In order to effectively identify vegetation from satellite images, one has to correlate those images with the ground-based observation, in the initial phase of such activity in any given area. This involves expensive field trips in the area by specialized personnel. Having a computerized system available to assist the job of identifying plants from firsthand information will help this process. Identifying a plant by looking at its visible features needs a fair amount of expertise. Yet, often this task is demanded upon many environmental workers for different reasons other than correlating with GIS data. For example, in order to make a decision whether a piece of land is wetland or not, a scientist of a governmental regulatory agency may have to determine whether some particular type of vegetation grows in that region. This paper talks about an ongoing project of developing an intelligent software for identifying plants. The expertise of identifying plants varies a great degree amongst human beings. Almost every layperson has the capability to identify at least a few plants, or a class of similar looking plants of his or her environment. On the other hand, a trained botanist is capable of identifying or discerning many more types of plants. However, the reasoning process behind such identification is more or less the same. This reasoning process involves gradually narrowing down possibilities by checking different discerning external features of plants. The only difference between an expert and a layperson is the amount of such distinguishing rules in the "knowledge-base" of the person. This type of reasoning process is called "forward reasoning" in the terminology of artificial intelligence. Specific algorithms exist for such reasoning process. Specialized softwares are developed based upon these algorithms to encode such reasoning processes. These softwares are called expert system tools. Programs written to do specific reasoning tasks (often using such expert system tools) are called expert systems-because they codify corresponding expertise. We are currently developing an expert system for identifying some plants. Delineating features for approximately forty plants growing in the wetland areas of the southeast region of the USA have been identified. A set of forward reasoning rules has been formulated to simulate an expert's reasoning process for identification of any of these plants. Using an expert system tool called CLIPS (developed by NASA Johnson Space Center), we are codifying this reasoning scheme. The reason for choosing CLIPS are (1) its simple and powerful forward reasoning capability, (2) its C language interface capability, and (3) involvement of some federal agency (BLM) in the project, since CLIPS is being promoted as standard expert system language within federal agencies. A major use of our program would be in environmental education. One of the ways to make people aware of, or appreciate, their environment is to "connect" them to their environment. A program which teaches and helps one to identify surrounding flora, and more importantly helps one to learn the basic delineating features of those plants, could be broadly educating. We are planning to use our program in science museums, public libraries, and schools. Other anticipated use is by field scientists in getting help to recognize vegetation in the process of classifying the environment of a region and standardizing GIS data. This paper would contain the plant identification scheme, which is the basis of our expert system, the design of the system and our experience in developing it, and possible use of it. We acknowledge the support from LBNL-JSU-AGMUS Science Consortium for this work. We are also indebted to Jackson office of the Federal Bureau of Land Management (BLM) for their cooperation, and U.J. Parikh for providing coordination between JSU and BLM for this project.
Topographic and plant community characteristics were measured at 400 plots located on the Westvaco Wildlife and Ecosystem Research Forest, a 8,430 acre tract near Elkins, West Virginia. Forest community types were identified by analyzing species abundance and basal area data using community classification and ordination techniques. A discriminant function was then developed that described the correlation of these forest types to topographic gradients in the landscape. Contour lines were scanned, converted to vectors and used to produce a 15 meter resolution digital elevation model for the study area. Surfaces of topographic conditions significantly correlated with plant communities were created from the DEM. The accuracy of the surfaces was tested by comparison with topographic data from field plots located using the global positioning system. Using these surfaces as variables, the discriminant function was implemented within a geographic information system and resulted in an image predicting the distribution of dominant community types over the landscape. This approach to landscape classification is being evaluated for its effectiveness in describing the potential productivity of forest sites and as a tool for a variety of ecological studies.
Many agencies and private companies now use GIS as a tool for producing maps of wildlife habitats and other natural resources. The next step in application of GIS is analysis of the database to address specific questions or hypotheses. Statistical programs such as S-Plus can be used to indicate significance of data relationships or show trends in certain variables. We present three examples in which GIS databases have proved useful in ecological applications. In the firs example, we examine relationships between water characteristics and species composition of plant communities associated with springs in the San Bernardino Mountains. Plots of the data in the form of Stiff diagrams illustrate variability in mineral composition of natural springs, associated with variability in vegetation. In the second example, we show application of GIS to vegetation and groundwater data obtained as part of a long-term riparian monitoring program in the Sierra Nevada. The data illustrate relationships between stream flow, depth to ground water, and distribution of wetland species. The third example shows habitat analysis for the Santa Ana River woolly star, an endangered plant species whose life cycle is linked to flooding frequency. One of the most important aspects of all three long-term monitoring projects is documentation and standardization, such that data collected in future years can be efficiently incorporated and compared to baseline conditions.