In the Central Coast of Brittish Columbia.

                                                                                                    The results of this model confirmed the prediction that the introduction of bias into the decision making process would affect the overall outcome.  The first analysis that biased human influences produced area of high quality habitat almost triple that of the neutral analysis.  The first MCE, when reclassed to highlight regions with habitat quality of 200 or more, had an area of  2,161,499 ha.  The second analysis produced an area of 639,791 ha of habitat above 200.  The final analysis when reclassed produced 890,765 ha of high quality habitat.  Without further exploration into qualifying the habitat rating scale in terms of what constitutes high quality habitat, the selection of 200 was arbitrary. The analysis that biased positive habitat criteria produced areas slightly higher than the neutral analysis.   These results are consistent with the literature that suggest any grizzly bear habitat model should include and appropriately weight both of these factors.  Limitations in this research design should not be understated.  This model lacked any information on specific bear location in the area or an attempt to validate the model by comparing it to known populations.  The omission of data sets on important criteria and the absence of model validation limit the usefulness of this model to predict habitat issues.  The analysis did illustrate that weighted pair wise comparisons used for the final MCE the results were significant.  A more thorough analysis would be necessary to arrive at conclusions as to the effectiveness of GIS in modeling grizzly bear habitat. 
     The process of verification is necessary because it determines how effectively a model represents the conditions that are being studied, if the factors have been combined correctly to represent proper factor interactions, and finally if the solution is acceptable to the users as a decision making tool  (DeMers, 2000).  One effective method for validating the results of GIS derived population estimates is hair capture and DNA profiling.  Mowat (2000) suggest that these techniques are promising in their ability to estimate distribution and abundance of bears applicable for conservation efforts and the decision making process.   Often this is achieved in habitat modeling by running the model in an area that is well documented.  One example is the Khutzeymateen Valley where extensive habitat and telemetry research has been undertaken and provincial studies draw heavily from this research.  Another example refers to a grizzly bear habitat model being built by the Craighead institute for the entire coast of BC that intends to ‘test’ the validity of its model by running the model on the intensely researched populations of grizzly bears in Montana and Yellowstone in conjunction with the US Fish and Wildlife Service. 
    The use of GIS to assist land managers in describing and classifying spatial data is a fundamental part of resource management.  Increasingly, there is an integration of human experience and local knowledge into both the data sets and the interpretation of the resultant data.  For example, although two of the watersheds studied in this analysis are highlighted as providing high quality grizzly bear habitat, field research in the area by the provincial government revealed that there are no female grizzly bears there (Hamilton, 2002, private correspondence). 
    The integration of local knowledge, confirmed sightings and hunting statistics are other important ways of acquiring information that can be introduced into the  dataset or used to test the model.  This realization is not new to issues of  modeling grizzly bear habitat and population densities, however the techniques for incorporating them into models is still being improved.  An obvious problem that can arise from the integration of other forms of knowledge into the mapping of resources is the high degree subjectivity inherent in this type of information.  An illustrating anecdote describes how a large forest company performed wildlife surveys for the spotted owl in the height of the old growth vs. spotted owl controversy of the early 90’s.   This company's biologists routinely identified fewer spotted owls than government biologists.  Had the forest been ‘re-mapped’ by each respective party, the resultant data would have been quite different, consistent with the objectives of each party.
     The creation of models and validation of the results that confirm the direction of an agency is undeniable.  Additionally, agencies with specific mandates will refer to studies that yielded compatible results and integrate these into their management policies.  For example, research suggesting that landscapes managed to most closely resemble the natural range of variability of the ecosystem ( Saxena, 1999) are embraced by institutions with continues timber harvesting as a central mandate.  Research suggesting the need for intact ecosystems absent of human influences are embraced by organizations with mandates for the preservation and resortation of wilderness.  Research confiming the effectiveness of estimating population densities, the need for management of 'problem bears' by hunting, and that these populations can be supported by hunting are supported by guide outfitters with an interest in grizzly bear hunting.  Each of these interest groups embraces different ideological and philosophical perspectives over the issues of  wilderness, the importance of upper level carnivores and the maintence of resource extractive cultures.  Discourses over differences in ideology are often less effective for implementing changes in policy than are critiques aimed at the science and the GIScience used to produce data supporting a perspective.   
    A further problem with the use of GIS in advising wildlife managers and the decision making process is that its complexity makes it difficult to understand the process by which error, bias and uncertainty are reflected in the maps and data.  This coupled with the influential force of maps and spatial data enforce the need to better express and document the ways in which bias does enter into the derivation of maps.  Meta data standards are making great progress with documentation of the original data sources as are efforts within the ministries highlighted here to standardize the assimilation of this data into databases.  However, one recurrent problem with GIS modeling is that the mapped results of the model are presented without showing how they have been derived and this derivation is necessary for their proper interpretation (IBM Report).
     Mandates and objectives of different ministries and institutions enter into the modeling process.  These influence the initial selection of criteria, the weight associated to the criteria and the ultimate analysis leading to a decision set.  An analysis conducted by Norheim (2002) determined that respective institutional cultures of a government agency and an ENGO had significant effects on the way that the two projects analyzing old growth in the Pacific Northwest were conducted and on the data that emerged.  He further recognized that neither data set on old growth was inherently more correct within their institutional context. Martin (2000) suggests that recommendations for implementation and evaluation of GIS can benefit from a broader theoretical foundation to support investigation, understanding and improvement.  He also states that there is much to be gained from understanding the important role that context plays in the configuring of GIS especially when similar GIS implementations produce different outcomes.   It may be unrealistic to expect that GIS and remote sensing technology can develop the absolute and unbiased answers that society often expects to questions that are inherently subjective (Norheim, 2002).  This exercise has emphasized this point.     

---

---