TYPES OF UNCERTAINTY
Fire-prone systems are inherently complex and imperfectly understood. As a result, management actions taken (or not taken) in these ecosystems can have uncertain consequences. The uncertainties that surround any given management issue are of two sorts. Some uncertainty result from a lack of knowledge and can be overcome through targeted research efforts. A very different type of uncertainty results from unpredictable stochastic variation of the system. In fire management, one of the greatest sources of this stochastic uncertainty is fire weather. For example, seemingly random shifts of wind are known to alter fire behavior or fuel break effectiveness with long-lasting consequences for vegetation. This stochastic uncertainty can be ignored, generalized, or rigorously exploited through probabilistic belief network modeling.
Forest management is increasingly concerned with understanding effects over long time periods and at broad spatial scales. Anticipating such long-term and broad-scale effects can be exceedingly difficult simply because the uncertainties associated with more places and longer periods of time can alter outcomes. For example, wildland fire is not readily predictable because it depends on future changes in vegetation, fuel and fire spread from many possible ignition sources. Stochastic changes in fire weather can make predicting fire behavior difficult even a few hours in advance. Long-term forecasts of fire effects are difficult because outcomes are often contingent on chance variation in a wide range of influential factors.
Back to Top
IMPERFECT SCIENTIFIC KNOWLEDGE
Science has markedly increased our knowledge of the nature and function of ecosystems. By focusing on ever-finer detail, reductionistic science and specialization has led to more detailed understanding. In parallel, more holistic science has led to more realistic modeling of nature by integrating these fragments of knowledge and putting knowledge in context. Yet after many decades of concerted scientific effort, scientists and managers continue to ask for more information on new species, relationships, and places that are poorly understood. A cynic could argue that science may be less effective at reducing uncertainty than it is capable of making us more aware of what we still do not know.
Given the vast amount of knowledge that is now available and the ever-pressing information needs of managers, decision makers need to know when they have enough science to support their decision and when more information is warranted. A substantial portion of available scientific information may be of little or marginal use for applied purposes for a number of reasons. Some of the key reasons why knowledge gaps exist are outlined below.
(1) Scientific research may be lacking on a subject or relationship of concern.
Research has been conducted on surprisingly obscure questions, yet there are always new ones that arise. Unanswered questions can surround narrowly focused biological issues or more holistic ecological ones. Knowledge about sensitive plant or animal species is often surprisingly incomplete, particularly for non-charismatic species. Before assuming that relevant research has not been conducted on a subject, managers should explore online databases, consult with a variety of experts or inquire on online listservs.
(2) Research was conducted, but it does not specifically relate to the management question at hand.
Understanding the relevance of existing research can be challenging for non-professional researchers. Managers may find a number of references on a general subject, but the specific question that researchers addressed (and the data included) may not meet the manager's specific need. In some cases, researchers may have needed data, but the published data is too general to be used for probabilistic modeling. The "raw data" may only be available by contacting the authors directly. In other instances, research may have been conducted at the wrong spatial or temporal scale to be of use. In this case, uncertainty does not result from a lack of research, but from the "wrong sort" of research. For example, scientists may now know more about the northern spotted owl than any other terrestrial species in North America, yet managers still struggle with questions involving fire and fuel management that may affect owl habitat. Conducting similar multi-million dollar research efforts for every sensitive component of the ecosystem is simply not possible. Instead, managers need to focus researcher's attention on the science that can will reduce the "critical uncertainties" that surround management decisions.
(3) Seemingly relevant research was conducted at a distant location, but its relevance to the project area is unknown.
Managers should use scientific information from distant locations carefully. It is common for managers to utilize information across a single vegetation type, but classifications based on a single dominant species often ignore differences in topography, climate or secondary vegetation. For example, ponderosa pine forests are extensive throughout the west and evidence of frequent fire is common throughout the range. Nonetheless, the climate of the interior northern Rockies is such that episodic crown fires may have been more "natural" in past centuries than in the Southwest or California. In order to know when distant research is relevant or not, similarities among sites should be critically examined. When distant research must be used, it is useful to compare a range of studies from different locations to know how much variability may exist across regions.
(4) Research may be available, but the rigor of the methods used to collect and analyze the data is questionable.
When decisions are based on critical information, that science should hold water. For managers, the “best available science” is that which best addresses their question of concern based on a sound methodology and analysis. In some cases, this may be research published in peer reviewed journals, but that information is more likely to not be local. In many cases, the "best available science" may be local information that is only available in proceedings, theses, or in unpublished reports. When building probability models, managers may need to rely on databases and other unpublished data. Possible concerns about the reliability of this information should be addressed upfront.
(5) Adequate information may be available in papers or databases, but managers' knowledge of how to mine and apply this data is limited.
It does not matter how much is known by “science” if knowledge is not accessible to people who can use it. Often times, a manager is unaware of existing research and uncertainty exists when it need not. This problem is increasingly common even among researchers because the number of publications is staggering. After a specific information need is identified, managers can examine online publication databases, consult with a variety of experts and even make specific inquiries on online listservs. Use of the vast amount of "in house" data is also challenging. Agency databases include surveys of vegetation and wildlife data, silvicultural records, FIA data, aerial photos, satellite images, and GIS layers. The first step in taking advantage of these data is to become more aware of what actually exists.
(6) There are competing sources of information having different results, and it is unclear which research is most relevant or reliable.
There is no easy way to assess the relative merits of two competing sources of information with different results. The scientific literature is replete with contradicting research, in part because the scientific process is designed to question itself. Older research is often less valuable than newer information, but classics can have enduring value. Many apparent conflicts reflect differences in geography or land use history. Although many authors claim that their research has relevance beyond their study area (this is often needed to appease the journal editor) it is up to the reader to decide how relevant a given research effort is for their particular area and management problem.
Back to Top
ON THE NATURE OF STOCHASTIC VARIABILITY
Uncertainties that result from a lack of knowledge can be partially overcome through targeted research, yet uncertainty that results from stochastic variation can only be better modeled. The best example of stochastic variability that affects fire and fuels management is fire weather. There are countless combinations of temperature, fuel moisture, wind speed, and wind direction that can alter fire behavior and fire effects. During many long-duration fire events, the most severe fire effects occur during just one or a few days due to a stochastic change in the weather. While the specific outcomes of a particular fire can not be readily predicted, the likelihood of a particular outcome is predictable through probabilistic analysis of historical weather data.
In addition to climate, other stochastic elements make long-term and broad scale planning difficult. For example, the exact location of a future lightning ignition is unknowable, regardless of scientific advances. Nevertheless, the most likely places of an ignition can be described probabilistically, and this added knowledge reduces uncertainty from random guess work. In addition to ignition locations, the effectiveness of fire suppression efforts includes many variables that are not readily predictable. Multiple ignitions may stretch suppression resources so thin that fire occurs when it would otherwise not. Due to stochastic variability in climate, fire spread and behavior, management outcomes may be very different than what is expected based on a few assumptions. To address this possible variability, multiple iterations of fire-related models is necessary. Through this process, the most likely outcomes become apparent, and decisions can be based on both what is most likely and what is possible.
Back to Top
DOES UNCERTAINTY MATTER?
The environment is imperfectly understood due to scientific and stochastic uncertainties, but an ecosystem does not need to be completely understood in order to be effectively managed. Belief network models help define which specific information is and is not needed. Moreover, sensitivity analyses provide information about the relative importance of different components of the models. These processes help managers identify “critical uncertainties” that, in essence, define the degree to which uncertainty really matters.
Here, the concept of "relative uncertainty" becomes important. Many people mistakingly think of scientific uncertainty in black and white, rather than as shades of gray; that is, something is either known or it is not known. In almost every situation, however, something really is known, such as the range of possible values, or the most likely value based on past experience. In a probabilistic belief network, relationships can be analyzed using a wide range of values without knowing the "true" answer. This approach lets managers explore how sensitive results are to different values without actually expending the time and effort to obtain precise data.
Back to Top
REDUCING UNCERTAINTY
Increased use of probability can reduce uncertainties associated with a wide range of forest management issues. Probability models that exploit stochastic variability can provide decision makers with both the range and likelihoods of different scenarios and the potential for these to affect outcomes. As a probability model is being developed, sensitivity analyses provide a method for identifying the relevance of scientific uncertainties. The great value of belief network modeling is that it is capable of dealing with both types of uncertainty in a transparent framework.
Back to Top
|
