Elements of Change 1995: Session I Reed

There is a strong co-dependence between global change models and the data that are nec essary to run them. Global change research is data limited and it is important to begin developing mechanisms to ensure that global change research can be properly supported by the necessary measurements and data sets. To that end, a research group at the U. S. Geological Survey's EROS Data Center (EDC) is working to produce products to support global change research, particularly global climate and ecosystem modeling efforts. Two of its contributions are 1) a land cover characteristics database, to which values of biophysical variables of importance to modelers can be assigned and 2) a seasonal characteristics database that describes the dynamics of the vegetated landscape.

The land cover needs of modelers vary in both their classification schemes and spatial scale. This is true for general circulation models (GCMs), ecosystem models, and mesoscale models. In developing the Land Cover Characteristics Database, EROS has taken into account the fact that data needs vary both within and between applications and that links must be made and compatibility strengthened with existing and future land cover classification systems. Flexibility and adaptability are the keys to the development of such a database. In order to ensure its use, the database must allow users to tailor it to their specific needs.

For the EDC Land Cover Characteristics Database, time series NDVI monthly maximum value composites taken from Advanced Very High Resolution Radiometer (AVHRR) satellite data are combined with digital Earth science information such as elevation, climate, and ecoregions, as well as with hard copy maps. Remote sensing data alone cannot separate land cover types sufficiently since distinct land cover types can have very similar spectral/temporal characteristics. Other information, such as spatial distribution and relationship to ecoregion, elevation, and climate, and coincidence to map information helps to distinguish these types and to assign labels to classes to differentiate regions with similar spectral characteristics.

Global change research is data limited and it is important to begin developing mechanisms to ensure that global change research can be properly supported by the necessary measurements and data sets.

The Land Cover Characteristics Database contains 159 distinct land cover regions. However, Reed points out, the product is more than a land cover map, it is a database. For each of the 159 classes there is AVHRR data, average biweekly NDVI, major land resource areas, elevation, and other environmental information. All of this information is supplied to users along with the land cover regions map so they have all the source information EROS used to create the database and can make adjustments to suit their own applications. Since many users have grown accustomed to working with particular land cover classification schemes, the 159 land cover types have been generalized into commonly used schemes such as the USGS Anderson Level 2 classes, those used for the Biosphere Atmosphere Transfer Scheme (BATS) and for the Simple Biosphere (SiB) model.

Now that the database is complete, a variety of operational applications are underway at various federal agencies. These include a national fire hazards assessment by the U. S. Forest Service, ecoregions mapping by the Environmental Protection Agency (EPA), water quality assessment by the U. S. Geological Survey, crop condition assessment by the U. S. Department of Agriculture, biogenic emissions modeling by the EPA, and weather forecasting by the National Weather Service.

On the whole, the results from model and operational applications are positive. The land cover database can be applied to a variety of models and has been shown to improve their results. The land cover characterization approach is a simple strategy based on how land cover data are used. It relates land cover to environmental functions and processes and is adaptable to a range of environmental modeling and assessment applications.

The results from the U. S. prototype have encouraged modelers who are now anticipating the completion of a global land cover characteristics data set. The Global Pathfinder 1-km AVHRR data set will make this global land cover database possible. Over thirty countries have been cooperating on this project since April 1992 to assemble global 10-day 1-km AVHRR coverage. Though a challenge, this effort is progressing well. There are currently ten global 10-day composites that are complete for the globe. Thirty-six 10-day composites covering April, 1992 to March, 1993 are complete for North America and South America. This data is on-line at the following World Wide Web address: http://sun1.cr.usgs.gov/landdaac/landdaac.html

For the global land cover characterization, AVHRR data will be used one continent at a time, beginning with North America, then South America with priorities for other continents yet to be determined. Cooperating partners in this project include the University of Nebraska - Lincoln, EPA, U. S. Forest Service, National Aeronautics and Space Administration (NASA), United Nations Environment Programme (UNEP), the International Geosphere Biosphere Programme (IGBP), National Autonomous University of Mexico (UNAM), University of Cordoba, Argentina, and others. As the effort progresses, more collaborators are expected to join the project.

The key to defining seasonal characteristics is to identify the onset of the growing season.

The North America land cover characterization based on one year of AVHRR data in 100 clusters is now in the preliminary labeling stage. A refining of these labels is underway, and new challenges seem to emerge all the time. One example is that in enlarging the scope from the conterminous U. S. to all of North America, the Yucatan region of Mexico appears as homogeneous because it has a higher NDVI than the rest of the continent. Such unexpected outcomes require new methods, such as applying an additional clustering procedure to that area. The North America land cover characteristics database is expected to be completed by the end of 1995 and the Western Hemisphere is expected to be completed by summer, 1996.

Another database developed at the EROS Data Center for use by the global modeling community is the prototype seasonal characteristics database for the conterminous United States. The objectives for this project are to:

The source data for this database are the biweekly maximum value NDVI composites. Pre-processing issues include recoding very low values (to eliminate clouds and non-vegetated surfaces), and smoothing of the time series to correct for sub-pixel cloud contamination. An iterative median smoother is used to retain peaks and eliminate valleys of the temporal NDVI curve.

The key to defining seasonal characteristics is to identify the onset of the growing season. The method relied upon for this work is the calculation of a moving average, borrowed from market trend analysis and involves calculating the average of the previous ten biweekly periods and comparing that value to the observed value for a biweekly period. The moving average acts as a predictor; when observed NDVI values depart markedly from predicted values (i. e., when the true values do not follow the trend), the onset of the growing season is occurring.

The event we call the "onset of growing season" is a description of ecosystem-level changes (due to coarse 1-km satellite resolution) and is not necessarily related to conventional, field-measured phenological events. Once the onset of the growing season is identified, additional seasonal characteristics can be derived. There are basically three families of seasonal characteristics: temporal, NDVI value, and curve feature. Temporal characteristics are time of onset of growing season, time of end of growing season, time of maximum NDVI, and duration of growing season. NDVI value characteristics are NDVI value at onset of growing season, NDVI value at end of growing season, maximum NDVI value, and range of NDVI values. Curve feature characteristics are rate of greenup, rate of senescence, time-integrated NDVI, and modality of growing seasons.

These datasets represent examples of how data producers and providers are working closely with the modeling community. This helps assure that the data fit the needs of the modelers rather than forcing modelers to adapt their research to accommodate the data.

Interesting results emerge from year-to-year comparisons. For example, 1993, a heavy precipitation year stands out from other years, due to large scale flooding in the Midwest. The difference in the response of natural (vigorous vegetative activity) vs. anthropogenic (seriously lowered vegetative activity) landscapes to the flooding are easily noted in these images. Another interesting result is the apparent "reverse green wave" in Great Plains grasslands where the more northern areas begin their growing season sooner than more southerly regions. This is caused by the different temperature requirements for growth by cool season C3 grasses in the north and warm season C4 grasses in the southern plains.

This seasonal characteristics database is currently being used by 10 university and government institutions to evaluate its utility in modeling applications. In addition, in a cooperative project involving Augustana College in Sioux Falls, South Dakota, EROS Data Center, and The Nature Conservancy has set up a network of ground observation sites in the Plains to assess the ground conditions that relate to the seasonal characteristics as observed by satellite. These two datasets represent examples of how data producers and providers are working closely with the modeling community. This helps assure that the data fit the needs of the modelers rather than forcing modelers to adapt their research to accommodate the data.