Recent Changes in the Configuration of the DOE AmeriFlux Network
and Their Effects on Network Representation

Analysis of the AmeriFlux network prior to these changes indicated that, while the central northeastern, southern, midwestern, and northwestern portions of the United States were relatively well-represented, environments within the Pacific Northwest, Sonoran, and south Texas were poorly represented by existing AmeriFlux tower sites (see http://research.esd.ornl.gov/~hnw/networks for details of this analysis).

Representation by the new AmeriFlux configuration is shown here.

Changes are subtle, but can be made more obvious by subtracting Ameriflux representation before the changes from AmeriFlux representation after the changes. Sites that were added or updated in position are shown in blue, and sites that are defunct or old positions are shown in red. Gray areas show gains in representation, red areas show losses in representation due to the changes in AmeriFlux network configuration.

The additional Mississippi site adds some representation throughout the lower Mississippi valley. The new Arizona site adds representation in east Texas and the Sonoran desert. The loss of the two sites in Oklahoma causes decreased representation in the Central Plains. The position adjustment for the Black Hills tower causes patchy losses in representation across the Central Western states. The representation of each site could be mapped by repeating the analysis for each site individually.

What if, instead of adding four new sites without statistical guidance, we wished to add a single new site, as directed by network representation? One way to approximate a good location for an additional site would be to multiply how well each region is already represented by AmeriFlux by the area of that region. Such an area-weighted map would show the explicit improvement in representation if an additional site were located within that region.

This map shows the direct area-weighted improvement in representation which would be gained if an additional site were located within each area. Darker green indicates a greater incremental gain in representation. Locating an additional AmeriFlux site in south Texas would result in the greatest direct increase in area-weighted representation.

We added an imaginary new AmeriFlux site in South Texas along Interstate 35, halfway between San Antonio and Laredo, and re-examined the incremental changes in the representation of this imaginary AmeriFlux network. So that they could be compared, we also dropped the same two defunct stations, and adjusted the position of the Black Hills station, as before.

The incremental changes in representation after adding a single site, as directed by the network analysis, are shown here. The marginal gains in representation by adding the single, directed site in Texas are equal to or slightly greater than the incremental increases obtained from the undirected addition of the four sites which have actually been added to the existing AmeriFlux network.

There are two components to additional representation that are gained when another site is added to an existing network. A direct component is the additional representation that is gained for adding that region itself. However, there is also an additional indirect gain from other regions which are similar to the added region. Thus, there is a tension between adding a very average site, which will have high similarity to many other sites (high indirect gains but low direct gains), and adding a very unusual site, which will have high direct gains but low similarity to other areas. The selection of additional sites should be made such that both direct and indirect gains in representation are maximized.

The decision to add an additional site in South Texas was made only on the basis of a direct gain calculation, and did not consider the indirect gains due to similarity with other regions. We believe that a "constrained clustering" algorithm, in which the already-established sites are fixed, but the additional sites are free to migrate, can be used to place any number of additional sites in such a way that both direct and indirect gains in additional network representation are maximized.

Network analysis is ecoregion-based, and occurs at the scale of the entire network. Local site characteristics such as land use and disturbance history are not considered in this analysis. Rather, the goal is to determine how well a set of network sites represent the population of environments that occur across the entire map.

For additional information contact:

William W. Hargrove

Oak Ridge National Laboratory

Environmental Sciences Division

Building 1507, Room 211

Mail Stop 6407

Oak Ridge, TN 37831-6407

(865) 241-2748

(865) 574-4665 fax

hnw@fire.esd.ornl.gov