교육데이터 활용•지원 서비스

The LANDFIRE (LF) 2001 Environmental Site Potential (ESP) product represents the vegetation that could be supported at a given site based on the biophysical environment. It is important to note that ESP is an abstract concept and represents neither current nor historical vegetation. Map units are named according to NatureServe's Ecological Systems classification, which is a nationally consistent set of mid-scale ecological units (Comer et al 2003). Usage of these classification units to describe environmental site potential, however, differs from the original intent of Ecological Systems as units of existing vegetation. As used in LF, map unit names represent the natural plant communities that would become established at late or climax stages of successional development in the absence of disturbance. They reflect the current climate and physical environment, as well as the competitive potential of native plant species. The ESP product is similar in concept to other approaches to classifying potential vegetation in the western United States, including habitat types (for example, Daubenmire 1968 and Pfister and others 1977) and plant associations (for example, Henderson and others 1989). To create the ESP product, LF assigned field plots to one of the ESP map unit classes. Assignments were based on presence and abundance of indicator plant species recorded on the plots, and on the ecological amplitude and competitive potential of these species. Plot locations were then intersected with a series of 30m spatially explicit gradient layers. Most of the gradient layers used in the predictive modeling of ESP are derived using the WX-BGC simulation model (Keane and Holsinger, in preparation; Keane et al 2002). WX-BGC simulations are based largely on spatially extrapolated weather data from DAYMET (Thornton et al 1997; Thornton and Running 1999) and on soils data in STATSGO (NRCS 1994). Additional indirect gradient layers, such as elevation, slope, and indices of topographic position, were also used. LF used data from plot locations to develop predictive classification tree models, using See5 data mining software (Quinlan 1993; Rulequest Research 1997), for each LF map zone. These decision trees were applied spatially to predict the ESP for every pixel across the landscape. Finally, ESP pixel values were, in some cases, modified based on a comparison with the LF Existing Vegetation Type (EVT) product created with the use of 30m Landsat ETM satellite imagery. LF made modifications only in non-vegetated areas (such as water, rock, snow, or ice) and where information in the EVT layer clearly enabled a better depiction of the ESP concept.