device, and to be treated in this fashion. Given that the needs and expectations of the science and nonscience collection of users are quite different, both must be taken into account in the design and operation of the EDC DAAC, if it is to be successful. Parenthetically, it seems possible that this broadening of the user community will extend to other EOS instruments as the data become available.

The Science Advisory Panel (SAP) is a formal, government-appointed body. Its current focus is on resource allocation, outreach, and user services for the AM platform, and it also helps the DAAC set priorities on holdings and reprocessing requests. Members feel that they have a good relationship with the DAAC and that the DAAC is fairly responsive to their recommendations.

The SAP is composed of instrument team scientists, researchers, including non-EOS researchers, and a user from the private sector. As noted above, however, the potential user community is much broader (including, for example, "remote sensing professionals") and is likely to expand further in the future.

The long-term health of the DAAC depends on a symbiotic relationship between the data operations activities and scientific applications. Data operations could benefit greatly from the participation of scientific users in the design of data delivery systems. Conversely, scientific users would benefit from understanding the constraints in the operations. The location of the DAAC within the

EROS Data Center creates an opportunity for such interactions. There is a rich resource of local scientists who are currently using data provided by the DAAC in their research. These scientists could provide a strong scientific backbone for DAAC activities.

However, the panel observed that interaction between DAAC personnel and the science community has been limited to date. The Science Advisory Panel serves as the primary linkage between the science community and the DAAC. Participation of DAAC personnel in scientific research is not actively encouraged, and there are no positions established for working scientists. Some limited efforts have been made to elicit feedback from scientists in the EROS Data Center on the design of Version 0, but these efforts have not received funding or priority. The observation that the DAAC's operations could be strengthened by involvement from scientific users was confirmed by DAAC managers. They expressed a desire to develop mechanisms that would foster interaction between DAAC personnel and the scientific users. In fact, the DAAC scientist distributed a draft "white paper" containing suggestions to expand the scientific involvement.

Examples of mechanisms that might foster the desired symbiotic relationship between the data operations and the scientific operations include the following:

• a visiting scientists program for research using DAAC data sets to be carried out in-house—the visiting scientists would then provide useful feedback on the design of the data delivery systems;

• involvement of DAAC personnel in research activities, particularly those being conducted within the EROS Data Center—this could be achieved through a short leave of absence from normal DAAC responsibilities or an allocation of a small percentage of the staff member's time to a research project;

• establishment of a small number of positions within the DAAC for scientists conducting research using DAAC products—the scientists would provide the perspective of the users in the design of DAAC operations and would be expected to be involved in such activities; and

• encouragement of day-to-day interaction between local scientists and DAAC personnel—this might be achieved through informal users' groups that would provide feedback to the DAAC or other mechanisms that promote informal interactions.

In particular, the panel encourages the DAAC to foster scientific connections with local EROS Data Center scientists who represent a pool of highly qualified users and could provide feedback regarding user services and scientific applications.

The successful construction and operation of the DAAC, as for any other service organization, is critically dependent upon (1) knowing to whom the service is to be provided and (2) knowing specifically what the goals of the service are to be. It is important to be as quantitative and specific as possible in answering both questions.

At scientific data centers, it is important that the user services group be able to answer questions of a scientific and technical nature. Thus, a scientific background for user services staff is highly desirable. At the EDC DAAC, the user services group appeared to have trouble answering questions at an impromptu demonstration for the panel, an impression that was confirmed by the SAP. The SAP also pointed out that the DAAC has no system for tracking user requests or the DAAC's responses, thus making it more difficult for the DAAC to serve its customers.

DAAC managers indicated that they plan to hire scientists to be part of the users services team. The panel concurs that improved technical support is needed and recommends that this plan be implemented as soon as possible.

The panel is concerned that, based on current plans, the ECS will not provide capabilities for reformatting options and geographical subsetting prior to the AM-1 and Landsat launches. These issues are likely to be of considerable importance to many users. As noted above, many users do not have experience with HDF-EOS format. In addition, the very large data volumes necessitate the capability for geographical subsetting because most users will not have the facilities or the bandwidth to download the full data sets. The latter applies particularly to global MODIS data sets. The panel therefore recommends that the DAAC develop plans for providing these formatting and subsetting capabilities, perhaps in consultation with other DAACs facing the same issues.

It is important for the DAAC to continually reach out to serve new users. For example, before the launch of Landsat 1, an announced goal was to make data available on request within three days of their acquisition. This was intended to serve a number of applications for which rapid perishability of the data's value was of key importance. This goal was never realized. The DAAC now has the opportunity to serve users who require data in near real time. These users are not currently a major component of the Landsat user community.

The panel observed that the DAAC provides all data free of charge and does not charge even for the cost of the media. In the EOS era, there is likely to be considerable demand for the data from a broad range of users in addition to scientists. This demand may create a backlog and make it difficult to provide the data to users in a timely fashion. The panel therefore concurs with a recent decision by the DAAC to charge for the costs of delivery and reproduction of popular data sets (e.g., GTOPO30). This decision would likely encourage users to order only the data they need.

Because the EDC DAAC has no heritage data sets, it has bought little hardware and done little development. Rather, its current focus is on data distribution, which is carried out through Version 0, an information system developed jointly by all the DAACs. The Version 0 system—hardware and software—will be phased out after the ECS is delivered. In the meantime, the EDC DAAC uses Version 0 to distribute several large data sets, including GTOPO30 elevation data, global AVHRR 1-km data, and NASA Landsat Pathfinder data. The panel viewed a demonstration of Version 0 and was impressed by the capabilities of the user interface, but felt that components of the system could be improved through user feedback.

In the EOS environment, the ECS contractors will supply the people, development, and hardware needed to archive and distribute the data. The complement of ECS equipment supplied to the EDC DAAC is extensive, and includes SGI and Sun processors, 3.3 TB of disk storage to support the processors, and STK and EMASS tape archive units. The hardware delivered by the ECS contractor is clearly adequate to handle the large data sets that the DAAC will soon be managing. However, the hardware, which was delivered in spring 1997 and sits idle, could become obsolete, probably in a few years. The EDC DAAC has to develop a long-term hardware plan that includes incremental upgrades during the lifetime of the project.

The EDC DAAC is concerned that the ECS will not have the capacity to create products, particularly MODIS products, to meet user demand. At the time of the review, the DAAC had not seen the ECS software, and even the ECS liaisons were uncertain about the software because some elements were being installed remotely. Six months later, the DAAC had begun to work with the ECS software, but found that significant, required capabilities of the system had not been delivered and that the stability of the system for processing data was poor. NASA apparently shares these concerns, and it now seems likely that the ECS will not be used for processing data from the AM-1 platform.

All Version 0 data (GTOPO30, AVHRR, SIR-C, Landsat Pathfinder, and aircraft scanner data) can be accessed through one of several Web pages for the DAAC. However, the data are mostly stored off-line, and the DAAC seems reluctant to move them on-line. To disseminate data on-line, the DAAC would have to develop tools to subset the large data sets and to enable users to download the resulting smaller data sets over the network. In the absence of such tools, data are mostly delivered on media, rather than over the network. In the panel's view, growing numbers of users will want to access data over the Web, and the development of subsetting tools (see Recommendation 5) should go a long way toward meeting this need. The panel recognizes that developing a subsetting capability is a tall order for large data sets and that the DAAC may have to meet this challenge by offering specific, restricted subsetting schemes and by restricting the volume of downloaded data per user.

The DAAC is connected to the ECS through the NASA Science Internet (NSI) router, which then connects to the ECS router by way of the Fiber Distrib-

uted Data Interface (FDDI). This connection has a bandwidth of 100 mbps. To connect to the Version 0 system, the NSI router links with the external router of the EROS Data Center. This connection is predominately Ethernet, which has a bandwidth of 10 mbps. External users access both the Version 0 and the ECS systems through a T1 link, which has a bandwidth of 1.5 mbps. Consequently, the T1 link is the bottleneck for remote access, and the DAAC plans to upgrade to increase bandwidth by the fall of 1998. Given that most data are delivered on media, current network bandwidth exceeds demand. However, low demand for on-line data sets may well be due to the lack of subsetting tools discussed above and, possibly, limits on bandwidth.

The DAAC manager, Lyndon Oleson, feels that the DAAC has little or no control over its fate—all decisions are made by ESDIS and the ECS contractor. Consequently, the DAAC has not yet formulated a clear vision for its future, but nonetheless seems optimistic that it will be able to fulfill its responsibilities. The DAAC offered no strategy for key elements of its operations, such as those involving scientists, characterizing the user community, subsetting data sets, and establishing performance measures. Without well-defined performance measures, the DAAC runs a risk of coming up short in the completion of its mission.

The EROS Data Center complex, which includes the DAAC, is a matrixed operation, and most staff work part-time for the DAAC as needed. Currently, some DAAC staff are working on ECS maintenance and operations. The DAAC hopes to acquire 50 more ECS staff in preparation for the Landsat and AM-1 launches, although this may not be necessary if the data are not processed at the DAAC. If and when the ECS contractors arrive, DAAC staff who currently work on Version 0 and ECS maintenance and operations will be reemployed by the EROS Data Center.

ECS contractors, including the ECS science and engineering liaisons, are not integrated into the DAAC but are kept separate from DAAC operations. They will receive a limited amount of training on DAAC issues, but most are expected to already have skills such as system administration and software development. A

discussion with the liaison and the head of maintenance and operations did not reveal any major concerns among the ECS contractors.

The DAAC's budget has grown from approximately $3 million in FY 1994 to $6.5 million in FY 1998, and is projected to reach $17.3 million by FY 2002, the latest year for which budget figures were available to the panel (Table 5.1). ECS hardware, software, and personnel account for about 30% of the current budget. The growth in the DAAC's budget is due partly to hardware acquisitions (e.g., FY 1997), and partly to an increase in operations and maintenance (e.g., FY 2001).

To illustrate its cost-effectiveness, the DAAC gave an example of how it had saved time and money developing Version 0 by leveraging off existing EROS Data Center capabilities. Producing the ASTER digital elevation model was also a good learning experience that will pay off for the DAAC in the AM-1 environment. The panel feels, however, that the DAAC would benefit by devising more quantitative measures of cost-effectiveness and tracking them over time. This is particularly important in an era of fiscal pressure on NASA budgets.

As noted above, NASA is likely to ask the DAACs or the science teams to process data from the AM-1 platform. Details on the MODIS, ASTER, and Landsat 7 science teams' contingency plans were not available to the panel, but the EDC DAAC's plan is to write the data to tape and then to inventory and shelve the tapes. Only the science teams would receive data in quantities suffi-

cient for calibration and validation activities. In this case, Level 2 MODIS data would likely arrive at least a year late from the GSFC DAAC (see Chapter 1, "Recent Developments"). In the panel's view, failure to deliver data to users would constitute a failure of the fundamental mission of the DAAC, and it is incumbent on the DAAC to work with the science teams to provide at least the processing and distributing functionality scheduled to be delivered in the ECS.

The DAAC is hosted by the USGS EROS Data Center, which provides the facility (NASA paid for the building extension that houses the DAAC) and most of the DAAC staff. The EROS Data Center will also provide the long-term archive of data from the EDC DAAC. NASA funds 98% of the DAAC's operational budget, and the USGS does not try to constrain the operations of the DAAC in any way.

The home pages of both the EDC DAAC and the EROS Data Center are linked, and users can access data from previous Landsat missions through the DAAC. The transition from DAAC to EROS Data Center home pages is seamless, and DAAC users must take care or they will order expensive data from the EROS Data Center, thinking that they are ordering freely distributed DAAC data.

The DAAC perceives ESDIS as being concerned exclusively with development. According to the EDC DAAC, ESDIS defines success as the delivery of the ECS, not the successful operation of the DAACs. Consequently, ESDIS discourages the DAAC from bringing up possible problems with the ECS that would lead to further slips in the schedule. The DAAC is also concerned about shifting responsibilities between Goddard and NASA Headquarters, which has led to confusion at the DAAC and a lack of continuity at ESDIS. Finally, the DAAC blamed ESDIS for several issues of concern to the panel, such as having no scientists on staff, not developing formats and subsetting tools that are suitable to the land processes community, and not considering the needs of applied users.

The panel agrees that a more operational focus at ESDIS is desirable, particularly since the first EOS satellite (TRMM) has already been launched. The panel also understands the need for the DAAC to satisfy the ESDIS requirements, which are intended to enable the DAACs to operate together as a system and to

keep costs down. However, the most effective data centers go beyond the stated requirements, and the panel encourages the DAAC to take the initiative in meeting the specialized needs of its user communities.

The EDC DAAC has a special relationship with the GSFC and ORNL DAACs. The GSFC DAAC will provide Level 2 MODIS data, which will be used by the EDC DAAC to create Level 2+ MODIS land products. The ORNL DAAC holds laboratory and field data, which complement the EDC DAAC's remote sensing holdings. Interactions with the other DAACs, however, are weak. Although it is recognized that the DAACs need to coordinate their activities, the EDC DAAC does not view the activities of the other DAACs as being relevant to its operations. Consequently, the DAAC is missing an opportunity to learn about new tools and techniques from the other DAACs. In particular, it seems that a closer relationship with other DAACs dealing with data relevant to land processes (e.g., NSIDC and ASF) could be beneficial.

The DAAC is counting on the ECS for its operations, but it did not have input into the ECS development and does not know what capabilities the ECS will eventually provide. The long-term ECS development is dealt with as a contract, not a partnership. The DAAC feels frustrated because it has little or no control over the ECS but must nevertheless implement and operate the system.

The ECS liaison has worked to convince the ECS contractor that the DAACs should be involved in decisions, especially on priorities. There are now some review boards with DAAC membership, and the ECS contractor is beginning to shuffle its deliverables and schedule to accommodate users.

The EDC DAAC is a world-class organization with a number of accomplishments in its six-year history. Particularly impressive to the panel was the DAAC's preparation and distribution of useful land processes data sets, such as AVHRR and GTOPO30. A valuable resource today, these data sets also demonstrate the DAAC's ability to handle the large data sets that will result from the upcoming EOS missions. The panel was also pleased by the DAAC's good relationship with its Science Advisory Panel, which bodes well for the DAAC's ability to meet the evolving needs of its scientific user community. On the other hand, the DAAC's future user community will likely be dominated by Landsat applications users. To serve their needs, the membership of the Scientific Advisory Panel will have

to be adjusted so that it includes commercial and governmental as well as scientific users.

As the EDC DAAC moves into the EOS era, it faces a number of challenges. Its greatest challenges, completing preparations for managing data from Landsat 7 and AM-1 data streams, and developing contingency plans for the inevitable failures in the processing system, have become more tractable because of significant launch delays. Rather than becoming complaisant, the DAAC should use the extra time to complete its readiness exercises and clarify its vision for serving its potentially large, diverse user community. To serve its users well, the DAAC needs to know who its current user communities are, how they are likely to change as new data sets and products become available, and what specialized services (e.g., subsetting, near-real-time data) they need. Consequently, improving user services through better training, adding staff with scientific backgrounds, and incorporating feedback from users on a day-to-day basis should be a part of the DAAC's vision. Finally, the vision should include a strategy for upgrading hardware on a regular basis to prevent it from becoming obsolete.

Because the DAAC is part of a larger EOSDIS, it must serve the needs of both the land processes and the broader earth science communities. EOSDIS was designed to foster interdisciplinary research, and for this goal to succeed, many nonspecialized researchers will use the data and products of the EDC DAAC. In the past, the DAAC has worked with ESDIS and the ECS contractor to customize the ECS to meet the specialized needs of the land processes community. Now that the ECS is nearing completion, the DAAC should work with its partners—ESDIS, the ECS contractor, and the other DAACs—to create EOSDIS.