Specific objectives for this study were 1) to evaluate agreement among experts on range plant species behavior and 2) to develop an agreement-based classification method for plant species responses. Declarative information at landscape scale was elicited from 7 role-suggested experts on expected responses to cattle grazing (none, moderate, or heavy) and fire (absent, applied in late summer or fall, or applied in late winter or spring) of 198 plant species from the Edwards Plateau in Texas. Kappa statistics and log-linear models were used to evaluate agreement. Conclusions from the survey are (1) low to moderate agreement on plant species responses to cattle grazing and fire was observed among 7 experts familiar with vegetation processes in the Edwards Plateau. This is in agreement with the low learnability of the domain. (2) Agreement was better and pattern of agreement was more consistent when scenarios were most familiar to the experts. (3) The use of different criteria for distinguishing among classes seems to be a significant source of lack of agreement. (4) Marginal disagreement could be reduced with training and it can be expected to be less significant for more specific (i.e., local) scenarios and vegetation compositions (i.e., on simplified domains). (5) Lack of information, rather than lack of agreement, seems to be the cause for the poor agreement observed in No fire and Summer/fall fire scenarios. (6) A procedure was developed that uses cumulative probability distributions to establish an objective criterion for identifying agreement among experts in ordinal scale. Graphical representations help to understand and evaluate relationships between the number of expert sources and their ability to distinguish among classes for a previously chosen accuracy. The authors emphasize that agreement rather than expertise was evaluated in this work. Feasible methods are needed to facilitate the evaluation of expertise in a domain characterized by moderate to low learnability. Empirical approaches based upon expert information may prove valuable in providing management decision support where such information would be difficult or unfeasible to generate from purely experimental alternatives.
Citations and enhanced abstracts for journals articles and documents focused on rangeland ecology and management. RSIS is a collaboration between Montana State University, University of Idaho, and University of Wyoming.