Rangeland Health: A Bayesian Approach

Ricardo M. Rodriguez Iglesias and Mort M. Kothmann
Department of Rangeland Ecology& Management
Texas A&M University

Rangeland health is a value-laden concept. There are no clearly identified indicators that we can measure, nor is their consensus on the values that should be selected as the basis for determining health. Linkages between values and indicators have not been established and currently there is no method available for assessing the health of natural resources. We make three assumptions in this paper. First, Ecosystem or rangeland health must be a recognizable attribute. That is, an expert should be able to classify the health of a specific site in the field. Without this, the concept of health is useless in natural resource management. Second, the social appeal of the concept will compel its use. And, third, ecology should provide a framework for the development of methodology for assessing health. The formal framework for rangeland health assessment will be presented in three parts: first, the conceptual components, second the spatial and temporal scales, and third a formalism for integration of monitoring with management. First we will address the conceptual components.

A few definitions will help to clarify some of the key terms we will use. An ecological object is a scale-dependent entity that is the focus of our interest in health assessment. Range sites and watersheds are examples of ecological objects. Values are the qualities which stakeholders consider worthwhile. They may include stability, diversity, carrying capacity for livestock and wildlife, watershed integrity, endangered species, etc. Values are not directly measurable in the field. Criteria are the assessment endpoints, which provide dimensions for values. Examples are watershed function, nutrient cycling, and soil stability. In many cases, criteria are not directly measurable and require the use of indicators. Indicators are measurement endpoints; i.e., things that can be measured in the field. They serve as surrogates to provide evidence for the status of criteria.

Whereas, range condition was focused primarily on assessing the amount of grazable forage available, rangeland health is definitely a multi-objective problem that must address numerous values. The appropriate scale to consider depends upon the objectives and values that form the basis for the assessment. Different problems require different scales. Values are considered to have a long temporal scale which transcends our lifetime; thus we place them in the temporal scale of centuries. Values may differ with spatial scale. However, some of the same values may transcend all of the spatial scales; e.g., sustainability.

Criteria are ecological concepts that are associated with the desired values. Their temporal scale is much shorter that that of values; i.e., minutes to a year. The status of criteria may be expected to change more frequently than values. The linkage between criteria and values is based on expert opinion and is not subject to direct measurement. Criteria are generally selected at spatial scales similar to those of the values they are intended to reflect.

Indicators are the measurement endpoints that can be assessed during a field survey. The relationship between indicators and criteria should be quantitatively evaluated and procedures applied to obtain the minimally redundant set of indicators required to assess the selected criteria. Note that while the temporal scale for indicators is short, their spatial scale varies in relation to the criteria being evaluated. For example, scale may vary from large-scale satellite photography to a point measurement on the ground. The spatial scales for linked indicators, criteria and values tend to be similar. Even when the value may be the same for different spatial scales, e.g. sustainability, the criteria and indicators used to assess the status of that value will likely be different for different spatial scales.

Spatial scales may be stratified with individual owners at the smallest scale increasing to county to state to country. Before beginning a monitoring program to assess rangeland health, it is important to identify the spatial scale for which answers are desired. Different spatial scales will generally require a different set of criteria and indicators.

The current approach to developing a health assessment procedure is to select and measure some indicators without first identifying the values and selecting and assessing the criteria that will be linked to the values. The link between criteria and indicators is established after the indicators have been measured using implicit comparisons. For example, measuring soil organic matter on grazed and ungrazed sites and assuming the grazed has unhealthy soil and the ungrazed has healthy soil. Since the links between indicators and criteria are not quantified in such a manner as to provide statistically valid relationships, some form of indexing of indicators is used to assess the status of the criteria. This provides no valid method for making inference from the measurement of indicators back to the status of the criteria, and the accuracy of the indicators cannot be evaluated.

Our approach is to first select the desired values. This requires agreement by representative stakeholders. Then criteria are selected to represent the values and agreement obtained as to the appropriateness of the criteria. Next experts are used to assess the status of the selected criteria on benchmark sites with emphasis on identifying a suite of problem profiles. Following the assessment of criteria by experts, measurements of selected indicators are made. This procedure allows the quantitative assessment of the relationships between indicators and criteria and the selection of the best set of minimally redundant indicators. These data will allow the development of a probabilistic expert system (PES) in the form of a Bayesian belief network (BBN) which can then be used to assess the status of criteria based upon field measurements of the selected indicators. The accuracy of indicators for assessing criteria can be evaluated. This approach provides an appropriate methodology for the assessment of values that represent rangeland health.

In summary, the proposed actions are: first, the values must be explicitly considered before selecting indicators and criteria, second, criteria are selected to reflect the values, and last, indicators are linked to the criteria using benchmark sites evaluated by experts. The accuracy of indicators is quantitatively assessed and a probabilistic expert system is developed for health assessment.

The pooling of expert knowledge for classifying benchmark sites a priori to measuring indicators is important. Classification trees may be used for identifying problem profiles that would result in a classification of unhealthy. Receiver-operator characteristics can be used for assessing the accuracy of indicators. And, finally the use of Bayesian belief networks for building probabilistic expert systems provides a viable quantitative method for linking quantitative and qualitative variables.

There are several advantages to the use of PES based on BBN. Using this approach expert and experimental information can be merged in an understandable and appropriate manner. It integrates both aspects of diagnosis and forecasting and allows one to operate under conditions where only partial evidence is available. The use of probabilities as the common currency of the model avoids the arbitrary weighting of quantitative variables that is in most current assessment models and allows evidence to be combined across scales.

A procedure can be developed in the context of monitoring and management for integrating ground point measurements with GIS to incorporate the assessment of spatially referenced indicators obtained from map data. PES may then be used to produce maps indicating the status of criteria for landscapes at various spatial scales. Such a model can also be linked with environmental and management components to provide forecasts of the probable consequences of selected management interventions under various environmental scenarios. Bayesian Belief Networks allow the user to play "what if" in different directions. For example, what might be the effects of burning and drought combined with heavy grazing on the status of rangeland health. Or, given that the current status is unhealthy, what combinations of management interventions and environmental conditions might promote a shift towards a healthy status. In summary, we believe that the proposed approach can provide direction to those charged with the responsibility of assessing the health of the nation's rangelands.

< Back to Agenda